{"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer_with_prompt","uri":"program://DMOSpeech2/module/src.duration_trainer_with_prompt#L1-L452","kind":"module","name":"src.duration_trainer_with_prompt","path":"src/duration_trainer_with_prompt.py","language":"python","start_line":1,"end_line":452,"context_start_line":1,"context_end_line":452,"code":"from __future__ import annotations\n\nimport gc\nimport os\n\nimport math\n\nimport torch\nimport torchaudio\nimport wandb\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nfrom ema_pytorch import EMA\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler, Subset # <-- Added Subset import\nfrom tqdm import tqdm\n\nimport torch.nn.functional as F\n\nfrom f5_tts.model import CFM\nfrom f5_tts.model.dataset import collate_fn, DynamicBatchSampler\nfrom f5_tts.model.utils import default, exists\n\n# trainer\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nSAMPLE_RATE = 24_000\n\n\nclass Trainer:\n def __init__(\n self,\n model,\n vocab_size,\n vocab_char_map,\n process_token_to_id=True,\n loss_fn='L1',\n lambda_L1=1,\n gumbel_tau=0.5,\n n_class=301,\n n_frame_per_class=10,\n epochs=15,\n learning_rate=1e-4,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n logger: str | None = \"wandb\", # \"wandb\" | \"tensorboard\" | None\n wandb_project=\"test_e2-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n last_per_steps=None,\n accelerate_kwargs: dict = dict(),\n ema_kwargs: dict = dict(),\n bnb_optimizer: bool = False,\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=False)\n\n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n print(f\"Using logger: {logger}\")\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n },\n )\n\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n self.model = model\n self.vocab_size = vocab_size\n self.vocab_char_map = vocab_char_map\n self.process_token_to_id = process_token_to_id\n assert loss_fn in ['L1', 'CE', 'L1_and_CE']\n self.loss_fn = loss_fn\n self.lambda_L1 = lambda_L1\n self.n_class = n_class\n self.n_frame_per_class = n_frame_per_class\n self.gumbel_tau = gumbel_tau\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n\n print(f'To load from {latest_checkpoint}.')\n\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n print(f'Loaded from {latest_checkpoint}.')\n\n if \"step\" in checkpoint:\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n \n del checkpoint\n gc.collect()\n\n print(f'Exit load_checkpoint.')\n\n return step\n\n\n def validate(self, valid_dataloader, global_step):\n \"\"\"\n Runs evaluation on the validation set, computes the average loss,\n and logs the average validation loss along with the CTC decoded strings.\n \"\"\"\n self.model.eval()\n total_valid_loss = 0.0\n total_sec_error = 0.0\n count = 0\n\n # Iterate over the validation dataloader\n with torch.no_grad():\n for batch in valid_dataloader:\n\n # Inputs\n prompt_mel = batch['pmt_mel_specs'].permute(0, 2, 1) # (B, L_mel, D)\n prompt_text = batch['pmt_text']\n text = batch['text']\n\n target_ids = list_str_to_idx(text, self.vocab_char_map).to(prompt_mel.device)\n target_ids = target_ids.masked_fill(target_ids==-1, vocab_size)\n\n prompt_ids = list_str_to_idx(prompt_text, self.vocab_char_map).to(prompt_mel.device)\n prompt_ids = prompt_ids.masked_fill(prompt_ids==-1, vocab_size)\n\n # Targets\n tar_lengths = batch['mel_lengths']\n\n # Forward\n predictions = SLP(target_ids=target_ids, prompt_ids=prompt_ids, prompt_mel=prompt_mel) # (B, C)\n\n if self.loss_fn == 'CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_labels = torch.argmax(est_length_logtis, dim=-1)\n loss = F.cross_entropy(est_length_logtis, tar_length_labels)\n \n est_lengths = est_length_labels * self.n_frame_per_class\n frame_error = (est_lengths.float() - tar_lengths.float()).abs().mean()\n sec_error = frame_error * 256 / 24000\n\n total_sec_error += sec_error.item()\n total_valid_loss += loss.item()\n count += 1\n\n avg_valid_loss = total_valid_loss / count if count > 0 else 0.0\n avg_valid_sec_error = total_sec_error / count if count > 0 else 0.0\n\n # Log validation metrics\n self.accelerator.log(\n {\n f\"valid_loss\": avg_valid_loss,\n f\"valid_sec_error\": avg_valid_sec_error\n }, \n step=global_step\n )\n \n self.model.train()\n\n\n def train(self, train_dataset: Dataset, valid_dataset: Dataset,\n num_workers=64, resumable_with_seed: int = None):\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n # Create training dataloader using the appropriate batching strategy\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )\n # Create validation dataloader (always sequential, no shuffling)\n valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n batch_size=self.batch_size,\n shuffle=False,\n )\n\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n\n sampler = SequentialSampler(valid_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n # Create validation dataloader (always sequential, no shuffling)\n valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True, \n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n \n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_steps = (\n self.num_warmup_updates * self.accelerator.num_processes\n ) # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_steps = len(train_dataloader) * self.epochs / self.grad_accumulation_steps\n decay_steps = total_steps - warmup_steps\n warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)\n decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)\n self.scheduler = SequentialLR(\n self.optimizer, schedulers=[warmup_scheduler, decay_scheduler], milestones=[warmup_steps]\n )\n train_dataloader, self.scheduler = self.accelerator.prepare(\n train_dataloader, self.scheduler\n ) # actual steps = 1 gpu steps / gpus\n start_step = self.load_checkpoint()\n global_step = start_step\n\n valid_dataloader = self.accelerator.prepare(valid_dataloader)\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar = tqdm(\n skipped_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n initial=skipped_batch,\n total=orig_epoch_step,\n )\n else:\n progress_bar = tqdm(\n train_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n )\n\n for batch in progress_bar:\n with self.accelerator.accumulate(self.model):\n # Inputs\n prompt_mel = batch['pmt_mel_specs'].permute(0, 2, 1) # (B, L_mel, D)\n prompt_text = batch['pmt_text']\n text = batch['text']\n\n target_ids = list_str_to_idx(text, self.vocab_char_map).to(prompt_mel.device)\n target_ids = target_ids.masked_fill(target_ids==-1, vocab_size)\n\n prompt_ids = list_str_to_idx(prompt_text, self.vocab_char_map).to(prompt_mel.device)\n prompt_ids = prompt_ids.masked_fill(prompt_ids==-1, vocab_size)\n\n # Targets\n tar_lengths = batch['mel_lengths']\n\n # Forward\n predictions = SLP(target_ids=target_ids, prompt_ids=prompt_ids, prompt_mel=prompt_mel) # (B, C)\n\n if self.loss_fn == 'CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_labels = torch.argmax(est_length_logtis, dim=-1)\n loss = F.cross_entropy(est_length_logtis, tar_length_labels)\n \n with torch.no_grad():\n est_lengths = est_length_labels * self.n_frame_per_class\n frame_error = (est_lengths.float() - tar_lengths.float()).abs().mean()\n sec_error = frame_error * 256 / 24000\n\n log_dict = {\n 'loss': loss.item(), \n 'loss_CE': loss.item(), \n 'sec_error': sec_error.item(),\n 'lr': self.scheduler.get_last_lr()[0]\n }\n\n else:\n raise NotImplementedError(self.loss_fn)\n\n \n self.accelerator.backward(loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n self.optimizer.step()\n self.scheduler.step()\n self.optimizer.zero_grad()\n\n global_step += 1\n\n if self.accelerator.is_local_main_process:\n self.accelerator.log(log_dict, step=global_step)\n progress_bar.set_postfix(step=str(global_step), loss=loss.item())\n\n if global_step % (self.save_per_updates * self.grad_accumulation_steps) == 0:\n self.save_checkpoint(global_step)\n # if self.log_samples and self.accelerator.is_local_main_process:\n # Run validation at the end of each epoch (only on the main process)\n if self.accelerator.is_local_main_process:\n self.validate(valid_dataloader, global_step)\n # if global_step % self.last_per_steps == 0:\n # self.save_checkpoint(global_step, last=True)\n\n self.save_checkpoint(global_step, last=True)\n self.accelerator.end_training()","source_hash":"20ff79d536aa888b20ce19adba2d43f1dcea96063d23f16c210e5b9fdd404874","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer_with_prompt.Trainer","uri":"program://DMOSpeech2/class/src.duration_trainer_with_prompt.Trainer#L39-L452","kind":"class","name":"Trainer","path":"src/duration_trainer_with_prompt.py","language":"python","start_line":39,"end_line":452,"context_start_line":19,"context_end_line":452,"code":"import torch.nn.functional as F\n\nfrom f5_tts.model import CFM\nfrom f5_tts.model.dataset import collate_fn, DynamicBatchSampler\nfrom f5_tts.model.utils import default, exists\n\n# trainer\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nSAMPLE_RATE = 24_000\n\n\nclass Trainer:\n def __init__(\n self,\n model,\n vocab_size,\n vocab_char_map,\n process_token_to_id=True,\n loss_fn='L1',\n lambda_L1=1,\n gumbel_tau=0.5,\n n_class=301,\n n_frame_per_class=10,\n epochs=15,\n learning_rate=1e-4,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n logger: str | None = \"wandb\", # \"wandb\" | \"tensorboard\" | None\n wandb_project=\"test_e2-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n last_per_steps=None,\n accelerate_kwargs: dict = dict(),\n ema_kwargs: dict = dict(),\n bnb_optimizer: bool = False,\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=False)\n\n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n print(f\"Using logger: {logger}\")\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n },\n )\n\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n self.model = model\n self.vocab_size = vocab_size\n self.vocab_char_map = vocab_char_map\n self.process_token_to_id = process_token_to_id\n assert loss_fn in ['L1', 'CE', 'L1_and_CE']\n self.loss_fn = loss_fn\n self.lambda_L1 = lambda_L1\n self.n_class = n_class\n self.n_frame_per_class = n_frame_per_class\n self.gumbel_tau = gumbel_tau\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n\n print(f'To load from {latest_checkpoint}.')\n\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n print(f'Loaded from {latest_checkpoint}.')\n\n if \"step\" in checkpoint:\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n \n del checkpoint\n gc.collect()\n\n print(f'Exit load_checkpoint.')\n\n return step\n\n\n def validate(self, valid_dataloader, global_step):\n \"\"\"\n Runs evaluation on the validation set, computes the average loss,\n and logs the average validation loss along with the CTC decoded strings.\n \"\"\"\n self.model.eval()\n total_valid_loss = 0.0\n total_sec_error = 0.0\n count = 0\n\n # Iterate over the validation dataloader\n with torch.no_grad():\n for batch in valid_dataloader:\n\n # Inputs\n prompt_mel = batch['pmt_mel_specs'].permute(0, 2, 1) # (B, L_mel, D)\n prompt_text = batch['pmt_text']\n text = batch['text']\n\n target_ids = list_str_to_idx(text, self.vocab_char_map).to(prompt_mel.device)\n target_ids = target_ids.masked_fill(target_ids==-1, vocab_size)\n\n prompt_ids = list_str_to_idx(prompt_text, self.vocab_char_map).to(prompt_mel.device)\n prompt_ids = prompt_ids.masked_fill(prompt_ids==-1, vocab_size)\n\n # Targets\n tar_lengths = batch['mel_lengths']\n\n # Forward\n predictions = SLP(target_ids=target_ids, prompt_ids=prompt_ids, prompt_mel=prompt_mel) # (B, C)\n\n if self.loss_fn == 'CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_labels = torch.argmax(est_length_logtis, dim=-1)\n loss = F.cross_entropy(est_length_logtis, tar_length_labels)\n \n est_lengths = est_length_labels * self.n_frame_per_class\n frame_error = (est_lengths.float() - tar_lengths.float()).abs().mean()\n sec_error = frame_error * 256 / 24000\n\n total_sec_error += sec_error.item()\n total_valid_loss += loss.item()\n count += 1\n\n avg_valid_loss = total_valid_loss / count if count > 0 else 0.0\n avg_valid_sec_error = total_sec_error / count if count > 0 else 0.0\n\n # Log validation metrics\n self.accelerator.log(\n {\n f\"valid_loss\": avg_valid_loss,\n f\"valid_sec_error\": avg_valid_sec_error\n }, \n step=global_step\n )\n \n self.model.train()\n\n\n def train(self, train_dataset: Dataset, valid_dataset: Dataset,\n num_workers=64, resumable_with_seed: int = None):\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n # Create training dataloader using the appropriate batching strategy\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )\n # Create validation dataloader (always sequential, no shuffling)\n valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n batch_size=self.batch_size,\n shuffle=False,\n )\n\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n\n sampler = SequentialSampler(valid_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n # Create validation dataloader (always sequential, no shuffling)\n valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True, \n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n \n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_steps = (\n self.num_warmup_updates * self.accelerator.num_processes\n ) # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_steps = len(train_dataloader) * self.epochs / self.grad_accumulation_steps\n decay_steps = total_steps - warmup_steps\n warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)\n decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)\n self.scheduler = SequentialLR(\n self.optimizer, schedulers=[warmup_scheduler, decay_scheduler], milestones=[warmup_steps]\n )\n train_dataloader, self.scheduler = self.accelerator.prepare(\n train_dataloader, self.scheduler\n ) # actual steps = 1 gpu steps / gpus\n start_step = self.load_checkpoint()\n global_step = start_step\n\n valid_dataloader = self.accelerator.prepare(valid_dataloader)\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar = tqdm(\n skipped_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n initial=skipped_batch,\n total=orig_epoch_step,\n )\n else:\n progress_bar = tqdm(\n train_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n )\n\n for batch in progress_bar:\n with self.accelerator.accumulate(self.model):\n # Inputs\n prompt_mel = batch['pmt_mel_specs'].permute(0, 2, 1) # (B, L_mel, D)\n prompt_text = batch['pmt_text']\n text = batch['text']\n\n target_ids = list_str_to_idx(text, self.vocab_char_map).to(prompt_mel.device)\n target_ids = target_ids.masked_fill(target_ids==-1, vocab_size)\n\n prompt_ids = list_str_to_idx(prompt_text, self.vocab_char_map).to(prompt_mel.device)\n prompt_ids = prompt_ids.masked_fill(prompt_ids==-1, vocab_size)\n\n # Targets\n tar_lengths = batch['mel_lengths']\n\n # Forward\n predictions = SLP(target_ids=target_ids, prompt_ids=prompt_ids, prompt_mel=prompt_mel) # (B, C)\n\n if self.loss_fn == 'CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_labels = torch.argmax(est_length_logtis, dim=-1)\n loss = F.cross_entropy(est_length_logtis, tar_length_labels)\n \n with torch.no_grad():\n est_lengths = est_length_labels * self.n_frame_per_class\n frame_error = (est_lengths.float() - tar_lengths.float()).abs().mean()\n sec_error = frame_error * 256 / 24000\n\n log_dict = {\n 'loss': loss.item(), \n 'loss_CE': loss.item(), \n 'sec_error': sec_error.item(),\n 'lr': self.scheduler.get_last_lr()[0]\n }\n\n else:\n raise NotImplementedError(self.loss_fn)\n\n \n self.accelerator.backward(loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n self.optimizer.step()\n self.scheduler.step()\n self.optimizer.zero_grad()\n\n global_step += 1\n\n if self.accelerator.is_local_main_process:\n self.accelerator.log(log_dict, step=global_step)\n progress_bar.set_postfix(step=str(global_step), loss=loss.item())\n\n if global_step % (self.save_per_updates * self.grad_accumulation_steps) == 0:\n self.save_checkpoint(global_step)\n # if self.log_samples and self.accelerator.is_local_main_process:\n # Run validation at the end of each epoch (only on the main process)\n if self.accelerator.is_local_main_process:\n self.validate(valid_dataloader, global_step)\n # if global_step % self.last_per_steps == 0:\n # self.save_checkpoint(global_step, last=True)\n\n self.save_checkpoint(global_step, last=True)\n self.accelerator.end_training()","source_hash":"20ff79d536aa888b20ce19adba2d43f1dcea96063d23f16c210e5b9fdd404874","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer_with_prompt.__init__","uri":"program://DMOSpeech2/function/src.duration_trainer_with_prompt.__init__#L40-L140","kind":"function","name":"__init__","path":"src/duration_trainer_with_prompt.py","language":"python","start_line":40,"end_line":140,"context_start_line":20,"context_end_line":160,"code":"\nfrom f5_tts.model import CFM\nfrom f5_tts.model.dataset import collate_fn, DynamicBatchSampler\nfrom f5_tts.model.utils import default, exists\n\n# trainer\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nSAMPLE_RATE = 24_000\n\n\nclass Trainer:\n def __init__(\n self,\n model,\n vocab_size,\n vocab_char_map,\n process_token_to_id=True,\n loss_fn='L1',\n lambda_L1=1,\n gumbel_tau=0.5,\n n_class=301,\n n_frame_per_class=10,\n epochs=15,\n learning_rate=1e-4,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n logger: str | None = \"wandb\", # \"wandb\" | \"tensorboard\" | None\n wandb_project=\"test_e2-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n last_per_steps=None,\n accelerate_kwargs: dict = dict(),\n ema_kwargs: dict = dict(),\n bnb_optimizer: bool = False,\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=False)\n\n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n print(f\"Using logger: {logger}\")\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n },\n )\n\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n self.model = model\n self.vocab_size = vocab_size\n self.vocab_char_map = vocab_char_map\n self.process_token_to_id = process_token_to_id\n assert loss_fn in ['L1', 'CE', 'L1_and_CE']\n self.loss_fn = loss_fn\n self.lambda_L1 = lambda_L1\n self.n_class = n_class\n self.n_frame_per_class = n_frame_per_class\n self.gumbel_tau = gumbel_tau\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")","source_hash":"20ff79d536aa888b20ce19adba2d43f1dcea96063d23f16c210e5b9fdd404874","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer_with_prompt.is_main","uri":"program://DMOSpeech2/function/src.duration_trainer_with_prompt.is_main#L143-L144","kind":"function","name":"is_main","path":"src/duration_trainer_with_prompt.py","language":"python","start_line":143,"end_line":144,"context_start_line":123,"context_end_line":164,"code":" self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)","source_hash":"20ff79d536aa888b20ce19adba2d43f1dcea96063d23f16c210e5b9fdd404874","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer_with_prompt.save_checkpoint","uri":"program://DMOSpeech2/function/src.duration_trainer_with_prompt.save_checkpoint#L146-L160","kind":"function","name":"save_checkpoint","path":"src/duration_trainer_with_prompt.py","language":"python","start_line":146,"end_line":160,"context_start_line":126,"context_end_line":180,"code":" self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n\n print(f'To load from {latest_checkpoint}.')\n","source_hash":"20ff79d536aa888b20ce19adba2d43f1dcea96063d23f16c210e5b9fdd404874","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer_with_prompt.load_checkpoint","uri":"program://DMOSpeech2/function/src.duration_trainer_with_prompt.load_checkpoint#L162-L211","kind":"function","name":"load_checkpoint","path":"src/duration_trainer_with_prompt.py","language":"python","start_line":162,"end_line":211,"context_start_line":142,"context_end_line":231,"code":" @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n\n print(f'To load from {latest_checkpoint}.')\n\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n print(f'Loaded from {latest_checkpoint}.')\n\n if \"step\" in checkpoint:\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n \n del checkpoint\n gc.collect()\n\n print(f'Exit load_checkpoint.')\n\n return step\n\n\n def validate(self, valid_dataloader, global_step):\n \"\"\"\n Runs evaluation on the validation set, computes the average loss,\n and logs the average validation loss along with the CTC decoded strings.\n \"\"\"\n self.model.eval()\n total_valid_loss = 0.0\n total_sec_error = 0.0\n count = 0\n\n # Iterate over the validation dataloader\n with torch.no_grad():\n for batch in valid_dataloader:\n\n # Inputs\n prompt_mel = batch['pmt_mel_specs'].permute(0, 2, 1) # (B, L_mel, D)\n prompt_text = batch['pmt_text']\n text = batch['text']","source_hash":"20ff79d536aa888b20ce19adba2d43f1dcea96063d23f16c210e5b9fdd404874","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer_with_prompt.validate","uri":"program://DMOSpeech2/function/src.duration_trainer_with_prompt.validate#L214-L272","kind":"function","name":"validate","path":"src/duration_trainer_with_prompt.py","language":"python","start_line":214,"end_line":272,"context_start_line":194,"context_end_line":292,"code":" if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n \n del checkpoint\n gc.collect()\n\n print(f'Exit load_checkpoint.')\n\n return step\n\n\n def validate(self, valid_dataloader, global_step):\n \"\"\"\n Runs evaluation on the validation set, computes the average loss,\n and logs the average validation loss along with the CTC decoded strings.\n \"\"\"\n self.model.eval()\n total_valid_loss = 0.0\n total_sec_error = 0.0\n count = 0\n\n # Iterate over the validation dataloader\n with torch.no_grad():\n for batch in valid_dataloader:\n\n # Inputs\n prompt_mel = batch['pmt_mel_specs'].permute(0, 2, 1) # (B, L_mel, D)\n prompt_text = batch['pmt_text']\n text = batch['text']\n\n target_ids = list_str_to_idx(text, self.vocab_char_map).to(prompt_mel.device)\n target_ids = target_ids.masked_fill(target_ids==-1, vocab_size)\n\n prompt_ids = list_str_to_idx(prompt_text, self.vocab_char_map).to(prompt_mel.device)\n prompt_ids = prompt_ids.masked_fill(prompt_ids==-1, vocab_size)\n\n # Targets\n tar_lengths = batch['mel_lengths']\n\n # Forward\n predictions = SLP(target_ids=target_ids, prompt_ids=prompt_ids, prompt_mel=prompt_mel) # (B, C)\n\n if self.loss_fn == 'CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_labels = torch.argmax(est_length_logtis, dim=-1)\n loss = F.cross_entropy(est_length_logtis, tar_length_labels)\n \n est_lengths = est_length_labels * self.n_frame_per_class\n frame_error = (est_lengths.float() - tar_lengths.float()).abs().mean()\n sec_error = frame_error * 256 / 24000\n\n total_sec_error += sec_error.item()\n total_valid_loss += loss.item()\n count += 1\n\n avg_valid_loss = total_valid_loss / count if count > 0 else 0.0\n avg_valid_sec_error = total_sec_error / count if count > 0 else 0.0\n\n # Log validation metrics\n self.accelerator.log(\n {\n f\"valid_loss\": avg_valid_loss,\n f\"valid_sec_error\": avg_valid_sec_error\n }, \n step=global_step\n )\n \n self.model.train()\n\n\n def train(self, train_dataset: Dataset, valid_dataset: Dataset,\n num_workers=64, resumable_with_seed: int = None):\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n # Create training dataloader using the appropriate batching strategy\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,","source_hash":"20ff79d536aa888b20ce19adba2d43f1dcea96063d23f16c210e5b9fdd404874","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer_with_prompt.train","uri":"program://DMOSpeech2/function/src.duration_trainer_with_prompt.train#L275-L452","kind":"function","name":"train","path":"src/duration_trainer_with_prompt.py","language":"python","start_line":275,"end_line":452,"context_start_line":255,"context_end_line":452,"code":"\n total_sec_error += sec_error.item()\n total_valid_loss += loss.item()\n count += 1\n\n avg_valid_loss = total_valid_loss / count if count > 0 else 0.0\n avg_valid_sec_error = total_sec_error / count if count > 0 else 0.0\n\n # Log validation metrics\n self.accelerator.log(\n {\n f\"valid_loss\": avg_valid_loss,\n f\"valid_sec_error\": avg_valid_sec_error\n }, \n step=global_step\n )\n \n self.model.train()\n\n\n def train(self, train_dataset: Dataset, valid_dataset: Dataset,\n num_workers=64, resumable_with_seed: int = None):\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n # Create training dataloader using the appropriate batching strategy\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )\n # Create validation dataloader (always sequential, no shuffling)\n valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n batch_size=self.batch_size,\n shuffle=False,\n )\n\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n\n sampler = SequentialSampler(valid_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n # Create validation dataloader (always sequential, no shuffling)\n valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True, \n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n \n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_steps = (\n self.num_warmup_updates * self.accelerator.num_processes\n ) # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_steps = len(train_dataloader) * self.epochs / self.grad_accumulation_steps\n decay_steps = total_steps - warmup_steps\n warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)\n decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)\n self.scheduler = SequentialLR(\n self.optimizer, schedulers=[warmup_scheduler, decay_scheduler], milestones=[warmup_steps]\n )\n train_dataloader, self.scheduler = self.accelerator.prepare(\n train_dataloader, self.scheduler\n ) # actual steps = 1 gpu steps / gpus\n start_step = self.load_checkpoint()\n global_step = start_step\n\n valid_dataloader = self.accelerator.prepare(valid_dataloader)\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar = tqdm(\n skipped_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n initial=skipped_batch,\n total=orig_epoch_step,\n )\n else:\n progress_bar = tqdm(\n train_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n )\n\n for batch in progress_bar:\n with self.accelerator.accumulate(self.model):\n # Inputs\n prompt_mel = batch['pmt_mel_specs'].permute(0, 2, 1) # (B, L_mel, D)\n prompt_text = batch['pmt_text']\n text = batch['text']\n\n target_ids = list_str_to_idx(text, self.vocab_char_map).to(prompt_mel.device)\n target_ids = target_ids.masked_fill(target_ids==-1, vocab_size)\n\n prompt_ids = list_str_to_idx(prompt_text, self.vocab_char_map).to(prompt_mel.device)\n prompt_ids = prompt_ids.masked_fill(prompt_ids==-1, vocab_size)\n\n # Targets\n tar_lengths = batch['mel_lengths']\n\n # Forward\n predictions = SLP(target_ids=target_ids, prompt_ids=prompt_ids, prompt_mel=prompt_mel) # (B, C)\n\n if self.loss_fn == 'CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_labels = torch.argmax(est_length_logtis, dim=-1)\n loss = F.cross_entropy(est_length_logtis, tar_length_labels)\n \n with torch.no_grad():\n est_lengths = est_length_labels * self.n_frame_per_class\n frame_error = (est_lengths.float() - tar_lengths.float()).abs().mean()\n sec_error = frame_error * 256 / 24000\n\n log_dict = {\n 'loss': loss.item(), \n 'loss_CE': loss.item(), \n 'sec_error': sec_error.item(),\n 'lr': self.scheduler.get_last_lr()[0]\n }\n\n else:\n raise NotImplementedError(self.loss_fn)\n\n \n self.accelerator.backward(loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n self.optimizer.step()\n self.scheduler.step()\n self.optimizer.zero_grad()\n\n global_step += 1\n\n if self.accelerator.is_local_main_process:\n self.accelerator.log(log_dict, step=global_step)\n progress_bar.set_postfix(step=str(global_step), loss=loss.item())\n\n if global_step % (self.save_per_updates * self.grad_accumulation_steps) == 0:\n self.save_checkpoint(global_step)\n # if self.log_samples and self.accelerator.is_local_main_process:\n # Run validation at the end of each epoch (only on the main process)\n if self.accelerator.is_local_main_process:\n self.validate(valid_dataloader, global_step)\n # if global_step % self.last_per_steps == 0:\n # self.save_checkpoint(global_step, last=True)\n\n self.save_checkpoint(global_step, last=True)\n self.accelerator.end_training()","source_hash":"20ff79d536aa888b20ce19adba2d43f1dcea96063d23f16c210e5b9fdd404874","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.dmd_trainer","uri":"program://DMOSpeech2/module/src.dmd_trainer#L1-L535","kind":"module","name":"src.dmd_trainer","path":"src/dmd_trainer.py","language":"python","start_line":1,"end_line":535,"context_start_line":1,"context_end_line":535,"code":"from __future__ import annotations\n\nimport os\nimport gc\nfrom tqdm import tqdm\nimport wandb\n\nimport torch\nimport torch.nn as nn\nfrom torch.optim import AdamW\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\n\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\n\nfrom unimodel import UniModel\nfrom f5_tts.model import CFM\nfrom f5_tts.model.utils import exists, default\nfrom f5_tts.model.dataset import DynamicBatchSampler, collate_fn\n\n\n# trainer\n\nimport math\n\nclass RunningStats:\n def __init__(self):\n self.count = 0\n self.mean = 0.0\n self.M2 = 0.0 # Sum of squared differences from the current mean\n\n def update(self, x):\n \"\"\"Update the running statistics with a new value x.\"\"\"\n self.count += 1\n delta = x - self.mean\n self.mean += delta / self.count\n delta2 = x - self.mean\n self.M2 += delta * delta2\n\n @property\n def variance(self):\n \"\"\"Return the sample variance. Returns NaN if fewer than two samples.\"\"\"\n return self.M2 / (self.count - 1) if self.count > 1 else float('nan')\n\n @property\n def std(self):\n \"\"\"Return the sample standard deviation.\"\"\"\n return math.sqrt(self.variance)\n\n\n\nclass Trainer:\n def __init__(\n self,\n model: UniModel,\n epochs,\n learning_rate,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n noise_scheduler: str | None = None,\n duration_predictor: torch.nn.Module | None = None,\n wandb_project=\"test_e2-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n last_per_steps=None,\n log_step=1000,\n accelerate_kwargs: dict = dict(),\n bnb_optimizer: bool = False,\n scale: float = 1.0,\n \n # training parameters for DMDSpeech\n num_student_step: int = 1,\n gen_update_ratio: int = 5,\n lambda_discriminator_loss: float = 1.0,\n lambda_generator_loss: float = 1.0,\n lambda_ctc_loss: float = 1.0,\n lambda_sim_loss: float = 1.0,\n\n num_GAN: int = 5000,\n num_D: int = 500,\n num_ctc: int = 5000,\n num_sim: int = 10000,\n num_simu: int = 1000,\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=True)\n\n logger = \"wandb\" if wandb.api.api_key else None\n print(f\"Using logger: {logger}\")\n\n self.accelerator = Accelerator(\n log_with=logger,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n if logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n \"noise_scheduler\": noise_scheduler,\n },\n )\n\n self.model = model\n\n self.scale = scale\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n self.noise_scheduler = noise_scheduler\n\n self.duration_predictor = duration_predictor\n \n self.log_step = log_step\n\n self.gen_update_ratio = gen_update_ratio # number of generator updates per guidance (fake score function and discriminator) update\n self.lambda_discriminator_loss = lambda_discriminator_loss # weight for discriminator loss (L_adv)\n self.lambda_generator_loss = lambda_generator_loss # weight for generator loss (L_adv)\n self.lambda_ctc_loss = lambda_ctc_loss # weight for ctc loss\n self.lambda_sim_loss = lambda_sim_loss # weight for similarity loss\n \n # create distillation schedule for student model\n self.student_steps = (\n torch.linspace(0.0, 1.0, num_student_step + 1)[:-1])\n \n self.GAN = model.guidance_model.gen_cls_loss # whether to use GAN training\n self.num_GAN = num_GAN # number of steps before adversarial training\n self.num_D = num_D # number of steps to train the discriminator before adversarial training \n self.num_ctc = num_ctc # number of steps before CTC training\n self.num_sim = num_sim # number of steps before similarity training\n self.num_simu = num_simu # number of steps before using simulated data\n\n # Assuming `self.model.fake_unet.parameters()` and `self.model.guidance_model.parameters()` are accessible\n if bnb_optimizer:\n import bitsandbytes as bnb\n self.optimizer_generator = bnb.optim.AdamW8bit(self.model.feedforward_model.parameters(), lr=learning_rate)\n self.optimizer_guidance = bnb.optim.AdamW8bit(self.model.guidance_model.parameters(), lr=learning_rate)\n else:\n self.optimizer_generator = AdamW(self.model.feedforward_model.parameters(), lr=learning_rate, eps=1e-7)\n self.optimizer_guidance = AdamW(self.model.guidance_model.parameters(), lr=learning_rate, eps=1e-7)\n\n self.model, self.optimizer_generator, self.optimizer_guidance = self.accelerator.prepare(self.model, self.optimizer_generator, self.optimizer_guidance)\n\n self.generator_norm = RunningStats()\n self.guidance_norm = RunningStats()\n\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_generator_state_dict=self.accelerator.unwrap_model(self.optimizer_generator).state_dict(),\n optimizer_guidance_state_dict=self.accelerator.unwrap_model(self.optimizer_guidance).state_dict(),\n scheduler_generator_state_dict=self.scheduler_generator.state_dict(),\n scheduler_guidance_state_dict=self.scheduler_guidance.state_dict(),\n step=step,\n )\n\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at step {step}\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not os.listdir(self.checkpoint_path)\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"], strict=False)\n # self.accelerator.unwrap_model(self.optimizer_generator).load_state_dict(checkpoint[\"optimizer_generator_state_dict\"])\n # self.accelerator.unwrap_model(self.optimizer_guidance).load_state_dict(checkpoint[\"optimizer_guidance_state_dict\"])\n # if self.scheduler_guidance:\n # self.scheduler_guidance.load_state_dict(checkpoint[\"scheduler_guidance_state_dict\"])\n # if self.scheduler_generator:\n # self.scheduler_generator.load_state_dict(checkpoint[\"scheduler_generator_state_dict\"])\n step = checkpoint[\"step\"]\n\n del checkpoint\n gc.collect()\n return step\n \n\n def train(self, train_dataset: Dataset, num_workers=64, resumable_with_seed: int = None, vocoder: nn.Module = None):\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n\n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_steps = (\n self.num_warmup_updates * self.accelerator.num_processes\n )\n \n # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_steps = len(train_dataloader) * self.epochs / self.grad_accumulation_steps\n decay_steps = total_steps - warmup_steps\n \n warmup_scheduler_generator = LinearLR(self.optimizer_generator, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps // (self.gen_update_ratio * self.grad_accumulation_steps))\n decay_scheduler_generator = LinearLR(self.optimizer_generator, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps // (self.gen_update_ratio * self.grad_accumulation_steps))\n self.scheduler_generator = SequentialLR(self.optimizer_generator, schedulers=[warmup_scheduler_generator, decay_scheduler_generator], milestones=[warmup_steps // (self.gen_update_ratio * self.grad_accumulation_steps)])\n\n warmup_scheduler_guidance = LinearLR(self.optimizer_guidance, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)\n decay_scheduler_guidance = LinearLR(self.optimizer_guidance, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)\n self.scheduler_guidance = SequentialLR(self.optimizer_guidance, schedulers=[warmup_scheduler_guidance, decay_scheduler_guidance], milestones=[warmup_steps])\n\n train_dataloader, self.scheduler_generator, self.scheduler_guidance = self.accelerator.prepare(\n train_dataloader, self.scheduler_generator, self.scheduler_guidance\n ) # actual steps = 1 gpu steps / gpus\n start_step = self.load_checkpoint()\n global_step = start_step\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar = tqdm(\n skipped_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n initial=skipped_batch,\n total=orig_epoch_step,\n )\n else:\n progress_bar = tqdm(\n train_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n )\n\n for batch in progress_bar:\n update_generator = global_step % self.gen_update_ratio == 0\n \n with self.accelerator.accumulate(self.model):\n metrics = {}\n text_inputs = batch[\"text\"]\n mel_spec = batch[\"mel\"].permute(0, 2, 1)\n mel_lengths = batch[\"mel_lengths\"]\n \n mel_spec = mel_spec / self.scale\n \n guidance_loss_dict, guidance_log_dict = self.model(inp=mel_spec, \n text=text_inputs, \n lens=mel_lengths, \n student_steps=self.student_steps,\n update_generator=False,\n use_simulated=global_step >= self.num_simu,\n )\n\n # if self.GAN and update_generator:\n # # only add discriminator loss if GAN is enabled and generator is being updated\n # guidance_cls_loss = guidance_loss_dict[\"guidance_cls_loss\"] * (self.lambda_discriminator_loss if global_step >= self.num_GAN and update_generator else 0)\n # metrics['loss/discriminator_loss'] = guidance_loss_dict[\"guidance_cls_loss\"]\n # self.accelerator.backward(guidance_cls_loss, retain_graph=True)\n \n # if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n # metrics['grad_norm_guidance'] = self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n guidance_loss = 0\n guidance_loss += guidance_loss_dict[\"loss_fake_mean\"]\n metrics['loss/fake_score'] = guidance_loss_dict[\"loss_fake_mean\"]\n metrics[\"loss/guidance_loss\"] = guidance_loss\n\n if self.GAN and update_generator:\n # only add discriminator loss if GAN is enabled and generator is being updated\n guidance_cls_loss = guidance_loss_dict[\"guidance_cls_loss\"] * (self.lambda_discriminator_loss if global_step >= self.num_GAN and update_generator else 0)\n metrics['loss/discriminator_loss'] = guidance_loss_dict[\"guidance_cls_loss\"]\n\n guidance_loss += guidance_cls_loss\n \n self.accelerator.backward(guidance_loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n metrics['grad_norm_guidance'] = self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n # if self.guidance_norm.count < 100:\n # self.guidance_norm.update(metrics['grad_norm_guidance'])\n\n # if metrics['grad_norm_guidance'] > self.guidance_norm.mean + 5 * self.guidance_norm.std:\n # self.optimizer_generator.zero_grad()\n # self.optimizer_guidance.zero_grad()\n # print(\"Gradient explosion detected. Skipping batch.\")\n # elif self.guidance_norm.count >= 100:\n # self.guidance_norm.update(metrics['grad_norm_guidance'])\n\n\n self.optimizer_guidance.step()\n self.scheduler_guidance.step()\n self.optimizer_guidance.zero_grad()\n self.optimizer_generator.zero_grad() # zero out the generator's gradient as well\n \n if update_generator:\n generator_loss_dict, generator_log_dict = self.model(inp=mel_spec, \n text=text_inputs, \n lens=mel_lengths, \n student_steps=self.student_steps,\n update_generator=True,\n use_simulated=global_step >= self.num_ctc,\n )\n # if self.GAN:\n # gen_cls_loss = generator_loss_dict[\"gen_cls_loss\"] * (self.lambda_generator_loss if global_step >= (self.num_GAN + self.num_D) and update_generator else 0)\n # metrics[\"loss/gen_cls_loss\"] = generator_loss_dict[\"gen_cls_loss\"]\n\n # self.accelerator.backward(gen_cls_loss, retain_graph=True)\n\n # if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n # metrics['grad_norm_generator'] = self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n generator_loss = 0\n generator_loss += generator_loss_dict[\"loss_dm\"]\n if \"loss_mse\" in generator_loss_dict:\n generator_loss += generator_loss_dict[\"loss_mse\"] \n generator_loss += generator_loss_dict[\"loss_ctc\"] * (self.lambda_ctc_loss if global_step >= self.num_ctc else 0)\n generator_loss += generator_loss_dict[\"loss_sim\"] * (self.lambda_sim_loss if global_step >= self.num_sim else 0)\n generator_loss += generator_loss_dict[\"loss_kl\"] * (self.lambda_ctc_loss if global_step >= self.num_ctc else 0)\n if self.GAN:\n# ... truncated ...","source_hash":"2f2618612fc8a72a725d379657533ec1e177fd919989a0850d8a287ee23d5b40","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.dmd_trainer.RunningStats","uri":"program://DMOSpeech2/class/src.dmd_trainer.RunningStats#L27-L49","kind":"class","name":"RunningStats","path":"src/dmd_trainer.py","language":"python","start_line":27,"end_line":49,"context_start_line":7,"context_end_line":69,"code":"\nimport torch\nimport torch.nn as nn\nfrom torch.optim import AdamW\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\n\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\n\nfrom unimodel import UniModel\nfrom f5_tts.model import CFM\nfrom f5_tts.model.utils import exists, default\nfrom f5_tts.model.dataset import DynamicBatchSampler, collate_fn\n\n\n# trainer\n\nimport math\n\nclass RunningStats:\n def __init__(self):\n self.count = 0\n self.mean = 0.0\n self.M2 = 0.0 # Sum of squared differences from the current mean\n\n def update(self, x):\n \"\"\"Update the running statistics with a new value x.\"\"\"\n self.count += 1\n delta = x - self.mean\n self.mean += delta / self.count\n delta2 = x - self.mean\n self.M2 += delta * delta2\n\n @property\n def variance(self):\n \"\"\"Return the sample variance. Returns NaN if fewer than two samples.\"\"\"\n return self.M2 / (self.count - 1) if self.count > 1 else float('nan')\n\n @property\n def std(self):\n \"\"\"Return the sample standard deviation.\"\"\"\n return math.sqrt(self.variance)\n\n\n\nclass Trainer:\n def __init__(\n self,\n model: UniModel,\n epochs,\n learning_rate,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n noise_scheduler: str | None = None,\n duration_predictor: torch.nn.Module | None = None,\n wandb_project=\"test_e2-tts\",","source_hash":"2f2618612fc8a72a725d379657533ec1e177fd919989a0850d8a287ee23d5b40","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.dmd_trainer.Trainer","uri":"program://DMOSpeech2/class/src.dmd_trainer.Trainer#L53-L533","kind":"class","name":"Trainer","path":"src/dmd_trainer.py","language":"python","start_line":53,"end_line":533,"context_start_line":33,"context_end_line":535,"code":" def update(self, x):\n \"\"\"Update the running statistics with a new value x.\"\"\"\n self.count += 1\n delta = x - self.mean\n self.mean += delta / self.count\n delta2 = x - self.mean\n self.M2 += delta * delta2\n\n @property\n def variance(self):\n \"\"\"Return the sample variance. Returns NaN if fewer than two samples.\"\"\"\n return self.M2 / (self.count - 1) if self.count > 1 else float('nan')\n\n @property\n def std(self):\n \"\"\"Return the sample standard deviation.\"\"\"\n return math.sqrt(self.variance)\n\n\n\nclass Trainer:\n def __init__(\n self,\n model: UniModel,\n epochs,\n learning_rate,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n noise_scheduler: str | None = None,\n duration_predictor: torch.nn.Module | None = None,\n wandb_project=\"test_e2-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n last_per_steps=None,\n log_step=1000,\n accelerate_kwargs: dict = dict(),\n bnb_optimizer: bool = False,\n scale: float = 1.0,\n \n # training parameters for DMDSpeech\n num_student_step: int = 1,\n gen_update_ratio: int = 5,\n lambda_discriminator_loss: float = 1.0,\n lambda_generator_loss: float = 1.0,\n lambda_ctc_loss: float = 1.0,\n lambda_sim_loss: float = 1.0,\n\n num_GAN: int = 5000,\n num_D: int = 500,\n num_ctc: int = 5000,\n num_sim: int = 10000,\n num_simu: int = 1000,\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=True)\n\n logger = \"wandb\" if wandb.api.api_key else None\n print(f\"Using logger: {logger}\")\n\n self.accelerator = Accelerator(\n log_with=logger,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n if logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n \"noise_scheduler\": noise_scheduler,\n },\n )\n\n self.model = model\n\n self.scale = scale\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n self.noise_scheduler = noise_scheduler\n\n self.duration_predictor = duration_predictor\n \n self.log_step = log_step\n\n self.gen_update_ratio = gen_update_ratio # number of generator updates per guidance (fake score function and discriminator) update\n self.lambda_discriminator_loss = lambda_discriminator_loss # weight for discriminator loss (L_adv)\n self.lambda_generator_loss = lambda_generator_loss # weight for generator loss (L_adv)\n self.lambda_ctc_loss = lambda_ctc_loss # weight for ctc loss\n self.lambda_sim_loss = lambda_sim_loss # weight for similarity loss\n \n # create distillation schedule for student model\n self.student_steps = (\n torch.linspace(0.0, 1.0, num_student_step + 1)[:-1])\n \n self.GAN = model.guidance_model.gen_cls_loss # whether to use GAN training\n self.num_GAN = num_GAN # number of steps before adversarial training\n self.num_D = num_D # number of steps to train the discriminator before adversarial training \n self.num_ctc = num_ctc # number of steps before CTC training\n self.num_sim = num_sim # number of steps before similarity training\n self.num_simu = num_simu # number of steps before using simulated data\n\n # Assuming `self.model.fake_unet.parameters()` and `self.model.guidance_model.parameters()` are accessible\n if bnb_optimizer:\n import bitsandbytes as bnb\n self.optimizer_generator = bnb.optim.AdamW8bit(self.model.feedforward_model.parameters(), lr=learning_rate)\n self.optimizer_guidance = bnb.optim.AdamW8bit(self.model.guidance_model.parameters(), lr=learning_rate)\n else:\n self.optimizer_generator = AdamW(self.model.feedforward_model.parameters(), lr=learning_rate, eps=1e-7)\n self.optimizer_guidance = AdamW(self.model.guidance_model.parameters(), lr=learning_rate, eps=1e-7)\n\n self.model, self.optimizer_generator, self.optimizer_guidance = self.accelerator.prepare(self.model, self.optimizer_generator, self.optimizer_guidance)\n\n self.generator_norm = RunningStats()\n self.guidance_norm = RunningStats()\n\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_generator_state_dict=self.accelerator.unwrap_model(self.optimizer_generator).state_dict(),\n optimizer_guidance_state_dict=self.accelerator.unwrap_model(self.optimizer_guidance).state_dict(),\n scheduler_generator_state_dict=self.scheduler_generator.state_dict(),\n scheduler_guidance_state_dict=self.scheduler_guidance.state_dict(),\n step=step,\n )\n\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at step {step}\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not os.listdir(self.checkpoint_path)\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"], strict=False)\n # self.accelerator.unwrap_model(self.optimizer_generator).load_state_dict(checkpoint[\"optimizer_generator_state_dict\"])\n # self.accelerator.unwrap_model(self.optimizer_guidance).load_state_dict(checkpoint[\"optimizer_guidance_state_dict\"])\n # if self.scheduler_guidance:\n # self.scheduler_guidance.load_state_dict(checkpoint[\"scheduler_guidance_state_dict\"])\n # if self.scheduler_generator:\n # self.scheduler_generator.load_state_dict(checkpoint[\"scheduler_generator_state_dict\"])\n step = checkpoint[\"step\"]\n\n del checkpoint\n gc.collect()\n return step\n \n\n def train(self, train_dataset: Dataset, num_workers=64, resumable_with_seed: int = None, vocoder: nn.Module = None):\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n\n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_steps = (\n self.num_warmup_updates * self.accelerator.num_processes\n )\n \n # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_steps = len(train_dataloader) * self.epochs / self.grad_accumulation_steps\n decay_steps = total_steps - warmup_steps\n \n warmup_scheduler_generator = LinearLR(self.optimizer_generator, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps // (self.gen_update_ratio * self.grad_accumulation_steps))\n decay_scheduler_generator = LinearLR(self.optimizer_generator, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps // (self.gen_update_ratio * self.grad_accumulation_steps))\n self.scheduler_generator = SequentialLR(self.optimizer_generator, schedulers=[warmup_scheduler_generator, decay_scheduler_generator], milestones=[warmup_steps // (self.gen_update_ratio * self.grad_accumulation_steps)])\n\n warmup_scheduler_guidance = LinearLR(self.optimizer_guidance, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)\n decay_scheduler_guidance = LinearLR(self.optimizer_guidance, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)\n self.scheduler_guidance = SequentialLR(self.optimizer_guidance, schedulers=[warmup_scheduler_guidance, decay_scheduler_guidance], milestones=[warmup_steps])\n\n train_dataloader, self.scheduler_generator, self.scheduler_guidance = self.accelerator.prepare(\n train_dataloader, self.scheduler_generator, self.scheduler_guidance\n ) # actual steps = 1 gpu steps / gpus\n start_step = self.load_checkpoint()\n global_step = start_step\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar = tqdm(\n skipped_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n initial=skipped_batch,\n total=orig_epoch_step,\n )\n else:\n progress_bar = tqdm(\n train_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n )\n\n for batch in progress_bar:\n update_generator = global_step % self.gen_update_ratio == 0\n \n with self.accelerator.accumulate(self.model):\n metrics = {}\n text_inputs = batch[\"text\"]\n mel_spec = batch[\"mel\"].permute(0, 2, 1)\n mel_lengths = batch[\"mel_lengths\"]\n \n mel_spec = mel_spec / self.scale\n \n guidance_loss_dict, guidance_log_dict = self.model(inp=mel_spec, \n text=text_inputs, \n lens=mel_lengths, \n student_steps=self.student_steps,\n update_generator=False,\n use_simulated=global_step >= self.num_simu,\n )\n\n # if self.GAN and update_generator:\n # # only add discriminator loss if GAN is enabled and generator is being updated\n # guidance_cls_loss = guidance_loss_dict[\"guidance_cls_loss\"] * (self.lambda_discriminator_loss if global_step >= self.num_GAN and update_generator else 0)\n # metrics['loss/discriminator_loss'] = guidance_loss_dict[\"guidance_cls_loss\"]\n # self.accelerator.backward(guidance_cls_loss, retain_graph=True)\n \n # if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n # metrics['grad_norm_guidance'] = self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n guidance_loss = 0\n guidance_loss += guidance_loss_dict[\"loss_fake_mean\"]\n metrics['loss/fake_score'] = guidance_loss_dict[\"loss_fake_mean\"]\n metrics[\"loss/guidance_loss\"] = guidance_loss\n\n if self.GAN and update_generator:\n # only add discriminator loss if GAN is enabled and generator is being updated\n guidance_cls_loss = guidance_loss_dict[\"guidance_cls_loss\"] * (self.lambda_discriminator_loss if global_step >= self.num_GAN and update_generator else 0)\n metrics['loss/discriminator_loss'] = guidance_loss_dict[\"guidance_cls_loss\"]\n\n guidance_loss += guidance_cls_loss\n \n self.accelerator.backward(guidance_loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n metrics['grad_norm_guidance'] = self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n # if self.guidance_norm.count < 100:\n # self.guidance_norm.update(metrics['grad_norm_guidance'])\n\n # if metrics['grad_norm_guidance'] > self.guidance_norm.mean + 5 * self.guidance_norm.std:\n # self.optimizer_generator.zero_grad()\n # self.optimizer_guidance.zero_grad()\n # print(\"Gradient explosion detected. Skipping batch.\")\n # elif self.guidance_norm.count >= 100:\n # self.guidance_norm.update(metrics['grad_norm_guidance'])\n\n\n self.optimizer_guidance.step()\n self.scheduler_guidance.step()\n self.optimizer_guidance.zero_grad()\n self.optimizer_generator.zero_grad() # zero out the generator's gradient as well\n \n if update_generator:\n generator_loss_dict, generator_log_dict = self.model(inp=mel_spec, \n text=text_inputs, \n lens=mel_lengths, \n student_steps=self.student_steps,\n update_generator=True,\n use_simulated=global_step >= self.num_ctc,\n )\n # if self.GAN:\n # gen_cls_loss = generator_loss_dict[\"gen_cls_loss\"] * (self.lambda_generator_loss if global_step >= (self.num_GAN + self.num_D) and update_generator else 0)\n # metrics[\"loss/gen_cls_loss\"] = generator_loss_dict[\"gen_cls_loss\"]\n\n # self.accelerator.backward(gen_cls_loss, retain_graph=True)\n\n # if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n # metrics['grad_norm_generator'] = self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n generator_loss = 0\n generator_loss += generator_loss_dict[\"loss_dm\"]\n if \"loss_mse\" in generator_loss_dict:\n generator_loss += generator_loss_dict[\"loss_mse\"] \n generator_loss += generator_loss_dict[\"loss_ctc\"] * (self.lambda_ctc_loss if global_step >= self.num_ctc else 0)\n generator_loss += generator_loss_dict[\"loss_sim\"] * (self.lambda_sim_loss if global_step >= self.num_sim else 0)\n generator_loss += generator_loss_dict[\"loss_kl\"] * (self.lambda_ctc_loss if global_step >= self.num_ctc else 0)\n if self.GAN:\n gen_cls_loss = generator_loss_dict[\"gen_cls_loss\"] * (self.lambda_generator_loss if global_step >= (self.num_GAN + self.num_D) and update_generator else 0)\n metrics[\"loss/gen_cls_loss\"] = generator_loss_dict[\"gen_cls_loss\"]\n generator_loss += gen_cls_loss\n\n metrics['loss/dm_loss'] = generator_loss_dict[\"loss_dm\"]\n metrics['loss/ctc_loss'] = generator_loss_dict[\"loss_ctc\"]\n\n metrics['loss/similarity_loss'] = generator_loss_dict[\"loss_sim\"]\n metrics['loss/generator_loss'] = generator_loss\n \n if \"loss_mse\" in generator_loss_dict and ge\n# ... truncated ...","source_hash":"2f2618612fc8a72a725d379657533ec1e177fd919989a0850d8a287ee23d5b40","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.dmd_trainer.__init__","uri":"program://DMOSpeech2/function/src.dmd_trainer.__init__#L54-L177","kind":"function","name":"__init__","path":"src/dmd_trainer.py","language":"python","start_line":54,"end_line":177,"context_start_line":34,"context_end_line":197,"code":" \"\"\"Update the running statistics with a new value x.\"\"\"\n self.count += 1\n delta = x - self.mean\n self.mean += delta / self.count\n delta2 = x - self.mean\n self.M2 += delta * delta2\n\n @property\n def variance(self):\n \"\"\"Return the sample variance. Returns NaN if fewer than two samples.\"\"\"\n return self.M2 / (self.count - 1) if self.count > 1 else float('nan')\n\n @property\n def std(self):\n \"\"\"Return the sample standard deviation.\"\"\"\n return math.sqrt(self.variance)\n\n\n\nclass Trainer:\n def __init__(\n self,\n model: UniModel,\n epochs,\n learning_rate,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n noise_scheduler: str | None = None,\n duration_predictor: torch.nn.Module | None = None,\n wandb_project=\"test_e2-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n last_per_steps=None,\n log_step=1000,\n accelerate_kwargs: dict = dict(),\n bnb_optimizer: bool = False,\n scale: float = 1.0,\n \n # training parameters for DMDSpeech\n num_student_step: int = 1,\n gen_update_ratio: int = 5,\n lambda_discriminator_loss: float = 1.0,\n lambda_generator_loss: float = 1.0,\n lambda_ctc_loss: float = 1.0,\n lambda_sim_loss: float = 1.0,\n\n num_GAN: int = 5000,\n num_D: int = 500,\n num_ctc: int = 5000,\n num_sim: int = 10000,\n num_simu: int = 1000,\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=True)\n\n logger = \"wandb\" if wandb.api.api_key else None\n print(f\"Using logger: {logger}\")\n\n self.accelerator = Accelerator(\n log_with=logger,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n if logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n \"noise_scheduler\": noise_scheduler,\n },\n )\n\n self.model = model\n\n self.scale = scale\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n self.noise_scheduler = noise_scheduler\n\n self.duration_predictor = duration_predictor\n \n self.log_step = log_step\n\n self.gen_update_ratio = gen_update_ratio # number of generator updates per guidance (fake score function and discriminator) update\n self.lambda_discriminator_loss = lambda_discriminator_loss # weight for discriminator loss (L_adv)\n self.lambda_generator_loss = lambda_generator_loss # weight for generator loss (L_adv)\n self.lambda_ctc_loss = lambda_ctc_loss # weight for ctc loss\n self.lambda_sim_loss = lambda_sim_loss # weight for similarity loss\n \n # create distillation schedule for student model\n self.student_steps = (\n torch.linspace(0.0, 1.0, num_student_step + 1)[:-1])\n \n self.GAN = model.guidance_model.gen_cls_loss # whether to use GAN training\n self.num_GAN = num_GAN # number of steps before adversarial training\n self.num_D = num_D # number of steps to train the discriminator before adversarial training \n self.num_ctc = num_ctc # number of steps before CTC training\n self.num_sim = num_sim # number of steps before similarity training\n self.num_simu = num_simu # number of steps before using simulated data\n\n # Assuming `self.model.fake_unet.parameters()` and `self.model.guidance_model.parameters()` are accessible\n if bnb_optimizer:\n import bitsandbytes as bnb\n self.optimizer_generator = bnb.optim.AdamW8bit(self.model.feedforward_model.parameters(), lr=learning_rate)\n self.optimizer_guidance = bnb.optim.AdamW8bit(self.model.guidance_model.parameters(), lr=learning_rate)\n else:\n self.optimizer_generator = AdamW(self.model.feedforward_model.parameters(), lr=learning_rate, eps=1e-7)\n self.optimizer_guidance = AdamW(self.model.guidance_model.parameters(), lr=learning_rate, eps=1e-7)\n\n self.model, self.optimizer_generator, self.optimizer_guidance = self.accelerator.prepare(self.model, self.optimizer_generator, self.optimizer_guidance)\n\n self.generator_norm = RunningStats()\n self.guidance_norm = RunningStats()\n\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_generator_state_dict=self.accelerator.unwrap_model(self.optimizer_generator).state_dict(),\n optimizer_guidance_state_dict=self.accelerator.unwrap_model(self.optimizer_guidance).state_dict(),\n scheduler_generator_state_dict=self.scheduler_generator.state_dict(),\n scheduler_guidance_state_dict=self.scheduler_guidance.state_dict(),\n step=step,\n )\n\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)","source_hash":"2f2618612fc8a72a725d379657533ec1e177fd919989a0850d8a287ee23d5b40","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.dmd_trainer.update","uri":"program://DMOSpeech2/function/src.dmd_trainer.update#L33-L39","kind":"function","name":"update","path":"src/dmd_trainer.py","language":"python","start_line":33,"end_line":39,"context_start_line":13,"context_end_line":59,"code":"\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\n\nfrom unimodel import UniModel\nfrom f5_tts.model import CFM\nfrom f5_tts.model.utils import exists, default\nfrom f5_tts.model.dataset import DynamicBatchSampler, collate_fn\n\n\n# trainer\n\nimport math\n\nclass RunningStats:\n def __init__(self):\n self.count = 0\n self.mean = 0.0\n self.M2 = 0.0 # Sum of squared differences from the current mean\n\n def update(self, x):\n \"\"\"Update the running statistics with a new value x.\"\"\"\n self.count += 1\n delta = x - self.mean\n self.mean += delta / self.count\n delta2 = x - self.mean\n self.M2 += delta * delta2\n\n @property\n def variance(self):\n \"\"\"Return the sample variance. Returns NaN if fewer than two samples.\"\"\"\n return self.M2 / (self.count - 1) if self.count > 1 else float('nan')\n\n @property\n def std(self):\n \"\"\"Return the sample standard deviation.\"\"\"\n return math.sqrt(self.variance)\n\n\n\nclass Trainer:\n def __init__(\n self,\n model: UniModel,\n epochs,\n learning_rate,\n num_warmup_updates=20000,","source_hash":"2f2618612fc8a72a725d379657533ec1e177fd919989a0850d8a287ee23d5b40","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.dmd_trainer.variance","uri":"program://DMOSpeech2/function/src.dmd_trainer.variance#L42-L44","kind":"function","name":"variance","path":"src/dmd_trainer.py","language":"python","start_line":42,"end_line":44,"context_start_line":22,"context_end_line":64,"code":"\n# trainer\n\nimport math\n\nclass RunningStats:\n def __init__(self):\n self.count = 0\n self.mean = 0.0\n self.M2 = 0.0 # Sum of squared differences from the current mean\n\n def update(self, x):\n \"\"\"Update the running statistics with a new value x.\"\"\"\n self.count += 1\n delta = x - self.mean\n self.mean += delta / self.count\n delta2 = x - self.mean\n self.M2 += delta * delta2\n\n @property\n def variance(self):\n \"\"\"Return the sample variance. Returns NaN if fewer than two samples.\"\"\"\n return self.M2 / (self.count - 1) if self.count > 1 else float('nan')\n\n @property\n def std(self):\n \"\"\"Return the sample standard deviation.\"\"\"\n return math.sqrt(self.variance)\n\n\n\nclass Trainer:\n def __init__(\n self,\n model: UniModel,\n epochs,\n learning_rate,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,","source_hash":"2f2618612fc8a72a725d379657533ec1e177fd919989a0850d8a287ee23d5b40","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.dmd_trainer.std","uri":"program://DMOSpeech2/function/src.dmd_trainer.std#L47-L49","kind":"function","name":"std","path":"src/dmd_trainer.py","language":"python","start_line":47,"end_line":49,"context_start_line":27,"context_end_line":69,"code":"class RunningStats:\n def __init__(self):\n self.count = 0\n self.mean = 0.0\n self.M2 = 0.0 # Sum of squared differences from the current mean\n\n def update(self, x):\n \"\"\"Update the running statistics with a new value x.\"\"\"\n self.count += 1\n delta = x - self.mean\n self.mean += delta / self.count\n delta2 = x - self.mean\n self.M2 += delta * delta2\n\n @property\n def variance(self):\n \"\"\"Return the sample variance. Returns NaN if fewer than two samples.\"\"\"\n return self.M2 / (self.count - 1) if self.count > 1 else float('nan')\n\n @property\n def std(self):\n \"\"\"Return the sample standard deviation.\"\"\"\n return math.sqrt(self.variance)\n\n\n\nclass Trainer:\n def __init__(\n self,\n model: UniModel,\n epochs,\n learning_rate,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n noise_scheduler: str | None = None,\n duration_predictor: torch.nn.Module | None = None,\n wandb_project=\"test_e2-tts\",","source_hash":"2f2618612fc8a72a725d379657533ec1e177fd919989a0850d8a287ee23d5b40","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.dmd_trainer.is_main","uri":"program://DMOSpeech2/function/src.dmd_trainer.is_main#L181-L182","kind":"function","name":"is_main","path":"src/dmd_trainer.py","language":"python","start_line":181,"end_line":182,"context_start_line":161,"context_end_line":202,"code":" self.num_ctc = num_ctc # number of steps before CTC training\n self.num_sim = num_sim # number of steps before similarity training\n self.num_simu = num_simu # number of steps before using simulated data\n\n # Assuming `self.model.fake_unet.parameters()` and `self.model.guidance_model.parameters()` are accessible\n if bnb_optimizer:\n import bitsandbytes as bnb\n self.optimizer_generator = bnb.optim.AdamW8bit(self.model.feedforward_model.parameters(), lr=learning_rate)\n self.optimizer_guidance = bnb.optim.AdamW8bit(self.model.guidance_model.parameters(), lr=learning_rate)\n else:\n self.optimizer_generator = AdamW(self.model.feedforward_model.parameters(), lr=learning_rate, eps=1e-7)\n self.optimizer_guidance = AdamW(self.model.guidance_model.parameters(), lr=learning_rate, eps=1e-7)\n\n self.model, self.optimizer_generator, self.optimizer_guidance = self.accelerator.prepare(self.model, self.optimizer_generator, self.optimizer_guidance)\n\n self.generator_norm = RunningStats()\n self.guidance_norm = RunningStats()\n\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_generator_state_dict=self.accelerator.unwrap_model(self.optimizer_generator).state_dict(),\n optimizer_guidance_state_dict=self.accelerator.unwrap_model(self.optimizer_guidance).state_dict(),\n scheduler_generator_state_dict=self.scheduler_generator.state_dict(),\n scheduler_guidance_state_dict=self.scheduler_guidance.state_dict(),\n step=step,\n )\n\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at step {step}\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")","source_hash":"2f2618612fc8a72a725d379657533ec1e177fd919989a0850d8a287ee23d5b40","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.dmd_trainer.save_checkpoint","uri":"program://DMOSpeech2/function/src.dmd_trainer.save_checkpoint#L184-L202","kind":"function","name":"save_checkpoint","path":"src/dmd_trainer.py","language":"python","start_line":184,"end_line":202,"context_start_line":164,"context_end_line":222,"code":"\n # Assuming `self.model.fake_unet.parameters()` and `self.model.guidance_model.parameters()` are accessible\n if bnb_optimizer:\n import bitsandbytes as bnb\n self.optimizer_generator = bnb.optim.AdamW8bit(self.model.feedforward_model.parameters(), lr=learning_rate)\n self.optimizer_guidance = bnb.optim.AdamW8bit(self.model.guidance_model.parameters(), lr=learning_rate)\n else:\n self.optimizer_generator = AdamW(self.model.feedforward_model.parameters(), lr=learning_rate, eps=1e-7)\n self.optimizer_guidance = AdamW(self.model.guidance_model.parameters(), lr=learning_rate, eps=1e-7)\n\n self.model, self.optimizer_generator, self.optimizer_guidance = self.accelerator.prepare(self.model, self.optimizer_generator, self.optimizer_guidance)\n\n self.generator_norm = RunningStats()\n self.guidance_norm = RunningStats()\n\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_generator_state_dict=self.accelerator.unwrap_model(self.optimizer_generator).state_dict(),\n optimizer_guidance_state_dict=self.accelerator.unwrap_model(self.optimizer_guidance).state_dict(),\n scheduler_generator_state_dict=self.scheduler_generator.state_dict(),\n scheduler_guidance_state_dict=self.scheduler_guidance.state_dict(),\n step=step,\n )\n\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at step {step}\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not os.listdir(self.checkpoint_path)\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n","source_hash":"2f2618612fc8a72a725d379657533ec1e177fd919989a0850d8a287ee23d5b40","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.dmd_trainer.load_checkpoint","uri":"program://DMOSpeech2/function/src.dmd_trainer.load_checkpoint#L204-L234","kind":"function","name":"load_checkpoint","path":"src/dmd_trainer.py","language":"python","start_line":204,"end_line":234,"context_start_line":184,"context_end_line":254,"code":" def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_generator_state_dict=self.accelerator.unwrap_model(self.optimizer_generator).state_dict(),\n optimizer_guidance_state_dict=self.accelerator.unwrap_model(self.optimizer_guidance).state_dict(),\n scheduler_generator_state_dict=self.scheduler_generator.state_dict(),\n scheduler_guidance_state_dict=self.scheduler_guidance.state_dict(),\n step=step,\n )\n\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at step {step}\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not os.listdir(self.checkpoint_path)\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"], strict=False)\n # self.accelerator.unwrap_model(self.optimizer_generator).load_state_dict(checkpoint[\"optimizer_generator_state_dict\"])\n # self.accelerator.unwrap_model(self.optimizer_guidance).load_state_dict(checkpoint[\"optimizer_guidance_state_dict\"])\n # if self.scheduler_guidance:\n # self.scheduler_guidance.load_state_dict(checkpoint[\"scheduler_guidance_state_dict\"])\n # if self.scheduler_generator:\n # self.scheduler_generator.load_state_dict(checkpoint[\"scheduler_generator_state_dict\"])\n step = checkpoint[\"step\"]\n\n del checkpoint\n gc.collect()\n return step\n \n\n def train(self, train_dataset: Dataset, num_workers=64, resumable_with_seed: int = None, vocoder: nn.Module = None):\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )","source_hash":"2f2618612fc8a72a725d379657533ec1e177fd919989a0850d8a287ee23d5b40","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.dmd_trainer.train","uri":"program://DMOSpeech2/function/src.dmd_trainer.train#L237-L533","kind":"function","name":"train","path":"src/dmd_trainer.py","language":"python","start_line":237,"end_line":533,"context_start_line":217,"context_end_line":535,"code":" [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"], strict=False)\n # self.accelerator.unwrap_model(self.optimizer_generator).load_state_dict(checkpoint[\"optimizer_generator_state_dict\"])\n # self.accelerator.unwrap_model(self.optimizer_guidance).load_state_dict(checkpoint[\"optimizer_guidance_state_dict\"])\n # if self.scheduler_guidance:\n # self.scheduler_guidance.load_state_dict(checkpoint[\"scheduler_guidance_state_dict\"])\n # if self.scheduler_generator:\n # self.scheduler_generator.load_state_dict(checkpoint[\"scheduler_generator_state_dict\"])\n step = checkpoint[\"step\"]\n\n del checkpoint\n gc.collect()\n return step\n \n\n def train(self, train_dataset: Dataset, num_workers=64, resumable_with_seed: int = None, vocoder: nn.Module = None):\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n\n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_steps = (\n self.num_warmup_updates * self.accelerator.num_processes\n )\n \n # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_steps = len(train_dataloader) * self.epochs / self.grad_accumulation_steps\n decay_steps = total_steps - warmup_steps\n \n warmup_scheduler_generator = LinearLR(self.optimizer_generator, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps // (self.gen_update_ratio * self.grad_accumulation_steps))\n decay_scheduler_generator = LinearLR(self.optimizer_generator, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps // (self.gen_update_ratio * self.grad_accumulation_steps))\n self.scheduler_generator = SequentialLR(self.optimizer_generator, schedulers=[warmup_scheduler_generator, decay_scheduler_generator], milestones=[warmup_steps // (self.gen_update_ratio * self.grad_accumulation_steps)])\n\n warmup_scheduler_guidance = LinearLR(self.optimizer_guidance, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)\n decay_scheduler_guidance = LinearLR(self.optimizer_guidance, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)\n self.scheduler_guidance = SequentialLR(self.optimizer_guidance, schedulers=[warmup_scheduler_guidance, decay_scheduler_guidance], milestones=[warmup_steps])\n\n train_dataloader, self.scheduler_generator, self.scheduler_guidance = self.accelerator.prepare(\n train_dataloader, self.scheduler_generator, self.scheduler_guidance\n ) # actual steps = 1 gpu steps / gpus\n start_step = self.load_checkpoint()\n global_step = start_step\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar = tqdm(\n skipped_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n initial=skipped_batch,\n total=orig_epoch_step,\n )\n else:\n progress_bar = tqdm(\n train_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n )\n\n for batch in progress_bar:\n update_generator = global_step % self.gen_update_ratio == 0\n \n with self.accelerator.accumulate(self.model):\n metrics = {}\n text_inputs = batch[\"text\"]\n mel_spec = batch[\"mel\"].permute(0, 2, 1)\n mel_lengths = batch[\"mel_lengths\"]\n \n mel_spec = mel_spec / self.scale\n \n guidance_loss_dict, guidance_log_dict = self.model(inp=mel_spec, \n text=text_inputs, \n lens=mel_lengths, \n student_steps=self.student_steps,\n update_generator=False,\n use_simulated=global_step >= self.num_simu,\n )\n\n # if self.GAN and update_generator:\n # # only add discriminator loss if GAN is enabled and generator is being updated\n # guidance_cls_loss = guidance_loss_dict[\"guidance_cls_loss\"] * (self.lambda_discriminator_loss if global_step >= self.num_GAN and update_generator else 0)\n # metrics['loss/discriminator_loss'] = guidance_loss_dict[\"guidance_cls_loss\"]\n # self.accelerator.backward(guidance_cls_loss, retain_graph=True)\n \n # if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n # metrics['grad_norm_guidance'] = self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n guidance_loss = 0\n guidance_loss += guidance_loss_dict[\"loss_fake_mean\"]\n metrics['loss/fake_score'] = guidance_loss_dict[\"loss_fake_mean\"]\n metrics[\"loss/guidance_loss\"] = guidance_loss\n\n if self.GAN and update_generator:\n # only add discriminator loss if GAN is enabled and generator is being updated\n guidance_cls_loss = guidance_loss_dict[\"guidance_cls_loss\"] * (self.lambda_discriminator_loss if global_step >= self.num_GAN and update_generator else 0)\n metrics['loss/discriminator_loss'] = guidance_loss_dict[\"guidance_cls_loss\"]\n\n guidance_loss += guidance_cls_loss\n \n self.accelerator.backward(guidance_loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n metrics['grad_norm_guidance'] = self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n # if self.guidance_norm.count < 100:\n # self.guidance_norm.update(metrics['grad_norm_guidance'])\n\n # if metrics['grad_norm_guidance'] > self.guidance_norm.mean + 5 * self.guidance_norm.std:\n # self.optimizer_generator.zero_grad()\n # self.optimizer_guidance.zero_grad()\n # print(\"Gradient explosion detected. Skipping batch.\")\n # elif self.guidance_norm.count >= 100:\n # self.guidance_norm.update(metrics['grad_norm_guidance'])\n\n\n self.optimizer_guidance.step()\n self.scheduler_guidance.step()\n self.optimizer_guidance.zero_grad()\n self.optimizer_generator.zero_grad() # zero out the generator's gradient as well\n \n if update_generator:\n generator_loss_dict, generator_log_dict = self.model(inp=mel_spec, \n text=text_inputs, \n lens=mel_lengths, \n student_steps=self.student_steps,\n update_generator=True,\n use_simulated=global_step >= self.num_ctc,\n )\n # if self.GAN:\n # gen_cls_loss = generator_loss_dict[\"gen_cls_loss\"] * (self.lambda_generator_loss if global_step >= (self.num_GAN + self.num_D) and update_generator else 0)\n # metrics[\"loss/gen_cls_loss\"] = generator_loss_dict[\"gen_cls_loss\"]\n\n # self.accelerator.backward(gen_cls_loss, retain_graph=True)\n\n # if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n # metrics['grad_norm_generator'] = self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n generator_loss = 0\n generator_loss += generator_loss_dict[\"loss_dm\"]\n if \"loss_mse\" in generator_loss_dict:\n generator_loss += generator_loss_dict[\"loss_mse\"] \n generator_loss += generator_loss_dict[\"loss_ctc\"] * (self.lambda_ctc_loss if global_step >= self.num_ctc else 0)\n generator_loss += generator_loss_dict[\"loss_sim\"] * (self.lambda_sim_loss if global_step >= self.num_sim else 0)\n generator_loss += generator_loss_dict[\"loss_kl\"] * (self.lambda_ctc_loss if global_step >= self.num_ctc else 0)\n if self.GAN:\n gen_cls_loss = generator_loss_dict[\"gen_cls_loss\"] * (self.lambda_generator_loss if global_step >= (self.num_GAN + self.num_D) and update_generator else 0)\n metrics[\"loss/gen_cls_loss\"] = generator_loss_dict[\"gen_cls_loss\"]\n generator_loss += gen_cls_loss\n\n metrics['loss/dm_loss'] = generator_loss_dict[\"loss_dm\"]\n metrics['loss/ctc_loss'] = generator_loss_dict[\"loss_ctc\"]\n\n metrics['loss/similarity_loss'] = generator_loss_dict[\"loss_sim\"]\n metrics['loss/generator_loss'] = generator_loss\n \n if \"loss_mse\" in generator_loss_dict and generator_loss_dict[\"loss_mse\"] != 0:\n metrics['loss/mse_loss'] = generator_loss_dict[\"loss_mse\"]\n if \"loss_kl\" in generator_loss_dict and generator_loss_dict[\"loss_kl\"] != 0:\n metrics['loss/kl_loss'] = generator_loss_dict[\"loss_kl\"]\n\n self.accelerator.backward(generator_loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n metrics['grad_norm_generator'] = self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n # self.generator_norm.update(metrics['grad_norm_generator'])\n \n # if metrics['grad_norm_generator'] > self.generator_norm.mean + 15 * self.generator_norm.std:\n # self.optimizer_generator.zero_grad()\n # self.optimizer_guidance.zero_grad()\n # update_generator = False\n # print(\"Gradient explosion detected. Skipping batch.\")\n\n if update_generator:\n self.optimizer_generator.step()\n self.scheduler_generator.step()\n self.optimizer_generator.zero_grad()\n self.optimizer_guidance.zero_grad() # zero out the guidance's gradient as well\n\n\n global_step += 1\n\n if self.accelerator.is_local_main_process:\n self.accelerator.log({**metrics,\n \"lr_generator\": self.scheduler_generator.get_last_lr()[0],\n \"lr_guidance\": self.scheduler_guidance.get_last_lr()[0],\n }\n , step=global_step)\n \n if global_step % self.log_step == 0 and self.accelerator.is_local_main_process and vocoder is not None:\n # log the first batch of the epoch\n with torch.no_grad():\n generator_input = generator_log_dict['generator_input'][0].unsqueeze(0).permute(0, 2, 1) * self.scale\n generator_input = vocoder.decode(generator_input.float().cpu())\n generator_input = wandb.Audio(\n generator_input.float().numpy().squeeze(),\n sample_rate=24000,\n caption=\"time: \" + str(generator_log_dict['time'][0].float().cpu().numpy())\n )\n\n generator_output = generator_log_dict['generator_output'][0].unsqueeze(0).permute(0, 2, 1) * self.scale\n generator_output = vocoder.decode(generator_output.float().cpu())\n generator_output = wandb.Audio(\n generator_output.float().numpy().squeeze(),\n sample_rate=24000,\n caption=\"time: \" + str(generator_log_dict['time'][0].float().cpu().numpy())\n )\n \n generator_cond = generator_log_dict['generator_cond'][0].unsqueeze(0).permute(0, 2, 1) * self.scale\n generator_cond = vocoder.decode(generator_cond.float().cpu())\n generator_cond = wandb.Audio(\n generator_cond.float().numpy().squeeze(),\n sample_rate=24000,\n caption=\"time: \" + str(generator_log_dict['time'][0].float().cpu().numpy())\n )\n \n ground_truth = generator_log_dict['ground_truth'][0].unsqueeze(0).permute(0, 2, 1) * self.scale\n ground_truth = vocoder.decode(ground_truth.float().cpu())\n ground_truth = wandb.Audio(\n ground_truth.float().numpy().squeeze(),\n sample_rate=24000,\n caption=\"time: \" + str(generator_log_dict['time'][0].float().cpu().numpy())\n )\n \n dmtrain_noisy_inp = generator_log_dict['dmtrain_noisy_inp'][0].unsqueeze(0).permute(0, 2, 1) * self.scale\n dmtrain_noisy_inp = vocoder.decode(dmtrain_noisy_inp.float().cpu())\n dmtrain_noisy_inp = wandb.Audio(\n dmtrain_noisy_inp.float().numpy().squeeze(),\n sample_rate=24000,\n caption=\"dmtrain_time: \" + str(generator_log_dict['dmtrain_time'][0].float().cpu().numpy())\n )\n \n dmtrain_pred_real_image = generator_log_dict['dmtrain_pred_real_image'][0].unsqueeze(0).permute(0, 2, 1) * self.scale\n dmtrain_pred_real_image = vocoder.decode(dmtrain_pred_real_image.float().cpu())\n dmtrain_pred_real_image = wandb.Audio(\n dmtrain_pred_real_image.float().numpy().squeeze(),\n sample_rate=24000,\n caption=\"dmtrain_time: \" + str(generator_log_dict['dmtrain_time'][0].float().cpu().numpy())\n )\n \n dmtrain_pred_fake_image = generator_log_dict['dmtrain_pred_fake_image'][0].unsqueeze(0).permute(0, 2, 1) * self.scale\n dmtrain_pred_fake_image = vocoder.decode(dmtrain_pred_fake_image.float().cpu())\n dmtrain_pred_fake_image = wandb.Audio(\n dmtrain_pred_fake_image.float().numpy().squeeze(),\n sample_rate=24000,\n caption=\"dmtrain_time: \" + str(generator_log_dict['dmtrain_time'][0].float().cpu().numpy())\n )\n \n \n self.accelerator.log({\"noisy_input\": generator_input, \n \"output\": generator_output,\n \"cond\": generator_cond,\n \"ground_truth\": ground_truth,\n \"dmtrain_noisy_inp\": dmtrain_noisy_inp,\n \"dmtrain_pred_real_image\": dmtrain_pred_real_image,\n \"dmtrain_pred_fake_image\": dmtrain_pred_fake_image,\n \n }, step=global_step)\n\n progress_bar.set_postfix(step=str(global_step), metrics=metrics)\n\n if global_step % (self.save_per_updates * self.grad_accumulation_steps) == 0:\n self.save_checkpoint(global_step)\n\n if global_step % self.last_per_steps == 0:\n self.save_checkpoint(global_step, last=True)\n\n self.save_checkpoint(global_step, last=True)\n\n self.accelerator.end_training()\n \n ","source_hash":"2f2618612fc8a72a725d379657533ec1e177fd919989a0850d8a287ee23d5b40","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model","uri":"program://DMOSpeech2/module/src.guidance_model#L1-L752","kind":"module","name":"src.guidance_model","path":"src/guidance_model.py","language":"python","start_line":1,"end_line":752,"context_start_line":1,"context_end_line":752,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\nfrom typing import Callable\nfrom random import random\nimport numpy as np\n\nimport torch\nfrom torch import nn\nimport torch.nn.functional as F\n\nfrom f5_tts.model import DiT\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nfrom discriminator_conformer import ConformerDiscirminator\nfrom ctcmodel import ConformerCTC\nfrom ecapa_tdnn import ECAPA_TDNN\n\nclass NoOpContext:\n def __enter__(self):\n pass\n\n def __exit__(self, *args):\n pass\n\ndef predict_flow(transformer, # flow model\n x, # noisy input\n cond, # mask (prompt mask + length mask)\n text, # text input\n time, # time step\n second_time=None,\n cfg_strength=1.0\n):\n pred = transformer(\n x=x, \n cond=cond, \n text=text, time=time, \n second_time=second_time,\n drop_audio_cond=False, \n drop_text=False\n )\n \n if cfg_strength < 1e-5:\n return pred\n \n null_pred = transformer(\n x=x, \n cond=cond, \n text=text, time=time, \n second_time=second_time,\n drop_audio_cond=True, \n drop_text=True\n )\n\n return pred + (pred - null_pred) * cfg_strength\n\ndef _kl_dist_func(x, y):\n log_probs = F.log_softmax(x, dim=2)\n target_probs = F.log_softmax(y, dim=2)\n return torch.nn.functional.kl_div(log_probs, target_probs, reduction=\"batchmean\", log_target=True)\n\n\nclass Guidance(nn.Module):\n def __init__(self, \n real_unet: DiT, # teacher flow model\n fake_unet: DiT, # student flow model\n\n use_fp16: bool = True,\n real_guidance_scale: float = 0.0, \n fake_guidance_scale: float = 0.0, \n gen_cls_loss: bool = False,\n \n sv_path_en: str = \"\",\n sv_path_zh: str = \"\",\n ctc_path: str = \"\",\n sway_coeff: float = 0.0,\n scale: float = 1.0,\n ):\n super().__init__()\n self.vocab_size = real_unet.vocab_size\n \n if ctc_path != \"\":\n model = ConformerCTC(vocab_size=real_unet.vocab_size, mel_dim=real_unet.mel_dim, num_heads=8, d_hid=512, nlayers=6)\n self.ctc_model = model.eval()\n self.ctc_model.requires_grad_(False)\n self.ctc_model.load_state_dict(torch.load(ctc_path, weights_only=True, map_location='cpu')['model_state_dict'])\n\n if sv_path_en != \"\":\n model = ECAPA_TDNN()\n self.sv_model_en = model.eval()\n self.sv_model_en.requires_grad_(False)\n self.sv_model_en.load_state_dict(torch.load(sv_path, weights_only=True, map_location='cpu')['model_state_dict'])\n\n if sv_path_zh != \"\":\n model = ECAPA_TDNN()\n self.sv_model_zh = model.eval()\n self.sv_model_zh.requires_grad_(False)\n self.sv_model_zh.load_state_dict(torch.load(sv_path_zh, weights_only=True, map_location='cpu')['model_state_dict'])\n\n self.scale = scale\n \n self.real_unet = real_unet\n self.real_unet.requires_grad_(False) # no update on the teacher model\n\n self.fake_unet = fake_unet\n self.fake_unet.requires_grad_(True) # update the student model\n \n self.real_guidance_scale = real_guidance_scale \n self.fake_guidance_scale = fake_guidance_scale\n \n assert self.fake_guidance_scale == 0, \"no guidance for fake\"\n\n self.use_fp16 = use_fp16\n\n self.gen_cls_loss = gen_cls_loss \n \n self.sway_coeff = sway_coeff\n \n if self.gen_cls_loss:\n self.cls_pred_branch = ConformerDiscirminator(\n input_dim=(self.fake_unet.depth + 1) * self.fake_unet.dim + 3 * 512, # 3 is the number of layers from the CTC model\n num_layers=3,\n channels=self.fake_unet.dim // 2,\n )\n\n self.cls_pred_branch.requires_grad_(True)\n \n self.network_context_manager = torch.autocast(device_type=\"cuda\", dtype=torch.float16) if self.use_fp16 else NoOpContext()\n\n\n from f5_tts.model.utils import get_tokenizer\n from torch.utils.data import DataLoader, Dataset, SequentialSampler\n from f5_tts.model.dataset import load_dataset \n from f5_tts.model.dataset import DynamicBatchSampler, collate_fn\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n\n self.vocab_char_map = vocab_char_map\n\n\n \n def compute_distribution_matching_loss(\n self, \n\n inp: float[\"b n d\"] | float[\"b nw\"], # mel or raw wave, ground truth latent\n text: int[\"b nt\"] | list[str], # text input\n *,\n second_time: torch.Tensor | None = None, # second time step for flow prediction\n rand_span_mask: bool[\"b n d\"] | bool[\"b nw\"] | None = None, # combined mask (prompt mask + padding mask)\n ):\n \"\"\"\n Compute DMD loss (L_DMD) between the student distribution and teacher distribution.\n Following the DMDSpeech logic:\n - Sample time t\n - Construct noisy input phi = (1 - t)*x0 + t*x1, where x0 is noise and x1 is inp\n - Predict flows with teacher (f_phi) and student (G_theta)\n - Compute gradient that aligns student distribution with teacher distribution\n\n The code is adapted from F5-TTS but conceptualized per DMD:\n L_DMD encourages p_theta to match p_data via the difference between teacher and student predictions.\n \"\"\"\n \n original_inp = inp\n \n with torch.no_grad():\n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n \n # mel is x1\n x1 = inp\n\n # x0 is gaussian noise\n x0 = torch.randn_like(x1)\n\n # time step\n time = torch.rand((batch,), dtype=dtype, device=device)\n \n # get flow\n t = time.unsqueeze(-1).unsqueeze(-1)\n # t = t + self.sway_coeff * (torch.cos(torch.pi / 2 * t) - 1 + t)\n sigma_t, alpha_t = (1 - t), t\n\n phi = (1 - t) * x0 + t * x1 # noisy x\n flow = x1 - x0 # flow target\n \n # only predict what is within the random mask span for infilling\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n \n # run at full precision as autocast and no_grad doesn't work well together \n with self.network_context_manager:\n pred_fake = predict_flow(\n self.fake_unet, \n phi, \n cond, # mask (prompt mask + length mask)\n text, # text input\n time, # time step\n second_time=second_time,\n cfg_strength=self.fake_guidance_scale\n )\n\n # pred = (x1 - x0), thus phi + (1-t) * pred = (1 - t) * x0 + t * x1 + (1 - t) * (x1 - x0) = (1 - t) * x1 + t * x1 = x1 \n pred_fake_image = phi + (1 - t) * pred_fake\n pred_fake_image[~rand_span_mask] = inp[~rand_span_mask]\n \n with self.network_context_manager:\n pred_real = predict_flow(\n self.real_unet, phi, cond, text, time, cfg_strength=self.real_guidance_scale\n )\n \n pred_real_image = phi + (1 - t) * pred_real\n pred_real_image[~rand_span_mask] = inp[~rand_span_mask]\n\n p_real = (inp - pred_real_image)\n p_fake = (inp - pred_fake_image)\n \n grad = (p_real - p_fake) / torch.abs(p_real).mean(dim=[1, 2], keepdim=True) \n grad = torch.nan_to_num(grad)\n \n # grad = grad / sigma_t # pred_fake - pred_real\n # grad = grad * (1 + sigma_t / alpha_t)\n \n # grad = grad / (1 + sigma_t / alpha_t) # noise\n # grad = grad / sigma_t # score difference\n # grad = grad * alpha_t\n # grad = grad * (sigma_t ** 2 / alpha_t)\n \n # grad = grad * (alpha_t + sigma_t ** 2 / alpha_t)\n \n # The DMD loss: MSE to move student distribution closer to teacher distribution\n # Only optimize over the masked region\n loss = 0.5 * F.mse_loss(original_inp.float(), (original_inp-grad).detach().float(), reduction=\"none\") * rand_span_mask.unsqueeze(\n -1\n )\n loss = loss.sum() / (rand_span_mask.sum() * grad.size(-1))\n \n loss_dict = {\n \"loss_dm\": loss \n }\n\n dm_log_dict = {\n \"dmtrain_time\": time.detach().float(),\n \"dmtrain_noisy_inp\": phi.detach().float(),\n \"dmtrain_pred_real_image\": pred_real_image.detach().float(),\n \"dmtrain_pred_fake_image\": pred_fake_image.detach().float(),\n \"dmtrain_grad\": grad.detach().float(),\n \"dmtrain_gradient_norm\": torch.norm(grad).item()\n }\n\n return loss_dict, dm_log_dict\n \n \n def compute_ctc_sv_loss(\n self,\n real_inp: torch.Tensor, # real data latent\n fake_inp: torch.Tensor, # student-generated data latent\n text: torch.Tensor,\n text_lens: torch.Tensor,\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n ):\n \"\"\"\n Compute CTC + SV loss for direct metric optimization, as described in DMDSpeech.\n - CTC loss reduces WER\n - SV loss improves speaker similarity\n\n Both CTC and SV models operate on latents.\n \"\"\"\n\n # compute CTC loss\n out, layer, ctc_loss = self.ctc_model(fake_inp * self.scale, text, text_lens) # lengths from rand_span_mask or known\n\n with torch.no_grad():\n real_out, real_layers, ctc_loss_test = self.ctc_model(real_inp * self.scale, text, text_lens)\n real_logits = real_out.log_softmax(dim=2)\n # emb_real = self.sv_model(real_inp * self.scale) # snippet from prompt region \n \n fake_logits = out.log_softmax(dim=2)\n kl_loss = F.kl_div(fake_logits, real_logits, reduction=\"mean\", log_target=True)\n \n # For SV:\n # Extract speaker embeddings from real (prompt) and fake:\n # emb_fake = self.sv_model(fake_inp * self.scale)\n # sv_loss = 1 - F.cosine_similarity(emb_real, emb_fake, dim=-1).mean()\n\n input_lengths = rand_span_mask.sum(axis=-1).cpu().numpy()\n prompt_lengths = real_inp.size(1) - rand_span_mask.sum(axis=-1).cpu().numpy()\n\n chunks_real = []\n chunks_fake = []\n mel_len = min([int(input_lengths.min().item() - 1), 300])\n\n for bib in range(len(input_lengths)):\n prompt_length = int(prompt_lengths[bib].item())\n mel_length = int(input_lengths[bib].item())\n mask = rand_span_mask[bib]\n mask = torch.where(mask)[0]\n\n prompt_start = mask[0].cpu().numpy()\n prompt_end = mask[-1].cpu().numpy()\n\n if prompt_end - mel_len <= prompt_start:\n random_start = np.random.randint(0, mel_length - mel_len)\n else:\n random_start = np.random.randint(prompt_start, prompt_end - mel_len)\n \n chunks_fake.append(fake_inp[bib, random_start:random_start + mel_len, :])\n chunks_real.append(real_inp[bib, :mel_len, :])\n\n chunks_real = torch.stack(chunks_real, dim=0)\n chunks_fake = torch.stack(chunks_fake, dim=0)\n\n with torch.no_grad():\n emb_real_en = self.sv_model_en(chunks_real * self.scale)\n emb_fake_en = self.sv_model_en(chunks_fake * self.scale)\n\n sv_loss_en = 1 - F.cosine_similarity(emb_real_en, emb_fake_en, dim=-1).mean()\n\n with torch.no_grad():\n emb_real_zh = self.sv_model_zh(chunks_real * self.scale)\n emb_fake_zh = self.sv_model_zh(chunks_fake * self.scale)\n\n sv_loss_zh = 1 - F.cosine_similarity(emb_real_zh, emb_fake_zh, dim=-1).mean()\n\n sv_loss = (sv_loss_en + sv_loss_zh) / 2\n\n return {\n \"loss_ctc\": ctc_loss,\n 'loss_kl': kl_loss,\n \"loss_sim\": sv_loss\n }, layer, real_layers\n\n \n \n def compute_loss_fake(\n self,\n inp: torch.Tensor, # student generator output \n text: torch.Tensor | list[str],\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n ):\n \"\"\"\n Compute flow loss for the fake flow model, which is trained to estimate the flow (score) of the student distribution.\n \n This is the same as L_diff in the paper. \n \"\"\"\n \n # Similar to distribution matching, but only train fake to predict flow directly\n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # Sample a time\n time = torch.rand((batch,), dtype=dtype, device=device)\n\n x1 = inp\n x0 = torch.randn_like(x1)\n t = time.unsqueeze(-1).unsqueeze(-1)\n \n phi = (1 - t) * x0 + t * x1\n flow = x1 - x0\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n\n with self.network_context_manager:\n pred = self.fake_unet(\n x=phi, \n cond=cond,\n text=text, \n time=time, \n second_time=second_time,\n drop_audio_cond=False, \n drop_text=False # make sure the cfg=1\n )\n\n # Compute MSE between predicted flow and actual flow, masked by rand_span_mask\n loss = F.mse_loss(pred, flow, reduction=\"none\")\n loss = loss[rand_span_mask].mean()\n \n loss_dict = {\n \"loss_fake_mean\": loss\n }\n log_dict = {\n \"faketrain_noisy_inp\": phi.detach().float(),\n \"faketrain_x1\": x1.detach().float(),\n \"faketrain_pred_flow\": pred.detach().float(),\n }\n\n return loss_dict, log_dict\n\n def compute_cls_logits(\n self,\n inp: torch.Tensor, # student generator output\n layer: torch.Tensor,\n text: torch.Tensor,\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n guidance: bool = False,\n ):\n '''\n Compute adversarial loss logits for the generator.\n \n This is used to compute L_adv in the paper.\n \n '''\n context_no_grad = torch.no_grad if guidance else NoOpContext\n\n with context_no_grad():\n # If we are not doing generator classification loss, return zeros\n if not self.gen_cls_loss:\n return torch.zeros_like(inp[..., 0]) # shape (b, n)\n\n # For classification, we need some representation:\n # We'll mimic the logic from compute_loss_fake\n \n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # Sample a time\n time = torch.rand((batch,), dtype=dtype, device=device)\n\n x1 = inp\n x0 = torch.randn_like(x1)\n t = time.unsqueeze(-1).unsqueeze(-1)\n \n phi = (1 - t) * x0 + t * x1\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n\n with self.network_context_manager:\n layers = self.fake_unet(\n x=phi, \n cond=cond,\n text=text, \n time=time, \n second_time=second_time,\n drop_audio_cond=False, \n drop_text=False, # make sure the cfg=1\n classify_mode=True\n )\n # layers = torch.stack(layers, dim=0)\n\n if guidance:\n layers = [layer.detach() for layer in layers]\n layer = layer[-3:] # only use the last 3 layers\n layer = [l.transpose(-1, -2) for l in layer]\n # layer = [F.interpolate(l, mode='nearest', scale_factor=4).transpose(-1, -2) for l in layer]\n if layer[0].size(1) < layers[0].size(1):\n layer = [F.pad(l, (0, 0, 0, layers[0].size(1) - l.size(1))) for l in layer]\n\n layers = layer + layers\n # logits: (b, 1)\n logits = self.cls_pred_branch(layers)\n\n return logits, layers\n\n\n def compute_generator_cls_loss(\n self,\n inp: torch.Tensor, # student generator output\n layer: torch.Tensor,\n real_layers: torch.Tensor,\n text: torch.Tensor,\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n mse_loss: bool = False,\n mse_inp: torch.Tensor | None = None,\n ):\n '''\n Compute the adversarial loss for the generator. \n '''\n \n # Compute classification loss for generator:\n if not self.gen_cls_loss:\n return {\"gen_cls_loss\": 0}\n\n logits, fake_layers = self.compute_cls_logits(inp, layer, text, rand_span_mask, second_time, guidance=False)\n\n loss = ((1 - logits) ** 2).mean()\n\n return {\"gen_cls_loss\": loss, \"loss_mse\": 0}\n \n def compute_guidance_cls_loss(\n self,\n fake_inp: torch.Tensor,\n text: torch.Tensor,\n rand_span_mask: torch.Tensor,\n real_data: dict,\n second_time: torch.Tensor | None = None,\n ):\n '''\n This function computes the adversarial loss for the discirminator.\n\n The discriminator is trained to classify the generator output as real or fake.\n '''\n\n with torch.no_grad():\n # get layers from CTC model\n _, layer = self.ctc_model(fake_inp * self.scale)\n\n logits_fake, _ = self.compute_cls_logits(fake_inp.detach(), layer, text, rand_span_mask, second_time, guidance=True)\n loss_fake = (logits_fake**2).mean()\n\n real_inp = real_data[\"inp\"]\n\n with torch.no_grad():\n # get layers from CTC model\n _, layer = self.ctc_model(real_inp * self.scale)\n \n logits_real, _ = self.compute_cls_logits(real_inp.detach(), layer, text, rand_span_mask, second_time, guidance=True)\n loss_real = ((1 - logits_real)**2).mean()\n\n classification_loss = loss_real + loss_fake\n\n loss_dict = {\n \"guidance_cls_loss\": classification_loss\n }\n log_dict = {\n \"pred_realism_on_real\": loss_real.detach().item(),\n \"pred_realism_on_fake\": loss_fake.detach().item()\n }\n\n return loss_dict, log_dict\n\n def generator_forward(\n self,\n inp: torch.Tensor,\n text: torch.Tensor,\n text_lens: torch.Tensor,\n text_normalized: to\n# ... truncated ...","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.NoOpContext","uri":"program://DMOSpeech2/class/src.guidance_model.NoOpContext#L34-L39","kind":"class","name":"NoOpContext","path":"src/guidance_model.py","language":"python","start_line":34,"end_line":39,"context_start_line":14,"context_end_line":59,"code":"\nimport torch\nfrom torch import nn\nimport torch.nn.functional as F\n\nfrom f5_tts.model import DiT\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nfrom discriminator_conformer import ConformerDiscirminator\nfrom ctcmodel import ConformerCTC\nfrom ecapa_tdnn import ECAPA_TDNN\n\nclass NoOpContext:\n def __enter__(self):\n pass\n\n def __exit__(self, *args):\n pass\n\ndef predict_flow(transformer, # flow model\n x, # noisy input\n cond, # mask (prompt mask + length mask)\n text, # text input\n time, # time step\n second_time=None,\n cfg_strength=1.0\n):\n pred = transformer(\n x=x, \n cond=cond, \n text=text, time=time, \n second_time=second_time,\n drop_audio_cond=False, \n drop_text=False\n )\n \n if cfg_strength < 1e-5:\n return pred","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.predict_flow","uri":"program://DMOSpeech2/function/src.guidance_model.predict_flow#L41-L70","kind":"function","name":"predict_flow","path":"src/guidance_model.py","language":"python","start_line":41,"end_line":70,"context_start_line":21,"context_end_line":90,"code":"from f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nfrom discriminator_conformer import ConformerDiscirminator\nfrom ctcmodel import ConformerCTC\nfrom ecapa_tdnn import ECAPA_TDNN\n\nclass NoOpContext:\n def __enter__(self):\n pass\n\n def __exit__(self, *args):\n pass\n\ndef predict_flow(transformer, # flow model\n x, # noisy input\n cond, # mask (prompt mask + length mask)\n text, # text input\n time, # time step\n second_time=None,\n cfg_strength=1.0\n):\n pred = transformer(\n x=x, \n cond=cond, \n text=text, time=time, \n second_time=second_time,\n drop_audio_cond=False, \n drop_text=False\n )\n \n if cfg_strength < 1e-5:\n return pred\n \n null_pred = transformer(\n x=x, \n cond=cond, \n text=text, time=time, \n second_time=second_time,\n drop_audio_cond=True, \n drop_text=True\n )\n\n return pred + (pred - null_pred) * cfg_strength\n\ndef _kl_dist_func(x, y):\n log_probs = F.log_softmax(x, dim=2)\n target_probs = F.log_softmax(y, dim=2)\n return torch.nn.functional.kl_div(log_probs, target_probs, reduction=\"batchmean\", log_target=True)\n\n\nclass Guidance(nn.Module):\n def __init__(self, \n real_unet: DiT, # teacher flow model\n fake_unet: DiT, # student flow model\n\n use_fp16: bool = True,\n real_guidance_scale: float = 0.0, \n fake_guidance_scale: float = 0.0, \n gen_cls_loss: bool = False,\n \n sv_path_en: str = \"\",\n sv_path_zh: str = \"\",\n ctc_path: str = \"\",","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model._kl_dist_func","uri":"program://DMOSpeech2/function/src.guidance_model._kl_dist_func#L72-L75","kind":"function","name":"_kl_dist_func","path":"src/guidance_model.py","language":"python","start_line":72,"end_line":75,"context_start_line":52,"context_end_line":95,"code":" text=text, time=time, \n second_time=second_time,\n drop_audio_cond=False, \n drop_text=False\n )\n \n if cfg_strength < 1e-5:\n return pred\n \n null_pred = transformer(\n x=x, \n cond=cond, \n text=text, time=time, \n second_time=second_time,\n drop_audio_cond=True, \n drop_text=True\n )\n\n return pred + (pred - null_pred) * cfg_strength\n\ndef _kl_dist_func(x, y):\n log_probs = F.log_softmax(x, dim=2)\n target_probs = F.log_softmax(y, dim=2)\n return torch.nn.functional.kl_div(log_probs, target_probs, reduction=\"batchmean\", log_target=True)\n\n\nclass Guidance(nn.Module):\n def __init__(self, \n real_unet: DiT, # teacher flow model\n fake_unet: DiT, # student flow model\n\n use_fp16: bool = True,\n real_guidance_scale: float = 0.0, \n fake_guidance_scale: float = 0.0, \n gen_cls_loss: bool = False,\n \n sv_path_en: str = \"\",\n sv_path_zh: str = \"\",\n ctc_path: str = \"\",\n sway_coeff: float = 0.0,\n scale: float = 1.0,\n ):\n super().__init__()\n self.vocab_size = real_unet.vocab_size","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.Guidance","uri":"program://DMOSpeech2/class/src.guidance_model.Guidance#L78-L665","kind":"class","name":"Guidance","path":"src/guidance_model.py","language":"python","start_line":78,"end_line":665,"context_start_line":58,"context_end_line":685,"code":" if cfg_strength < 1e-5:\n return pred\n \n null_pred = transformer(\n x=x, \n cond=cond, \n text=text, time=time, \n second_time=second_time,\n drop_audio_cond=True, \n drop_text=True\n )\n\n return pred + (pred - null_pred) * cfg_strength\n\ndef _kl_dist_func(x, y):\n log_probs = F.log_softmax(x, dim=2)\n target_probs = F.log_softmax(y, dim=2)\n return torch.nn.functional.kl_div(log_probs, target_probs, reduction=\"batchmean\", log_target=True)\n\n\nclass Guidance(nn.Module):\n def __init__(self, \n real_unet: DiT, # teacher flow model\n fake_unet: DiT, # student flow model\n\n use_fp16: bool = True,\n real_guidance_scale: float = 0.0, \n fake_guidance_scale: float = 0.0, \n gen_cls_loss: bool = False,\n \n sv_path_en: str = \"\",\n sv_path_zh: str = \"\",\n ctc_path: str = \"\",\n sway_coeff: float = 0.0,\n scale: float = 1.0,\n ):\n super().__init__()\n self.vocab_size = real_unet.vocab_size\n \n if ctc_path != \"\":\n model = ConformerCTC(vocab_size=real_unet.vocab_size, mel_dim=real_unet.mel_dim, num_heads=8, d_hid=512, nlayers=6)\n self.ctc_model = model.eval()\n self.ctc_model.requires_grad_(False)\n self.ctc_model.load_state_dict(torch.load(ctc_path, weights_only=True, map_location='cpu')['model_state_dict'])\n\n if sv_path_en != \"\":\n model = ECAPA_TDNN()\n self.sv_model_en = model.eval()\n self.sv_model_en.requires_grad_(False)\n self.sv_model_en.load_state_dict(torch.load(sv_path, weights_only=True, map_location='cpu')['model_state_dict'])\n\n if sv_path_zh != \"\":\n model = ECAPA_TDNN()\n self.sv_model_zh = model.eval()\n self.sv_model_zh.requires_grad_(False)\n self.sv_model_zh.load_state_dict(torch.load(sv_path_zh, weights_only=True, map_location='cpu')['model_state_dict'])\n\n self.scale = scale\n \n self.real_unet = real_unet\n self.real_unet.requires_grad_(False) # no update on the teacher model\n\n self.fake_unet = fake_unet\n self.fake_unet.requires_grad_(True) # update the student model\n \n self.real_guidance_scale = real_guidance_scale \n self.fake_guidance_scale = fake_guidance_scale\n \n assert self.fake_guidance_scale == 0, \"no guidance for fake\"\n\n self.use_fp16 = use_fp16\n\n self.gen_cls_loss = gen_cls_loss \n \n self.sway_coeff = sway_coeff\n \n if self.gen_cls_loss:\n self.cls_pred_branch = ConformerDiscirminator(\n input_dim=(self.fake_unet.depth + 1) * self.fake_unet.dim + 3 * 512, # 3 is the number of layers from the CTC model\n num_layers=3,\n channels=self.fake_unet.dim // 2,\n )\n\n self.cls_pred_branch.requires_grad_(True)\n \n self.network_context_manager = torch.autocast(device_type=\"cuda\", dtype=torch.float16) if self.use_fp16 else NoOpContext()\n\n\n from f5_tts.model.utils import get_tokenizer\n from torch.utils.data import DataLoader, Dataset, SequentialSampler\n from f5_tts.model.dataset import load_dataset \n from f5_tts.model.dataset import DynamicBatchSampler, collate_fn\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n\n self.vocab_char_map = vocab_char_map\n\n\n \n def compute_distribution_matching_loss(\n self, \n\n inp: float[\"b n d\"] | float[\"b nw\"], # mel or raw wave, ground truth latent\n text: int[\"b nt\"] | list[str], # text input\n *,\n second_time: torch.Tensor | None = None, # second time step for flow prediction\n rand_span_mask: bool[\"b n d\"] | bool[\"b nw\"] | None = None, # combined mask (prompt mask + padding mask)\n ):\n \"\"\"\n Compute DMD loss (L_DMD) between the student distribution and teacher distribution.\n Following the DMDSpeech logic:\n - Sample time t\n - Construct noisy input phi = (1 - t)*x0 + t*x1, where x0 is noise and x1 is inp\n - Predict flows with teacher (f_phi) and student (G_theta)\n - Compute gradient that aligns student distribution with teacher distribution\n\n The code is adapted from F5-TTS but conceptualized per DMD:\n L_DMD encourages p_theta to match p_data via the difference between teacher and student predictions.\n \"\"\"\n \n original_inp = inp\n \n with torch.no_grad():\n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n \n # mel is x1\n x1 = inp\n\n # x0 is gaussian noise\n x0 = torch.randn_like(x1)\n\n # time step\n time = torch.rand((batch,), dtype=dtype, device=device)\n \n # get flow\n t = time.unsqueeze(-1).unsqueeze(-1)\n # t = t + self.sway_coeff * (torch.cos(torch.pi / 2 * t) - 1 + t)\n sigma_t, alpha_t = (1 - t), t\n\n phi = (1 - t) * x0 + t * x1 # noisy x\n flow = x1 - x0 # flow target\n \n # only predict what is within the random mask span for infilling\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n \n # run at full precision as autocast and no_grad doesn't work well together \n with self.network_context_manager:\n pred_fake = predict_flow(\n self.fake_unet, \n phi, \n cond, # mask (prompt mask + length mask)\n text, # text input\n time, # time step\n second_time=second_time,\n cfg_strength=self.fake_guidance_scale\n )\n\n # pred = (x1 - x0), thus phi + (1-t) * pred = (1 - t) * x0 + t * x1 + (1 - t) * (x1 - x0) = (1 - t) * x1 + t * x1 = x1 \n pred_fake_image = phi + (1 - t) * pred_fake\n pred_fake_image[~rand_span_mask] = inp[~rand_span_mask]\n \n with self.network_context_manager:\n pred_real = predict_flow(\n self.real_unet, phi, cond, text, time, cfg_strength=self.real_guidance_scale\n )\n \n pred_real_image = phi + (1 - t) * pred_real\n pred_real_image[~rand_span_mask] = inp[~rand_span_mask]\n\n p_real = (inp - pred_real_image)\n p_fake = (inp - pred_fake_image)\n \n grad = (p_real - p_fake) / torch.abs(p_real).mean(dim=[1, 2], keepdim=True) \n grad = torch.nan_to_num(grad)\n \n # grad = grad / sigma_t # pred_fake - pred_real\n # grad = grad * (1 + sigma_t / alpha_t)\n \n # grad = grad / (1 + sigma_t / alpha_t) # noise\n # grad = grad / sigma_t # score difference\n # grad = grad * alpha_t\n # grad = grad * (sigma_t ** 2 / alpha_t)\n \n # grad = grad * (alpha_t + sigma_t ** 2 / alpha_t)\n \n # The DMD loss: MSE to move student distribution closer to teacher distribution\n # Only optimize over the masked region\n loss = 0.5 * F.mse_loss(original_inp.float(), (original_inp-grad).detach().float(), reduction=\"none\") * rand_span_mask.unsqueeze(\n -1\n )\n loss = loss.sum() / (rand_span_mask.sum() * grad.size(-1))\n \n loss_dict = {\n \"loss_dm\": loss \n }\n\n dm_log_dict = {\n \"dmtrain_time\": time.detach().float(),\n \"dmtrain_noisy_inp\": phi.detach().float(),\n \"dmtrain_pred_real_image\": pred_real_image.detach().float(),\n \"dmtrain_pred_fake_image\": pred_fake_image.detach().float(),\n \"dmtrain_grad\": grad.detach().float(),\n \"dmtrain_gradient_norm\": torch.norm(grad).item()\n }\n\n return loss_dict, dm_log_dict\n \n \n def compute_ctc_sv_loss(\n self,\n real_inp: torch.Tensor, # real data latent\n fake_inp: torch.Tensor, # student-generated data latent\n text: torch.Tensor,\n text_lens: torch.Tensor,\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n ):\n \"\"\"\n Compute CTC + SV loss for direct metric optimization, as described in DMDSpeech.\n - CTC loss reduces WER\n - SV loss improves speaker similarity\n\n Both CTC and SV models operate on latents.\n \"\"\"\n\n # compute CTC loss\n out, layer, ctc_loss = self.ctc_model(fake_inp * self.scale, text, text_lens) # lengths from rand_span_mask or known\n\n with torch.no_grad():\n real_out, real_layers, ctc_loss_test = self.ctc_model(real_inp * self.scale, text, text_lens)\n real_logits = real_out.log_softmax(dim=2)\n # emb_real = self.sv_model(real_inp * self.scale) # snippet from prompt region \n \n fake_logits = out.log_softmax(dim=2)\n kl_loss = F.kl_div(fake_logits, real_logits, reduction=\"mean\", log_target=True)\n \n # For SV:\n # Extract speaker embeddings from real (prompt) and fake:\n # emb_fake = self.sv_model(fake_inp * self.scale)\n # sv_loss = 1 - F.cosine_similarity(emb_real, emb_fake, dim=-1).mean()\n\n input_lengths = rand_span_mask.sum(axis=-1).cpu().numpy()\n prompt_lengths = real_inp.size(1) - rand_span_mask.sum(axis=-1).cpu().numpy()\n\n chunks_real = []\n chunks_fake = []\n mel_len = min([int(input_lengths.min().item() - 1), 300])\n\n for bib in range(len(input_lengths)):\n prompt_length = int(prompt_lengths[bib].item())\n mel_length = int(input_lengths[bib].item())\n mask = rand_span_mask[bib]\n mask = torch.where(mask)[0]\n\n prompt_start = mask[0].cpu().numpy()\n prompt_end = mask[-1].cpu().numpy()\n\n if prompt_end - mel_len <= prompt_start:\n random_start = np.random.randint(0, mel_length - mel_len)\n else:\n random_start = np.random.randint(prompt_start, prompt_end - mel_len)\n \n chunks_fake.append(fake_inp[bib, random_start:random_start + mel_len, :])\n chunks_real.append(real_inp[bib, :mel_len, :])\n\n chunks_real = torch.stack(chunks_real, dim=0)\n chunks_fake = torch.stack(chunks_fake, dim=0)\n\n with torch.no_grad():\n emb_real_en = self.sv_model_en(chunks_real * self.scale)\n emb_fake_en = self.sv_model_en(chunks_fake * self.scale)\n\n sv_loss_en = 1 - F.cosine_similarity(emb_real_en, emb_fake_en, dim=-1).mean()\n\n with torch.no_grad():\n emb_real_zh = self.sv_model_zh(chunks_real * self.scale)\n emb_fake_zh = self.sv_model_zh(chunks_fake * self.scale)\n\n sv_loss_zh = 1 - F.cosine_similarity(emb_real_zh, emb_fake_zh, dim=-1).mean()\n\n sv_loss = (sv_loss_en + sv_loss_zh) / 2\n\n return {\n \"loss_ctc\": ctc_loss,\n 'loss_kl': kl_loss,\n \"loss_sim\": sv_loss\n }, layer, real_layers\n\n \n \n def compute_loss_fake(\n self,\n inp: torch.Tensor, # student generator output \n text: torch.Tensor | list[str],\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n ):\n \"\"\"\n Compute flow loss for the fake flow model, which is trained to estimate the flow (score) of the student distribution.\n \n This is the same as L_diff in the paper. \n \"\"\"\n \n # Similar to distribution matching, but only train fake to predict flow directly\n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # Sample a time\n time = torch.rand((batch,), dtype=dtype, device=device)\n\n x1 = inp\n x0 = torch.randn_like(x1)\n t = time.unsqueeze(-1).unsqueeze(-1)\n \n phi = (1 - t) * x0 + t * x1\n flow = x1 - x0\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n\n with self.network_context_manager:\n pred = self.fake_unet(\n x=phi, \n cond=cond,\n text=text, \n time=time, \n second_time=second_time,\n drop_audio_cond=False, \n drop_text=False # make sure the cfg=1\n )\n\n # Compute MSE between predicted flow and actual flow, masked by rand_span_mask\n loss = F.mse_loss(pred, flow, reduction=\"none\")\n loss = loss[rand_span_mask].mean()\n \n loss_dict = {\n \"loss_fake_mean\": loss\n }\n log_dict = {\n \"faketrain_noisy_inp\": phi.detach().float(),\n \"faketrain_x1\": x1.detach().float(),\n \"faketrain_pred_flow\": pred.detach().float(),\n }\n\n return loss_dict, log_dict\n\n def compute_cls_logits(\n self,\n inp: torch.Tensor, # student generator output\n layer: torch.Tensor,\n text: torch.Tensor,\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n guidance: bool = False,\n ):\n '''\n Compute adversarial loss logits for the generator.\n \n This is used to compute L_adv in the paper.\n \n '''\n context_no_grad = torch.no_grad if guidance else NoOpContext\n\n with context_no_grad():\n # If we are not doing generator classification loss, return zeros\n if not self.gen_cls_loss:\n return torch.zeros_like(inp[..., 0]) # shape (b, n)\n\n # For classification, we need some representation:\n # We'll mimic the logic from compute_loss_fake\n \n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # Sample a time\n time = torch.rand((batch,), dtype=dtype, device=device)\n\n x1 = inp\n x0 = torch.randn_like(x1)\n t = time.unsqueeze(-1).unsqueeze(-1)\n \n phi = (1 - t) * x0 + t * x1\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n\n with self.network_context_manager:\n layers = self.fake_unet(\n x=phi, \n cond=cond,\n text=text, \n time=time, \n second_time=second_time,\n drop_audio_cond=False, \n drop_text=False, # make sure the cfg=1\n classify_mode=True\n )\n # layers = torch.stack(layers, dim=0)\n\n if guidance:\n layers = [layer.detach() for layer in layers]\n layer = layer[-3:] # only use the last 3 layers\n layer = [l.transpose(-1, -2) for l in layer]\n # layer = [F.interpolate(l, mode='nearest', scale_factor=4).transpose(-1, -2) for l in layer]\n if layer[0].size(1) < layers[0].size(1):\n layer = [F.pad(l, (0, 0, 0, layers[0].size(1) - l.size(1))) for l in layer]\n\n layers = layer + layers\n # logits: (b, 1)\n logits = self.cls_pred_branch(layers)\n\n return logits, layers\n\n\n def compute_generator_cls_loss(\n self,\n inp: torch.Tensor, # student generator output\n layer: torch.Tensor,\n real_layers: torch.Tensor,\n text: torch.Tensor,\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n mse_loss: bool = False,\n mse_inp: torch.Tensor | None = None,\n ):\n '''\n Compute the adversarial loss for the generator. \n '''\n \n # Compute classification loss for generator:\n if not self.gen_cls_loss:\n return {\"gen_cls_loss\": 0}\n\n logits, fake_layers = self.compute_cls_logits(inp, layer, text, rand_span_mask, second_time, guidance=False)\n\n loss = ((1 - logits) ** 2).mean()\n\n return {\"gen_cls_loss\": loss, \"loss_mse\": 0}\n \n def compute_guidance_cls_loss(\n self,\n fake_inp: torch.Tensor,\n text: torch.Tensor,\n rand_span_mask: torch.Tensor,\n real_data: dict,\n second_time: torch.Tensor | None = None,\n ):\n '''\n This function computes the adversarial loss for the discirminator.\n\n The discriminator is trained to classify the generator output as real or fake.\n '''\n\n with torch.no_grad():\n # get layers from CTC model\n _, layer = self.ctc_model(fake_inp * self.scale)\n\n logits_fake, _ = self.compute_cls_logits(fake_inp.detach(), layer, text, rand_span_mask, second_time, guidance=True)\n loss_fake = (logits_fake**2).mean()\n\n real_inp = real_data[\"inp\"]\n\n with torch.no_grad():\n # get layers from CTC model\n _, layer = self.ctc_model(real_inp * self.scale)\n \n logits_real, _ = self.compute_cls_logits(real_inp.detach(), layer, text, rand_span_mask, second_time, guidance=True)\n loss_real = ((1 - logits_real)**2).mean()\n\n classification_loss = loss_real + loss_fake\n\n loss_dict = {\n \"guidance_cls_loss\": classification_loss\n }\n log_dict = {\n \"pred_realism_on_real\": loss_real.detach().item(),\n \"pred_realism_on_fake\": loss_fake.detach().item()\n }\n\n return loss_dict, log_dict\n\n def generator_forward(\n self,\n inp: torch.Tensor,\n text: torch.Tensor,\n text_lens: torch.Tensor,\n text_normalized: torch.Tensor,\n text_normalized_lens: torch.Tensor,\n rand_span_mask: torch.Tensor,\n real_data: dict | None = None, # ground truth data (primarily prompt) to compute SV loss\n second_time: torch.Tensor | None = None,\n mse_loss: bool = False,\n ):\n '''\n Forward pass for the generator.\n \n This function computes the loss for the generator, which includes:\n - Distribution matching loss (L_DMD)\n - Adversarial generator loss (L_adv(G; D))\n - CTC/SV loss (L_ctc + L_sv)\n '''\n \n # 1. Compute DM loss\n dm_loss_dict, dm_log_dict = self.compute_distribution_matching_loss(inp, text, rand_span_mask=rand_span_mask, second_time=second_time)\n\n ctc_sv_loss_dict = {}\n cls_loss_dict = {}\n\n # 2. Compute optional CTC/SV loss if real_data provided\n if real_data is not None:\n real_inp = real_data[\"inp\"]\n ctc_sv_loss_dict, layer, real_layers = self.compute_ctc_sv_loss(real_inp, inp, text_normalized, text_normalized_lens, rand_span_mask, second_time=second_time)\n\n # 3. Compute optional classification los\n# ... truncated ...","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.__enter__","uri":"program://DMOSpeech2/function/src.guidance_model.__enter__#L35-L36","kind":"function","name":"__enter__","path":"src/guidance_model.py","language":"python","start_line":35,"end_line":36,"context_start_line":15,"context_end_line":56,"code":"import torch\nfrom torch import nn\nimport torch.nn.functional as F\n\nfrom f5_tts.model import DiT\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nfrom discriminator_conformer import ConformerDiscirminator\nfrom ctcmodel import ConformerCTC\nfrom ecapa_tdnn import ECAPA_TDNN\n\nclass NoOpContext:\n def __enter__(self):\n pass\n\n def __exit__(self, *args):\n pass\n\ndef predict_flow(transformer, # flow model\n x, # noisy input\n cond, # mask (prompt mask + length mask)\n text, # text input\n time, # time step\n second_time=None,\n cfg_strength=1.0\n):\n pred = transformer(\n x=x, \n cond=cond, \n text=text, time=time, \n second_time=second_time,\n drop_audio_cond=False, \n drop_text=False\n )","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.__exit__","uri":"program://DMOSpeech2/function/src.guidance_model.__exit__#L38-L39","kind":"function","name":"__exit__","path":"src/guidance_model.py","language":"python","start_line":38,"end_line":39,"context_start_line":18,"context_end_line":59,"code":"\nfrom f5_tts.model import DiT\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nfrom discriminator_conformer import ConformerDiscirminator\nfrom ctcmodel import ConformerCTC\nfrom ecapa_tdnn import ECAPA_TDNN\n\nclass NoOpContext:\n def __enter__(self):\n pass\n\n def __exit__(self, *args):\n pass\n\ndef predict_flow(transformer, # flow model\n x, # noisy input\n cond, # mask (prompt mask + length mask)\n text, # text input\n time, # time step\n second_time=None,\n cfg_strength=1.0\n):\n pred = transformer(\n x=x, \n cond=cond, \n text=text, time=time, \n second_time=second_time,\n drop_audio_cond=False, \n drop_text=False\n )\n \n if cfg_strength < 1e-5:\n return pred","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.__init__","uri":"program://DMOSpeech2/function/src.guidance_model.__init__#L79-L162","kind":"function","name":"__init__","path":"src/guidance_model.py","language":"python","start_line":79,"end_line":162,"context_start_line":59,"context_end_line":182,"code":" return pred\n \n null_pred = transformer(\n x=x, \n cond=cond, \n text=text, time=time, \n second_time=second_time,\n drop_audio_cond=True, \n drop_text=True\n )\n\n return pred + (pred - null_pred) * cfg_strength\n\ndef _kl_dist_func(x, y):\n log_probs = F.log_softmax(x, dim=2)\n target_probs = F.log_softmax(y, dim=2)\n return torch.nn.functional.kl_div(log_probs, target_probs, reduction=\"batchmean\", log_target=True)\n\n\nclass Guidance(nn.Module):\n def __init__(self, \n real_unet: DiT, # teacher flow model\n fake_unet: DiT, # student flow model\n\n use_fp16: bool = True,\n real_guidance_scale: float = 0.0, \n fake_guidance_scale: float = 0.0, \n gen_cls_loss: bool = False,\n \n sv_path_en: str = \"\",\n sv_path_zh: str = \"\",\n ctc_path: str = \"\",\n sway_coeff: float = 0.0,\n scale: float = 1.0,\n ):\n super().__init__()\n self.vocab_size = real_unet.vocab_size\n \n if ctc_path != \"\":\n model = ConformerCTC(vocab_size=real_unet.vocab_size, mel_dim=real_unet.mel_dim, num_heads=8, d_hid=512, nlayers=6)\n self.ctc_model = model.eval()\n self.ctc_model.requires_grad_(False)\n self.ctc_model.load_state_dict(torch.load(ctc_path, weights_only=True, map_location='cpu')['model_state_dict'])\n\n if sv_path_en != \"\":\n model = ECAPA_TDNN()\n self.sv_model_en = model.eval()\n self.sv_model_en.requires_grad_(False)\n self.sv_model_en.load_state_dict(torch.load(sv_path, weights_only=True, map_location='cpu')['model_state_dict'])\n\n if sv_path_zh != \"\":\n model = ECAPA_TDNN()\n self.sv_model_zh = model.eval()\n self.sv_model_zh.requires_grad_(False)\n self.sv_model_zh.load_state_dict(torch.load(sv_path_zh, weights_only=True, map_location='cpu')['model_state_dict'])\n\n self.scale = scale\n \n self.real_unet = real_unet\n self.real_unet.requires_grad_(False) # no update on the teacher model\n\n self.fake_unet = fake_unet\n self.fake_unet.requires_grad_(True) # update the student model\n \n self.real_guidance_scale = real_guidance_scale \n self.fake_guidance_scale = fake_guidance_scale\n \n assert self.fake_guidance_scale == 0, \"no guidance for fake\"\n\n self.use_fp16 = use_fp16\n\n self.gen_cls_loss = gen_cls_loss \n \n self.sway_coeff = sway_coeff\n \n if self.gen_cls_loss:\n self.cls_pred_branch = ConformerDiscirminator(\n input_dim=(self.fake_unet.depth + 1) * self.fake_unet.dim + 3 * 512, # 3 is the number of layers from the CTC model\n num_layers=3,\n channels=self.fake_unet.dim // 2,\n )\n\n self.cls_pred_branch.requires_grad_(True)\n \n self.network_context_manager = torch.autocast(device_type=\"cuda\", dtype=torch.float16) if self.use_fp16 else NoOpContext()\n\n\n from f5_tts.model.utils import get_tokenizer\n from torch.utils.data import DataLoader, Dataset, SequentialSampler\n from f5_tts.model.dataset import load_dataset \n from f5_tts.model.dataset import DynamicBatchSampler, collate_fn\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n\n self.vocab_char_map = vocab_char_map\n\n\n \n def compute_distribution_matching_loss(\n self, \n\n inp: float[\"b n d\"] | float[\"b nw\"], # mel or raw wave, ground truth latent\n text: int[\"b nt\"] | list[str], # text input\n *,\n second_time: torch.Tensor | None = None, # second time step for flow prediction\n rand_span_mask: bool[\"b n d\"] | bool[\"b nw\"] | None = None, # combined mask (prompt mask + padding mask)\n ):\n \"\"\"\n Compute DMD loss (L_DMD) between the student distribution and teacher distribution.\n Following the DMDSpeech logic:\n - Sample time t\n - Construct noisy input phi = (1 - t)*x0 + t*x1, where x0 is noise and x1 is inp\n - Predict flows with teacher (f_phi) and student (G_theta)\n - Compute gradient that aligns student distribution with teacher distribution\n","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.compute_distribution_matching_loss","uri":"program://DMOSpeech2/function/src.guidance_model.compute_distribution_matching_loss#L166-L272","kind":"function","name":"compute_distribution_matching_loss","path":"src/guidance_model.py","language":"python","start_line":166,"end_line":272,"context_start_line":146,"context_end_line":292,"code":" from f5_tts.model.utils import get_tokenizer\n from torch.utils.data import DataLoader, Dataset, SequentialSampler\n from f5_tts.model.dataset import load_dataset \n from f5_tts.model.dataset import DynamicBatchSampler, collate_fn\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n\n self.vocab_char_map = vocab_char_map\n\n\n \n def compute_distribution_matching_loss(\n self, \n\n inp: float[\"b n d\"] | float[\"b nw\"], # mel or raw wave, ground truth latent\n text: int[\"b nt\"] | list[str], # text input\n *,\n second_time: torch.Tensor | None = None, # second time step for flow prediction\n rand_span_mask: bool[\"b n d\"] | bool[\"b nw\"] | None = None, # combined mask (prompt mask + padding mask)\n ):\n \"\"\"\n Compute DMD loss (L_DMD) between the student distribution and teacher distribution.\n Following the DMDSpeech logic:\n - Sample time t\n - Construct noisy input phi = (1 - t)*x0 + t*x1, where x0 is noise and x1 is inp\n - Predict flows with teacher (f_phi) and student (G_theta)\n - Compute gradient that aligns student distribution with teacher distribution\n\n The code is adapted from F5-TTS but conceptualized per DMD:\n L_DMD encourages p_theta to match p_data via the difference between teacher and student predictions.\n \"\"\"\n \n original_inp = inp\n \n with torch.no_grad():\n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n \n # mel is x1\n x1 = inp\n\n # x0 is gaussian noise\n x0 = torch.randn_like(x1)\n\n # time step\n time = torch.rand((batch,), dtype=dtype, device=device)\n \n # get flow\n t = time.unsqueeze(-1).unsqueeze(-1)\n # t = t + self.sway_coeff * (torch.cos(torch.pi / 2 * t) - 1 + t)\n sigma_t, alpha_t = (1 - t), t\n\n phi = (1 - t) * x0 + t * x1 # noisy x\n flow = x1 - x0 # flow target\n \n # only predict what is within the random mask span for infilling\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n \n # run at full precision as autocast and no_grad doesn't work well together \n with self.network_context_manager:\n pred_fake = predict_flow(\n self.fake_unet, \n phi, \n cond, # mask (prompt mask + length mask)\n text, # text input\n time, # time step\n second_time=second_time,\n cfg_strength=self.fake_guidance_scale\n )\n\n # pred = (x1 - x0), thus phi + (1-t) * pred = (1 - t) * x0 + t * x1 + (1 - t) * (x1 - x0) = (1 - t) * x1 + t * x1 = x1 \n pred_fake_image = phi + (1 - t) * pred_fake\n pred_fake_image[~rand_span_mask] = inp[~rand_span_mask]\n \n with self.network_context_manager:\n pred_real = predict_flow(\n self.real_unet, phi, cond, text, time, cfg_strength=self.real_guidance_scale\n )\n \n pred_real_image = phi + (1 - t) * pred_real\n pred_real_image[~rand_span_mask] = inp[~rand_span_mask]\n\n p_real = (inp - pred_real_image)\n p_fake = (inp - pred_fake_image)\n \n grad = (p_real - p_fake) / torch.abs(p_real).mean(dim=[1, 2], keepdim=True) \n grad = torch.nan_to_num(grad)\n \n # grad = grad / sigma_t # pred_fake - pred_real\n # grad = grad * (1 + sigma_t / alpha_t)\n \n # grad = grad / (1 + sigma_t / alpha_t) # noise\n # grad = grad / sigma_t # score difference\n # grad = grad * alpha_t\n # grad = grad * (sigma_t ** 2 / alpha_t)\n \n # grad = grad * (alpha_t + sigma_t ** 2 / alpha_t)\n \n # The DMD loss: MSE to move student distribution closer to teacher distribution\n # Only optimize over the masked region\n loss = 0.5 * F.mse_loss(original_inp.float(), (original_inp-grad).detach().float(), reduction=\"none\") * rand_span_mask.unsqueeze(\n -1\n )\n loss = loss.sum() / (rand_span_mask.sum() * grad.size(-1))\n \n loss_dict = {\n \"loss_dm\": loss \n }\n\n dm_log_dict = {\n \"dmtrain_time\": time.detach().float(),\n \"dmtrain_noisy_inp\": phi.detach().float(),\n \"dmtrain_pred_real_image\": pred_real_image.detach().float(),\n \"dmtrain_pred_fake_image\": pred_fake_image.detach().float(),\n \"dmtrain_grad\": grad.detach().float(),\n \"dmtrain_gradient_norm\": torch.norm(grad).item()\n }\n\n return loss_dict, dm_log_dict\n \n \n def compute_ctc_sv_loss(\n self,\n real_inp: torch.Tensor, # real data latent\n fake_inp: torch.Tensor, # student-generated data latent\n text: torch.Tensor,\n text_lens: torch.Tensor,\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n ):\n \"\"\"\n Compute CTC + SV loss for direct metric optimization, as described in DMDSpeech.\n - CTC loss reduces WER\n - SV loss improves speaker similarity\n\n Both CTC and SV models operate on latents.\n \"\"\"\n\n # compute CTC loss","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.compute_ctc_sv_loss","uri":"program://DMOSpeech2/function/src.guidance_model.compute_ctc_sv_loss#L275-L353","kind":"function","name":"compute_ctc_sv_loss","path":"src/guidance_model.py","language":"python","start_line":275,"end_line":353,"context_start_line":255,"context_end_line":373,"code":" -1\n )\n loss = loss.sum() / (rand_span_mask.sum() * grad.size(-1))\n \n loss_dict = {\n \"loss_dm\": loss \n }\n\n dm_log_dict = {\n \"dmtrain_time\": time.detach().float(),\n \"dmtrain_noisy_inp\": phi.detach().float(),\n \"dmtrain_pred_real_image\": pred_real_image.detach().float(),\n \"dmtrain_pred_fake_image\": pred_fake_image.detach().float(),\n \"dmtrain_grad\": grad.detach().float(),\n \"dmtrain_gradient_norm\": torch.norm(grad).item()\n }\n\n return loss_dict, dm_log_dict\n \n \n def compute_ctc_sv_loss(\n self,\n real_inp: torch.Tensor, # real data latent\n fake_inp: torch.Tensor, # student-generated data latent\n text: torch.Tensor,\n text_lens: torch.Tensor,\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n ):\n \"\"\"\n Compute CTC + SV loss for direct metric optimization, as described in DMDSpeech.\n - CTC loss reduces WER\n - SV loss improves speaker similarity\n\n Both CTC and SV models operate on latents.\n \"\"\"\n\n # compute CTC loss\n out, layer, ctc_loss = self.ctc_model(fake_inp * self.scale, text, text_lens) # lengths from rand_span_mask or known\n\n with torch.no_grad():\n real_out, real_layers, ctc_loss_test = self.ctc_model(real_inp * self.scale, text, text_lens)\n real_logits = real_out.log_softmax(dim=2)\n # emb_real = self.sv_model(real_inp * self.scale) # snippet from prompt region \n \n fake_logits = out.log_softmax(dim=2)\n kl_loss = F.kl_div(fake_logits, real_logits, reduction=\"mean\", log_target=True)\n \n # For SV:\n # Extract speaker embeddings from real (prompt) and fake:\n # emb_fake = self.sv_model(fake_inp * self.scale)\n # sv_loss = 1 - F.cosine_similarity(emb_real, emb_fake, dim=-1).mean()\n\n input_lengths = rand_span_mask.sum(axis=-1).cpu().numpy()\n prompt_lengths = real_inp.size(1) - rand_span_mask.sum(axis=-1).cpu().numpy()\n\n chunks_real = []\n chunks_fake = []\n mel_len = min([int(input_lengths.min().item() - 1), 300])\n\n for bib in range(len(input_lengths)):\n prompt_length = int(prompt_lengths[bib].item())\n mel_length = int(input_lengths[bib].item())\n mask = rand_span_mask[bib]\n mask = torch.where(mask)[0]\n\n prompt_start = mask[0].cpu().numpy()\n prompt_end = mask[-1].cpu().numpy()\n\n if prompt_end - mel_len <= prompt_start:\n random_start = np.random.randint(0, mel_length - mel_len)\n else:\n random_start = np.random.randint(prompt_start, prompt_end - mel_len)\n \n chunks_fake.append(fake_inp[bib, random_start:random_start + mel_len, :])\n chunks_real.append(real_inp[bib, :mel_len, :])\n\n chunks_real = torch.stack(chunks_real, dim=0)\n chunks_fake = torch.stack(chunks_fake, dim=0)\n\n with torch.no_grad():\n emb_real_en = self.sv_model_en(chunks_real * self.scale)\n emb_fake_en = self.sv_model_en(chunks_fake * self.scale)\n\n sv_loss_en = 1 - F.cosine_similarity(emb_real_en, emb_fake_en, dim=-1).mean()\n\n with torch.no_grad():\n emb_real_zh = self.sv_model_zh(chunks_real * self.scale)\n emb_fake_zh = self.sv_model_zh(chunks_fake * self.scale)\n\n sv_loss_zh = 1 - F.cosine_similarity(emb_real_zh, emb_fake_zh, dim=-1).mean()\n\n sv_loss = (sv_loss_en + sv_loss_zh) / 2\n\n return {\n \"loss_ctc\": ctc_loss,\n 'loss_kl': kl_loss,\n \"loss_sim\": sv_loss\n }, layer, real_layers\n\n \n \n def compute_loss_fake(\n self,\n inp: torch.Tensor, # student generator output \n text: torch.Tensor | list[str],\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n ):\n \"\"\"\n Compute flow loss for the fake flow model, which is trained to estimate the flow (score) of the student distribution.\n \n This is the same as L_diff in the paper. \n \"\"\"\n \n # Similar to distribution matching, but only train fake to predict flow directly\n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n\n if isinstance(text, list):","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.compute_loss_fake","uri":"program://DMOSpeech2/function/src.guidance_model.compute_loss_fake#L357-L415","kind":"function","name":"compute_loss_fake","path":"src/guidance_model.py","language":"python","start_line":357,"end_line":415,"context_start_line":337,"context_end_line":435,"code":" emb_fake_en = self.sv_model_en(chunks_fake * self.scale)\n\n sv_loss_en = 1 - F.cosine_similarity(emb_real_en, emb_fake_en, dim=-1).mean()\n\n with torch.no_grad():\n emb_real_zh = self.sv_model_zh(chunks_real * self.scale)\n emb_fake_zh = self.sv_model_zh(chunks_fake * self.scale)\n\n sv_loss_zh = 1 - F.cosine_similarity(emb_real_zh, emb_fake_zh, dim=-1).mean()\n\n sv_loss = (sv_loss_en + sv_loss_zh) / 2\n\n return {\n \"loss_ctc\": ctc_loss,\n 'loss_kl': kl_loss,\n \"loss_sim\": sv_loss\n }, layer, real_layers\n\n \n \n def compute_loss_fake(\n self,\n inp: torch.Tensor, # student generator output \n text: torch.Tensor | list[str],\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n ):\n \"\"\"\n Compute flow loss for the fake flow model, which is trained to estimate the flow (score) of the student distribution.\n \n This is the same as L_diff in the paper. \n \"\"\"\n \n # Similar to distribution matching, but only train fake to predict flow directly\n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # Sample a time\n time = torch.rand((batch,), dtype=dtype, device=device)\n\n x1 = inp\n x0 = torch.randn_like(x1)\n t = time.unsqueeze(-1).unsqueeze(-1)\n \n phi = (1 - t) * x0 + t * x1\n flow = x1 - x0\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n\n with self.network_context_manager:\n pred = self.fake_unet(\n x=phi, \n cond=cond,\n text=text, \n time=time, \n second_time=second_time,\n drop_audio_cond=False, \n drop_text=False # make sure the cfg=1\n )\n\n # Compute MSE between predicted flow and actual flow, masked by rand_span_mask\n loss = F.mse_loss(pred, flow, reduction=\"none\")\n loss = loss[rand_span_mask].mean()\n \n loss_dict = {\n \"loss_fake_mean\": loss\n }\n log_dict = {\n \"faketrain_noisy_inp\": phi.detach().float(),\n \"faketrain_x1\": x1.detach().float(),\n \"faketrain_pred_flow\": pred.detach().float(),\n }\n\n return loss_dict, log_dict\n\n def compute_cls_logits(\n self,\n inp: torch.Tensor, # student generator output\n layer: torch.Tensor,\n text: torch.Tensor,\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n guidance: bool = False,\n ):\n '''\n Compute adversarial loss logits for the generator.\n \n This is used to compute L_adv in the paper.\n \n '''\n context_no_grad = torch.no_grad if guidance else NoOpContext\n\n with context_no_grad():\n # If we are not doing generator classification loss, return zeros","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.compute_cls_logits","uri":"program://DMOSpeech2/function/src.guidance_model.compute_cls_logits#L417-L485","kind":"function","name":"compute_cls_logits","path":"src/guidance_model.py","language":"python","start_line":417,"end_line":485,"context_start_line":397,"context_end_line":505,"code":" second_time=second_time,\n drop_audio_cond=False, \n drop_text=False # make sure the cfg=1\n )\n\n # Compute MSE between predicted flow and actual flow, masked by rand_span_mask\n loss = F.mse_loss(pred, flow, reduction=\"none\")\n loss = loss[rand_span_mask].mean()\n \n loss_dict = {\n \"loss_fake_mean\": loss\n }\n log_dict = {\n \"faketrain_noisy_inp\": phi.detach().float(),\n \"faketrain_x1\": x1.detach().float(),\n \"faketrain_pred_flow\": pred.detach().float(),\n }\n\n return loss_dict, log_dict\n\n def compute_cls_logits(\n self,\n inp: torch.Tensor, # student generator output\n layer: torch.Tensor,\n text: torch.Tensor,\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n guidance: bool = False,\n ):\n '''\n Compute adversarial loss logits for the generator.\n \n This is used to compute L_adv in the paper.\n \n '''\n context_no_grad = torch.no_grad if guidance else NoOpContext\n\n with context_no_grad():\n # If we are not doing generator classification loss, return zeros\n if not self.gen_cls_loss:\n return torch.zeros_like(inp[..., 0]) # shape (b, n)\n\n # For classification, we need some representation:\n # We'll mimic the logic from compute_loss_fake\n \n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # Sample a time\n time = torch.rand((batch,), dtype=dtype, device=device)\n\n x1 = inp\n x0 = torch.randn_like(x1)\n t = time.unsqueeze(-1).unsqueeze(-1)\n \n phi = (1 - t) * x0 + t * x1\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n\n with self.network_context_manager:\n layers = self.fake_unet(\n x=phi, \n cond=cond,\n text=text, \n time=time, \n second_time=second_time,\n drop_audio_cond=False, \n drop_text=False, # make sure the cfg=1\n classify_mode=True\n )\n # layers = torch.stack(layers, dim=0)\n\n if guidance:\n layers = [layer.detach() for layer in layers]\n layer = layer[-3:] # only use the last 3 layers\n layer = [l.transpose(-1, -2) for l in layer]\n # layer = [F.interpolate(l, mode='nearest', scale_factor=4).transpose(-1, -2) for l in layer]\n if layer[0].size(1) < layers[0].size(1):\n layer = [F.pad(l, (0, 0, 0, layers[0].size(1) - l.size(1))) for l in layer]\n\n layers = layer + layers\n # logits: (b, 1)\n logits = self.cls_pred_branch(layers)\n\n return logits, layers\n\n\n def compute_generator_cls_loss(\n self,\n inp: torch.Tensor, # student generator output\n layer: torch.Tensor,\n real_layers: torch.Tensor,\n text: torch.Tensor,\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n mse_loss: bool = False,\n mse_inp: torch.Tensor | None = None,\n ):\n '''\n Compute the adversarial loss for the generator. \n '''\n \n # Compute classification loss for generator:\n if not self.gen_cls_loss:\n return {\"gen_cls_loss\": 0}","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.compute_generator_cls_loss","uri":"program://DMOSpeech2/function/src.guidance_model.compute_generator_cls_loss#L488-L511","kind":"function","name":"compute_generator_cls_loss","path":"src/guidance_model.py","language":"python","start_line":488,"end_line":511,"context_start_line":468,"context_end_line":531,"code":" drop_text=False, # make sure the cfg=1\n classify_mode=True\n )\n # layers = torch.stack(layers, dim=0)\n\n if guidance:\n layers = [layer.detach() for layer in layers]\n layer = layer[-3:] # only use the last 3 layers\n layer = [l.transpose(-1, -2) for l in layer]\n # layer = [F.interpolate(l, mode='nearest', scale_factor=4).transpose(-1, -2) for l in layer]\n if layer[0].size(1) < layers[0].size(1):\n layer = [F.pad(l, (0, 0, 0, layers[0].size(1) - l.size(1))) for l in layer]\n\n layers = layer + layers\n # logits: (b, 1)\n logits = self.cls_pred_branch(layers)\n\n return logits, layers\n\n\n def compute_generator_cls_loss(\n self,\n inp: torch.Tensor, # student generator output\n layer: torch.Tensor,\n real_layers: torch.Tensor,\n text: torch.Tensor,\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n mse_loss: bool = False,\n mse_inp: torch.Tensor | None = None,\n ):\n '''\n Compute the adversarial loss for the generator. \n '''\n \n # Compute classification loss for generator:\n if not self.gen_cls_loss:\n return {\"gen_cls_loss\": 0}\n\n logits, fake_layers = self.compute_cls_logits(inp, layer, text, rand_span_mask, second_time, guidance=False)\n\n loss = ((1 - logits) ** 2).mean()\n\n return {\"gen_cls_loss\": loss, \"loss_mse\": 0}\n \n def compute_guidance_cls_loss(\n self,\n fake_inp: torch.Tensor,\n text: torch.Tensor,\n rand_span_mask: torch.Tensor,\n real_data: dict,\n second_time: torch.Tensor | None = None,\n ):\n '''\n This function computes the adversarial loss for the discirminator.\n\n The discriminator is trained to classify the generator output as real or fake.\n '''\n\n with torch.no_grad():\n # get layers from CTC model\n _, layer = self.ctc_model(fake_inp * self.scale)\n\n logits_fake, _ = self.compute_cls_logits(fake_inp.detach(), layer, text, rand_span_mask, second_time, guidance=True)","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.compute_guidance_cls_loss","uri":"program://DMOSpeech2/function/src.guidance_model.compute_guidance_cls_loss#L513-L553","kind":"function","name":"compute_guidance_cls_loss","path":"src/guidance_model.py","language":"python","start_line":513,"end_line":553,"context_start_line":493,"context_end_line":573,"code":" text: torch.Tensor,\n rand_span_mask: torch.Tensor,\n second_time: torch.Tensor | None = None,\n mse_loss: bool = False,\n mse_inp: torch.Tensor | None = None,\n ):\n '''\n Compute the adversarial loss for the generator. \n '''\n \n # Compute classification loss for generator:\n if not self.gen_cls_loss:\n return {\"gen_cls_loss\": 0}\n\n logits, fake_layers = self.compute_cls_logits(inp, layer, text, rand_span_mask, second_time, guidance=False)\n\n loss = ((1 - logits) ** 2).mean()\n\n return {\"gen_cls_loss\": loss, \"loss_mse\": 0}\n \n def compute_guidance_cls_loss(\n self,\n fake_inp: torch.Tensor,\n text: torch.Tensor,\n rand_span_mask: torch.Tensor,\n real_data: dict,\n second_time: torch.Tensor | None = None,\n ):\n '''\n This function computes the adversarial loss for the discirminator.\n\n The discriminator is trained to classify the generator output as real or fake.\n '''\n\n with torch.no_grad():\n # get layers from CTC model\n _, layer = self.ctc_model(fake_inp * self.scale)\n\n logits_fake, _ = self.compute_cls_logits(fake_inp.detach(), layer, text, rand_span_mask, second_time, guidance=True)\n loss_fake = (logits_fake**2).mean()\n\n real_inp = real_data[\"inp\"]\n\n with torch.no_grad():\n # get layers from CTC model\n _, layer = self.ctc_model(real_inp * self.scale)\n \n logits_real, _ = self.compute_cls_logits(real_inp.detach(), layer, text, rand_span_mask, second_time, guidance=True)\n loss_real = ((1 - logits_real)**2).mean()\n\n classification_loss = loss_real + loss_fake\n\n loss_dict = {\n \"guidance_cls_loss\": classification_loss\n }\n log_dict = {\n \"pred_realism_on_real\": loss_real.detach().item(),\n \"pred_realism_on_fake\": loss_fake.detach().item()\n }\n\n return loss_dict, log_dict\n\n def generator_forward(\n self,\n inp: torch.Tensor,\n text: torch.Tensor,\n text_lens: torch.Tensor,\n text_normalized: torch.Tensor,\n text_normalized_lens: torch.Tensor,\n rand_span_mask: torch.Tensor,\n real_data: dict | None = None, # ground truth data (primarily prompt) to compute SV loss\n second_time: torch.Tensor | None = None,\n mse_loss: bool = False,\n ):\n '''\n Forward pass for the generator.\n \n This function computes the loss for the generator, which includes:\n - Distribution matching loss (L_DMD)\n - Adversarial generator loss (L_adv(G; D))\n - CTC/SV loss (L_ctc + L_sv)","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.generator_forward","uri":"program://DMOSpeech2/function/src.guidance_model.generator_forward#L555-L600","kind":"function","name":"generator_forward","path":"src/guidance_model.py","language":"python","start_line":555,"end_line":600,"context_start_line":535,"context_end_line":620,"code":"\n with torch.no_grad():\n # get layers from CTC model\n _, layer = self.ctc_model(real_inp * self.scale)\n \n logits_real, _ = self.compute_cls_logits(real_inp.detach(), layer, text, rand_span_mask, second_time, guidance=True)\n loss_real = ((1 - logits_real)**2).mean()\n\n classification_loss = loss_real + loss_fake\n\n loss_dict = {\n \"guidance_cls_loss\": classification_loss\n }\n log_dict = {\n \"pred_realism_on_real\": loss_real.detach().item(),\n \"pred_realism_on_fake\": loss_fake.detach().item()\n }\n\n return loss_dict, log_dict\n\n def generator_forward(\n self,\n inp: torch.Tensor,\n text: torch.Tensor,\n text_lens: torch.Tensor,\n text_normalized: torch.Tensor,\n text_normalized_lens: torch.Tensor,\n rand_span_mask: torch.Tensor,\n real_data: dict | None = None, # ground truth data (primarily prompt) to compute SV loss\n second_time: torch.Tensor | None = None,\n mse_loss: bool = False,\n ):\n '''\n Forward pass for the generator.\n \n This function computes the loss for the generator, which includes:\n - Distribution matching loss (L_DMD)\n - Adversarial generator loss (L_adv(G; D))\n - CTC/SV loss (L_ctc + L_sv)\n '''\n \n # 1. Compute DM loss\n dm_loss_dict, dm_log_dict = self.compute_distribution_matching_loss(inp, text, rand_span_mask=rand_span_mask, second_time=second_time)\n\n ctc_sv_loss_dict = {}\n cls_loss_dict = {}\n\n # 2. Compute optional CTC/SV loss if real_data provided\n if real_data is not None:\n real_inp = real_data[\"inp\"]\n ctc_sv_loss_dict, layer, real_layers = self.compute_ctc_sv_loss(real_inp, inp, text_normalized, text_normalized_lens, rand_span_mask, second_time=second_time)\n\n # 3. Compute optional classification loss\n if self.gen_cls_loss:\n cls_loss_dict = self.compute_generator_cls_loss(inp, layer, real_layers, text,\n rand_span_mask=rand_span_mask, \n second_time=second_time,\n mse_inp = real_data[\"inp\"] if real_data is not None else None,\n mse_loss = mse_loss,\n )\n\n\n loss_dict = {**dm_loss_dict, **cls_loss_dict, **ctc_sv_loss_dict}\n log_dict = {**dm_log_dict}\n\n return loss_dict, log_dict\n\n def guidance_forward(\n self,\n fake_inp: torch.Tensor,\n text: torch.Tensor,\n text_lens: torch.Tensor,\n rand_span_mask: torch.Tensor,\n real_data: dict | None = None,\n second_time: torch.Tensor | None = None,\n ):\n '''\n Forward pass for the guidnce module (discriminator + fake flow function).\n \n This function computes the loss for the guidance module, which includes:\n - Flow matching loss (L_diff)\n - Adversarial discrminator loss (L_adv(D; G))\n \n '''\n \n # Compute fake loss (like epsilon prediction loss in Guidance)","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.guidance_forward","uri":"program://DMOSpeech2/function/src.guidance_model.guidance_forward#L602-L632","kind":"function","name":"guidance_forward","path":"src/guidance_model.py","language":"python","start_line":602,"end_line":632,"context_start_line":582,"context_end_line":652,"code":" # 2. Compute optional CTC/SV loss if real_data provided\n if real_data is not None:\n real_inp = real_data[\"inp\"]\n ctc_sv_loss_dict, layer, real_layers = self.compute_ctc_sv_loss(real_inp, inp, text_normalized, text_normalized_lens, rand_span_mask, second_time=second_time)\n\n # 3. Compute optional classification loss\n if self.gen_cls_loss:\n cls_loss_dict = self.compute_generator_cls_loss(inp, layer, real_layers, text,\n rand_span_mask=rand_span_mask, \n second_time=second_time,\n mse_inp = real_data[\"inp\"] if real_data is not None else None,\n mse_loss = mse_loss,\n )\n\n\n loss_dict = {**dm_loss_dict, **cls_loss_dict, **ctc_sv_loss_dict}\n log_dict = {**dm_log_dict}\n\n return loss_dict, log_dict\n\n def guidance_forward(\n self,\n fake_inp: torch.Tensor,\n text: torch.Tensor,\n text_lens: torch.Tensor,\n rand_span_mask: torch.Tensor,\n real_data: dict | None = None,\n second_time: torch.Tensor | None = None,\n ):\n '''\n Forward pass for the guidnce module (discriminator + fake flow function).\n \n This function computes the loss for the guidance module, which includes:\n - Flow matching loss (L_diff)\n - Adversarial discrminator loss (L_adv(D; G))\n \n '''\n \n # Compute fake loss (like epsilon prediction loss in Guidance)\n fake_loss_dict, fake_log_dict = self.compute_loss_fake(fake_inp, text, rand_span_mask=rand_span_mask, second_time=second_time)\n\n # If gen_cls_loss, compute guidance cls loss\n cls_loss_dict = {}\n cls_log_dict = {}\n if self.gen_cls_loss and real_data is not None:\n cls_loss_dict, cls_log_dict = self.compute_guidance_cls_loss(fake_inp, text, rand_span_mask, real_data, second_time=second_time)\n\n loss_dict = {**fake_loss_dict, **cls_loss_dict}\n log_dict = {**fake_log_dict, **cls_log_dict}\n\n return loss_dict, log_dict\n \n def forward(\n self,\n generator_turn=False,\n guidance_turn=False,\n generator_data_dict=None,\n guidance_data_dict=None\n ):\n if generator_turn:\n loss_dict, log_dict = self.generator_forward(\n inp=generator_data_dict[\"inp\"],\n text=generator_data_dict[\"text\"],\n text_lens=generator_data_dict[\"text_lens\"],\n text_normalized=generator_data_dict[\"text_normalized\"],\n text_normalized_lens=generator_data_dict[\"text_normalized_lens\"],\n rand_span_mask=generator_data_dict[\"rand_span_mask\"],\n real_data=generator_data_dict.get(\"real_data\", None),\n second_time=generator_data_dict.get(\"second_time\", None),\n mse_loss=generator_data_dict.get(\"mse_loss\", False),\n )","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.guidance_model.forward","uri":"program://DMOSpeech2/function/src.guidance_model.forward#L634-L665","kind":"function","name":"forward","path":"src/guidance_model.py","language":"python","start_line":634,"end_line":665,"context_start_line":614,"context_end_line":685,"code":" This function computes the loss for the guidance module, which includes:\n - Flow matching loss (L_diff)\n - Adversarial discrminator loss (L_adv(D; G))\n \n '''\n \n # Compute fake loss (like epsilon prediction loss in Guidance)\n fake_loss_dict, fake_log_dict = self.compute_loss_fake(fake_inp, text, rand_span_mask=rand_span_mask, second_time=second_time)\n\n # If gen_cls_loss, compute guidance cls loss\n cls_loss_dict = {}\n cls_log_dict = {}\n if self.gen_cls_loss and real_data is not None:\n cls_loss_dict, cls_log_dict = self.compute_guidance_cls_loss(fake_inp, text, rand_span_mask, real_data, second_time=second_time)\n\n loss_dict = {**fake_loss_dict, **cls_loss_dict}\n log_dict = {**fake_log_dict, **cls_log_dict}\n\n return loss_dict, log_dict\n \n def forward(\n self,\n generator_turn=False,\n guidance_turn=False,\n generator_data_dict=None,\n guidance_data_dict=None\n ):\n if generator_turn:\n loss_dict, log_dict = self.generator_forward(\n inp=generator_data_dict[\"inp\"],\n text=generator_data_dict[\"text\"],\n text_lens=generator_data_dict[\"text_lens\"],\n text_normalized=generator_data_dict[\"text_normalized\"],\n text_normalized_lens=generator_data_dict[\"text_normalized_lens\"],\n rand_span_mask=generator_data_dict[\"rand_span_mask\"],\n real_data=generator_data_dict.get(\"real_data\", None),\n second_time=generator_data_dict.get(\"second_time\", None),\n mse_loss=generator_data_dict.get(\"mse_loss\", False),\n )\n elif guidance_turn:\n loss_dict, log_dict = self.guidance_forward(\n fake_inp=guidance_data_dict[\"inp\"],\n text=guidance_data_dict[\"text\"],\n text_lens=guidance_data_dict[\"text_lens\"],\n rand_span_mask=guidance_data_dict[\"rand_span_mask\"],\n real_data=guidance_data_dict.get(\"real_data\", None),\n second_time=guidance_data_dict.get(\"second_time\", None),\n )\n else:\n raise NotImplementedError(\"Must specify either generator_turn or guidance_turn\")\n\n return loss_dict, log_dict\n\n\n\n \nif __name__ == \"__main__\":\n from f5_tts.model.utils import get_tokenizer\n\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n \n ","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer","uri":"program://DMOSpeech2/module/src.duration_trainer#L1-L563","kind":"module","name":"src.duration_trainer","path":"src/duration_trainer.py","language":"python","start_line":1,"end_line":563,"context_start_line":1,"context_end_line":563,"code":"from __future__ import annotations\n\nimport gc\nimport os\n\nimport math\n\nimport torch\nimport torchaudio\nimport wandb\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nfrom ema_pytorch import EMA\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler, Subset # <-- Added Subset import\nfrom tqdm import tqdm\n\nimport torch.nn.functional as F\n\nfrom f5_tts.model import CFM\nfrom f5_tts.model.dataset import collate_fn, DynamicBatchSampler\nfrom f5_tts.model.utils import default, exists\n\nfrom duration_predictor import calculate_remaining_lengths\n\n# trainer\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nSAMPLE_RATE = 24_000\n\n\ndef masked_l1_loss(est_lengths, tar_lengths):\n first_zero_idx = (tar_lengths == 0).int().argmax(dim=1) \n B, L = tar_lengths.shape\n range_tensor = torch.arange(L, device=tar_lengths.device).expand(B, L) \n mask = range_tensor <= first_zero_idx[:, None] # Include the first 0\n loss = F.l1_loss(est_lengths, tar_lengths, reduction='none') # (B, L)\n loss = loss * mask # Zero out ignored positions\n loss = loss.sum() / mask.sum() # Normalize by valid elements\n return loss\n\n\ndef masked_cross_entropy_loss(est_length_logits, tar_length_labels):\n first_zero_idx = (tar_length_labels == 0).int().argmax(dim=1)\n B, L = tar_length_labels.shape\n range_tensor = torch.arange(L, device=tar_length_labels.device).expand(B, L)\n mask = range_tensor <= first_zero_idx[:, None] # Include the first 0\n loss = F.cross_entropy(\n est_length_logits.reshape(-1, est_length_logits.size(-1)), \n tar_length_labels.reshape(-1), \n reduction='none'\n ).reshape(B, L)\n loss = loss * mask\n loss = loss.sum() / mask.sum()\n return loss\n\n\nclass Trainer:\n def __init__(\n self,\n model,\n vocab_size,\n vocab_char_map,\n process_token_to_id=True,\n loss_fn='L1',\n lambda_L1=1,\n gumbel_tau=0.5,\n n_class=301,\n n_frame_per_class=10,\n epochs=15,\n learning_rate=1e-4,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n logger: str | None = \"wandb\", # \"wandb\" | \"tensorboard\" | None\n wandb_project=\"test_e2-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n last_per_steps=None,\n accelerate_kwargs: dict = dict(),\n ema_kwargs: dict = dict(),\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=False)\n\n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n print(f\"Using logger: {logger}\")\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n },\n )\n\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n self.model = model\n self.vocab_size = vocab_size\n self.vocab_char_map = vocab_char_map\n self.process_token_to_id = process_token_to_id\n assert loss_fn in ['L1', 'CE', 'L1_and_CE']\n self.loss_fn = loss_fn\n self.lambda_L1 = lambda_L1\n self.n_class = n_class\n self.n_frame_per_class = n_frame_per_class\n self.gumbel_tau = gumbel_tau\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n\n print(f'To load from {latest_checkpoint}.')\n\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n print(f'Loaded from {latest_checkpoint}.')\n\n if \"step\" in checkpoint:\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n \n del checkpoint\n gc.collect()\n\n print(f'Exit load_checkpoint.')\n\n return step\n\n\n def validate(self, valid_dataloader, global_step):\n \"\"\"\n Runs evaluation on the validation set, computes the average loss,\n and logs the average validation loss along with the CTC decoded strings.\n \"\"\"\n self.model.eval()\n total_valid_loss = 0.0\n total_sec_error = 0.0\n count = 0\n # Iterate over the validation dataloader\n with torch.no_grad():\n for batch in valid_dataloader:\n # Inputs\n mel = batch['mel'].permute(0, 2, 1) # (B, L_mel, D)\n text = batch['text']\n\n if self.process_token_to_id:\n text_ids = list_str_to_idx(text, self.vocab_char_map).to(mel.device)\n text_ids = text_ids.masked_fill(text_ids==-1, self.vocab_size)\n else:\n text_ids = text\n\n # Targets\n mel_lengths = batch['mel_lengths']\n tar_lengths = calculate_remaining_lengths(mel_lengths)\n predictions = self.model(text_ids=text_ids, mel=mel)\n\n if self.loss_fn == 'L1':\n est_lengths = predictions\n loss = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n frame_error = loss\n\n elif self.loss_fn == 'CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_labels = torch.argmax(est_length_logtis, dim=-1)\n loss = masked_cross_entropy_loss(\n est_length_logits=est_length_logtis, tar_length_labels=tar_length_labels\n )\n est_lengths = est_length_labels * self.n_frame_per_class\n frame_error = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n\n elif self.loss_fn == 'L1_and_CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_1hots = F.gumbel_softmax(\n est_length_logtis, tau=self.gumbel_tau, hard=True, dim=-1\n )\n length_values = torch.arange(\n self.n_class, device=est_length_1hots.device\n ).float() * self.n_frame_per_class\n est_lengths = (est_length_1hots * length_values).sum(-1)\n\n loss_CE = masked_cross_entropy_loss(\n est_length_logits=est_length_logtis, tar_length_labels=tar_length_labels\n )\n\n loss_L1 = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n\n loss = loss_CE + self.lambda_L1 * loss_L1\n\n frame_error = loss_L1\n\n else:\n raise NotImplementedError(self.loss_fn)\n\n sec_error = frame_error * 256 / 24000\n total_sec_error += sec_error.item()\n total_valid_loss += loss.item()\n count += 1\n\n avg_valid_loss = total_valid_loss / count if count > 0 else 0.0\n avg_valid_sec_error = total_sec_error / count if count > 0 else 0.0\n # Log validation metrics\n self.accelerator.log(\n {\n f\"valid_loss\": avg_valid_loss,\n f\"valid_sec_error\": avg_valid_sec_error\n }, \n step=global_step\n )\n \n self.model.train()\n\n\n def train(self, train_dataset: Dataset, valid_dataset: Dataset,\n num_workers=64, resumable_with_seed: int = None):\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n # Create training dataloader using the appropriate batching strategy\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )\n # Create validation dataloader (always sequential, no shuffling)\n valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n batch_size=self.batch_size,\n shuffle=False,\n )\n\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n\n sampler = SequentialSampler(valid_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n # Create validation dataloader (always sequential, no shuffling)\n valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True, \n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n \n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_steps = (\n self.num_warmup_updates * self.accelerator.num_processes\n ) # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_steps = len(train_dataloader) * self.epochs / self.grad_accumulation_steps\n decay_steps = total_steps - warmup_steps\n warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)\n decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)\n self.scheduler = SequentialLR(\n self.optimizer, schedulers=[warmup_scheduler, decay_scheduler], milestones=[warmup_steps]\n )\n train_dataloader, self.scheduler = self.accelerator.prepare(\n train_dataloader, self.scheduler\n ) # actual steps = 1 gpu steps / gpus\n start_step = self.load_checkpoint()\n global_step = start_step\n\n valid_dataloader = self.accelerator.prepare(valid_dataloader)\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar = tqdm(\n skipped_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n initial=skipped_batch,\n total=orig_epoch_step,\n )\n else:\n progress_bar = tqdm(\n train_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n )\n\n for batch in progress_bar:\n with self.accelerator.accumulate(self.model):\n # Inputs\n mel = batch['mel'].permute(0, 2, 1) # (B, L_mel, D)\n text = batch['text']\n\n if self.process_token_to_id:\n text_ids = list_str_to_idx(text, self.vocab_char_map).to(mel.device)\n text_ids = text_ids.masked_fill(text_ids==-1, self.vocab_size)\n else:\n text_ids = text\n\n # Targets\n mel_lengths = batch['mel_lengths']\n tar_lengths = calculate_remaining_lengths(mel_lengths)\n predictions = self.model(text_ids=text_ids, mel=mel)\n\n if self.loss_fn == 'L1':\n est_lengths = predictions\n loss = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n\n with torch.no_grad():\n frame_error = loss\n sec_error = frame_error * 256 / 24000\n\n log_dict = {\n 'loss': loss.item(), \n 'loss_L1': loss.item(), \n 'sec_error': sec_error.item(),\n 'lr': self.scheduler.get_last_lr()[0]\n }\n\n elif self.loss_fn == 'CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_labels = torch.argmax(est_length_logtis, dim=-1)\n loss = masked_cross_entropy_loss(\n est_length_logits=est_length_logtis, tar_length_labels=tar_length_labels\n )\n with torch.no_grad():\n est_lengths = est_length_labels * self.n_frame_per_class\n frame_error = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n sec_error = frame_error * 256 / 24000\n\n log_dict = {\n# ... truncated ...","source_hash":"2d47b5df9155cbe6db309e3617bab7f7fbf0c03498c4a19ddf3e3229f1ea24c1","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer.masked_l1_loss","uri":"program://DMOSpeech2/function/src.duration_trainer.masked_l1_loss#L41-L49","kind":"function","name":"masked_l1_loss","path":"src/duration_trainer.py","language":"python","start_line":41,"end_line":49,"context_start_line":21,"context_end_line":69,"code":"from f5_tts.model import CFM\nfrom f5_tts.model.dataset import collate_fn, DynamicBatchSampler\nfrom f5_tts.model.utils import default, exists\n\nfrom duration_predictor import calculate_remaining_lengths\n\n# trainer\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nSAMPLE_RATE = 24_000\n\n\ndef masked_l1_loss(est_lengths, tar_lengths):\n first_zero_idx = (tar_lengths == 0).int().argmax(dim=1) \n B, L = tar_lengths.shape\n range_tensor = torch.arange(L, device=tar_lengths.device).expand(B, L) \n mask = range_tensor <= first_zero_idx[:, None] # Include the first 0\n loss = F.l1_loss(est_lengths, tar_lengths, reduction='none') # (B, L)\n loss = loss * mask # Zero out ignored positions\n loss = loss.sum() / mask.sum() # Normalize by valid elements\n return loss\n\n\ndef masked_cross_entropy_loss(est_length_logits, tar_length_labels):\n first_zero_idx = (tar_length_labels == 0).int().argmax(dim=1)\n B, L = tar_length_labels.shape\n range_tensor = torch.arange(L, device=tar_length_labels.device).expand(B, L)\n mask = range_tensor <= first_zero_idx[:, None] # Include the first 0\n loss = F.cross_entropy(\n est_length_logits.reshape(-1, est_length_logits.size(-1)), \n tar_length_labels.reshape(-1), \n reduction='none'\n ).reshape(B, L)\n loss = loss * mask\n loss = loss.sum() / mask.sum()\n return loss\n\n\nclass Trainer:\n def __init__(\n self,","source_hash":"2d47b5df9155cbe6db309e3617bab7f7fbf0c03498c4a19ddf3e3229f1ea24c1","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer.masked_cross_entropy_loss","uri":"program://DMOSpeech2/function/src.duration_trainer.masked_cross_entropy_loss#L52-L64","kind":"function","name":"masked_cross_entropy_loss","path":"src/duration_trainer.py","language":"python","start_line":52,"end_line":64,"context_start_line":32,"context_end_line":84,"code":" list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nSAMPLE_RATE = 24_000\n\n\ndef masked_l1_loss(est_lengths, tar_lengths):\n first_zero_idx = (tar_lengths == 0).int().argmax(dim=1) \n B, L = tar_lengths.shape\n range_tensor = torch.arange(L, device=tar_lengths.device).expand(B, L) \n mask = range_tensor <= first_zero_idx[:, None] # Include the first 0\n loss = F.l1_loss(est_lengths, tar_lengths, reduction='none') # (B, L)\n loss = loss * mask # Zero out ignored positions\n loss = loss.sum() / mask.sum() # Normalize by valid elements\n return loss\n\n\ndef masked_cross_entropy_loss(est_length_logits, tar_length_labels):\n first_zero_idx = (tar_length_labels == 0).int().argmax(dim=1)\n B, L = tar_length_labels.shape\n range_tensor = torch.arange(L, device=tar_length_labels.device).expand(B, L)\n mask = range_tensor <= first_zero_idx[:, None] # Include the first 0\n loss = F.cross_entropy(\n est_length_logits.reshape(-1, est_length_logits.size(-1)), \n tar_length_labels.reshape(-1), \n reduction='none'\n ).reshape(B, L)\n loss = loss * mask\n loss = loss.sum() / mask.sum()\n return loss\n\n\nclass Trainer:\n def __init__(\n self,\n model,\n vocab_size,\n vocab_char_map,\n process_token_to_id=True,\n loss_fn='L1',\n lambda_L1=1,\n gumbel_tau=0.5,\n n_class=301,\n n_frame_per_class=10,\n epochs=15,\n learning_rate=1e-4,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,","source_hash":"2d47b5df9155cbe6db309e3617bab7f7fbf0c03498c4a19ddf3e3229f1ea24c1","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer.Trainer","uri":"program://DMOSpeech2/class/src.duration_trainer.Trainer#L67-L563","kind":"class","name":"Trainer","path":"src/duration_trainer.py","language":"python","start_line":67,"end_line":563,"context_start_line":47,"context_end_line":563,"code":" loss = loss * mask # Zero out ignored positions\n loss = loss.sum() / mask.sum() # Normalize by valid elements\n return loss\n\n\ndef masked_cross_entropy_loss(est_length_logits, tar_length_labels):\n first_zero_idx = (tar_length_labels == 0).int().argmax(dim=1)\n B, L = tar_length_labels.shape\n range_tensor = torch.arange(L, device=tar_length_labels.device).expand(B, L)\n mask = range_tensor <= first_zero_idx[:, None] # Include the first 0\n loss = F.cross_entropy(\n est_length_logits.reshape(-1, est_length_logits.size(-1)), \n tar_length_labels.reshape(-1), \n reduction='none'\n ).reshape(B, L)\n loss = loss * mask\n loss = loss.sum() / mask.sum()\n return loss\n\n\nclass Trainer:\n def __init__(\n self,\n model,\n vocab_size,\n vocab_char_map,\n process_token_to_id=True,\n loss_fn='L1',\n lambda_L1=1,\n gumbel_tau=0.5,\n n_class=301,\n n_frame_per_class=10,\n epochs=15,\n learning_rate=1e-4,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n logger: str | None = \"wandb\", # \"wandb\" | \"tensorboard\" | None\n wandb_project=\"test_e2-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n last_per_steps=None,\n accelerate_kwargs: dict = dict(),\n ema_kwargs: dict = dict(),\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=False)\n\n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n print(f\"Using logger: {logger}\")\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n },\n )\n\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n self.model = model\n self.vocab_size = vocab_size\n self.vocab_char_map = vocab_char_map\n self.process_token_to_id = process_token_to_id\n assert loss_fn in ['L1', 'CE', 'L1_and_CE']\n self.loss_fn = loss_fn\n self.lambda_L1 = lambda_L1\n self.n_class = n_class\n self.n_frame_per_class = n_frame_per_class\n self.gumbel_tau = gumbel_tau\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n\n print(f'To load from {latest_checkpoint}.')\n\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n print(f'Loaded from {latest_checkpoint}.')\n\n if \"step\" in checkpoint:\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n \n del checkpoint\n gc.collect()\n\n print(f'Exit load_checkpoint.')\n\n return step\n\n\n def validate(self, valid_dataloader, global_step):\n \"\"\"\n Runs evaluation on the validation set, computes the average loss,\n and logs the average validation loss along with the CTC decoded strings.\n \"\"\"\n self.model.eval()\n total_valid_loss = 0.0\n total_sec_error = 0.0\n count = 0\n # Iterate over the validation dataloader\n with torch.no_grad():\n for batch in valid_dataloader:\n # Inputs\n mel = batch['mel'].permute(0, 2, 1) # (B, L_mel, D)\n text = batch['text']\n\n if self.process_token_to_id:\n text_ids = list_str_to_idx(text, self.vocab_char_map).to(mel.device)\n text_ids = text_ids.masked_fill(text_ids==-1, self.vocab_size)\n else:\n text_ids = text\n\n # Targets\n mel_lengths = batch['mel_lengths']\n tar_lengths = calculate_remaining_lengths(mel_lengths)\n predictions = self.model(text_ids=text_ids, mel=mel)\n\n if self.loss_fn == 'L1':\n est_lengths = predictions\n loss = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n frame_error = loss\n\n elif self.loss_fn == 'CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_labels = torch.argmax(est_length_logtis, dim=-1)\n loss = masked_cross_entropy_loss(\n est_length_logits=est_length_logtis, tar_length_labels=tar_length_labels\n )\n est_lengths = est_length_labels * self.n_frame_per_class\n frame_error = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n\n elif self.loss_fn == 'L1_and_CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_1hots = F.gumbel_softmax(\n est_length_logtis, tau=self.gumbel_tau, hard=True, dim=-1\n )\n length_values = torch.arange(\n self.n_class, device=est_length_1hots.device\n ).float() * self.n_frame_per_class\n est_lengths = (est_length_1hots * length_values).sum(-1)\n\n loss_CE = masked_cross_entropy_loss(\n est_length_logits=est_length_logtis, tar_length_labels=tar_length_labels\n )\n\n loss_L1 = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n\n loss = loss_CE + self.lambda_L1 * loss_L1\n\n frame_error = loss_L1\n\n else:\n raise NotImplementedError(self.loss_fn)\n\n sec_error = frame_error * 256 / 24000\n total_sec_error += sec_error.item()\n total_valid_loss += loss.item()\n count += 1\n\n avg_valid_loss = total_valid_loss / count if count > 0 else 0.0\n avg_valid_sec_error = total_sec_error / count if count > 0 else 0.0\n # Log validation metrics\n self.accelerator.log(\n {\n f\"valid_loss\": avg_valid_loss,\n f\"valid_sec_error\": avg_valid_sec_error\n }, \n step=global_step\n )\n \n self.model.train()\n\n\n def train(self, train_dataset: Dataset, valid_dataset: Dataset,\n num_workers=64, resumable_with_seed: int = None):\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n # Create training dataloader using the appropriate batching strategy\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )\n # Create validation dataloader (always sequential, no shuffling)\n valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n batch_size=self.batch_size,\n shuffle=False,\n )\n\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n\n sampler = SequentialSampler(valid_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n # Create validation dataloader (always sequential, no shuffling)\n valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True, \n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n \n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_steps = (\n self.num_warmup_updates * self.accelerator.num_processes\n ) # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_steps = len(train_dataloader) * self.epochs / self.grad_accumulation_steps\n decay_steps = total_steps - warmup_steps\n warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)\n decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)\n self.scheduler = SequentialLR(\n self.optimizer, schedulers=[warmup_scheduler, decay_scheduler], milestones=[warmup_steps]\n )\n train_dataloader, self.scheduler = self.accelerator.prepare(\n train_dataloader, self.scheduler\n ) # actual steps = 1 gpu steps / gpus\n start_step = self.load_checkpoint()\n global_step = start_step\n\n valid_dataloader = self.accelerator.prepare(valid_dataloader)\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar = tqdm(\n skipped_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n initial=skipped_batch,\n total=orig_epoch_step,\n )\n else:\n progress_bar = tqdm(\n train_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n )\n\n for batch in progress_bar:\n with self.accelerator.accumulate(self.model):\n # Inputs\n mel = batch['mel'].permute(0, 2, 1) # (B, L_mel, D)\n text = batch['text']\n\n if self.process_token_to_id:\n text_ids = list_str_to_idx(text, self.vocab_char_map).to(mel.device)\n text_ids = text_ids.masked_fill(text_ids==-1, self.vocab_size)\n else:\n text_ids = text\n\n # Targets\n mel_lengths = batch['mel_lengths']\n tar_lengths = calculate_remaining_lengths(mel_lengths)\n predictions = self.model(text_ids=text_ids, mel=mel)\n\n if self.loss_fn == 'L1':\n est_lengths = predictions\n loss = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n\n with torch.no_grad():\n frame_error = loss\n sec_error = frame_error * 256 / 24000\n\n log_dict = {\n 'loss': loss.item(), \n 'loss_L1': loss.item(), \n 'sec_error': sec_error.item(),\n 'lr': self.scheduler.get_last_lr()[0]\n }\n\n elif self.loss_fn == 'CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_labels = torch.argmax(est_length_logtis, dim=-1)\n loss = masked_cross_entropy_loss(\n est_length_logits=est_length_logtis, tar_length_labels=tar_length_labels\n )\n with torch.no_grad():\n est_lengths = est_length_labels * self.n_frame_per_class\n frame_error = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n sec_error = frame_error * 256 / 24000\n\n log_dict = {\n 'loss': loss.item(), \n 'loss_CE': loss.item(), \n 'sec_error': sec_error.item(),\n 'lr': self.scheduler.get_last_lr()[0]\n }\n\n elif self.loss_fn == 'L1_and_CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_1hots = F.gumbel_softmax(\n est_length_logtis, tau=self.gumbel_tau, hard=True, dim=-1\n )\n length_values = torch.arange(\n self.n_class, device=est_length_1hots.device\n ).float() * self.n_frame_per_class\n est_lengths = (est_length_1hots * length_values).sum(-1)\n\n loss_CE = masked_cross_entropy_loss(\n est_length_logits=est_length_logtis, tar_length_labels=tar_length_labels\n )\n\n loss_L1 = masked_l1_loss(\n est_lengths=e\n# ... truncated ...","source_hash":"2d47b5df9155cbe6db309e3617bab7f7fbf0c03498c4a19ddf3e3229f1ea24c1","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer.__init__","uri":"program://DMOSpeech2/function/src.duration_trainer.__init__#L68-L167","kind":"function","name":"__init__","path":"src/duration_trainer.py","language":"python","start_line":68,"end_line":167,"context_start_line":48,"context_end_line":187,"code":" loss = loss.sum() / mask.sum() # Normalize by valid elements\n return loss\n\n\ndef masked_cross_entropy_loss(est_length_logits, tar_length_labels):\n first_zero_idx = (tar_length_labels == 0).int().argmax(dim=1)\n B, L = tar_length_labels.shape\n range_tensor = torch.arange(L, device=tar_length_labels.device).expand(B, L)\n mask = range_tensor <= first_zero_idx[:, None] # Include the first 0\n loss = F.cross_entropy(\n est_length_logits.reshape(-1, est_length_logits.size(-1)), \n tar_length_labels.reshape(-1), \n reduction='none'\n ).reshape(B, L)\n loss = loss * mask\n loss = loss.sum() / mask.sum()\n return loss\n\n\nclass Trainer:\n def __init__(\n self,\n model,\n vocab_size,\n vocab_char_map,\n process_token_to_id=True,\n loss_fn='L1',\n lambda_L1=1,\n gumbel_tau=0.5,\n n_class=301,\n n_frame_per_class=10,\n epochs=15,\n learning_rate=1e-4,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n logger: str | None = \"wandb\", # \"wandb\" | \"tensorboard\" | None\n wandb_project=\"test_e2-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n last_per_steps=None,\n accelerate_kwargs: dict = dict(),\n ema_kwargs: dict = dict(),\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=False)\n\n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n print(f\"Using logger: {logger}\")\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n },\n )\n\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n self.model = model\n self.vocab_size = vocab_size\n self.vocab_char_map = vocab_char_map\n self.process_token_to_id = process_token_to_id\n assert loss_fn in ['L1', 'CE', 'L1_and_CE']\n self.loss_fn = loss_fn\n self.lambda_L1 = lambda_L1\n self.n_class = n_class\n self.n_frame_per_class = n_frame_per_class\n self.gumbel_tau = gumbel_tau\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")","source_hash":"2d47b5df9155cbe6db309e3617bab7f7fbf0c03498c4a19ddf3e3229f1ea24c1","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer.is_main","uri":"program://DMOSpeech2/function/src.duration_trainer.is_main#L170-L171","kind":"function","name":"is_main","path":"src/duration_trainer.py","language":"python","start_line":170,"end_line":171,"context_start_line":150,"context_end_line":191,"code":" self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)","source_hash":"2d47b5df9155cbe6db309e3617bab7f7fbf0c03498c4a19ddf3e3229f1ea24c1","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer.save_checkpoint","uri":"program://DMOSpeech2/function/src.duration_trainer.save_checkpoint#L173-L187","kind":"function","name":"save_checkpoint","path":"src/duration_trainer.py","language":"python","start_line":173,"end_line":187,"context_start_line":153,"context_end_line":207,"code":" self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n\n print(f'To load from {latest_checkpoint}.')\n","source_hash":"2d47b5df9155cbe6db309e3617bab7f7fbf0c03498c4a19ddf3e3229f1ea24c1","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer.load_checkpoint","uri":"program://DMOSpeech2/function/src.duration_trainer.load_checkpoint#L189-L238","kind":"function","name":"load_checkpoint","path":"src/duration_trainer.py","language":"python","start_line":189,"end_line":238,"context_start_line":169,"context_end_line":258,"code":" @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n\n print(f'To load from {latest_checkpoint}.')\n\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n print(f'Loaded from {latest_checkpoint}.')\n\n if \"step\" in checkpoint:\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n \n del checkpoint\n gc.collect()\n\n print(f'Exit load_checkpoint.')\n\n return step\n\n\n def validate(self, valid_dataloader, global_step):\n \"\"\"\n Runs evaluation on the validation set, computes the average loss,\n and logs the average validation loss along with the CTC decoded strings.\n \"\"\"\n self.model.eval()\n total_valid_loss = 0.0\n total_sec_error = 0.0\n count = 0\n # Iterate over the validation dataloader\n with torch.no_grad():\n for batch in valid_dataloader:\n # Inputs\n mel = batch['mel'].permute(0, 2, 1) # (B, L_mel, D)\n text = batch['text']\n\n if self.process_token_to_id:\n text_ids = list_str_to_idx(text, self.vocab_char_map).to(mel.device)","source_hash":"2d47b5df9155cbe6db309e3617bab7f7fbf0c03498c4a19ddf3e3229f1ea24c1","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer.validate","uri":"program://DMOSpeech2/function/src.duration_trainer.validate#L241-L331","kind":"function","name":"validate","path":"src/duration_trainer.py","language":"python","start_line":241,"end_line":331,"context_start_line":221,"context_end_line":351,"code":" if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n \n del checkpoint\n gc.collect()\n\n print(f'Exit load_checkpoint.')\n\n return step\n\n\n def validate(self, valid_dataloader, global_step):\n \"\"\"\n Runs evaluation on the validation set, computes the average loss,\n and logs the average validation loss along with the CTC decoded strings.\n \"\"\"\n self.model.eval()\n total_valid_loss = 0.0\n total_sec_error = 0.0\n count = 0\n # Iterate over the validation dataloader\n with torch.no_grad():\n for batch in valid_dataloader:\n # Inputs\n mel = batch['mel'].permute(0, 2, 1) # (B, L_mel, D)\n text = batch['text']\n\n if self.process_token_to_id:\n text_ids = list_str_to_idx(text, self.vocab_char_map).to(mel.device)\n text_ids = text_ids.masked_fill(text_ids==-1, self.vocab_size)\n else:\n text_ids = text\n\n # Targets\n mel_lengths = batch['mel_lengths']\n tar_lengths = calculate_remaining_lengths(mel_lengths)\n predictions = self.model(text_ids=text_ids, mel=mel)\n\n if self.loss_fn == 'L1':\n est_lengths = predictions\n loss = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n frame_error = loss\n\n elif self.loss_fn == 'CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_labels = torch.argmax(est_length_logtis, dim=-1)\n loss = masked_cross_entropy_loss(\n est_length_logits=est_length_logtis, tar_length_labels=tar_length_labels\n )\n est_lengths = est_length_labels * self.n_frame_per_class\n frame_error = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n\n elif self.loss_fn == 'L1_and_CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_1hots = F.gumbel_softmax(\n est_length_logtis, tau=self.gumbel_tau, hard=True, dim=-1\n )\n length_values = torch.arange(\n self.n_class, device=est_length_1hots.device\n ).float() * self.n_frame_per_class\n est_lengths = (est_length_1hots * length_values).sum(-1)\n\n loss_CE = masked_cross_entropy_loss(\n est_length_logits=est_length_logtis, tar_length_labels=tar_length_labels\n )\n\n loss_L1 = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n\n loss = loss_CE + self.lambda_L1 * loss_L1\n\n frame_error = loss_L1\n\n else:\n raise NotImplementedError(self.loss_fn)\n\n sec_error = frame_error * 256 / 24000\n total_sec_error += sec_error.item()\n total_valid_loss += loss.item()\n count += 1\n\n avg_valid_loss = total_valid_loss / count if count > 0 else 0.0\n avg_valid_sec_error = total_sec_error / count if count > 0 else 0.0\n # Log validation metrics\n self.accelerator.log(\n {\n f\"valid_loss\": avg_valid_loss,\n f\"valid_sec_error\": avg_valid_sec_error\n }, \n step=global_step\n )\n \n self.model.train()\n\n\n def train(self, train_dataset: Dataset, valid_dataset: Dataset,\n num_workers=64, resumable_with_seed: int = None):\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n # Create training dataloader using the appropriate batching strategy\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,","source_hash":"2d47b5df9155cbe6db309e3617bab7f7fbf0c03498c4a19ddf3e3229f1ea24c1","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_trainer.train","uri":"program://DMOSpeech2/function/src.duration_trainer.train#L334-L563","kind":"function","name":"train","path":"src/duration_trainer.py","language":"python","start_line":334,"end_line":563,"context_start_line":314,"context_end_line":563,"code":"\n sec_error = frame_error * 256 / 24000\n total_sec_error += sec_error.item()\n total_valid_loss += loss.item()\n count += 1\n\n avg_valid_loss = total_valid_loss / count if count > 0 else 0.0\n avg_valid_sec_error = total_sec_error / count if count > 0 else 0.0\n # Log validation metrics\n self.accelerator.log(\n {\n f\"valid_loss\": avg_valid_loss,\n f\"valid_sec_error\": avg_valid_sec_error\n }, \n step=global_step\n )\n \n self.model.train()\n\n\n def train(self, train_dataset: Dataset, valid_dataset: Dataset,\n num_workers=64, resumable_with_seed: int = None):\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n # Create training dataloader using the appropriate batching strategy\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )\n # Create validation dataloader (always sequential, no shuffling)\n valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n batch_size=self.batch_size,\n shuffle=False,\n )\n\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n\n sampler = SequentialSampler(valid_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n # Create validation dataloader (always sequential, no shuffling)\n valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True, \n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n \n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_steps = (\n self.num_warmup_updates * self.accelerator.num_processes\n ) # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_steps = len(train_dataloader) * self.epochs / self.grad_accumulation_steps\n decay_steps = total_steps - warmup_steps\n warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)\n decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)\n self.scheduler = SequentialLR(\n self.optimizer, schedulers=[warmup_scheduler, decay_scheduler], milestones=[warmup_steps]\n )\n train_dataloader, self.scheduler = self.accelerator.prepare(\n train_dataloader, self.scheduler\n ) # actual steps = 1 gpu steps / gpus\n start_step = self.load_checkpoint()\n global_step = start_step\n\n valid_dataloader = self.accelerator.prepare(valid_dataloader)\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar = tqdm(\n skipped_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n initial=skipped_batch,\n total=orig_epoch_step,\n )\n else:\n progress_bar = tqdm(\n train_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n )\n\n for batch in progress_bar:\n with self.accelerator.accumulate(self.model):\n # Inputs\n mel = batch['mel'].permute(0, 2, 1) # (B, L_mel, D)\n text = batch['text']\n\n if self.process_token_to_id:\n text_ids = list_str_to_idx(text, self.vocab_char_map).to(mel.device)\n text_ids = text_ids.masked_fill(text_ids==-1, self.vocab_size)\n else:\n text_ids = text\n\n # Targets\n mel_lengths = batch['mel_lengths']\n tar_lengths = calculate_remaining_lengths(mel_lengths)\n predictions = self.model(text_ids=text_ids, mel=mel)\n\n if self.loss_fn == 'L1':\n est_lengths = predictions\n loss = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n\n with torch.no_grad():\n frame_error = loss\n sec_error = frame_error * 256 / 24000\n\n log_dict = {\n 'loss': loss.item(), \n 'loss_L1': loss.item(), \n 'sec_error': sec_error.item(),\n 'lr': self.scheduler.get_last_lr()[0]\n }\n\n elif self.loss_fn == 'CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_labels = torch.argmax(est_length_logtis, dim=-1)\n loss = masked_cross_entropy_loss(\n est_length_logits=est_length_logtis, tar_length_labels=tar_length_labels\n )\n with torch.no_grad():\n est_lengths = est_length_labels * self.n_frame_per_class\n frame_error = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n sec_error = frame_error * 256 / 24000\n\n log_dict = {\n 'loss': loss.item(), \n 'loss_CE': loss.item(), \n 'sec_error': sec_error.item(),\n 'lr': self.scheduler.get_last_lr()[0]\n }\n\n elif self.loss_fn == 'L1_and_CE':\n tar_length_labels = (tar_lengths // self.n_frame_per_class) \\\n .clamp(min=0, max=self.n_class-1) # [0, 1, ..., n_class-1]\n est_length_logtis = predictions\n est_length_1hots = F.gumbel_softmax(\n est_length_logtis, tau=self.gumbel_tau, hard=True, dim=-1\n )\n length_values = torch.arange(\n self.n_class, device=est_length_1hots.device\n ).float() * self.n_frame_per_class\n est_lengths = (est_length_1hots * length_values).sum(-1)\n\n loss_CE = masked_cross_entropy_loss(\n est_length_logits=est_length_logtis, tar_length_labels=tar_length_labels\n )\n\n loss_L1 = masked_l1_loss(\n est_lengths=est_lengths, tar_lengths=tar_lengths\n )\n\n loss = loss_CE + self.lambda_L1 * loss_L1\n\n with torch.no_grad():\n frame_error = loss_L1\n sec_error = frame_error * 256 / 24000\n\n log_dict = {\n 'loss': loss.item(), \n 'loss_L1': loss_L1.item(), \n 'loss_CE': loss_CE.item(), \n 'sec_error': sec_error.item(),\n 'lr': self.scheduler.get_last_lr()[0]\n }\n\n else:\n raise NotImplementedError(self.loss_fn)\n\n \n self.accelerator.backward(loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n self.optimizer.step()\n self.scheduler.step()\n self.optimizer.zero_grad()\n\n global_step += 1\n\n if self.accelerator.is_local_main_process:\n self.accelerator.log(log_dict, step=global_step)\n progress_bar.set_postfix(step=str(global_step), loss=loss.item())\n\n if global_step % (self.save_per_updates * self.grad_accumulation_steps) == 0:\n self.save_checkpoint(global_step)\n # if self.log_samples and self.accelerator.is_local_main_process:\n # Run validation at the end of each epoch (only on the main process)\n if self.accelerator.is_local_main_process:\n self.validate(valid_dataloader, global_step)\n # if global_step % self.last_per_steps == 0:\n # self.save_checkpoint(global_step, last=True)\n\n self.save_checkpoint(global_step, last=True)\n self.accelerator.end_training()","source_hash":"2d47b5df9155cbe6db309e3617bab7f7fbf0c03498c4a19ddf3e3229f1ea24c1","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ecapa_tdnn","uri":"program://DMOSpeech2/module/src.ecapa_tdnn#L1-L268","kind":"module","name":"src.ecapa_tdnn","path":"src/ecapa_tdnn.py","language":"python","start_line":1,"end_line":268,"context_start_line":1,"context_end_line":268,"code":"# part of the code is borrowed from https://github.com/lawlict/ECAPA-TDNN\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torchaudio.transforms as trans\nfrom ctcmodel import ConformerCTC\n# from ctcmodel_nopool import ConformerCTC as ConformerCTCNoPool\nfrom pathlib import Path\n\n''' Res2Conv1d + BatchNorm1d + ReLU\n'''\n\n\nclass Res2Conv1dReluBn(nn.Module):\n '''\n in_channels == out_channels == channels\n '''\n\n def __init__(self, channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True, scale=4):\n super().__init__()\n assert channels % scale == 0, \"{} % {} != 0\".format(channels, scale)\n self.scale = scale\n self.width = channels // scale\n self.nums = scale if scale == 1 else scale - 1\n\n self.convs = []\n self.bns = []\n for i in range(self.nums):\n self.convs.append(nn.Conv1d(self.width, self.width, kernel_size, stride, padding, dilation, bias=bias))\n self.bns.append(nn.BatchNorm1d(self.width))\n self.convs = nn.ModuleList(self.convs)\n self.bns = nn.ModuleList(self.bns)\n\n def forward(self, x):\n out = []\n spx = torch.split(x, self.width, 1)\n for i in range(self.nums):\n if i == 0:\n sp = spx[i]\n else:\n sp = sp + spx[i]\n # Order: conv -> relu -> bn\n sp = self.convs[i](sp)\n sp = self.bns[i](F.relu(sp))\n out.append(sp)\n if self.scale != 1:\n out.append(spx[self.nums])\n out = torch.cat(out, dim=1)\n\n return out\n\n\n''' Conv1d + BatchNorm1d + ReLU\n'''\n\n\nclass Conv1dReluBn(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True):\n super().__init__()\n self.conv = nn.Conv1d(in_channels, out_channels, kernel_size, stride, padding, dilation, bias=bias)\n self.bn = nn.BatchNorm1d(out_channels)\n\n def forward(self, x):\n return self.bn(F.relu(self.conv(x)))\n\n\n''' The SE connection of 1D case.\n'''\n\n\nclass SE_Connect(nn.Module):\n def __init__(self, channels, se_bottleneck_dim=128):\n super().__init__()\n self.linear1 = nn.Linear(channels, se_bottleneck_dim)\n self.linear2 = nn.Linear(se_bottleneck_dim, channels)\n\n def forward(self, x):\n out = x.mean(dim=2)\n out = F.relu(self.linear1(out))\n out = torch.sigmoid(self.linear2(out))\n out = x * out.unsqueeze(2)\n\n return out\n\n\n''' SE-Res2Block of the ECAPA-TDNN architecture.\n'''\n\nclass SE_Res2Block(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size, stride, padding, dilation, scale, se_bottleneck_dim):\n super().__init__()\n self.Conv1dReluBn1 = Conv1dReluBn(in_channels, out_channels, kernel_size=1, stride=1, padding=0)\n self.Res2Conv1dReluBn = Res2Conv1dReluBn(out_channels, kernel_size, stride, padding, dilation, scale=scale)\n self.Conv1dReluBn2 = Conv1dReluBn(out_channels, out_channels, kernel_size=1, stride=1, padding=0)\n self.SE_Connect = SE_Connect(out_channels, se_bottleneck_dim)\n\n self.shortcut = None\n if in_channels != out_channels:\n self.shortcut = nn.Conv1d(\n in_channels=in_channels,\n out_channels=out_channels,\n kernel_size=1,\n )\n\n def forward(self, x):\n residual = x\n if self.shortcut:\n residual = self.shortcut(x)\n\n x = self.Conv1dReluBn1(x)\n x = self.Res2Conv1dReluBn(x)\n x = self.Conv1dReluBn2(x)\n x = self.SE_Connect(x)\n\n return x + residual\n\n\n''' Attentive weighted mean and standard deviation pooling.\n'''\n\nclass AttentiveStatsPool(nn.Module):\n def __init__(self, in_dim, attention_channels=128, global_context_att=False):\n super().__init__()\n self.global_context_att = global_context_att\n\n # Use Conv1d with stride == 1 rather than Linear, then we don't need to transpose inputs.\n if global_context_att:\n self.linear1 = nn.Conv1d(in_dim * 3, attention_channels, kernel_size=1) # equals W and b in the paper\n else:\n self.linear1 = nn.Conv1d(in_dim, attention_channels, kernel_size=1) # equals W and b in the paper\n self.linear2 = nn.Conv1d(attention_channels, in_dim, kernel_size=1) # equals V and k in the paper\n\n def forward(self, x):\n\n if self.global_context_att:\n context_mean = torch.mean(x, dim=-1, keepdim=True).expand_as(x)\n context_std = torch.sqrt(torch.var(x, dim=-1, keepdim=True) + 1e-10).expand_as(x)\n x_in = torch.cat((x, context_mean, context_std), dim=1)\n else:\n x_in = x\n\n # DON'T use ReLU here! In experiments, I find ReLU hard to converge.\n alpha = torch.tanh(self.linear1(x_in))\n # alpha = F.relu(self.linear1(x_in))\n alpha = torch.softmax(self.linear2(alpha), dim=2)\n mean = torch.sum(alpha * x, dim=2)\n residuals = torch.sum(alpha * (x ** 2), dim=2) - mean ** 2\n std = torch.sqrt(residuals.clamp(min=1e-9))\n return torch.cat([mean, std], dim=1)\n\n\nclass ECAPA_TDNN(nn.Module):\n def __init__(self, channels=512, emb_dim=512, \n global_context_att=False, use_fp16=True,\n ctc_cls=ConformerCTC,\n ctc_path='/data4/F5TTS/ckpts/F5TTS_norm_ASR_vocos_pinyin_Emilia_ZH_EN/model_last.pt',\n ctc_args={'vocab_size': 2545, 'mel_dim': 100, 'num_heads': 8, 'd_hid': 512, 'nlayers': 6},\n ctc_no_grad=False\n ):\n super().__init__()\n if ctc_path != None:\n ctc_path = Path(ctc_path)\n model = ctc_cls(**ctc_args)\n state_dict = torch.load(ctc_path, map_location='cpu')\n model.load_state_dict(state_dict['model_state_dict'])\n print(f\"Initialized pretrained ConformerCTC backbone from {ctc_path}.\")\n else:\n raise ValueError(ctc_path)\n\n self.ctc_model = model\n self.ctc_model.out.requires_grad_(False)\n \n if ctc_cls == ConformerCTC:\n self.feat_num = ctc_args['nlayers'] + 2 + 1\n # elif ctc_cls == ConformerCTCNoPool:\n # self.feat_num = ctc_args['nlayers'] + 1\n else:\n raise ValueError(ctc_cls)\n feat_dim = ctc_args['d_hid']\n\n self.emb_dim = emb_dim\n \n self.feature_weight = nn.Parameter(torch.zeros(self.feat_num))\n self.instance_norm = nn.InstanceNorm1d(feat_dim)\n\n # self.channels = [channels] * 4 + [channels * 3]\n self.channels = [channels] * 4 + [1536]\n\n self.layer1 = Conv1dReluBn(feat_dim, self.channels[0], kernel_size=5, padding=2)\n self.layer2 = SE_Res2Block(self.channels[0], self.channels[1], kernel_size=3, stride=1, padding=2, dilation=2, scale=8, se_bottleneck_dim=128)\n self.layer3 = SE_Res2Block(self.channels[1], self.channels[2], kernel_size=3, stride=1, padding=3, dilation=3, scale=8, se_bottleneck_dim=128)\n self.layer4 = SE_Res2Block(self.channels[2], self.channels[3], kernel_size=3, stride=1, padding=4, dilation=4, scale=8, se_bottleneck_dim=128)\n\n # self.conv = nn.Conv1d(self.channels[-1], self.channels[-1], kernel_size=1)\n cat_channels = channels * 3\n self.conv = nn.Conv1d(cat_channels, self.channels[-1], kernel_size=1)\n self.pooling = AttentiveStatsPool(self.channels[-1], attention_channels=128, global_context_att=global_context_att)\n self.bn = nn.BatchNorm1d(self.channels[-1] * 2)\n self.linear = nn.Linear(self.channels[-1] * 2, emb_dim)\n\n if ctc_no_grad:\n for param in self.ctc_model.parameters():\n param.requires_grad = False\n self.ctc_model = self.ctc_model.eval()\n else:\n self.ctc_model = self.ctc_model.train()\n self.ctc_no_grad = ctc_no_grad\n print('ctc_no_grad: ', self.ctc_no_grad)\n\n def forward(self, latent, input_lengths, return_asr=False):\n if self.ctc_no_grad:\n with torch.no_grad():\n asr, h = self.ctc_model(latent, input_lengths)\n else:\n asr, h = self.ctc_model(latent, input_lengths)\n \n x = torch.stack(h, dim=0)\n norm_weights = F.softmax(self.feature_weight, dim=-1).unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)\n x = (norm_weights * x).sum(dim=0)\n x = x + 1e-6\n # x = torch.transpose(x, 1, 2) + 1e-6\n \n x = self.instance_norm(x)\n # x = torch.transpose(x, 1, 2)\n\n out1 = self.layer1(x)\n out2 = self.layer2(out1)\n out3 = self.layer3(out2)\n out4 = self.layer4(out3)\n\n out = torch.cat([out2, out3, out4], dim=1)\n out = F.relu(self.conv(out))\n out = self.bn(self.pooling(out))\n out = self.linear(out)\n\n if return_asr:\n return out, asr\n return out\n\nif __name__ == \"__main__\":\n from diffspeech.ldm.model import DiT\n from diffspeech.data.collate import get_mask_from_lengths\n from diffspeech.tools.text.vocab import IPA\n\n bsz = 3\n\n # Sample ipa\n ipa_lens = torch.randint(10, 50, (bsz,)).cuda()\n ipa_mask = get_mask_from_lengths(ipa_lens).cuda()\n ipa = torch.randint(0, len(IPA.vocab), (bsz, ipa_mask.size(-1))).cuda()\n\n # Sample latent\n latent_lens = torch.randint(50, 250, (bsz,)).cuda()\n latent_mask = get_mask_from_lengths(latent_lens).cuda()\n latent = torch.randn(bsz, latent_mask.size(-1), 64).cuda()\n\n # Sample prompt\n prompt_mask = get_mask_from_lengths(\n (latent_lens * 0.25).long(), max_len=latent_mask.size(-1)\n ).cuda()\n prompt_latent = latent * prompt_mask.unsqueeze(-1)\n\n model = ECAPA_TDNN(emb_dim=512).cuda()\n\n emb = model(latent, latent_mask.sum(axis=-1))\n\n print(emb.shape)","source_hash":"2856f9a59c60f45ed4d23d46faef7d769bec4babf5bc0157bd1d32de30046a4f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ecapa_tdnn.Res2Conv1dReluBn","uri":"program://DMOSpeech2/class/src.ecapa_tdnn.Res2Conv1dReluBn#L15-L51","kind":"class","name":"Res2Conv1dReluBn","path":"src/ecapa_tdnn.py","language":"python","start_line":15,"end_line":51,"context_start_line":1,"context_end_line":71,"code":"# part of the code is borrowed from https://github.com/lawlict/ECAPA-TDNN\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torchaudio.transforms as trans\nfrom ctcmodel import ConformerCTC\n# from ctcmodel_nopool import ConformerCTC as ConformerCTCNoPool\nfrom pathlib import Path\n\n''' Res2Conv1d + BatchNorm1d + ReLU\n'''\n\n\nclass Res2Conv1dReluBn(nn.Module):\n '''\n in_channels == out_channels == channels\n '''\n\n def __init__(self, channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True, scale=4):\n super().__init__()\n assert channels % scale == 0, \"{} % {} != 0\".format(channels, scale)\n self.scale = scale\n self.width = channels // scale\n self.nums = scale if scale == 1 else scale - 1\n\n self.convs = []\n self.bns = []\n for i in range(self.nums):\n self.convs.append(nn.Conv1d(self.width, self.width, kernel_size, stride, padding, dilation, bias=bias))\n self.bns.append(nn.BatchNorm1d(self.width))\n self.convs = nn.ModuleList(self.convs)\n self.bns = nn.ModuleList(self.bns)\n\n def forward(self, x):\n out = []\n spx = torch.split(x, self.width, 1)\n for i in range(self.nums):\n if i == 0:\n sp = spx[i]\n else:\n sp = sp + spx[i]\n # Order: conv -> relu -> bn\n sp = self.convs[i](sp)\n sp = self.bns[i](F.relu(sp))\n out.append(sp)\n if self.scale != 1:\n out.append(spx[self.nums])\n out = torch.cat(out, dim=1)\n\n return out\n\n\n''' Conv1d + BatchNorm1d + ReLU\n'''\n\n\nclass Conv1dReluBn(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True):\n super().__init__()\n self.conv = nn.Conv1d(in_channels, out_channels, kernel_size, stride, padding, dilation, bias=bias)\n self.bn = nn.BatchNorm1d(out_channels)\n\n def forward(self, x):\n return self.bn(F.relu(self.conv(x)))\n\n\n''' The SE connection of 1D case.\n'''\n\n","source_hash":"2856f9a59c60f45ed4d23d46faef7d769bec4babf5bc0157bd1d32de30046a4f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ecapa_tdnn.Conv1dReluBn","uri":"program://DMOSpeech2/class/src.ecapa_tdnn.Conv1dReluBn#L58-L65","kind":"class","name":"Conv1dReluBn","path":"src/ecapa_tdnn.py","language":"python","start_line":58,"end_line":65,"context_start_line":38,"context_end_line":85,"code":" for i in range(self.nums):\n if i == 0:\n sp = spx[i]\n else:\n sp = sp + spx[i]\n # Order: conv -> relu -> bn\n sp = self.convs[i](sp)\n sp = self.bns[i](F.relu(sp))\n out.append(sp)\n if self.scale != 1:\n out.append(spx[self.nums])\n out = torch.cat(out, dim=1)\n\n return out\n\n\n''' Conv1d + BatchNorm1d + ReLU\n'''\n\n\nclass Conv1dReluBn(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True):\n super().__init__()\n self.conv = nn.Conv1d(in_channels, out_channels, kernel_size, stride, padding, dilation, bias=bias)\n self.bn = nn.BatchNorm1d(out_channels)\n\n def forward(self, x):\n return self.bn(F.relu(self.conv(x)))\n\n\n''' The SE connection of 1D case.\n'''\n\n\nclass SE_Connect(nn.Module):\n def __init__(self, channels, se_bottleneck_dim=128):\n super().__init__()\n self.linear1 = nn.Linear(channels, se_bottleneck_dim)\n self.linear2 = nn.Linear(se_bottleneck_dim, channels)\n\n def forward(self, x):\n out = x.mean(dim=2)\n out = F.relu(self.linear1(out))\n out = torch.sigmoid(self.linear2(out))\n out = x * out.unsqueeze(2)\n\n return out\n","source_hash":"2856f9a59c60f45ed4d23d46faef7d769bec4babf5bc0157bd1d32de30046a4f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ecapa_tdnn.SE_Connect","uri":"program://DMOSpeech2/class/src.ecapa_tdnn.SE_Connect#L72-L84","kind":"class","name":"SE_Connect","path":"src/ecapa_tdnn.py","language":"python","start_line":72,"end_line":84,"context_start_line":52,"context_end_line":104,"code":"\n\n''' Conv1d + BatchNorm1d + ReLU\n'''\n\n\nclass Conv1dReluBn(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True):\n super().__init__()\n self.conv = nn.Conv1d(in_channels, out_channels, kernel_size, stride, padding, dilation, bias=bias)\n self.bn = nn.BatchNorm1d(out_channels)\n\n def forward(self, x):\n return self.bn(F.relu(self.conv(x)))\n\n\n''' The SE connection of 1D case.\n'''\n\n\nclass SE_Connect(nn.Module):\n def __init__(self, channels, se_bottleneck_dim=128):\n super().__init__()\n self.linear1 = nn.Linear(channels, se_bottleneck_dim)\n self.linear2 = nn.Linear(se_bottleneck_dim, channels)\n\n def forward(self, x):\n out = x.mean(dim=2)\n out = F.relu(self.linear1(out))\n out = torch.sigmoid(self.linear2(out))\n out = x * out.unsqueeze(2)\n\n return out\n\n\n''' SE-Res2Block of the ECAPA-TDNN architecture.\n'''\n\nclass SE_Res2Block(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size, stride, padding, dilation, scale, se_bottleneck_dim):\n super().__init__()\n self.Conv1dReluBn1 = Conv1dReluBn(in_channels, out_channels, kernel_size=1, stride=1, padding=0)\n self.Res2Conv1dReluBn = Res2Conv1dReluBn(out_channels, kernel_size, stride, padding, dilation, scale=scale)\n self.Conv1dReluBn2 = Conv1dReluBn(out_channels, out_channels, kernel_size=1, stride=1, padding=0)\n self.SE_Connect = SE_Connect(out_channels, se_bottleneck_dim)\n\n self.shortcut = None\n if in_channels != out_channels:\n self.shortcut = nn.Conv1d(\n in_channels=in_channels,\n out_channels=out_channels,\n kernel_size=1,\n )","source_hash":"2856f9a59c60f45ed4d23d46faef7d769bec4babf5bc0157bd1d32de30046a4f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ecapa_tdnn.SE_Res2Block","uri":"program://DMOSpeech2/class/src.ecapa_tdnn.SE_Res2Block#L90-L116","kind":"class","name":"SE_Res2Block","path":"src/ecapa_tdnn.py","language":"python","start_line":90,"end_line":116,"context_start_line":70,"context_end_line":136,"code":"\n\nclass SE_Connect(nn.Module):\n def __init__(self, channels, se_bottleneck_dim=128):\n super().__init__()\n self.linear1 = nn.Linear(channels, se_bottleneck_dim)\n self.linear2 = nn.Linear(se_bottleneck_dim, channels)\n\n def forward(self, x):\n out = x.mean(dim=2)\n out = F.relu(self.linear1(out))\n out = torch.sigmoid(self.linear2(out))\n out = x * out.unsqueeze(2)\n\n return out\n\n\n''' SE-Res2Block of the ECAPA-TDNN architecture.\n'''\n\nclass SE_Res2Block(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size, stride, padding, dilation, scale, se_bottleneck_dim):\n super().__init__()\n self.Conv1dReluBn1 = Conv1dReluBn(in_channels, out_channels, kernel_size=1, stride=1, padding=0)\n self.Res2Conv1dReluBn = Res2Conv1dReluBn(out_channels, kernel_size, stride, padding, dilation, scale=scale)\n self.Conv1dReluBn2 = Conv1dReluBn(out_channels, out_channels, kernel_size=1, stride=1, padding=0)\n self.SE_Connect = SE_Connect(out_channels, se_bottleneck_dim)\n\n self.shortcut = None\n if in_channels != out_channels:\n self.shortcut = nn.Conv1d(\n in_channels=in_channels,\n out_channels=out_channels,\n kernel_size=1,\n )\n\n def forward(self, x):\n residual = x\n if self.shortcut:\n residual = self.shortcut(x)\n\n x = self.Conv1dReluBn1(x)\n x = self.Res2Conv1dReluBn(x)\n x = self.Conv1dReluBn2(x)\n x = self.SE_Connect(x)\n\n return x + residual\n\n\n''' Attentive weighted mean and standard deviation pooling.\n'''\n\nclass AttentiveStatsPool(nn.Module):\n def __init__(self, in_dim, attention_channels=128, global_context_att=False):\n super().__init__()\n self.global_context_att = global_context_att\n\n # Use Conv1d with stride == 1 rather than Linear, then we don't need to transpose inputs.\n if global_context_att:\n self.linear1 = nn.Conv1d(in_dim * 3, attention_channels, kernel_size=1) # equals W and b in the paper\n else:\n self.linear1 = nn.Conv1d(in_dim, attention_channels, kernel_size=1) # equals W and b in the paper\n self.linear2 = nn.Conv1d(attention_channels, in_dim, kernel_size=1) # equals V and k in the paper\n\n def forward(self, x):\n\n if self.global_context_att:","source_hash":"2856f9a59c60f45ed4d23d46faef7d769bec4babf5bc0157bd1d32de30046a4f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ecapa_tdnn.AttentiveStatsPool","uri":"program://DMOSpeech2/class/src.ecapa_tdnn.AttentiveStatsPool#L122-L150","kind":"class","name":"AttentiveStatsPool","path":"src/ecapa_tdnn.py","language":"python","start_line":122,"end_line":150,"context_start_line":102,"context_end_line":170,"code":" out_channels=out_channels,\n kernel_size=1,\n )\n\n def forward(self, x):\n residual = x\n if self.shortcut:\n residual = self.shortcut(x)\n\n x = self.Conv1dReluBn1(x)\n x = self.Res2Conv1dReluBn(x)\n x = self.Conv1dReluBn2(x)\n x = self.SE_Connect(x)\n\n return x + residual\n\n\n''' Attentive weighted mean and standard deviation pooling.\n'''\n\nclass AttentiveStatsPool(nn.Module):\n def __init__(self, in_dim, attention_channels=128, global_context_att=False):\n super().__init__()\n self.global_context_att = global_context_att\n\n # Use Conv1d with stride == 1 rather than Linear, then we don't need to transpose inputs.\n if global_context_att:\n self.linear1 = nn.Conv1d(in_dim * 3, attention_channels, kernel_size=1) # equals W and b in the paper\n else:\n self.linear1 = nn.Conv1d(in_dim, attention_channels, kernel_size=1) # equals W and b in the paper\n self.linear2 = nn.Conv1d(attention_channels, in_dim, kernel_size=1) # equals V and k in the paper\n\n def forward(self, x):\n\n if self.global_context_att:\n context_mean = torch.mean(x, dim=-1, keepdim=True).expand_as(x)\n context_std = torch.sqrt(torch.var(x, dim=-1, keepdim=True) + 1e-10).expand_as(x)\n x_in = torch.cat((x, context_mean, context_std), dim=1)\n else:\n x_in = x\n\n # DON'T use ReLU here! In experiments, I find ReLU hard to converge.\n alpha = torch.tanh(self.linear1(x_in))\n # alpha = F.relu(self.linear1(x_in))\n alpha = torch.softmax(self.linear2(alpha), dim=2)\n mean = torch.sum(alpha * x, dim=2)\n residuals = torch.sum(alpha * (x ** 2), dim=2) - mean ** 2\n std = torch.sqrt(residuals.clamp(min=1e-9))\n return torch.cat([mean, std], dim=1)\n\n\nclass ECAPA_TDNN(nn.Module):\n def __init__(self, channels=512, emb_dim=512, \n global_context_att=False, use_fp16=True,\n ctc_cls=ConformerCTC,\n ctc_path='/data4/F5TTS/ckpts/F5TTS_norm_ASR_vocos_pinyin_Emilia_ZH_EN/model_last.pt',\n ctc_args={'vocab_size': 2545, 'mel_dim': 100, 'num_heads': 8, 'd_hid': 512, 'nlayers': 6},\n ctc_no_grad=False\n ):\n super().__init__()\n if ctc_path != None:\n ctc_path = Path(ctc_path)\n model = ctc_cls(**ctc_args)\n state_dict = torch.load(ctc_path, map_location='cpu')\n model.load_state_dict(state_dict['model_state_dict'])\n print(f\"Initialized pretrained ConformerCTC backbone from {ctc_path}.\")\n else:\n raise ValueError(ctc_path)\n","source_hash":"2856f9a59c60f45ed4d23d46faef7d769bec4babf5bc0157bd1d32de30046a4f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ecapa_tdnn.ECAPA_TDNN","uri":"program://DMOSpeech2/class/src.ecapa_tdnn.ECAPA_TDNN#L153-L239","kind":"class","name":"ECAPA_TDNN","path":"src/ecapa_tdnn.py","language":"python","start_line":153,"end_line":239,"context_start_line":133,"context_end_line":259,"code":"\n def forward(self, x):\n\n if self.global_context_att:\n context_mean = torch.mean(x, dim=-1, keepdim=True).expand_as(x)\n context_std = torch.sqrt(torch.var(x, dim=-1, keepdim=True) + 1e-10).expand_as(x)\n x_in = torch.cat((x, context_mean, context_std), dim=1)\n else:\n x_in = x\n\n # DON'T use ReLU here! In experiments, I find ReLU hard to converge.\n alpha = torch.tanh(self.linear1(x_in))\n # alpha = F.relu(self.linear1(x_in))\n alpha = torch.softmax(self.linear2(alpha), dim=2)\n mean = torch.sum(alpha * x, dim=2)\n residuals = torch.sum(alpha * (x ** 2), dim=2) - mean ** 2\n std = torch.sqrt(residuals.clamp(min=1e-9))\n return torch.cat([mean, std], dim=1)\n\n\nclass ECAPA_TDNN(nn.Module):\n def __init__(self, channels=512, emb_dim=512, \n global_context_att=False, use_fp16=True,\n ctc_cls=ConformerCTC,\n ctc_path='/data4/F5TTS/ckpts/F5TTS_norm_ASR_vocos_pinyin_Emilia_ZH_EN/model_last.pt',\n ctc_args={'vocab_size': 2545, 'mel_dim': 100, 'num_heads': 8, 'd_hid': 512, 'nlayers': 6},\n ctc_no_grad=False\n ):\n super().__init__()\n if ctc_path != None:\n ctc_path = Path(ctc_path)\n model = ctc_cls(**ctc_args)\n state_dict = torch.load(ctc_path, map_location='cpu')\n model.load_state_dict(state_dict['model_state_dict'])\n print(f\"Initialized pretrained ConformerCTC backbone from {ctc_path}.\")\n else:\n raise ValueError(ctc_path)\n\n self.ctc_model = model\n self.ctc_model.out.requires_grad_(False)\n \n if ctc_cls == ConformerCTC:\n self.feat_num = ctc_args['nlayers'] + 2 + 1\n # elif ctc_cls == ConformerCTCNoPool:\n # self.feat_num = ctc_args['nlayers'] + 1\n else:\n raise ValueError(ctc_cls)\n feat_dim = ctc_args['d_hid']\n\n self.emb_dim = emb_dim\n \n self.feature_weight = nn.Parameter(torch.zeros(self.feat_num))\n self.instance_norm = nn.InstanceNorm1d(feat_dim)\n\n # self.channels = [channels] * 4 + [channels * 3]\n self.channels = [channels] * 4 + [1536]\n\n self.layer1 = Conv1dReluBn(feat_dim, self.channels[0], kernel_size=5, padding=2)\n self.layer2 = SE_Res2Block(self.channels[0], self.channels[1], kernel_size=3, stride=1, padding=2, dilation=2, scale=8, se_bottleneck_dim=128)\n self.layer3 = SE_Res2Block(self.channels[1], self.channels[2], kernel_size=3, stride=1, padding=3, dilation=3, scale=8, se_bottleneck_dim=128)\n self.layer4 = SE_Res2Block(self.channels[2], self.channels[3], kernel_size=3, stride=1, padding=4, dilation=4, scale=8, se_bottleneck_dim=128)\n\n # self.conv = nn.Conv1d(self.channels[-1], self.channels[-1], kernel_size=1)\n cat_channels = channels * 3\n self.conv = nn.Conv1d(cat_channels, self.channels[-1], kernel_size=1)\n self.pooling = AttentiveStatsPool(self.channels[-1], attention_channels=128, global_context_att=global_context_att)\n self.bn = nn.BatchNorm1d(self.channels[-1] * 2)\n self.linear = nn.Linear(self.channels[-1] * 2, emb_dim)\n\n if ctc_no_grad:\n for param in self.ctc_model.parameters():\n param.requires_grad = False\n self.ctc_model = self.ctc_model.eval()\n else:\n self.ctc_model = self.ctc_model.train()\n self.ctc_no_grad = ctc_no_grad\n print('ctc_no_grad: ', self.ctc_no_grad)\n\n def forward(self, latent, input_lengths, return_asr=False):\n if self.ctc_no_grad:\n with torch.no_grad():\n asr, h = self.ctc_model(latent, input_lengths)\n else:\n asr, h = self.ctc_model(latent, input_lengths)\n \n x = torch.stack(h, dim=0)\n norm_weights = F.softmax(self.feature_weight, dim=-1).unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)\n x = (norm_weights * x).sum(dim=0)\n x = x + 1e-6\n # x = torch.transpose(x, 1, 2) + 1e-6\n \n x = self.instance_norm(x)\n # x = torch.transpose(x, 1, 2)\n\n out1 = self.layer1(x)\n out2 = self.layer2(out1)\n out3 = self.layer3(out2)\n out4 = self.layer4(out3)\n\n out = torch.cat([out2, out3, out4], dim=1)\n out = F.relu(self.conv(out))\n out = self.bn(self.pooling(out))\n out = self.linear(out)\n\n if return_asr:\n return out, asr\n return out\n\nif __name__ == \"__main__\":\n from diffspeech.ldm.model import DiT\n from diffspeech.data.collate import get_mask_from_lengths\n from diffspeech.tools.text.vocab import IPA\n\n bsz = 3\n\n # Sample ipa\n ipa_lens = torch.randint(10, 50, (bsz,)).cuda()\n ipa_mask = get_mask_from_lengths(ipa_lens).cuda()\n ipa = torch.randint(0, len(IPA.vocab), (bsz, ipa_mask.size(-1))).cuda()\n\n # Sample latent\n latent_lens = torch.randint(50, 250, (bsz,)).cuda()\n latent_mask = get_mask_from_lengths(latent_lens).cuda()\n latent = torch.randn(bsz, latent_mask.size(-1), 64).cuda()\n\n # Sample prompt\n prompt_mask = get_mask_from_lengths(","source_hash":"2856f9a59c60f45ed4d23d46faef7d769bec4babf5bc0157bd1d32de30046a4f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ecapa_tdnn.__init__","uri":"program://DMOSpeech2/function/src.ecapa_tdnn.__init__#L154-L209","kind":"function","name":"__init__","path":"src/ecapa_tdnn.py","language":"python","start_line":154,"end_line":209,"context_start_line":134,"context_end_line":229,"code":" def forward(self, x):\n\n if self.global_context_att:\n context_mean = torch.mean(x, dim=-1, keepdim=True).expand_as(x)\n context_std = torch.sqrt(torch.var(x, dim=-1, keepdim=True) + 1e-10).expand_as(x)\n x_in = torch.cat((x, context_mean, context_std), dim=1)\n else:\n x_in = x\n\n # DON'T use ReLU here! In experiments, I find ReLU hard to converge.\n alpha = torch.tanh(self.linear1(x_in))\n # alpha = F.relu(self.linear1(x_in))\n alpha = torch.softmax(self.linear2(alpha), dim=2)\n mean = torch.sum(alpha * x, dim=2)\n residuals = torch.sum(alpha * (x ** 2), dim=2) - mean ** 2\n std = torch.sqrt(residuals.clamp(min=1e-9))\n return torch.cat([mean, std], dim=1)\n\n\nclass ECAPA_TDNN(nn.Module):\n def __init__(self, channels=512, emb_dim=512, \n global_context_att=False, use_fp16=True,\n ctc_cls=ConformerCTC,\n ctc_path='/data4/F5TTS/ckpts/F5TTS_norm_ASR_vocos_pinyin_Emilia_ZH_EN/model_last.pt',\n ctc_args={'vocab_size': 2545, 'mel_dim': 100, 'num_heads': 8, 'd_hid': 512, 'nlayers': 6},\n ctc_no_grad=False\n ):\n super().__init__()\n if ctc_path != None:\n ctc_path = Path(ctc_path)\n model = ctc_cls(**ctc_args)\n state_dict = torch.load(ctc_path, map_location='cpu')\n model.load_state_dict(state_dict['model_state_dict'])\n print(f\"Initialized pretrained ConformerCTC backbone from {ctc_path}.\")\n else:\n raise ValueError(ctc_path)\n\n self.ctc_model = model\n self.ctc_model.out.requires_grad_(False)\n \n if ctc_cls == ConformerCTC:\n self.feat_num = ctc_args['nlayers'] + 2 + 1\n # elif ctc_cls == ConformerCTCNoPool:\n # self.feat_num = ctc_args['nlayers'] + 1\n else:\n raise ValueError(ctc_cls)\n feat_dim = ctc_args['d_hid']\n\n self.emb_dim = emb_dim\n \n self.feature_weight = nn.Parameter(torch.zeros(self.feat_num))\n self.instance_norm = nn.InstanceNorm1d(feat_dim)\n\n # self.channels = [channels] * 4 + [channels * 3]\n self.channels = [channels] * 4 + [1536]\n\n self.layer1 = Conv1dReluBn(feat_dim, self.channels[0], kernel_size=5, padding=2)\n self.layer2 = SE_Res2Block(self.channels[0], self.channels[1], kernel_size=3, stride=1, padding=2, dilation=2, scale=8, se_bottleneck_dim=128)\n self.layer3 = SE_Res2Block(self.channels[1], self.channels[2], kernel_size=3, stride=1, padding=3, dilation=3, scale=8, se_bottleneck_dim=128)\n self.layer4 = SE_Res2Block(self.channels[2], self.channels[3], kernel_size=3, stride=1, padding=4, dilation=4, scale=8, se_bottleneck_dim=128)\n\n # self.conv = nn.Conv1d(self.channels[-1], self.channels[-1], kernel_size=1)\n cat_channels = channels * 3\n self.conv = nn.Conv1d(cat_channels, self.channels[-1], kernel_size=1)\n self.pooling = AttentiveStatsPool(self.channels[-1], attention_channels=128, global_context_att=global_context_att)\n self.bn = nn.BatchNorm1d(self.channels[-1] * 2)\n self.linear = nn.Linear(self.channels[-1] * 2, emb_dim)\n\n if ctc_no_grad:\n for param in self.ctc_model.parameters():\n param.requires_grad = False\n self.ctc_model = self.ctc_model.eval()\n else:\n self.ctc_model = self.ctc_model.train()\n self.ctc_no_grad = ctc_no_grad\n print('ctc_no_grad: ', self.ctc_no_grad)\n\n def forward(self, latent, input_lengths, return_asr=False):\n if self.ctc_no_grad:\n with torch.no_grad():\n asr, h = self.ctc_model(latent, input_lengths)\n else:\n asr, h = self.ctc_model(latent, input_lengths)\n \n x = torch.stack(h, dim=0)\n norm_weights = F.softmax(self.feature_weight, dim=-1).unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)\n x = (norm_weights * x).sum(dim=0)\n x = x + 1e-6\n # x = torch.transpose(x, 1, 2) + 1e-6\n \n x = self.instance_norm(x)\n # x = torch.transpose(x, 1, 2)\n\n out1 = self.layer1(x)\n out2 = self.layer2(out1)\n out3 = self.layer3(out2)","source_hash":"2856f9a59c60f45ed4d23d46faef7d769bec4babf5bc0157bd1d32de30046a4f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ecapa_tdnn.forward","uri":"program://DMOSpeech2/function/src.ecapa_tdnn.forward#L211-L239","kind":"function","name":"forward","path":"src/ecapa_tdnn.py","language":"python","start_line":211,"end_line":239,"context_start_line":191,"context_end_line":259,"code":" self.layer2 = SE_Res2Block(self.channels[0], self.channels[1], kernel_size=3, stride=1, padding=2, dilation=2, scale=8, se_bottleneck_dim=128)\n self.layer3 = SE_Res2Block(self.channels[1], self.channels[2], kernel_size=3, stride=1, padding=3, dilation=3, scale=8, se_bottleneck_dim=128)\n self.layer4 = SE_Res2Block(self.channels[2], self.channels[3], kernel_size=3, stride=1, padding=4, dilation=4, scale=8, se_bottleneck_dim=128)\n\n # self.conv = nn.Conv1d(self.channels[-1], self.channels[-1], kernel_size=1)\n cat_channels = channels * 3\n self.conv = nn.Conv1d(cat_channels, self.channels[-1], kernel_size=1)\n self.pooling = AttentiveStatsPool(self.channels[-1], attention_channels=128, global_context_att=global_context_att)\n self.bn = nn.BatchNorm1d(self.channels[-1] * 2)\n self.linear = nn.Linear(self.channels[-1] * 2, emb_dim)\n\n if ctc_no_grad:\n for param in self.ctc_model.parameters():\n param.requires_grad = False\n self.ctc_model = self.ctc_model.eval()\n else:\n self.ctc_model = self.ctc_model.train()\n self.ctc_no_grad = ctc_no_grad\n print('ctc_no_grad: ', self.ctc_no_grad)\n\n def forward(self, latent, input_lengths, return_asr=False):\n if self.ctc_no_grad:\n with torch.no_grad():\n asr, h = self.ctc_model(latent, input_lengths)\n else:\n asr, h = self.ctc_model(latent, input_lengths)\n \n x = torch.stack(h, dim=0)\n norm_weights = F.softmax(self.feature_weight, dim=-1).unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)\n x = (norm_weights * x).sum(dim=0)\n x = x + 1e-6\n # x = torch.transpose(x, 1, 2) + 1e-6\n \n x = self.instance_norm(x)\n # x = torch.transpose(x, 1, 2)\n\n out1 = self.layer1(x)\n out2 = self.layer2(out1)\n out3 = self.layer3(out2)\n out4 = self.layer4(out3)\n\n out = torch.cat([out2, out3, out4], dim=1)\n out = F.relu(self.conv(out))\n out = self.bn(self.pooling(out))\n out = self.linear(out)\n\n if return_asr:\n return out, asr\n return out\n\nif __name__ == \"__main__\":\n from diffspeech.ldm.model import DiT\n from diffspeech.data.collate import get_mask_from_lengths\n from diffspeech.tools.text.vocab import IPA\n\n bsz = 3\n\n # Sample ipa\n ipa_lens = torch.randint(10, 50, (bsz,)).cuda()\n ipa_mask = get_mask_from_lengths(ipa_lens).cuda()\n ipa = torch.randint(0, len(IPA.vocab), (bsz, ipa_mask.size(-1))).cuda()\n\n # Sample latent\n latent_lens = torch.randint(50, 250, (bsz,)).cuda()\n latent_mask = get_mask_from_lengths(latent_lens).cuda()\n latent = torch.randn(bsz, latent_mask.size(-1), 64).cuda()\n\n # Sample prompt\n prompt_mask = get_mask_from_lengths(","source_hash":"2856f9a59c60f45ed4d23d46faef7d769bec4babf5bc0157bd1d32de30046a4f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.unimodel","uri":"program://DMOSpeech2/module/src.unimodel#L1-L327","kind":"module","name":"src.unimodel","path":"src/unimodel.py","language":"python","start_line":1,"end_line":327,"context_start_line":1,"context_end_line":327,"code":"from __future__ import annotations\nfrom typing import Callable\nfrom random import random\n\nimport contextlib\n\nfrom torch import nn\nimport torch \nimport copy\nimport os\n\nfrom f5_tts.model import DiT, UNetT\nfrom pathlib import Path\nfrom guidance_model import Guidance\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n sample_consecutive_steps,\n sample_from_list,\n)\n\nclass UniModel(nn.Module):\n def __init__(self, \n model: DiT, # teacher model (dit model)\n checkpoint_path: str = \"\",\n second_time: bool = True,\n use_fp16: bool = True,\n real_guidance_scale: float = 2.0, \n fake_guidance_scale: float = 0.0, \n gen_cls_loss: bool = False,\n sway_coeff: float = -1.0,\n vocab_char_map: dict[str, int] | None = None,\n frac_lengths_mask: tuple[float, float] = (0.7, 1.0)):\n \n super().__init__()\n \n if checkpoint_path != \"\":\n if \"model_last.pt\" in os.listdir(checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n checkpoint = torch.load(f\"{checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n if \"scale\" in checkpoint:\n self.scale = checkpoint[\"scale\"]\n else:\n self.scale = 1.0\n print(f\"Loaded teacher model with scale: {self.scale}\")\n\n if \"step\" in checkpoint:\n state = checkpoint[\"model_state_dict\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n state = checkpoint[\"model_state_dict\"]\n\n # only load the DiT module\n filtered_state_dict = {\n k.replace(\"transformer.\", \"\"): v\n for k, v in state.items()\n if k.startswith(\"transformer.\")\n }\n\n model.load_state_dict(filtered_state_dict, strict=False)\n else:\n self.scale = 1.0\n \n real_unet = copy.deepcopy(model)\n real_unet.time_embed2 = None\n \n fake_unet = copy.deepcopy(model)\n \n # Instantiate Guidance, which internally uses real_unet and fake_unet initialized from the teacher\n self.guidance_model = Guidance(\n real_unet=real_unet,\n fake_unet=fake_unet,\n use_fp16=use_fp16,\n real_guidance_scale=real_guidance_scale,\n fake_guidance_scale=fake_guidance_scale,\n gen_cls_loss=gen_cls_loss,\n sway_coeff=sway_coeff,\n )\n \n self.feedforward_model = copy.deepcopy(model) # initialize the student model\n self.feedforward_model.requires_grad_(True)\n self.feedforward_model.time_embed2 = None\n\n self.vocab_char_map = vocab_char_map\n self.frac_lengths_mask = frac_lengths_mask\n \n self.second_time = second_time # fake_unet.time_embed2 is not None\n\n def forward(self,\n inp: float[\"b n d\"], # mel\n text: int[\"b nt\"] | list[str],\n *,\n lens: int[\"b\"] | None = None,\n student_steps: list[int] = [0, 0.25, 0.5, 0.75],\n update_generator: bool = False,\n ):\n \"\"\"\n Forward pass that routes to either generator_forward or guidance_forward\n in the Guidance class, depending on the arguments.\n\n Parameters:\n -----------\n generator_turn: bool\n If True, run the generator forward pass (distribution matching loss, etc.)\n guidance_turn: bool\n If True, run the guidance forward pass (fake loss, cls loss, etc.)\n data_dict: dict\n Input dictionary containing the necessary keys for the forward passes.\n Expected keys may include:\n \"inp\": Tensor (B, N, D) - input mel or latent\n \"text\": Tensor or list[str] - text input\n \"rand_span_mask\": Tensor (B, N) - boolean mask\n \"real_data\": dict with keys like:\n \"inp\", \"text\", \"rand_span_mask\"\n \n Returns:\n --------\n loss_dict: dict[str, Tensor]\n Dictionary of losses.\n log_dict: dict[str, Tensor or float]\n Dictionary of logging tensors or values.\n \"\"\"\n \n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n\n # handle text as string\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # lens and mask\n if not exists(lens):\n lens = torch.full((batch,), seq_len, device=device)\n\n mask = lens_to_mask(lens, length=seq_len) # useless here, as collate_fn will pad to max length in batch\n\n # sample from the list of student steps\n time = sample_from_list(student_steps, batch).to(device)\n c_time, p_time = sample_consecutive_steps(student_steps)\n time = torch.ones_like(time) * c_time\n p_time = torch.ones_like(time) * p_time\n\n frac_lengths = torch.zeros((batch,), device=device).float().uniform_(*self.frac_lengths_mask)\n rand_span_mask = mask_from_frac_lengths(lens, frac_lengths)\n \n if exists(mask):\n rand_span_mask &= mask\n \n\n # # use generated output from previous step as input\n with torch.no_grad():\n x1 = inp\n x0 = torch.randn_like(x1)\n t = p_time.unsqueeze(-1).unsqueeze(-1)\n phi = (1 - t) * x0 + t * x1\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n \n pred = self.feedforward_model(\n x=phi, \n cond=cond,\n text=text, \n time=p_time, \n drop_audio_cond=False, \n drop_text=False # make sure the cfg=1\n ) # flow prediction\n \n # predicted mel spectrogram\n output = phi + (1 - t) * pred \n output[~rand_span_mask] = inp[~rand_span_mask]\n \n # forward diffusion\n x1 = output\n x0 = torch.randn_like(x1)\n t = time.unsqueeze(-1).unsqueeze(-1)\n phi = (1 - t) * x0 + t * x1\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n \n with torch.no_grad() if not update_generator else contextlib.nullcontext():\n pred = self.feedforward_model(\n x=phi, \n cond=cond,\n text=text, \n time=time, \n drop_audio_cond=False, \n drop_text=False # make sure no cfg is used \n )\n \n # predicted mel spectrogram\n output = phi + (1 - t) * pred\n output[~rand_span_mask] = inp[~rand_span_mask]\n \n if update_generator:\n generator_data_dict = {\n \"inp\": output,\n \"text\": text,\n \"rand_span_mask\": rand_span_mask,\n \"second_time\": time if self.second_time else None,\n \"mse_loss\": time.mean() == student_steps[-1].mean(),\n \"real_data\": {\n \"inp\": inp,\n \"text\": text,\n \"rand_span_mask\": rand_span_mask\n }\n }\n \n # avoid any side effects of gradient accumulation\n # self.guidance_model.requires_grad_(False)\n # self.feedforward_model.requires_grad_(True)\n generator_loss_dict, generator_log_dict = self.guidance_model(\n generator_turn=True,\n guidance_turn=False,\n generator_data_dict=generator_data_dict,\n guidance_data_dict=None\n )\n \n generator_log_dict['ground_truth'] = x1\n generator_log_dict['generator_input'] = phi\n generator_log_dict['generator_output'] = output\n generator_log_dict['generator_cond'] = cond\n generator_log_dict['time'] = time\n \n return generator_loss_dict, generator_log_dict\n else:\n guidance_data_dict = {\n \"inp\": output.detach(),\n \"text\": text,\n \"rand_span_mask\": rand_span_mask,\n \"second_time\": time if self.second_time else None,\n \"real_data\": {\n \"inp\": inp,\n \"text\": text,\n \"rand_span_mask\": rand_span_mask\n }\n }\n \n # avoid any side effects of gradient accumulation\n # self.feedforward_model.requires_grad_(False)\n # self.guidance_model.requires_grad_(True)\n guidance_loss_dict, guidance_log_dict = self.guidance_model(\n generator_turn=False,\n guidance_turn=True,\n generator_data_dict=None,\n guidance_data_dict=guidance_data_dict\n )\n # self.feedforward_model.requires_grad_(True)\n \n return guidance_loss_dict, guidance_log_dict\n \n # return guidance_loss_dict, guidance_log_dict, generator_loss_dict, generator_log_dict\n \n\nif __name__ == \"__main__\":\n \n from f5_tts.model.utils import get_tokenizer\n from torch.utils.data import DataLoader, Dataset, SequentialSampler\n from f5_tts.model.dataset import load_dataset \n from f5_tts.model.dataset import DynamicBatchSampler, collate_fn\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n\n dit = DiT(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4, text_num_embeds=vocab_size, mel_dim=100)\n \n model = UniModel(dit, \n checkpoint_path=\"/data4/F5TTS/ckpts/F5TTS_Base_norm_flow_8GPU_vocos_pinyin_Emilia_ZH_EN\",\n gen_cls_loss=True,\n vocab_char_map=vocab_char_map,\n frac_lengths_mask=(0.7, 1.0)\n ).cuda()\n \n # batch = next(iter(train_dataloader))\n # torch.save(batch, \"batch.pt\")\n batch = torch.load(\"batch.pt\")\n inp, text, lens = batch[\"mel\"].permute(0, 2, 1).cuda(), batch[\"text\"], batch[\"mel_lengths\"].cuda() \n\n \n # text = [\"hello world\"] * bsz\n # lens = torch.randint(1, 1000, (bsz,)).cuda()\n # inp = torch.randn(bsz, lens.max(), 100).cuda()\n with torch.autocast(device_type=\"cuda\", dtype=torch.float16):\n num_student_step = 4\n\n guidance_loss_dict, guidance_log_dict = model(inp, text, lens=lens, update_generator=False, student_steps=(torch.linspace(0.0, 1.0, num_student_step + 1)[:-1]))\n\n generator_loss_dict, generator_log_dict = model(inp, text, lens=lens, update_generator=True, student_steps=(torch.linspace(0.0, 1.0, num_student_step + 1)[:-1]))\n \n print(guidance_loss_dict)\n print(generator_loss_dict)\n \n guidance_loss = 0\n guidance_loss += guidance_loss_dict[\"loss_fake_mean\"]\n guidance_loss += guidance_loss_dict[\"guidance_cls_loss\"]\n\n generator_loss = 0\n generator_loss += generator_loss_dict[\"loss_dm\"]\n generator_loss += generator_loss_dict[\"loss_ctc\"]\n generator_loss += generator_loss_dict[\"loss_sim\"]\n generator_loss += generator_loss_dict[\"gen_cls_loss\"]\n generator_loss += generator_loss_dict[\"loss_mse\"]\n\n guidance_loss.backward()\n generator_loss.backward()","source_hash":"d0a2447ed4e6999217aa5d1177a1def4c9688256e5ff0c12fe4a7cd040b19fbd","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.unimodel.UniModel","uri":"program://DMOSpeech2/class/src.unimodel.UniModel#L26-L264","kind":"class","name":"UniModel","path":"src/unimodel.py","language":"python","start_line":26,"end_line":264,"context_start_line":6,"context_end_line":284,"code":"\nfrom torch import nn\nimport torch \nimport copy\nimport os\n\nfrom f5_tts.model import DiT, UNetT\nfrom pathlib import Path\nfrom guidance_model import Guidance\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n sample_consecutive_steps,\n sample_from_list,\n)\n\nclass UniModel(nn.Module):\n def __init__(self, \n model: DiT, # teacher model (dit model)\n checkpoint_path: str = \"\",\n second_time: bool = True,\n use_fp16: bool = True,\n real_guidance_scale: float = 2.0, \n fake_guidance_scale: float = 0.0, \n gen_cls_loss: bool = False,\n sway_coeff: float = -1.0,\n vocab_char_map: dict[str, int] | None = None,\n frac_lengths_mask: tuple[float, float] = (0.7, 1.0)):\n \n super().__init__()\n \n if checkpoint_path != \"\":\n if \"model_last.pt\" in os.listdir(checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n checkpoint = torch.load(f\"{checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n if \"scale\" in checkpoint:\n self.scale = checkpoint[\"scale\"]\n else:\n self.scale = 1.0\n print(f\"Loaded teacher model with scale: {self.scale}\")\n\n if \"step\" in checkpoint:\n state = checkpoint[\"model_state_dict\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n state = checkpoint[\"model_state_dict\"]\n\n # only load the DiT module\n filtered_state_dict = {\n k.replace(\"transformer.\", \"\"): v\n for k, v in state.items()\n if k.startswith(\"transformer.\")\n }\n\n model.load_state_dict(filtered_state_dict, strict=False)\n else:\n self.scale = 1.0\n \n real_unet = copy.deepcopy(model)\n real_unet.time_embed2 = None\n \n fake_unet = copy.deepcopy(model)\n \n # Instantiate Guidance, which internally uses real_unet and fake_unet initialized from the teacher\n self.guidance_model = Guidance(\n real_unet=real_unet,\n fake_unet=fake_unet,\n use_fp16=use_fp16,\n real_guidance_scale=real_guidance_scale,\n fake_guidance_scale=fake_guidance_scale,\n gen_cls_loss=gen_cls_loss,\n sway_coeff=sway_coeff,\n )\n \n self.feedforward_model = copy.deepcopy(model) # initialize the student model\n self.feedforward_model.requires_grad_(True)\n self.feedforward_model.time_embed2 = None\n\n self.vocab_char_map = vocab_char_map\n self.frac_lengths_mask = frac_lengths_mask\n \n self.second_time = second_time # fake_unet.time_embed2 is not None\n\n def forward(self,\n inp: float[\"b n d\"], # mel\n text: int[\"b nt\"] | list[str],\n *,\n lens: int[\"b\"] | None = None,\n student_steps: list[int] = [0, 0.25, 0.5, 0.75],\n update_generator: bool = False,\n ):\n \"\"\"\n Forward pass that routes to either generator_forward or guidance_forward\n in the Guidance class, depending on the arguments.\n\n Parameters:\n -----------\n generator_turn: bool\n If True, run the generator forward pass (distribution matching loss, etc.)\n guidance_turn: bool\n If True, run the guidance forward pass (fake loss, cls loss, etc.)\n data_dict: dict\n Input dictionary containing the necessary keys for the forward passes.\n Expected keys may include:\n \"inp\": Tensor (B, N, D) - input mel or latent\n \"text\": Tensor or list[str] - text input\n \"rand_span_mask\": Tensor (B, N) - boolean mask\n \"real_data\": dict with keys like:\n \"inp\", \"text\", \"rand_span_mask\"\n \n Returns:\n --------\n loss_dict: dict[str, Tensor]\n Dictionary of losses.\n log_dict: dict[str, Tensor or float]\n Dictionary of logging tensors or values.\n \"\"\"\n \n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n\n # handle text as string\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # lens and mask\n if not exists(lens):\n lens = torch.full((batch,), seq_len, device=device)\n\n mask = lens_to_mask(lens, length=seq_len) # useless here, as collate_fn will pad to max length in batch\n\n # sample from the list of student steps\n time = sample_from_list(student_steps, batch).to(device)\n c_time, p_time = sample_consecutive_steps(student_steps)\n time = torch.ones_like(time) * c_time\n p_time = torch.ones_like(time) * p_time\n\n frac_lengths = torch.zeros((batch,), device=device).float().uniform_(*self.frac_lengths_mask)\n rand_span_mask = mask_from_frac_lengths(lens, frac_lengths)\n \n if exists(mask):\n rand_span_mask &= mask\n \n\n # # use generated output from previous step as input\n with torch.no_grad():\n x1 = inp\n x0 = torch.randn_like(x1)\n t = p_time.unsqueeze(-1).unsqueeze(-1)\n phi = (1 - t) * x0 + t * x1\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n \n pred = self.feedforward_model(\n x=phi, \n cond=cond,\n text=text, \n time=p_time, \n drop_audio_cond=False, \n drop_text=False # make sure the cfg=1\n ) # flow prediction\n \n # predicted mel spectrogram\n output = phi + (1 - t) * pred \n output[~rand_span_mask] = inp[~rand_span_mask]\n \n # forward diffusion\n x1 = output\n x0 = torch.randn_like(x1)\n t = time.unsqueeze(-1).unsqueeze(-1)\n phi = (1 - t) * x0 + t * x1\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n \n with torch.no_grad() if not update_generator else contextlib.nullcontext():\n pred = self.feedforward_model(\n x=phi, \n cond=cond,\n text=text, \n time=time, \n drop_audio_cond=False, \n drop_text=False # make sure no cfg is used \n )\n \n # predicted mel spectrogram\n output = phi + (1 - t) * pred\n output[~rand_span_mask] = inp[~rand_span_mask]\n \n if update_generator:\n generator_data_dict = {\n \"inp\": output,\n \"text\": text,\n \"rand_span_mask\": rand_span_mask,\n \"second_time\": time if self.second_time else None,\n \"mse_loss\": time.mean() == student_steps[-1].mean(),\n \"real_data\": {\n \"inp\": inp,\n \"text\": text,\n \"rand_span_mask\": rand_span_mask\n }\n }\n \n # avoid any side effects of gradient accumulation\n # self.guidance_model.requires_grad_(False)\n # self.feedforward_model.requires_grad_(True)\n generator_loss_dict, generator_log_dict = self.guidance_model(\n generator_turn=True,\n guidance_turn=False,\n generator_data_dict=generator_data_dict,\n guidance_data_dict=None\n )\n \n generator_log_dict['ground_truth'] = x1\n generator_log_dict['generator_input'] = phi\n generator_log_dict['generator_output'] = output\n generator_log_dict['generator_cond'] = cond\n generator_log_dict['time'] = time\n \n return generator_loss_dict, generator_log_dict\n else:\n guidance_data_dict = {\n \"inp\": output.detach(),\n \"text\": text,\n \"rand_span_mask\": rand_span_mask,\n \"second_time\": time if self.second_time else None,\n \"real_data\": {\n \"inp\": inp,\n \"text\": text,\n \"rand_span_mask\": rand_span_mask\n }\n }\n \n # avoid any side effects of gradient accumulation\n # self.feedforward_model.requires_grad_(False)\n # self.guidance_model.requires_grad_(True)\n guidance_loss_dict, guidance_log_dict = self.guidance_model(\n generator_turn=False,\n guidance_turn=True,\n generator_data_dict=None,\n guidance_data_dict=guidance_data_dict\n )\n # self.feedforward_model.requires_grad_(True)\n \n return guidance_loss_dict, guidance_log_dict\n \n # return guidance_loss_dict, guidance_log_dict, generator_loss_dict, generator_log_dict\n \n\nif __name__ == \"__main__\":\n \n from f5_tts.model.utils import get_tokenizer\n from torch.utils.data import DataLoader, Dataset, SequentialSampler\n from f5_tts.model.dataset import load_dataset \n from f5_tts.model.dataset import DynamicBatchSampler, collate_fn\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name","source_hash":"d0a2447ed4e6999217aa5d1177a1def4c9688256e5ff0c12fe4a7cd040b19fbd","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.unimodel.__init__","uri":"program://DMOSpeech2/function/src.unimodel.__init__#L27-L101","kind":"function","name":"__init__","path":"src/unimodel.py","language":"python","start_line":27,"end_line":101,"context_start_line":7,"context_end_line":121,"code":"from torch import nn\nimport torch \nimport copy\nimport os\n\nfrom f5_tts.model import DiT, UNetT\nfrom pathlib import Path\nfrom guidance_model import Guidance\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n sample_consecutive_steps,\n sample_from_list,\n)\n\nclass UniModel(nn.Module):\n def __init__(self, \n model: DiT, # teacher model (dit model)\n checkpoint_path: str = \"\",\n second_time: bool = True,\n use_fp16: bool = True,\n real_guidance_scale: float = 2.0, \n fake_guidance_scale: float = 0.0, \n gen_cls_loss: bool = False,\n sway_coeff: float = -1.0,\n vocab_char_map: dict[str, int] | None = None,\n frac_lengths_mask: tuple[float, float] = (0.7, 1.0)):\n \n super().__init__()\n \n if checkpoint_path != \"\":\n if \"model_last.pt\" in os.listdir(checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n checkpoint = torch.load(f\"{checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n if \"scale\" in checkpoint:\n self.scale = checkpoint[\"scale\"]\n else:\n self.scale = 1.0\n print(f\"Loaded teacher model with scale: {self.scale}\")\n\n if \"step\" in checkpoint:\n state = checkpoint[\"model_state_dict\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n state = checkpoint[\"model_state_dict\"]\n\n # only load the DiT module\n filtered_state_dict = {\n k.replace(\"transformer.\", \"\"): v\n for k, v in state.items()\n if k.startswith(\"transformer.\")\n }\n\n model.load_state_dict(filtered_state_dict, strict=False)\n else:\n self.scale = 1.0\n \n real_unet = copy.deepcopy(model)\n real_unet.time_embed2 = None\n \n fake_unet = copy.deepcopy(model)\n \n # Instantiate Guidance, which internally uses real_unet and fake_unet initialized from the teacher\n self.guidance_model = Guidance(\n real_unet=real_unet,\n fake_unet=fake_unet,\n use_fp16=use_fp16,\n real_guidance_scale=real_guidance_scale,\n fake_guidance_scale=fake_guidance_scale,\n gen_cls_loss=gen_cls_loss,\n sway_coeff=sway_coeff,\n )\n \n self.feedforward_model = copy.deepcopy(model) # initialize the student model\n self.feedforward_model.requires_grad_(True)\n self.feedforward_model.time_embed2 = None\n\n self.vocab_char_map = vocab_char_map\n self.frac_lengths_mask = frac_lengths_mask\n \n self.second_time = second_time # fake_unet.time_embed2 is not None\n\n def forward(self,\n inp: float[\"b n d\"], # mel\n text: int[\"b nt\"] | list[str],\n *,\n lens: int[\"b\"] | None = None,\n student_steps: list[int] = [0, 0.25, 0.5, 0.75],\n update_generator: bool = False,\n ):\n \"\"\"\n Forward pass that routes to either generator_forward or guidance_forward\n in the Guidance class, depending on the arguments.\n\n Parameters:\n -----------\n generator_turn: bool\n If True, run the generator forward pass (distribution matching loss, etc.)\n guidance_turn: bool\n If True, run the guidance forward pass (fake loss, cls loss, etc.)\n data_dict: dict","source_hash":"d0a2447ed4e6999217aa5d1177a1def4c9688256e5ff0c12fe4a7cd040b19fbd","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.unimodel.forward","uri":"program://DMOSpeech2/function/src.unimodel.forward#L103-L264","kind":"function","name":"forward","path":"src/unimodel.py","language":"python","start_line":103,"end_line":264,"context_start_line":83,"context_end_line":284,"code":" # Instantiate Guidance, which internally uses real_unet and fake_unet initialized from the teacher\n self.guidance_model = Guidance(\n real_unet=real_unet,\n fake_unet=fake_unet,\n use_fp16=use_fp16,\n real_guidance_scale=real_guidance_scale,\n fake_guidance_scale=fake_guidance_scale,\n gen_cls_loss=gen_cls_loss,\n sway_coeff=sway_coeff,\n )\n \n self.feedforward_model = copy.deepcopy(model) # initialize the student model\n self.feedforward_model.requires_grad_(True)\n self.feedforward_model.time_embed2 = None\n\n self.vocab_char_map = vocab_char_map\n self.frac_lengths_mask = frac_lengths_mask\n \n self.second_time = second_time # fake_unet.time_embed2 is not None\n\n def forward(self,\n inp: float[\"b n d\"], # mel\n text: int[\"b nt\"] | list[str],\n *,\n lens: int[\"b\"] | None = None,\n student_steps: list[int] = [0, 0.25, 0.5, 0.75],\n update_generator: bool = False,\n ):\n \"\"\"\n Forward pass that routes to either generator_forward or guidance_forward\n in the Guidance class, depending on the arguments.\n\n Parameters:\n -----------\n generator_turn: bool\n If True, run the generator forward pass (distribution matching loss, etc.)\n guidance_turn: bool\n If True, run the guidance forward pass (fake loss, cls loss, etc.)\n data_dict: dict\n Input dictionary containing the necessary keys for the forward passes.\n Expected keys may include:\n \"inp\": Tensor (B, N, D) - input mel or latent\n \"text\": Tensor or list[str] - text input\n \"rand_span_mask\": Tensor (B, N) - boolean mask\n \"real_data\": dict with keys like:\n \"inp\", \"text\", \"rand_span_mask\"\n \n Returns:\n --------\n loss_dict: dict[str, Tensor]\n Dictionary of losses.\n log_dict: dict[str, Tensor or float]\n Dictionary of logging tensors or values.\n \"\"\"\n \n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n\n # handle text as string\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # lens and mask\n if not exists(lens):\n lens = torch.full((batch,), seq_len, device=device)\n\n mask = lens_to_mask(lens, length=seq_len) # useless here, as collate_fn will pad to max length in batch\n\n # sample from the list of student steps\n time = sample_from_list(student_steps, batch).to(device)\n c_time, p_time = sample_consecutive_steps(student_steps)\n time = torch.ones_like(time) * c_time\n p_time = torch.ones_like(time) * p_time\n\n frac_lengths = torch.zeros((batch,), device=device).float().uniform_(*self.frac_lengths_mask)\n rand_span_mask = mask_from_frac_lengths(lens, frac_lengths)\n \n if exists(mask):\n rand_span_mask &= mask\n \n\n # # use generated output from previous step as input\n with torch.no_grad():\n x1 = inp\n x0 = torch.randn_like(x1)\n t = p_time.unsqueeze(-1).unsqueeze(-1)\n phi = (1 - t) * x0 + t * x1\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n \n pred = self.feedforward_model(\n x=phi, \n cond=cond,\n text=text, \n time=p_time, \n drop_audio_cond=False, \n drop_text=False # make sure the cfg=1\n ) # flow prediction\n \n # predicted mel spectrogram\n output = phi + (1 - t) * pred \n output[~rand_span_mask] = inp[~rand_span_mask]\n \n # forward diffusion\n x1 = output\n x0 = torch.randn_like(x1)\n t = time.unsqueeze(-1).unsqueeze(-1)\n phi = (1 - t) * x0 + t * x1\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n \n with torch.no_grad() if not update_generator else contextlib.nullcontext():\n pred = self.feedforward_model(\n x=phi, \n cond=cond,\n text=text, \n time=time, \n drop_audio_cond=False, \n drop_text=False # make sure no cfg is used \n )\n \n # predicted mel spectrogram\n output = phi + (1 - t) * pred\n output[~rand_span_mask] = inp[~rand_span_mask]\n \n if update_generator:\n generator_data_dict = {\n \"inp\": output,\n \"text\": text,\n \"rand_span_mask\": rand_span_mask,\n \"second_time\": time if self.second_time else None,\n \"mse_loss\": time.mean() == student_steps[-1].mean(),\n \"real_data\": {\n \"inp\": inp,\n \"text\": text,\n \"rand_span_mask\": rand_span_mask\n }\n }\n \n # avoid any side effects of gradient accumulation\n # self.guidance_model.requires_grad_(False)\n # self.feedforward_model.requires_grad_(True)\n generator_loss_dict, generator_log_dict = self.guidance_model(\n generator_turn=True,\n guidance_turn=False,\n generator_data_dict=generator_data_dict,\n guidance_data_dict=None\n )\n \n generator_log_dict['ground_truth'] = x1\n generator_log_dict['generator_input'] = phi\n generator_log_dict['generator_output'] = output\n generator_log_dict['generator_cond'] = cond\n generator_log_dict['time'] = time\n \n return generator_loss_dict, generator_log_dict\n else:\n guidance_data_dict = {\n \"inp\": output.detach(),\n \"text\": text,\n \"rand_span_mask\": rand_span_mask,\n \"second_time\": time if self.second_time else None,\n \"real_data\": {\n \"inp\": inp,\n \"text\": text,\n \"rand_span_mask\": rand_span_mask\n }\n }\n \n # avoid any side effects of gradient accumulation\n # self.feedforward_model.requires_grad_(False)\n # self.guidance_model.requires_grad_(True)\n guidance_loss_dict, guidance_log_dict = self.guidance_model(\n generator_turn=False,\n guidance_turn=True,\n generator_data_dict=None,\n guidance_data_dict=guidance_data_dict\n )\n # self.feedforward_model.requires_grad_(True)\n \n return guidance_loss_dict, guidance_log_dict\n \n # return guidance_loss_dict, guidance_log_dict, generator_loss_dict, generator_log_dict\n \n\nif __name__ == \"__main__\":\n \n from f5_tts.model.utils import get_tokenizer\n from torch.utils.data import DataLoader, Dataset, SequentialSampler\n from f5_tts.model.dataset import load_dataset \n from f5_tts.model.dataset import DynamicBatchSampler, collate_fn\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name","source_hash":"d0a2447ed4e6999217aa5d1177a1def4c9688256e5ff0c12fe4a7cd040b19fbd","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_predictor","uri":"program://DMOSpeech2/module/src.duration_predictor#L1-L84","kind":"module","name":"src.duration_predictor","path":"src/duration_predictor.py","language":"python","start_line":1,"end_line":84,"context_start_line":1,"context_end_line":84,"code":"import torch\nimport torch.nn as nn\n\n# from tts_encode import tts_encode\n\ndef calculate_remaining_lengths(mel_lengths):\n B = mel_lengths.shape[0]\n max_L = mel_lengths.max().item() # Get the maximum length in the batch\n\n # Create a range tensor: shape (max_L,), [0, 1, 2, ..., max_L-1]\n range_tensor = torch.arange(max_L, device=mel_lengths.device).expand(B, max_L)\n\n # Compute targets using broadcasting: (L-1) - range_tensor\n remain_lengths = (mel_lengths[:, None] - 1 - range_tensor).clamp(min=0)\n\n return remain_lengths\n\n\nclass PositionalEncoding(nn.Module):\n def __init__(self, hidden_dim, max_len=4096):\n super().__init__()\n pe = torch.zeros(max_len, hidden_dim)\n position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)\n div_term = torch.exp(torch.arange(0, hidden_dim, 2).float() * (-torch.log(torch.tensor(10000.0)) / hidden_dim))\n pe[:, 0::2] = torch.sin(position * div_term)\n pe[:, 1::2] = torch.cos(position * div_term)\n self.pe = pe.unsqueeze(0) # Shape: (1, max_len, hidden_dim)\n\n def forward(self, x):\n x = x + self.pe[:, :x.size(1)].to(x.device)\n return x\n\n\nclass SpeechLengthPredictor(nn.Module):\n\n def __init__(self, \n vocab_size=2545, n_mel=100, hidden_dim=256, \n n_text_layer=4, n_cross_layer=4, n_head=8,\n output_dim=1,\n ):\n super().__init__()\n \n # Text Encoder: Embedding + Transformer Layers\n self.text_embedder = nn.Embedding(vocab_size+1, hidden_dim, padding_idx=vocab_size)\n self.text_pe = PositionalEncoding(hidden_dim)\n encoder_layer = nn.TransformerEncoderLayer(\n d_model=hidden_dim, nhead=n_head, dim_feedforward=hidden_dim*2, batch_first=True\n )\n self.text_encoder = nn.TransformerEncoder(encoder_layer, num_layers=n_text_layer)\n \n # Mel Spectrogram Embedder\n self.mel_embedder = nn.Linear(n_mel, hidden_dim)\n self.mel_pe = PositionalEncoding(hidden_dim)\n\n # Transformer Decoder Layers with Cross-Attention in Every Layer\n decoder_layer = nn.TransformerDecoderLayer(\n d_model=hidden_dim, nhead=n_head, dim_feedforward=hidden_dim*2, batch_first=True\n )\n self.decoder = nn.TransformerDecoder(decoder_layer, num_layers=n_cross_layer)\n \n # Final Classification Layer\n self.predictor = nn.Linear(hidden_dim, output_dim)\n\n def forward(self, text_ids, mel):\n # Encode text\n text_embedded = self.text_pe(self.text_embedder(text_ids))\n text_features = self.text_encoder(text_embedded) # (B, L_text, D)\n \n # Encode Mel spectrogram\n mel_features = self.mel_pe(self.mel_embedder(mel)) # (B, L_mel, D)\n \n # Causal Masking for Decoder\n seq_len = mel_features.size(1)\n causal_mask = torch.triu(torch.ones(seq_len, seq_len), diagonal=1).bool().to(mel.device)\n # causal_mask = torch.triu(\n # torch.full((seq_len, seq_len), float('-inf'), device=mel.device), diagonal=1\n # )\n\n # Transformer Decoder with Cross-Attention in Each Layer\n decoder_out = self.decoder(mel_features, text_features, tgt_mask=causal_mask)\n \n # Length Prediction\n length_logits = self.predictor(decoder_out).squeeze(-1)\n return length_logits","source_hash":"9db72e598a7cc564ed6b6fb1eb634d42d4853ed6797b38a5461c6550d0491a8e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_predictor.calculate_remaining_lengths","uri":"program://DMOSpeech2/function/src.duration_predictor.calculate_remaining_lengths#L6-L16","kind":"function","name":"calculate_remaining_lengths","path":"src/duration_predictor.py","language":"python","start_line":6,"end_line":16,"context_start_line":1,"context_end_line":36,"code":"import torch\nimport torch.nn as nn\n\n# from tts_encode import tts_encode\n\ndef calculate_remaining_lengths(mel_lengths):\n B = mel_lengths.shape[0]\n max_L = mel_lengths.max().item() # Get the maximum length in the batch\n\n # Create a range tensor: shape (max_L,), [0, 1, 2, ..., max_L-1]\n range_tensor = torch.arange(max_L, device=mel_lengths.device).expand(B, max_L)\n\n # Compute targets using broadcasting: (L-1) - range_tensor\n remain_lengths = (mel_lengths[:, None] - 1 - range_tensor).clamp(min=0)\n\n return remain_lengths\n\n\nclass PositionalEncoding(nn.Module):\n def __init__(self, hidden_dim, max_len=4096):\n super().__init__()\n pe = torch.zeros(max_len, hidden_dim)\n position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)\n div_term = torch.exp(torch.arange(0, hidden_dim, 2).float() * (-torch.log(torch.tensor(10000.0)) / hidden_dim))\n pe[:, 0::2] = torch.sin(position * div_term)\n pe[:, 1::2] = torch.cos(position * div_term)\n self.pe = pe.unsqueeze(0) # Shape: (1, max_len, hidden_dim)\n\n def forward(self, x):\n x = x + self.pe[:, :x.size(1)].to(x.device)\n return x\n\n\nclass SpeechLengthPredictor(nn.Module):\n\n def __init__(self, ","source_hash":"9db72e598a7cc564ed6b6fb1eb634d42d4853ed6797b38a5461c6550d0491a8e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_predictor.PositionalEncoding","uri":"program://DMOSpeech2/class/src.duration_predictor.PositionalEncoding#L19-L31","kind":"class","name":"PositionalEncoding","path":"src/duration_predictor.py","language":"python","start_line":19,"end_line":31,"context_start_line":1,"context_end_line":51,"code":"import torch\nimport torch.nn as nn\n\n# from tts_encode import tts_encode\n\ndef calculate_remaining_lengths(mel_lengths):\n B = mel_lengths.shape[0]\n max_L = mel_lengths.max().item() # Get the maximum length in the batch\n\n # Create a range tensor: shape (max_L,), [0, 1, 2, ..., max_L-1]\n range_tensor = torch.arange(max_L, device=mel_lengths.device).expand(B, max_L)\n\n # Compute targets using broadcasting: (L-1) - range_tensor\n remain_lengths = (mel_lengths[:, None] - 1 - range_tensor).clamp(min=0)\n\n return remain_lengths\n\n\nclass PositionalEncoding(nn.Module):\n def __init__(self, hidden_dim, max_len=4096):\n super().__init__()\n pe = torch.zeros(max_len, hidden_dim)\n position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)\n div_term = torch.exp(torch.arange(0, hidden_dim, 2).float() * (-torch.log(torch.tensor(10000.0)) / hidden_dim))\n pe[:, 0::2] = torch.sin(position * div_term)\n pe[:, 1::2] = torch.cos(position * div_term)\n self.pe = pe.unsqueeze(0) # Shape: (1, max_len, hidden_dim)\n\n def forward(self, x):\n x = x + self.pe[:, :x.size(1)].to(x.device)\n return x\n\n\nclass SpeechLengthPredictor(nn.Module):\n\n def __init__(self, \n vocab_size=2545, n_mel=100, hidden_dim=256, \n n_text_layer=4, n_cross_layer=4, n_head=8,\n output_dim=1,\n ):\n super().__init__()\n \n # Text Encoder: Embedding + Transformer Layers\n self.text_embedder = nn.Embedding(vocab_size+1, hidden_dim, padding_idx=vocab_size)\n self.text_pe = PositionalEncoding(hidden_dim)\n encoder_layer = nn.TransformerEncoderLayer(\n d_model=hidden_dim, nhead=n_head, dim_feedforward=hidden_dim*2, batch_first=True\n )\n self.text_encoder = nn.TransformerEncoder(encoder_layer, num_layers=n_text_layer)\n \n # Mel Spectrogram Embedder","source_hash":"9db72e598a7cc564ed6b6fb1eb634d42d4853ed6797b38a5461c6550d0491a8e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_predictor.SpeechLengthPredictor","uri":"program://DMOSpeech2/class/src.duration_predictor.SpeechLengthPredictor#L34-L84","kind":"class","name":"SpeechLengthPredictor","path":"src/duration_predictor.py","language":"python","start_line":34,"end_line":84,"context_start_line":14,"context_end_line":84,"code":" remain_lengths = (mel_lengths[:, None] - 1 - range_tensor).clamp(min=0)\n\n return remain_lengths\n\n\nclass PositionalEncoding(nn.Module):\n def __init__(self, hidden_dim, max_len=4096):\n super().__init__()\n pe = torch.zeros(max_len, hidden_dim)\n position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)\n div_term = torch.exp(torch.arange(0, hidden_dim, 2).float() * (-torch.log(torch.tensor(10000.0)) / hidden_dim))\n pe[:, 0::2] = torch.sin(position * div_term)\n pe[:, 1::2] = torch.cos(position * div_term)\n self.pe = pe.unsqueeze(0) # Shape: (1, max_len, hidden_dim)\n\n def forward(self, x):\n x = x + self.pe[:, :x.size(1)].to(x.device)\n return x\n\n\nclass SpeechLengthPredictor(nn.Module):\n\n def __init__(self, \n vocab_size=2545, n_mel=100, hidden_dim=256, \n n_text_layer=4, n_cross_layer=4, n_head=8,\n output_dim=1,\n ):\n super().__init__()\n \n # Text Encoder: Embedding + Transformer Layers\n self.text_embedder = nn.Embedding(vocab_size+1, hidden_dim, padding_idx=vocab_size)\n self.text_pe = PositionalEncoding(hidden_dim)\n encoder_layer = nn.TransformerEncoderLayer(\n d_model=hidden_dim, nhead=n_head, dim_feedforward=hidden_dim*2, batch_first=True\n )\n self.text_encoder = nn.TransformerEncoder(encoder_layer, num_layers=n_text_layer)\n \n # Mel Spectrogram Embedder\n self.mel_embedder = nn.Linear(n_mel, hidden_dim)\n self.mel_pe = PositionalEncoding(hidden_dim)\n\n # Transformer Decoder Layers with Cross-Attention in Every Layer\n decoder_layer = nn.TransformerDecoderLayer(\n d_model=hidden_dim, nhead=n_head, dim_feedforward=hidden_dim*2, batch_first=True\n )\n self.decoder = nn.TransformerDecoder(decoder_layer, num_layers=n_cross_layer)\n \n # Final Classification Layer\n self.predictor = nn.Linear(hidden_dim, output_dim)\n\n def forward(self, text_ids, mel):\n # Encode text\n text_embedded = self.text_pe(self.text_embedder(text_ids))\n text_features = self.text_encoder(text_embedded) # (B, L_text, D)\n \n # Encode Mel spectrogram\n mel_features = self.mel_pe(self.mel_embedder(mel)) # (B, L_mel, D)\n \n # Causal Masking for Decoder\n seq_len = mel_features.size(1)\n causal_mask = torch.triu(torch.ones(seq_len, seq_len), diagonal=1).bool().to(mel.device)\n # causal_mask = torch.triu(\n # torch.full((seq_len, seq_len), float('-inf'), device=mel.device), diagonal=1\n # )\n\n # Transformer Decoder with Cross-Attention in Each Layer\n decoder_out = self.decoder(mel_features, text_features, tgt_mask=causal_mask)\n \n # Length Prediction\n length_logits = self.predictor(decoder_out).squeeze(-1)\n return length_logits","source_hash":"9db72e598a7cc564ed6b6fb1eb634d42d4853ed6797b38a5461c6550d0491a8e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_predictor.__init__","uri":"program://DMOSpeech2/function/src.duration_predictor.__init__#L36-L62","kind":"function","name":"__init__","path":"src/duration_predictor.py","language":"python","start_line":36,"end_line":62,"context_start_line":16,"context_end_line":82,"code":" return remain_lengths\n\n\nclass PositionalEncoding(nn.Module):\n def __init__(self, hidden_dim, max_len=4096):\n super().__init__()\n pe = torch.zeros(max_len, hidden_dim)\n position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)\n div_term = torch.exp(torch.arange(0, hidden_dim, 2).float() * (-torch.log(torch.tensor(10000.0)) / hidden_dim))\n pe[:, 0::2] = torch.sin(position * div_term)\n pe[:, 1::2] = torch.cos(position * div_term)\n self.pe = pe.unsqueeze(0) # Shape: (1, max_len, hidden_dim)\n\n def forward(self, x):\n x = x + self.pe[:, :x.size(1)].to(x.device)\n return x\n\n\nclass SpeechLengthPredictor(nn.Module):\n\n def __init__(self, \n vocab_size=2545, n_mel=100, hidden_dim=256, \n n_text_layer=4, n_cross_layer=4, n_head=8,\n output_dim=1,\n ):\n super().__init__()\n \n # Text Encoder: Embedding + Transformer Layers\n self.text_embedder = nn.Embedding(vocab_size+1, hidden_dim, padding_idx=vocab_size)\n self.text_pe = PositionalEncoding(hidden_dim)\n encoder_layer = nn.TransformerEncoderLayer(\n d_model=hidden_dim, nhead=n_head, dim_feedforward=hidden_dim*2, batch_first=True\n )\n self.text_encoder = nn.TransformerEncoder(encoder_layer, num_layers=n_text_layer)\n \n # Mel Spectrogram Embedder\n self.mel_embedder = nn.Linear(n_mel, hidden_dim)\n self.mel_pe = PositionalEncoding(hidden_dim)\n\n # Transformer Decoder Layers with Cross-Attention in Every Layer\n decoder_layer = nn.TransformerDecoderLayer(\n d_model=hidden_dim, nhead=n_head, dim_feedforward=hidden_dim*2, batch_first=True\n )\n self.decoder = nn.TransformerDecoder(decoder_layer, num_layers=n_cross_layer)\n \n # Final Classification Layer\n self.predictor = nn.Linear(hidden_dim, output_dim)\n\n def forward(self, text_ids, mel):\n # Encode text\n text_embedded = self.text_pe(self.text_embedder(text_ids))\n text_features = self.text_encoder(text_embedded) # (B, L_text, D)\n \n # Encode Mel spectrogram\n mel_features = self.mel_pe(self.mel_embedder(mel)) # (B, L_mel, D)\n \n # Causal Masking for Decoder\n seq_len = mel_features.size(1)\n causal_mask = torch.triu(torch.ones(seq_len, seq_len), diagonal=1).bool().to(mel.device)\n # causal_mask = torch.triu(\n # torch.full((seq_len, seq_len), float('-inf'), device=mel.device), diagonal=1\n # )\n\n # Transformer Decoder with Cross-Attention in Each Layer\n decoder_out = self.decoder(mel_features, text_features, tgt_mask=causal_mask)\n \n # Length Prediction","source_hash":"9db72e598a7cc564ed6b6fb1eb634d42d4853ed6797b38a5461c6550d0491a8e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.duration_predictor.forward","uri":"program://DMOSpeech2/function/src.duration_predictor.forward#L64-L84","kind":"function","name":"forward","path":"src/duration_predictor.py","language":"python","start_line":64,"end_line":84,"context_start_line":44,"context_end_line":84,"code":" self.text_embedder = nn.Embedding(vocab_size+1, hidden_dim, padding_idx=vocab_size)\n self.text_pe = PositionalEncoding(hidden_dim)\n encoder_layer = nn.TransformerEncoderLayer(\n d_model=hidden_dim, nhead=n_head, dim_feedforward=hidden_dim*2, batch_first=True\n )\n self.text_encoder = nn.TransformerEncoder(encoder_layer, num_layers=n_text_layer)\n \n # Mel Spectrogram Embedder\n self.mel_embedder = nn.Linear(n_mel, hidden_dim)\n self.mel_pe = PositionalEncoding(hidden_dim)\n\n # Transformer Decoder Layers with Cross-Attention in Every Layer\n decoder_layer = nn.TransformerDecoderLayer(\n d_model=hidden_dim, nhead=n_head, dim_feedforward=hidden_dim*2, batch_first=True\n )\n self.decoder = nn.TransformerDecoder(decoder_layer, num_layers=n_cross_layer)\n \n # Final Classification Layer\n self.predictor = nn.Linear(hidden_dim, output_dim)\n\n def forward(self, text_ids, mel):\n # Encode text\n text_embedded = self.text_pe(self.text_embedder(text_ids))\n text_features = self.text_encoder(text_embedded) # (B, L_text, D)\n \n # Encode Mel spectrogram\n mel_features = self.mel_pe(self.mel_embedder(mel)) # (B, L_mel, D)\n \n # Causal Masking for Decoder\n seq_len = mel_features.size(1)\n causal_mask = torch.triu(torch.ones(seq_len, seq_len), diagonal=1).bool().to(mel.device)\n # causal_mask = torch.triu(\n # torch.full((seq_len, seq_len), float('-inf'), device=mel.device), diagonal=1\n # )\n\n # Transformer Decoder with Cross-Attention in Each Layer\n decoder_out = self.decoder(mel_features, text_features, tgt_mask=causal_mask)\n \n # Length Prediction\n length_logits = self.predictor(decoder_out).squeeze(-1)\n return length_logits","source_hash":"9db72e598a7cc564ed6b6fb1eb634d42d4853ed6797b38a5461c6550d0491a8e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.grpo_duration_trainer","uri":"program://DMOSpeech2/module/src.grpo_duration_trainer#L1-L729","kind":"module","name":"src.grpo_duration_trainer","path":"src/grpo_duration_trainer.py","language":"python","start_line":1,"end_line":729,"context_start_line":1,"context_end_line":729,"code":"import os\nimport gc\nimport json\nimport random\nimport time\nimport io\nimport copy\nfrom typing import List, Dict, Any, Optional, Callable, Tuple, Union\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler, Subset\nfrom tqdm import tqdm\n\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nimport wandb\n\nfrom f5_tts.model.dataset import collate_fn, DynamicBatchSampler\nfrom f5_tts.model.utils import list_str_to_idx\n\n# torch.autograd.set_detect_anomaly(True)\n# os.environ['HYDRA_FULL_ERROR'] = 'True'\n\n\ndef safe_sample(logits, temperature=1.0):\n \"\"\"\n logits: Tensor of shape (B, n_class)\n temperature: Sampling temperature (higher => more random)\n \"\"\"\n # Apply temperature scaling\n scaled_logits = logits / temperature\n \n # Compute categorical distribution\n probs = F.softmax(scaled_logits, dim=-1)\n \n # Sample from the distribution once per batch element\n samples = torch.multinomial(probs, num_samples=1) # (B, 1)\n \n # Convert to one-hot encoding\n one_hot_samples = torch.zeros_like(probs).scatter_(1, samples, 1)\n \n return one_hot_samples\n\n\nclass GRPODurationTrainer:\n \"\"\"\n Trainer class that implements GRPO (Generative Reinforcement Learning from Preference Optimization)\n for a duration predictor in text-to-speech synthesis.\n \"\"\"\n def __init__(\n self,\n model, # Duration predictor model\n inference_fn, # Function to generate speech\n reward_fn, # Function to compute rewards from generated speech\n \n vocab_size: int, # Size of the vocabulary\n vocab_char_map: dict, # Mapping from characters to token IDs\n\n # Duration model parameters\n n_class: int = 301, # Number of duration classes\n n_frame_per_class: int = 10, # Number of frames per class\n gumbel_tau: int = 0.7,\n \n # GRPO parameters\n beta: float = 0.04, # KL regularization weight\n clip_param: float = 0.2, # PPO clip parameter\n num_pre_samples: int = 8, # Number of samples per prompt\n compute_gen_logps: bool = True, # Whether to compute generation log probabilities\n \n # Training parameters\n learning_rate: float = 5e-6,\n num_warmup_updates: int = 10000,\n save_per_updates: int = 10000,\n checkpoint_path: Optional[str] = None,\n all_steps: int = 100000, # Total training steps\n \n # Batch parameters\n batch_size: int = 8,\n batch_size_type: str = \"sample\",\n max_samples: int = 16,\n grad_accumulation_steps: int = 2,\n max_grad_norm: float = 1.0,\n \n # Logging parameters\n logger: Optional[str] = \"wandb\",\n wandb_project: str = \"tts-duration-grpo\",\n wandb_run_name: str = \"grpo_run\",\n wandb_resume_id: Optional[str] = None,\n \n accelerate_kwargs: dict = dict(),\n ):\n # Initialize accelerator for distributed training\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=False)\n \n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n print(f\"Using logger: {logger}\")\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if wandb_resume_id:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"beta\": beta,\n \"clip_param\": clip_param,\n \"num_pre_samples\": num_pre_samples,\n \"n_class\": n_class,\n \"n_frame_per_class\": n_frame_per_class,\n \"all_steps\": all_steps,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n },\n )\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n # Store model, inference function, and reward function\n self.model = model\n \n # Create reference model (frozen clone of the initial model)\n self.ref_model = copy.deepcopy(model)\n for param in self.ref_model.parameters():\n param.requires_grad = False\n self.ref_model.eval()\n \n # prepare inference_fn\n self.inference_fn = inference_fn\n self.inference_fn.scale = self.inference_fn.scale.to(self.accelerator.device)\n self.inference_fn.tts_model = self.inference_fn.tts_model.to(self.accelerator.device)\n # prepare reward_fn\n self.reward_fn = reward_fn\n \n # Store vocabulary and mapping\n self.vocab_size = vocab_size\n self.vocab_char_map = vocab_char_map\n\n # Store duration model parameters\n self.n_class = n_class\n self.n_frame_per_class = n_frame_per_class\n self.gumbel_tau = gumbel_tau\n \n # Store GRPO parameters\n self.beta = beta\n self.clip_param = clip_param\n self.num_pre_samples = num_pre_samples\n self.compute_gen_logps = compute_gen_logps\n \n # Store training parameters\n self.learning_rate = learning_rate\n self.num_warmup_updates: int = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.checkpoint_path = checkpoint_path or f\"ckpts/{wandb_run_name}\"\n self.all_steps = all_steps\n \n # Store batch parameters\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n \n # Initialize optimizer\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n \n # Prepare model and optimizer with accelerator\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n self.ref_model = self.accelerator.prepare(self.ref_model)\n self.reward_fn, self.inference_fn = self.accelerator.prepare(self.reward_fn, self.inference_fn) \n \n # GRPO batch queue\n self.batch_queue = []\n \n # Store distributed rank\n self.rank = torch.distributed.get_rank() if torch.distributed.is_initialized() else 0\n\n self.device = f'cuda:{self.rank}'\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n \n def save_checkpoint(self, step, last=False):\n \"\"\"Save model and optimizer state\"\"\"\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict() if hasattr(self, 'scheduler') else None,\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n \n def load_checkpoint(self):\n \"\"\"Load latest checkpoint if available\"\"\"\n if (\n not self.checkpoint_path\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n\n print(f'Loading checkpoint: {latest_checkpoint}')\n checkpoint = torch.load(\n f\"{self.checkpoint_path}/{latest_checkpoint}\", \n weights_only=True, \n map_location=\"cpu\"\n )\n\n if \"step\" in checkpoint:\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if hasattr(self, 'scheduler') and checkpoint[\"scheduler_state_dict\"]:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n \n del checkpoint\n gc.collect()\n \n print(f'Successfully loaded checkpoint at step {step}')\n return step\n \n @torch.no_grad()\n def get_ref_logps(self, text_ids, mel, sampled_classes):\n \"\"\"\n Get log probabilities from the reference model for the sampled classes\n \"\"\"\n B = text_ids.shape[0]\n K = self.num_pre_samples\n with torch.no_grad():\n ref_logits = self.ref_model(text_ids=text_ids, mel=mel)[:, -1, :]\n ref_logits = ref_logits.unsqueeze(1).repeat(1, K, 1).view(B*K, -1)\n ref_log_probs = F.log_softmax(ref_logits, dim=-1)\n ref_logps = torch.gather(\n ref_log_probs, \n dim=-1, \n index=sampled_classes.unsqueeze(-1)\n ).squeeze(-1)\n return ref_logps\n \n @torch.no_grad()\n def generate_duration_samples(self, batch_inputs):\n \"\"\"\n Generate multiple duration predictions from the model for each input\n and evaluate them using the inference function and reward model\n \n Args:\n batch_inputs: Dictionary with text, prompt audio, etc.\n \n Returns:\n Dictionary with duration samples, rewards, and reference logits\n \"\"\"\n\n if self.rank == 0:\n print(\"Generating duration samples...\")\n \n # all_logits = []\n all_text_ids = []\n all_mels = []\n all_sampled_classes = []\n all_durations = []\n all_rewards = []\n all_gen_logps = []\n\n all_ctc_loss = []\n all_sv_loss = []\n\n # Fetch batch inputs\n # prompt_mel = batch_inputs['mel'].permute(0, 2, 1).to(self.device)\n prompt_mel = batch_inputs['mel'].permute(0, 2, 1) # (B, T, 100)\n prompt_text = batch_inputs['text']\n\n batch_size = prompt_mel.shape[0]\n\n # Shift text to unpair 'mel' and 'text'; The shifted text will be synthesized\n target_text = batch_inputs['target_text']\n target_text_lengths = torch.LongTensor([len(t) for t in target_text]).to(prompt_mel.device)\n try:\n full_text = [prompt+[' ']+target for prompt, target in zip(prompt_text, target_text)]\n except:\n target_text = [batch_inputs['text'][-1]] + batch_inputs['text'][:-1]\n target_text_lengths = batch_inputs['text_lengths'].clone().roll(1, 0)\n full_text = [prompt+[' ']+target for prompt, target in zip(prompt_text, target_text)]\n\n # Goes to reward model\n target_text_ids = list_str_to_idx(target_text, self.vocab_char_map).to(self.accelerator.device) # to device, the dataloader only gives list\n\n # Goes to duration model and TTS\n full_text_ids = list_str_to_idx(full_text, self.vocab_char_map).to(self.accelerator.device)\n\n # Deepcopy to separate text_ids for SLP and TTS\n slp_text_ids = full_text_ids.detach().clone()\n slp_text_ids = slp_text_ids.masked_fill(slp_text_ids==-1, self.vocab_size) # (B, L)\n\n # Pre-compute duration logits\n K = self.num_pre_samples\n B, T, _ = prompt_mel.shape\n _, L = slp_text_ids.shape\n # prompt_mel_k_repeats = prompt_mel.unsqueeze(1).repeat(1, K, 1, 1) # (B, K, T, 100)\n # slp_text_ids_k_repeats = slp_text_ids.unsqueeze(1).repeat(1, K, 1) # (B, K, L)\n\n # Run model once for B inputs\n old_logits = self.model(\n text_ids=slp_text_ids, # (B, L)\n mel=prompt_mel # (B, T, 100)\n )[:, -1, :] # (B, n_class)\n\n # Repeat each result K times along batch dimension\n old_logits = old_logits.unsqueeze(1).repeat(1, K, 1) # (B, K, n_class)\n # logits_nograd = logits_grad.detach().clone().view(B, K, -1) # (B, K, n_class)\n\n for _full_text_ids, _target_text_ids, _target_text_lengths, \\\n _prompt_mel, _old_logits in zip(\n full_text_ids, target_text_ids, target_text_lengths, \n prompt_mel, old_logits\n ):\n\n duration_sample = F.gumbel_softmax(_old_logits, tau=self.gumbel_tau, hard=True, dim=-1)\n duration2frames = torch.arange(self.n_class).float().to(self.accelerator.device) * self.n_frame_per_class\n est_frames = (duration_sample * duration2frames).sum(-1) # (K, )\n\n # Compute log probabilities of the samples\n sampled_classes = duration_sample.argmax(dim=-1)\n log_probs = F.log_softmax(_old_logits, dim=-1)\n gen_logps = torch.gather(\n log_probs, \n dim=-1, \n index=sampled_classes.unsqueeze(-1)\n ).squeeze(-1) # Shape: [K, n_class]\n \n # Generate speech using the sampled durations\n sampled_rewards = []\n\n for i in range(K):\n cur_duration = est_frames[i]\n if cur_duration == 0:\n cur_duration = cur_duration + 50 # prevent 0 duration\n infer_full_text_ids = _full_text_ids.unsqueeze(0)\n infer_prompt_mel = _prompt_mel.unsqueeze(0)\n cur_duration = cur_duration.unsqueeze(0)\n infer_target_text_ids = _target_text_ids.unsqueeze(0)\n infer_target_text_lengths = _target_text_lengths.unsqueeze(0)\n with torch.inference_mode():\n try:\n _est_mel = self.inference_fn(\n full_text_ids=infer_full_text_ids, \n prompt_mel=infer_prompt_mel, \n target_duration=cur_duration, \n teacher_steps=0\n )\n _est_mel = _est_mel.permute(0, 2, 1) # (1, T, 100)\n \n loss_dict = self.reward_fn(\n prompt_mel=infer_prompt_mel,\n est_mel=_est_mel,\n target_text_id=infer_target_text_ids,\n target_text_length=infer_target_text_lengths\n )\n # #TODO reweight the loss for reward\n reward_sim = loss_dict['loss_sim'] # 0 to 1\n reward_ctc = loss_dict['loss_ctc']\n reward = -(reward_ctc + reward_sim * 3)\n all_ctc_loss.append(reward_ctc)\n all_sv_loss.append(reward_sim)\n except Exception as e:\n if self.rank == 0:\n print(f\"Error in speech synthesis: {e}\")\n reward = torch.tensor(-1.0).to(cur_duration.device)\n \n sampled_rewards.append(reward)\n # list with length of K\n sampled_rewards = torch.stack(sampled_rewards) # (K, )\n # Normalize rewards\n if (sampled_rewards.max() - sampled_rewards.min()).item() > 1e-6:\n sampled_rewards = (sampled_rewards - sampled_rewards.mean()) / (sampled_rewards.std() + 1e-8)\n\n # Store all data\n # all_logits.append(duration_logits)\n # all_text_ids.append(duration_input_expanded[\"text_ids\"])\n # all_mels.append(duration_input_expanded[\"mel\"])\n all_sampled_classes.append(sampled_classes)\n all_durations.append(est_frames)\n all_gen_logps.append(gen_logps)\n all_rewards.extend(sampled_rewards) # list with length of B*K\n \n # Concatenate all data\n # logits = torch.cat(all_logits, dim=0)\n # text_ids = torch.cat(all_text_ids, dim=0)\n # mels = torch.cat(all_mels, dim=0)\n sampled_classes = torch.cat(all_sampled_classes, dim=0)\n durations = torch.cat(all_durations, dim=0)\n rewards = torch.stack(all_rewards) # use stack to keep the same device of elements\n gen_logps = torch.cat(all_gen_logps, dim=0)\n\n ctc_losses = torch.stack(all_ctc_loss)\n sv_losses = torch.stack(all_sv_loss)\n \n if self.is_main:\n self.accelerator.log({\n \"ctc_loss\": ctc_losses.mean().item(),\n \"sv_loss\": sv_losses.mean().item(),\n \"reward\": rewards.mean().item(),\n \"reward_min\": rewards.min().item(),\n \"reward_max\": rewards.max().item(),\n }, step=self.global_step)\n\n # # Normalize rewards\n # if (rewards.max() - rewards.min()).item() > 1e-6:\n # rewards = (rewards - rewards.mean()) / (rewards.std() + 1e-8)\n\n ref_logps = self.get_ref_logps(slp_text_ids, prompt_mel, sampled_classes)\n\n # Create batch dict similar to Qwen2.5 implementation\n batch_outputs = {\n # \"logits\": logits_grad,\n \"text_ids\": slp_text_ids,\n \"prompt_mel\": prompt_mel,\n \"rewards\": rewards,\n \"refs\": ref_logps,\n \"sampled_classes\": sampled_classes,\n \"durations\": durations,\n }\n \n if self.compute_gen_logps:\n batch_outputs[\"gen_logps\"] = gen_logps\n \n if self.rank == 0:\n print(f\"Generated {len(rewards)} samples with reward min/mean/max: {rewards.min().item():.4f}/{rewards.mean().item():.4f}/{rewards.max().item():.4f}\")\n \n return batch_outputs\n \n def GRPO_step(self, batch):\n \"\"\"\n Perform a GRPO update step\n \n Args:\n batch: Dictionary with inputs, rewards, reference logits, etc.\n \n Returns:\n Loss value\n \"\"\"\n # Extract batch data\n # NOTE: why .unsqueeze(1) ???\n rewards = batch['rewards'] #.unsqueeze(1)\n ref_logps = batch['refs'] # (B)\n sampled_classes = batch['sampled_classes'] # (B)\n prompt_mel = batch['prompt_mel']\n text_ids = batch['text_ids']\n\n # Forward pass to get current model logits\n K = self.num_pre_samples\n B, T, _ = prompt_mel.shape\n _, L = text_ids.shape\n cur_logits = self.model(\n# ... truncated ...","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.grpo_duration_trainer.safe_sample","uri":"program://DMOSpeech2/function/src.grpo_duration_trainer.safe_sample#L29-L46","kind":"function","name":"safe_sample","path":"src/grpo_duration_trainer.py","language":"python","start_line":29,"end_line":46,"context_start_line":9,"context_end_line":66,"code":"\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler, Subset\nfrom tqdm import tqdm\n\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nimport wandb\n\nfrom f5_tts.model.dataset import collate_fn, DynamicBatchSampler\nfrom f5_tts.model.utils import list_str_to_idx\n\n# torch.autograd.set_detect_anomaly(True)\n# os.environ['HYDRA_FULL_ERROR'] = 'True'\n\n\ndef safe_sample(logits, temperature=1.0):\n \"\"\"\n logits: Tensor of shape (B, n_class)\n temperature: Sampling temperature (higher => more random)\n \"\"\"\n # Apply temperature scaling\n scaled_logits = logits / temperature\n \n # Compute categorical distribution\n probs = F.softmax(scaled_logits, dim=-1)\n \n # Sample from the distribution once per batch element\n samples = torch.multinomial(probs, num_samples=1) # (B, 1)\n \n # Convert to one-hot encoding\n one_hot_samples = torch.zeros_like(probs).scatter_(1, samples, 1)\n \n return one_hot_samples\n\n\nclass GRPODurationTrainer:\n \"\"\"\n Trainer class that implements GRPO (Generative Reinforcement Learning from Preference Optimization)\n for a duration predictor in text-to-speech synthesis.\n \"\"\"\n def __init__(\n self,\n model, # Duration predictor model\n inference_fn, # Function to generate speech\n reward_fn, # Function to compute rewards from generated speech\n \n vocab_size: int, # Size of the vocabulary\n vocab_char_map: dict, # Mapping from characters to token IDs\n\n # Duration model parameters\n n_class: int = 301, # Number of duration classes\n n_frame_per_class: int = 10, # Number of frames per class\n gumbel_tau: int = 0.7,","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.grpo_duration_trainer.GRPODurationTrainer","uri":"program://DMOSpeech2/class/src.grpo_duration_trainer.GRPODurationTrainer#L49-L729","kind":"class","name":"GRPODurationTrainer","path":"src/grpo_duration_trainer.py","language":"python","start_line":49,"end_line":729,"context_start_line":29,"context_end_line":729,"code":"def safe_sample(logits, temperature=1.0):\n \"\"\"\n logits: Tensor of shape (B, n_class)\n temperature: Sampling temperature (higher => more random)\n \"\"\"\n # Apply temperature scaling\n scaled_logits = logits / temperature\n \n # Compute categorical distribution\n probs = F.softmax(scaled_logits, dim=-1)\n \n # Sample from the distribution once per batch element\n samples = torch.multinomial(probs, num_samples=1) # (B, 1)\n \n # Convert to one-hot encoding\n one_hot_samples = torch.zeros_like(probs).scatter_(1, samples, 1)\n \n return one_hot_samples\n\n\nclass GRPODurationTrainer:\n \"\"\"\n Trainer class that implements GRPO (Generative Reinforcement Learning from Preference Optimization)\n for a duration predictor in text-to-speech synthesis.\n \"\"\"\n def __init__(\n self,\n model, # Duration predictor model\n inference_fn, # Function to generate speech\n reward_fn, # Function to compute rewards from generated speech\n \n vocab_size: int, # Size of the vocabulary\n vocab_char_map: dict, # Mapping from characters to token IDs\n\n # Duration model parameters\n n_class: int = 301, # Number of duration classes\n n_frame_per_class: int = 10, # Number of frames per class\n gumbel_tau: int = 0.7,\n \n # GRPO parameters\n beta: float = 0.04, # KL regularization weight\n clip_param: float = 0.2, # PPO clip parameter\n num_pre_samples: int = 8, # Number of samples per prompt\n compute_gen_logps: bool = True, # Whether to compute generation log probabilities\n \n # Training parameters\n learning_rate: float = 5e-6,\n num_warmup_updates: int = 10000,\n save_per_updates: int = 10000,\n checkpoint_path: Optional[str] = None,\n all_steps: int = 100000, # Total training steps\n \n # Batch parameters\n batch_size: int = 8,\n batch_size_type: str = \"sample\",\n max_samples: int = 16,\n grad_accumulation_steps: int = 2,\n max_grad_norm: float = 1.0,\n \n # Logging parameters\n logger: Optional[str] = \"wandb\",\n wandb_project: str = \"tts-duration-grpo\",\n wandb_run_name: str = \"grpo_run\",\n wandb_resume_id: Optional[str] = None,\n \n accelerate_kwargs: dict = dict(),\n ):\n # Initialize accelerator for distributed training\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=False)\n \n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n print(f\"Using logger: {logger}\")\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if wandb_resume_id:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"beta\": beta,\n \"clip_param\": clip_param,\n \"num_pre_samples\": num_pre_samples,\n \"n_class\": n_class,\n \"n_frame_per_class\": n_frame_per_class,\n \"all_steps\": all_steps,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n },\n )\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n # Store model, inference function, and reward function\n self.model = model\n \n # Create reference model (frozen clone of the initial model)\n self.ref_model = copy.deepcopy(model)\n for param in self.ref_model.parameters():\n param.requires_grad = False\n self.ref_model.eval()\n \n # prepare inference_fn\n self.inference_fn = inference_fn\n self.inference_fn.scale = self.inference_fn.scale.to(self.accelerator.device)\n self.inference_fn.tts_model = self.inference_fn.tts_model.to(self.accelerator.device)\n # prepare reward_fn\n self.reward_fn = reward_fn\n \n # Store vocabulary and mapping\n self.vocab_size = vocab_size\n self.vocab_char_map = vocab_char_map\n\n # Store duration model parameters\n self.n_class = n_class\n self.n_frame_per_class = n_frame_per_class\n self.gumbel_tau = gumbel_tau\n \n # Store GRPO parameters\n self.beta = beta\n self.clip_param = clip_param\n self.num_pre_samples = num_pre_samples\n self.compute_gen_logps = compute_gen_logps\n \n # Store training parameters\n self.learning_rate = learning_rate\n self.num_warmup_updates: int = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.checkpoint_path = checkpoint_path or f\"ckpts/{wandb_run_name}\"\n self.all_steps = all_steps\n \n # Store batch parameters\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n \n # Initialize optimizer\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n \n # Prepare model and optimizer with accelerator\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n self.ref_model = self.accelerator.prepare(self.ref_model)\n self.reward_fn, self.inference_fn = self.accelerator.prepare(self.reward_fn, self.inference_fn) \n \n # GRPO batch queue\n self.batch_queue = []\n \n # Store distributed rank\n self.rank = torch.distributed.get_rank() if torch.distributed.is_initialized() else 0\n\n self.device = f'cuda:{self.rank}'\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n \n def save_checkpoint(self, step, last=False):\n \"\"\"Save model and optimizer state\"\"\"\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict() if hasattr(self, 'scheduler') else None,\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n \n def load_checkpoint(self):\n \"\"\"Load latest checkpoint if available\"\"\"\n if (\n not self.checkpoint_path\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n\n print(f'Loading checkpoint: {latest_checkpoint}')\n checkpoint = torch.load(\n f\"{self.checkpoint_path}/{latest_checkpoint}\", \n weights_only=True, \n map_location=\"cpu\"\n )\n\n if \"step\" in checkpoint:\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if hasattr(self, 'scheduler') and checkpoint[\"scheduler_state_dict\"]:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n \n del checkpoint\n gc.collect()\n \n print(f'Successfully loaded checkpoint at step {step}')\n return step\n \n @torch.no_grad()\n def get_ref_logps(self, text_ids, mel, sampled_classes):\n \"\"\"\n Get log probabilities from the reference model for the sampled classes\n \"\"\"\n B = text_ids.shape[0]\n K = self.num_pre_samples\n with torch.no_grad():\n ref_logits = self.ref_model(text_ids=text_ids, mel=mel)[:, -1, :]\n ref_logits = ref_logits.unsqueeze(1).repeat(1, K, 1).view(B*K, -1)\n ref_log_probs = F.log_softmax(ref_logits, dim=-1)\n ref_logps = torch.gather(\n ref_log_probs, \n dim=-1, \n index=sampled_classes.unsqueeze(-1)\n ).squeeze(-1)\n return ref_logps\n \n @torch.no_grad()\n def generate_duration_samples(self, batch_inputs):\n \"\"\"\n Generate multiple duration predictions from the model for each input\n and evaluate them using the inference function and reward model\n \n Args:\n batch_inputs: Dictionary with text, prompt audio, etc.\n \n Returns:\n Dictionary with duration samples, rewards, and reference logits\n \"\"\"\n\n if self.rank == 0:\n print(\"Generating duration samples...\")\n \n # all_logits = []\n all_text_ids = []\n all_mels = []\n all_sampled_classes = []\n all_durations = []\n all_rewards = []\n all_gen_logps = []\n\n all_ctc_loss = []\n all_sv_loss = []\n\n # Fetch batch inputs\n # prompt_mel = batch_inputs['mel'].permute(0, 2, 1).to(self.device)\n prompt_mel = batch_inputs['mel'].permute(0, 2, 1) # (B, T, 100)\n prompt_text = batch_inputs['text']\n\n batch_size = prompt_mel.shape[0]\n\n # Shift text to unpair 'mel' and 'text'; The shifted text will be synthesized\n target_text = batch_inputs['target_text']\n target_text_lengths = torch.LongTensor([len(t) for t in target_text]).to(prompt_mel.device)\n try:\n full_text = [prompt+[' ']+target for prompt, target in zip(prompt_text, target_text)]\n except:\n target_text = [batch_inputs['text'][-1]] + batch_inputs['text'][:-1]\n target_text_lengths = batch_inputs['text_lengths'].clone().roll(1, 0)\n full_text = [prompt+[' ']+target for prompt, target in zip(prompt_text, target_text)]\n\n # Goes to reward model\n target_text_ids = list_str_to_idx(target_text, self.vocab_char_map).to(self.accelerator.device) # to device, the dataloader only gives list\n\n # Goes to duration model and TTS\n full_text_ids = list_str_to_idx(full_text, self.vocab_char_map).to(self.accelerator.device)\n\n # Deepcopy to separate text_ids for SLP and TTS\n slp_text_ids = full_text_ids.detach().clone()\n slp_text_ids = slp_text_ids.masked_fill(slp_text_ids==-1, self.vocab_size) # (B, L)\n\n # Pre-compute duration logits\n K = self.num_pre_samples\n B, T, _ = prompt_mel.shape\n _, L = slp_text_ids.shape\n # prompt_mel_k_repeats = prompt_mel.unsqueeze(1).repeat(1, K, 1, 1) # (B, K, T, 100)\n # slp_text_ids_k_repeats = slp_text_ids.unsqueeze(1).repeat(1, K, 1) # (B, K, L)\n\n # Run model once for B inputs\n old_logits = self.model(\n text_ids=slp_text_ids, # (B, L)\n mel=prompt_mel # (B, T, 100)\n )[:, -1, :] # (B, n_class)\n\n # Repeat each result K times along batch dimension\n old_logits = old_logits.unsqueeze(1).repeat(1, K, 1) # (B, K, n_class)\n # logits_nograd = logits_grad.detach().clone().view(B, K, -1) # (B, K, n_class)\n\n for _full_text_ids, _target_text_ids, _target_text_lengths, \\\n _prompt_mel, _old_logits in zip(\n full_text_ids, target_text_ids, target_text_lengths, \n prompt_mel, old_logits\n ):\n\n duration_sample = F.gumbel_softmax(_old_logits, tau=self.gumbel_tau, hard=True, dim=-1)\n duration2frames = torch.arange(self.n_class).float().to(self.accelerator.device) * self.n_frame_per_class\n est_frames = (duration_sample * duration2frames).sum(-1) # (K, )\n\n # Compute log probabilities of the samples\n sampled_classes = duration_sample.argmax(dim=-1)\n log_probs = F.log_softmax(_old_logits, dim=-1)\n gen_logps = torch.gather(\n log_probs, \n dim=-1, \n index=sampled_classes.unsqueeze(-1)\n ).squeeze(-1) # Shape: [K, n_class]\n \n # Generate speech using the sampled durations\n sampled_rewards = []\n\n for i in range(K):\n cur_duration = est_frames[i]\n if cur_duration == 0:\n cur_duration = cur_duration + 50 # prevent 0 duration\n infer_full_text_ids = _full_text_ids.unsqueeze(0)\n infer_prompt_mel = _prompt_mel.unsqueeze(0)\n cur_duration = cur_duration.unsqueeze(0)\n infer_target_text_ids = _target_text_ids.unsqueeze(0)\n infer_target_text_lengths = _target_text_lengths.unsqueeze(0)\n with torch.inference_mode():\n try:\n _est_mel = self.inference_fn(\n full_text_ids=infer_full_text_ids, \n prompt_mel=infer_prompt_mel, \n target_duration=cur_duration, \n teacher_steps=0\n )\n _est_mel = _est_mel.permute(0, 2, 1) # (1, T, 100)\n \n loss_dict = self.reward_fn(\n prompt_mel=infer_prompt_mel,\n est_mel=_est_mel,\n target_text_id=infer_target_text_ids,\n target_text_length=infer_target_text_lengths\n )\n # #TODO reweight the loss for reward\n reward_sim = loss_dict['loss_sim'] # 0 to 1\n reward_ctc = loss_dict['loss_ctc']\n reward = -(reward_ctc + reward_sim * 3)\n all_ctc_loss.append(reward_ctc)\n all_sv_loss.append(reward_sim)\n except Exception as e:\n if self.rank == 0:\n print(f\"Error in speech synthesis: {e}\")\n reward = torch.tensor(-1.0).to(cur_duration.device)\n \n sampled_rewards.append(reward)\n # list with length of K\n sampled_rewards = torch.stack(sampled_rewards) # (K, )\n # Normalize rewards\n if (sampled_rewards.max() - sampled_rewards.min()).item() > 1e-6:\n sampled_rewards = (sampled_rewards - sampled_rewards.mean()) / (sampled_rewards.std() + 1e-8)\n\n # Store all data\n # all_logits.append(duration_logits)\n # all_text_ids.append(duration_input_expanded[\"text_ids\"])\n # all_mels.append(duration_input_expanded[\"mel\"])\n all_sampled_classes.append(sampled_classes)\n all_durations.append(est_frames)\n all_gen_logps.append(gen_logps)\n all_rewards.extend(sampled_rewards) # list with length of B*K\n \n # Concatenate all data\n # logits = torch.cat(all_logits, dim=0)\n # text_ids = torch.cat(all_text_ids, dim=0)\n # mels = torch.cat(all_mels, dim=0)\n sampled_classes = torch.cat(all_sampled_classes, dim=0)\n durations = torch.cat(all_durations, dim=0)\n rewards = torch.stack(all_rewards) # use stack to keep the same device of elements\n gen_logps = torch.cat(all_gen_logps, dim=0)\n\n ctc_losses = torch.stack(all_ctc_loss)\n sv_losses = torch.stack(all_sv_loss)\n \n if self.is_main:\n self.accelerator.log({\n \"ctc_loss\": ctc_losses.mean().item(),\n \"sv_loss\": sv_losses.mean().item(),\n \"reward\": rewards.mean().item(),\n \"reward_min\": rewards.min().item(),\n \"reward_max\": rewards.max().item(),\n }, step=self.global_step)\n\n # # Normalize rewards\n # if (rewards.max() - rewards.min()).item() > 1e-6:\n # rewards = (rewards - rewards.mean()) / (rewards.std() + 1e-8)\n\n ref_logps = self.get_ref_logps(slp_text_ids, prompt_mel, sampled_classes)\n\n # Create batch dict similar to Qwen2.5 implementation\n batch_outputs = {\n # \"logits\": logits_grad,\n \"text_ids\": slp_text_ids,\n \"prompt_mel\": prompt_mel,\n \"rewards\": rewards,\n \"refs\": ref_logps,\n \"sampled_classes\": sampled_classes,\n \"durations\": durations,\n }\n \n if self.compute_gen_logps:\n batch_outputs[\"gen_logps\"] = gen_logps\n \n if self.rank == 0:\n print(f\"Generated {len(rewards)} samples with reward min/mean/max: {rewards.min().item():.4f}/{rewards.mean().item():.4f}/{rewards.max().item():.4f}\")\n \n return batch_outputs\n \n def GRPO_step(self, batch):\n \"\"\"\n Perform a GRPO update step\n \n Args:\n batch: Dictionary with inputs, rewards, reference logits, etc.\n \n Returns:\n Loss value\n \"\"\"\n # Extract batch data\n # NOTE: why .unsqueeze(1) ???\n rewards = batch['rewards'] #.unsqueeze(1)\n ref_logps = batch['refs'] # (B)\n sampled_classes = batch['sampled_classes'] # (B)\n prompt_mel = batch['prompt_mel']\n text_ids = batch['text_ids']\n\n # Forward pass to get current model logits\n K = self.num_pre_samples\n B, T, _ = prompt_mel.shape\n _, L = text_ids.shape\n cur_logits = self.model(\n text_ids=text_ids, # (B, L)\n mel=prompt_mel # (B, T, 100)\n )[:, -1, :]\n cur_logits = cur_logits.unsqueeze(1).repeat(1, K, 1).view(B*K, -1) \n\n # Compute current log probabilities for sampled actions\n log_probs = F.log_softmax(cur_logits, dim=-1)\n cur_logps = torch.gather(\n log_probs, \n dim=-1, \n index=sampled_classes.unsqueeze(-1)\n ).squeeze(-1) # (B)\n\n # KL divergence computation (same as in Qwen2.5 code)\n # KL = exp(ref - cur) - (ref - cur) - 1\n kl_div = torch.exp(ref_logps - cur_logps) - (ref_logps - cur_logps) - 1 # (B)\n \n # Compute probability ratio for PPO\n# ... truncated ...","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.grpo_duration_trainer.__init__","uri":"program://DMOSpeech2/function/src.grpo_duration_trainer.__init__#L54-L198","kind":"function","name":"__init__","path":"src/grpo_duration_trainer.py","language":"python","start_line":54,"end_line":198,"context_start_line":34,"context_end_line":218,"code":" # Apply temperature scaling\n scaled_logits = logits / temperature\n \n # Compute categorical distribution\n probs = F.softmax(scaled_logits, dim=-1)\n \n # Sample from the distribution once per batch element\n samples = torch.multinomial(probs, num_samples=1) # (B, 1)\n \n # Convert to one-hot encoding\n one_hot_samples = torch.zeros_like(probs).scatter_(1, samples, 1)\n \n return one_hot_samples\n\n\nclass GRPODurationTrainer:\n \"\"\"\n Trainer class that implements GRPO (Generative Reinforcement Learning from Preference Optimization)\n for a duration predictor in text-to-speech synthesis.\n \"\"\"\n def __init__(\n self,\n model, # Duration predictor model\n inference_fn, # Function to generate speech\n reward_fn, # Function to compute rewards from generated speech\n \n vocab_size: int, # Size of the vocabulary\n vocab_char_map: dict, # Mapping from characters to token IDs\n\n # Duration model parameters\n n_class: int = 301, # Number of duration classes\n n_frame_per_class: int = 10, # Number of frames per class\n gumbel_tau: int = 0.7,\n \n # GRPO parameters\n beta: float = 0.04, # KL regularization weight\n clip_param: float = 0.2, # PPO clip parameter\n num_pre_samples: int = 8, # Number of samples per prompt\n compute_gen_logps: bool = True, # Whether to compute generation log probabilities\n \n # Training parameters\n learning_rate: float = 5e-6,\n num_warmup_updates: int = 10000,\n save_per_updates: int = 10000,\n checkpoint_path: Optional[str] = None,\n all_steps: int = 100000, # Total training steps\n \n # Batch parameters\n batch_size: int = 8,\n batch_size_type: str = \"sample\",\n max_samples: int = 16,\n grad_accumulation_steps: int = 2,\n max_grad_norm: float = 1.0,\n \n # Logging parameters\n logger: Optional[str] = \"wandb\",\n wandb_project: str = \"tts-duration-grpo\",\n wandb_run_name: str = \"grpo_run\",\n wandb_resume_id: Optional[str] = None,\n \n accelerate_kwargs: dict = dict(),\n ):\n # Initialize accelerator for distributed training\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=False)\n \n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n print(f\"Using logger: {logger}\")\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if wandb_resume_id:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"beta\": beta,\n \"clip_param\": clip_param,\n \"num_pre_samples\": num_pre_samples,\n \"n_class\": n_class,\n \"n_frame_per_class\": n_frame_per_class,\n \"all_steps\": all_steps,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n },\n )\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n # Store model, inference function, and reward function\n self.model = model\n \n # Create reference model (frozen clone of the initial model)\n self.ref_model = copy.deepcopy(model)\n for param in self.ref_model.parameters():\n param.requires_grad = False\n self.ref_model.eval()\n \n # prepare inference_fn\n self.inference_fn = inference_fn\n self.inference_fn.scale = self.inference_fn.scale.to(self.accelerator.device)\n self.inference_fn.tts_model = self.inference_fn.tts_model.to(self.accelerator.device)\n # prepare reward_fn\n self.reward_fn = reward_fn\n \n # Store vocabulary and mapping\n self.vocab_size = vocab_size\n self.vocab_char_map = vocab_char_map\n\n # Store duration model parameters\n self.n_class = n_class\n self.n_frame_per_class = n_frame_per_class\n self.gumbel_tau = gumbel_tau\n \n # Store GRPO parameters\n self.beta = beta\n self.clip_param = clip_param\n self.num_pre_samples = num_pre_samples\n self.compute_gen_logps = compute_gen_logps\n \n # Store training parameters\n self.learning_rate = learning_rate\n self.num_warmup_updates: int = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.checkpoint_path = checkpoint_path or f\"ckpts/{wandb_run_name}\"\n self.all_steps = all_steps\n \n # Store batch parameters\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n \n # Initialize optimizer\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n \n # Prepare model and optimizer with accelerator\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n self.ref_model = self.accelerator.prepare(self.ref_model)\n self.reward_fn, self.inference_fn = self.accelerator.prepare(self.reward_fn, self.inference_fn) \n \n # GRPO batch queue\n self.batch_queue = []\n \n # Store distributed rank\n self.rank = torch.distributed.get_rank() if torch.distributed.is_initialized() else 0\n\n self.device = f'cuda:{self.rank}'\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n \n def save_checkpoint(self, step, last=False):\n \"\"\"Save model and optimizer state\"\"\"\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict() if hasattr(self, 'scheduler') else None,\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.grpo_duration_trainer.is_main","uri":"program://DMOSpeech2/function/src.grpo_duration_trainer.is_main#L201-L202","kind":"function","name":"is_main","path":"src/grpo_duration_trainer.py","language":"python","start_line":201,"end_line":202,"context_start_line":181,"context_end_line":222,"code":" self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n \n # Initialize optimizer\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n \n # Prepare model and optimizer with accelerator\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n self.ref_model = self.accelerator.prepare(self.ref_model)\n self.reward_fn, self.inference_fn = self.accelerator.prepare(self.reward_fn, self.inference_fn) \n \n # GRPO batch queue\n self.batch_queue = []\n \n # Store distributed rank\n self.rank = torch.distributed.get_rank() if torch.distributed.is_initialized() else 0\n\n self.device = f'cuda:{self.rank}'\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n \n def save_checkpoint(self, step, last=False):\n \"\"\"Save model and optimizer state\"\"\"\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict() if hasattr(self, 'scheduler') else None,\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n \n def load_checkpoint(self):\n \"\"\"Load latest checkpoint if available\"\"\"","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.grpo_duration_trainer.save_checkpoint","uri":"program://DMOSpeech2/function/src.grpo_duration_trainer.save_checkpoint#L204-L219","kind":"function","name":"save_checkpoint","path":"src/grpo_duration_trainer.py","language":"python","start_line":204,"end_line":219,"context_start_line":184,"context_end_line":239,"code":" # Initialize optimizer\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n \n # Prepare model and optimizer with accelerator\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n self.ref_model = self.accelerator.prepare(self.ref_model)\n self.reward_fn, self.inference_fn = self.accelerator.prepare(self.reward_fn, self.inference_fn) \n \n # GRPO batch queue\n self.batch_queue = []\n \n # Store distributed rank\n self.rank = torch.distributed.get_rank() if torch.distributed.is_initialized() else 0\n\n self.device = f'cuda:{self.rank}'\n \n @property\n def is_main(self):\n return self.accelerator.is_main_process\n \n def save_checkpoint(self, step, last=False):\n \"\"\"Save model and optimizer state\"\"\"\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict() if hasattr(self, 'scheduler') else None,\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n \n def load_checkpoint(self):\n \"\"\"Load latest checkpoint if available\"\"\"\n if (\n not self.checkpoint_path\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n\n print(f'Loading checkpoint: {latest_checkpoint}')","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.grpo_duration_trainer.load_checkpoint","uri":"program://DMOSpeech2/function/src.grpo_duration_trainer.load_checkpoint#L221-L260","kind":"function","name":"load_checkpoint","path":"src/grpo_duration_trainer.py","language":"python","start_line":221,"end_line":260,"context_start_line":201,"context_end_line":280,"code":" def is_main(self):\n return self.accelerator.is_main_process\n \n def save_checkpoint(self, step, last=False):\n \"\"\"Save model and optimizer state\"\"\"\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n scheduler_state_dict=self.scheduler.state_dict() if hasattr(self, 'scheduler') else None,\n step=step,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n \n def load_checkpoint(self):\n \"\"\"Load latest checkpoint if available\"\"\"\n if (\n not self.checkpoint_path\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n\n print(f'Loading checkpoint: {latest_checkpoint}')\n checkpoint = torch.load(\n f\"{self.checkpoint_path}/{latest_checkpoint}\", \n weights_only=True, \n map_location=\"cpu\"\n )\n\n if \"step\" in checkpoint:\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if hasattr(self, 'scheduler') and checkpoint[\"scheduler_state_dict\"]:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n \n del checkpoint\n gc.collect()\n \n print(f'Successfully loaded checkpoint at step {step}')\n return step\n \n @torch.no_grad()\n def get_ref_logps(self, text_ids, mel, sampled_classes):\n \"\"\"\n Get log probabilities from the reference model for the sampled classes\n \"\"\"\n B = text_ids.shape[0]\n K = self.num_pre_samples\n with torch.no_grad():\n ref_logits = self.ref_model(text_ids=text_ids, mel=mel)[:, -1, :]\n ref_logits = ref_logits.unsqueeze(1).repeat(1, K, 1).view(B*K, -1)\n ref_log_probs = F.log_softmax(ref_logits, dim=-1)\n ref_logps = torch.gather(\n ref_log_probs, \n dim=-1, \n index=sampled_classes.unsqueeze(-1)\n ).squeeze(-1)\n return ref_logps\n \n @torch.no_grad()","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.grpo_duration_trainer.get_ref_logps","uri":"program://DMOSpeech2/function/src.grpo_duration_trainer.get_ref_logps#L263-L278","kind":"function","name":"get_ref_logps","path":"src/grpo_duration_trainer.py","language":"python","start_line":263,"end_line":278,"context_start_line":243,"context_end_line":298,"code":" map_location=\"cpu\"\n )\n\n if \"step\" in checkpoint:\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if hasattr(self, 'scheduler') and checkpoint[\"scheduler_state_dict\"]:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n \n del checkpoint\n gc.collect()\n \n print(f'Successfully loaded checkpoint at step {step}')\n return step\n \n @torch.no_grad()\n def get_ref_logps(self, text_ids, mel, sampled_classes):\n \"\"\"\n Get log probabilities from the reference model for the sampled classes\n \"\"\"\n B = text_ids.shape[0]\n K = self.num_pre_samples\n with torch.no_grad():\n ref_logits = self.ref_model(text_ids=text_ids, mel=mel)[:, -1, :]\n ref_logits = ref_logits.unsqueeze(1).repeat(1, K, 1).view(B*K, -1)\n ref_log_probs = F.log_softmax(ref_logits, dim=-1)\n ref_logps = torch.gather(\n ref_log_probs, \n dim=-1, \n index=sampled_classes.unsqueeze(-1)\n ).squeeze(-1)\n return ref_logps\n \n @torch.no_grad()\n def generate_duration_samples(self, batch_inputs):\n \"\"\"\n Generate multiple duration predictions from the model for each input\n and evaluate them using the inference function and reward model\n \n Args:\n batch_inputs: Dictionary with text, prompt audio, etc.\n \n Returns:\n Dictionary with duration samples, rewards, and reference logits\n \"\"\"\n\n if self.rank == 0:\n print(\"Generating duration samples...\")\n \n # all_logits = []\n all_text_ids = []\n all_mels = []","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.grpo_duration_trainer.generate_duration_samples","uri":"program://DMOSpeech2/function/src.grpo_duration_trainer.generate_duration_samples#L281-L469","kind":"function","name":"generate_duration_samples","path":"src/grpo_duration_trainer.py","language":"python","start_line":281,"end_line":469,"context_start_line":261,"context_end_line":489,"code":" \n @torch.no_grad()\n def get_ref_logps(self, text_ids, mel, sampled_classes):\n \"\"\"\n Get log probabilities from the reference model for the sampled classes\n \"\"\"\n B = text_ids.shape[0]\n K = self.num_pre_samples\n with torch.no_grad():\n ref_logits = self.ref_model(text_ids=text_ids, mel=mel)[:, -1, :]\n ref_logits = ref_logits.unsqueeze(1).repeat(1, K, 1).view(B*K, -1)\n ref_log_probs = F.log_softmax(ref_logits, dim=-1)\n ref_logps = torch.gather(\n ref_log_probs, \n dim=-1, \n index=sampled_classes.unsqueeze(-1)\n ).squeeze(-1)\n return ref_logps\n \n @torch.no_grad()\n def generate_duration_samples(self, batch_inputs):\n \"\"\"\n Generate multiple duration predictions from the model for each input\n and evaluate them using the inference function and reward model\n \n Args:\n batch_inputs: Dictionary with text, prompt audio, etc.\n \n Returns:\n Dictionary with duration samples, rewards, and reference logits\n \"\"\"\n\n if self.rank == 0:\n print(\"Generating duration samples...\")\n \n # all_logits = []\n all_text_ids = []\n all_mels = []\n all_sampled_classes = []\n all_durations = []\n all_rewards = []\n all_gen_logps = []\n\n all_ctc_loss = []\n all_sv_loss = []\n\n # Fetch batch inputs\n # prompt_mel = batch_inputs['mel'].permute(0, 2, 1).to(self.device)\n prompt_mel = batch_inputs['mel'].permute(0, 2, 1) # (B, T, 100)\n prompt_text = batch_inputs['text']\n\n batch_size = prompt_mel.shape[0]\n\n # Shift text to unpair 'mel' and 'text'; The shifted text will be synthesized\n target_text = batch_inputs['target_text']\n target_text_lengths = torch.LongTensor([len(t) for t in target_text]).to(prompt_mel.device)\n try:\n full_text = [prompt+[' ']+target for prompt, target in zip(prompt_text, target_text)]\n except:\n target_text = [batch_inputs['text'][-1]] + batch_inputs['text'][:-1]\n target_text_lengths = batch_inputs['text_lengths'].clone().roll(1, 0)\n full_text = [prompt+[' ']+target for prompt, target in zip(prompt_text, target_text)]\n\n # Goes to reward model\n target_text_ids = list_str_to_idx(target_text, self.vocab_char_map).to(self.accelerator.device) # to device, the dataloader only gives list\n\n # Goes to duration model and TTS\n full_text_ids = list_str_to_idx(full_text, self.vocab_char_map).to(self.accelerator.device)\n\n # Deepcopy to separate text_ids for SLP and TTS\n slp_text_ids = full_text_ids.detach().clone()\n slp_text_ids = slp_text_ids.masked_fill(slp_text_ids==-1, self.vocab_size) # (B, L)\n\n # Pre-compute duration logits\n K = self.num_pre_samples\n B, T, _ = prompt_mel.shape\n _, L = slp_text_ids.shape\n # prompt_mel_k_repeats = prompt_mel.unsqueeze(1).repeat(1, K, 1, 1) # (B, K, T, 100)\n # slp_text_ids_k_repeats = slp_text_ids.unsqueeze(1).repeat(1, K, 1) # (B, K, L)\n\n # Run model once for B inputs\n old_logits = self.model(\n text_ids=slp_text_ids, # (B, L)\n mel=prompt_mel # (B, T, 100)\n )[:, -1, :] # (B, n_class)\n\n # Repeat each result K times along batch dimension\n old_logits = old_logits.unsqueeze(1).repeat(1, K, 1) # (B, K, n_class)\n # logits_nograd = logits_grad.detach().clone().view(B, K, -1) # (B, K, n_class)\n\n for _full_text_ids, _target_text_ids, _target_text_lengths, \\\n _prompt_mel, _old_logits in zip(\n full_text_ids, target_text_ids, target_text_lengths, \n prompt_mel, old_logits\n ):\n\n duration_sample = F.gumbel_softmax(_old_logits, tau=self.gumbel_tau, hard=True, dim=-1)\n duration2frames = torch.arange(self.n_class).float().to(self.accelerator.device) * self.n_frame_per_class\n est_frames = (duration_sample * duration2frames).sum(-1) # (K, )\n\n # Compute log probabilities of the samples\n sampled_classes = duration_sample.argmax(dim=-1)\n log_probs = F.log_softmax(_old_logits, dim=-1)\n gen_logps = torch.gather(\n log_probs, \n dim=-1, \n index=sampled_classes.unsqueeze(-1)\n ).squeeze(-1) # Shape: [K, n_class]\n \n # Generate speech using the sampled durations\n sampled_rewards = []\n\n for i in range(K):\n cur_duration = est_frames[i]\n if cur_duration == 0:\n cur_duration = cur_duration + 50 # prevent 0 duration\n infer_full_text_ids = _full_text_ids.unsqueeze(0)\n infer_prompt_mel = _prompt_mel.unsqueeze(0)\n cur_duration = cur_duration.unsqueeze(0)\n infer_target_text_ids = _target_text_ids.unsqueeze(0)\n infer_target_text_lengths = _target_text_lengths.unsqueeze(0)\n with torch.inference_mode():\n try:\n _est_mel = self.inference_fn(\n full_text_ids=infer_full_text_ids, \n prompt_mel=infer_prompt_mel, \n target_duration=cur_duration, \n teacher_steps=0\n )\n _est_mel = _est_mel.permute(0, 2, 1) # (1, T, 100)\n \n loss_dict = self.reward_fn(\n prompt_mel=infer_prompt_mel,\n est_mel=_est_mel,\n target_text_id=infer_target_text_ids,\n target_text_length=infer_target_text_lengths\n )\n # #TODO reweight the loss for reward\n reward_sim = loss_dict['loss_sim'] # 0 to 1\n reward_ctc = loss_dict['loss_ctc']\n reward = -(reward_ctc + reward_sim * 3)\n all_ctc_loss.append(reward_ctc)\n all_sv_loss.append(reward_sim)\n except Exception as e:\n if self.rank == 0:\n print(f\"Error in speech synthesis: {e}\")\n reward = torch.tensor(-1.0).to(cur_duration.device)\n \n sampled_rewards.append(reward)\n # list with length of K\n sampled_rewards = torch.stack(sampled_rewards) # (K, )\n # Normalize rewards\n if (sampled_rewards.max() - sampled_rewards.min()).item() > 1e-6:\n sampled_rewards = (sampled_rewards - sampled_rewards.mean()) / (sampled_rewards.std() + 1e-8)\n\n # Store all data\n # all_logits.append(duration_logits)\n # all_text_ids.append(duration_input_expanded[\"text_ids\"])\n # all_mels.append(duration_input_expanded[\"mel\"])\n all_sampled_classes.append(sampled_classes)\n all_durations.append(est_frames)\n all_gen_logps.append(gen_logps)\n all_rewards.extend(sampled_rewards) # list with length of B*K\n \n # Concatenate all data\n # logits = torch.cat(all_logits, dim=0)\n # text_ids = torch.cat(all_text_ids, dim=0)\n # mels = torch.cat(all_mels, dim=0)\n sampled_classes = torch.cat(all_sampled_classes, dim=0)\n durations = torch.cat(all_durations, dim=0)\n rewards = torch.stack(all_rewards) # use stack to keep the same device of elements\n gen_logps = torch.cat(all_gen_logps, dim=0)\n\n ctc_losses = torch.stack(all_ctc_loss)\n sv_losses = torch.stack(all_sv_loss)\n \n if self.is_main:\n self.accelerator.log({\n \"ctc_loss\": ctc_losses.mean().item(),\n \"sv_loss\": sv_losses.mean().item(),\n \"reward\": rewards.mean().item(),\n \"reward_min\": rewards.min().item(),\n \"reward_max\": rewards.max().item(),\n }, step=self.global_step)\n\n # # Normalize rewards\n # if (rewards.max() - rewards.min()).item() > 1e-6:\n # rewards = (rewards - rewards.mean()) / (rewards.std() + 1e-8)\n\n ref_logps = self.get_ref_logps(slp_text_ids, prompt_mel, sampled_classes)\n\n # Create batch dict similar to Qwen2.5 implementation\n batch_outputs = {\n # \"logits\": logits_grad,\n \"text_ids\": slp_text_ids,\n \"prompt_mel\": prompt_mel,\n \"rewards\": rewards,\n \"refs\": ref_logps,\n \"sampled_classes\": sampled_classes,\n \"durations\": durations,\n }\n \n if self.compute_gen_logps:\n batch_outputs[\"gen_logps\"] = gen_logps\n \n if self.rank == 0:\n print(f\"Generated {len(rewards)} samples with reward min/mean/max: {rewards.min().item():.4f}/{rewards.mean().item():.4f}/{rewards.max().item():.4f}\")\n \n return batch_outputs\n \n def GRPO_step(self, batch):\n \"\"\"\n Perform a GRPO update step\n \n Args:\n batch: Dictionary with inputs, rewards, reference logits, etc.\n \n Returns:\n Loss value\n \"\"\"\n # Extract batch data\n # NOTE: why .unsqueeze(1) ???\n rewards = batch['rewards'] #.unsqueeze(1)\n ref_logps = batch['refs'] # (B)\n sampled_classes = batch['sampled_classes'] # (B)\n prompt_mel = batch['prompt_mel']\n text_ids = batch['text_ids']\n\n # Forward pass to get current model logits","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.grpo_duration_trainer.GRPO_step","uri":"program://DMOSpeech2/function/src.grpo_duration_trainer.GRPO_step#L471-L525","kind":"function","name":"GRPO_step","path":"src/grpo_duration_trainer.py","language":"python","start_line":471,"end_line":525,"context_start_line":451,"context_end_line":545,"code":"\n # Create batch dict similar to Qwen2.5 implementation\n batch_outputs = {\n # \"logits\": logits_grad,\n \"text_ids\": slp_text_ids,\n \"prompt_mel\": prompt_mel,\n \"rewards\": rewards,\n \"refs\": ref_logps,\n \"sampled_classes\": sampled_classes,\n \"durations\": durations,\n }\n \n if self.compute_gen_logps:\n batch_outputs[\"gen_logps\"] = gen_logps\n \n if self.rank == 0:\n print(f\"Generated {len(rewards)} samples with reward min/mean/max: {rewards.min().item():.4f}/{rewards.mean().item():.4f}/{rewards.max().item():.4f}\")\n \n return batch_outputs\n \n def GRPO_step(self, batch):\n \"\"\"\n Perform a GRPO update step\n \n Args:\n batch: Dictionary with inputs, rewards, reference logits, etc.\n \n Returns:\n Loss value\n \"\"\"\n # Extract batch data\n # NOTE: why .unsqueeze(1) ???\n rewards = batch['rewards'] #.unsqueeze(1)\n ref_logps = batch['refs'] # (B)\n sampled_classes = batch['sampled_classes'] # (B)\n prompt_mel = batch['prompt_mel']\n text_ids = batch['text_ids']\n\n # Forward pass to get current model logits\n K = self.num_pre_samples\n B, T, _ = prompt_mel.shape\n _, L = text_ids.shape\n cur_logits = self.model(\n text_ids=text_ids, # (B, L)\n mel=prompt_mel # (B, T, 100)\n )[:, -1, :]\n cur_logits = cur_logits.unsqueeze(1).repeat(1, K, 1).view(B*K, -1) \n\n # Compute current log probabilities for sampled actions\n log_probs = F.log_softmax(cur_logits, dim=-1)\n cur_logps = torch.gather(\n log_probs, \n dim=-1, \n index=sampled_classes.unsqueeze(-1)\n ).squeeze(-1) # (B)\n\n # KL divergence computation (same as in Qwen2.5 code)\n # KL = exp(ref - cur) - (ref - cur) - 1\n kl_div = torch.exp(ref_logps - cur_logps) - (ref_logps - cur_logps) - 1 # (B)\n \n # Compute probability ratio for PPO\n if \"gen_logps\" in batch:\n gen_logps = batch['gen_logps']\n ratio = torch.exp(cur_logps - gen_logps)\n clipped_ratio = torch.clamp(ratio, 1 - self.clip_param, 1 + self.clip_param)\n loss = torch.min(ratio * rewards, clipped_ratio * rewards)\n else:\n # Simplification if gen_logps not available\n loss = torch.exp(cur_logps - cur_logps.detach()) * rewards\n \n # Final GRPO loss with KL regularization\n loss = -(loss - self.beta * kl_div) # (B)\n loss = loss.mean()\n \n return loss\n \n def get_batch(self):\n \"\"\"Get a batch from the queue or return None if empty\"\"\"\n if not self.batch_queue:\n return None\n return self.batch_queue.pop(0)\n \n def generate_mode(self, num_batches=5):\n \"\"\"\n Generate samples and add them to the batch queue\n \n Args:\n dataset: Dataset to sample from\n num_batches: Number of batches to generate\n \"\"\"\n if self.rank == 0:\n print(\"Entering generate mode...\")\n \n tic = time.time()\n for _ in range(num_batches):","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.grpo_duration_trainer.get_batch","uri":"program://DMOSpeech2/function/src.grpo_duration_trainer.get_batch#L527-L531","kind":"function","name":"get_batch","path":"src/grpo_duration_trainer.py","language":"python","start_line":527,"end_line":531,"context_start_line":507,"context_end_line":551,"code":" # KL divergence computation (same as in Qwen2.5 code)\n # KL = exp(ref - cur) - (ref - cur) - 1\n kl_div = torch.exp(ref_logps - cur_logps) - (ref_logps - cur_logps) - 1 # (B)\n \n # Compute probability ratio for PPO\n if \"gen_logps\" in batch:\n gen_logps = batch['gen_logps']\n ratio = torch.exp(cur_logps - gen_logps)\n clipped_ratio = torch.clamp(ratio, 1 - self.clip_param, 1 + self.clip_param)\n loss = torch.min(ratio * rewards, clipped_ratio * rewards)\n else:\n # Simplification if gen_logps not available\n loss = torch.exp(cur_logps - cur_logps.detach()) * rewards\n \n # Final GRPO loss with KL regularization\n loss = -(loss - self.beta * kl_div) # (B)\n loss = loss.mean()\n \n return loss\n \n def get_batch(self):\n \"\"\"Get a batch from the queue or return None if empty\"\"\"\n if not self.batch_queue:\n return None\n return self.batch_queue.pop(0)\n \n def generate_mode(self, num_batches=5):\n \"\"\"\n Generate samples and add them to the batch queue\n \n Args:\n dataset: Dataset to sample from\n num_batches: Number of batches to generate\n \"\"\"\n if self.rank == 0:\n print(\"Entering generate mode...\")\n \n tic = time.time()\n for _ in range(num_batches):\n try:\n batch_inputs = next(self.train_iterator)\n except StopIteration:\n self.train_iterator = iter(self.train_dataloader)\n batch_inputs = next(self.train_iterator)\n","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.grpo_duration_trainer.generate_mode","uri":"program://DMOSpeech2/function/src.grpo_duration_trainer.generate_mode#L533-L564","kind":"function","name":"generate_mode","path":"src/grpo_duration_trainer.py","language":"python","start_line":533,"end_line":564,"context_start_line":513,"context_end_line":584,"code":" gen_logps = batch['gen_logps']\n ratio = torch.exp(cur_logps - gen_logps)\n clipped_ratio = torch.clamp(ratio, 1 - self.clip_param, 1 + self.clip_param)\n loss = torch.min(ratio * rewards, clipped_ratio * rewards)\n else:\n # Simplification if gen_logps not available\n loss = torch.exp(cur_logps - cur_logps.detach()) * rewards\n \n # Final GRPO loss with KL regularization\n loss = -(loss - self.beta * kl_div) # (B)\n loss = loss.mean()\n \n return loss\n \n def get_batch(self):\n \"\"\"Get a batch from the queue or return None if empty\"\"\"\n if not self.batch_queue:\n return None\n return self.batch_queue.pop(0)\n \n def generate_mode(self, num_batches=5):\n \"\"\"\n Generate samples and add them to the batch queue\n \n Args:\n dataset: Dataset to sample from\n num_batches: Number of batches to generate\n \"\"\"\n if self.rank == 0:\n print(\"Entering generate mode...\")\n \n tic = time.time()\n for _ in range(num_batches):\n try:\n batch_inputs = next(self.train_iterator)\n except StopIteration:\n self.train_iterator = iter(self.train_dataloader)\n batch_inputs = next(self.train_iterator)\n\n # Generate samples and compute rewards\n batch_outputs = self.generate_duration_samples(batch_inputs)\n # Check if batch has sufficient reward diversity\n rewards = batch_outputs[\"rewards\"]\n if (rewards.max() - rewards.min()).item() < 0.01:\n if self.rank == 0:\n print(\"Skipping batch with low reward diversity\")\n continue\n # Add batch to queue\n self.batch_queue.append(batch_outputs)\n \n if self.rank == 0:\n print(f\"Exiting generate mode: {time.time() - tic:.3f}s\")\n \n def train(self, train_dataset, valid_dataset=None, num_workers=64, resumable_with_seed=666):\n \"\"\"\n Train the model using GRPO\n \n Args:\n train_dataset: Training dataset\n valid_dataset: Validation dataset (optional)\n num_workers: Number of workers for data loading\n \"\"\"\n\n # Create training dataloader using the appropriate batching strategy\n if self.batch_size_type == \"sample\":\n self.train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.grpo_duration_trainer.train","uri":"program://DMOSpeech2/function/src.grpo_duration_trainer.train#L566-L729","kind":"function","name":"train","path":"src/grpo_duration_trainer.py","language":"python","start_line":566,"end_line":729,"context_start_line":546,"context_end_line":729,"code":" try:\n batch_inputs = next(self.train_iterator)\n except StopIteration:\n self.train_iterator = iter(self.train_dataloader)\n batch_inputs = next(self.train_iterator)\n\n # Generate samples and compute rewards\n batch_outputs = self.generate_duration_samples(batch_inputs)\n # Check if batch has sufficient reward diversity\n rewards = batch_outputs[\"rewards\"]\n if (rewards.max() - rewards.min()).item() < 0.01:\n if self.rank == 0:\n print(\"Skipping batch with low reward diversity\")\n continue\n # Add batch to queue\n self.batch_queue.append(batch_outputs)\n \n if self.rank == 0:\n print(f\"Exiting generate mode: {time.time() - tic:.3f}s\")\n \n def train(self, train_dataset, valid_dataset=None, num_workers=64, resumable_with_seed=666):\n \"\"\"\n Train the model using GRPO\n \n Args:\n train_dataset: Training dataset\n valid_dataset: Validation dataset (optional)\n num_workers: Number of workers for data loading\n \"\"\"\n\n # Create training dataloader using the appropriate batching strategy\n if self.batch_size_type == \"sample\":\n self.train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )\n # Create validation dataloader (always sequential, no shuffling)\n self.valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n batch_size=self.batch_size,\n shuffle=False,\n )\n\n self.train_iterator = iter(self.train_dataloader)\n self.valid_iterator = iter(self.valid_dataloader)\n \n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n self.train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n\n sampler = SequentialSampler(valid_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n # Create validation dataloader (always sequential, no shuffling)\n self.valid_dataloader = DataLoader(\n valid_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True, \n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n \n self.train_dataloader, self.valid_dataloader = self.accelerator.prepare(self.train_dataloader, self.valid_dataloader)\n\n self.train_iterator = iter(self.train_dataloader)\n self.valid_iterator = iter(self.valid_dataloader)\n\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n \n\n # Setup schedulers\n warmup_steps = self.num_warmup_updates * self.accelerator.num_processes\n total_steps = self.all_steps\n decay_steps = total_steps - warmup_steps\n \n warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)\n decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)\n \n self.scheduler = SequentialLR(\n self.optimizer, \n schedulers=[warmup_scheduler, decay_scheduler], \n milestones=[warmup_steps]\n )\n \n self.scheduler = self.accelerator.prepare(self.scheduler)\n \n # Load checkpoint if available\n start_step = self.load_checkpoint()\n self.global_step = start_step\n \n # Generate initial batches\n self.generate_mode()\n \n # Training loop\n progress = range(1, self.all_steps + 1)\n \n # Skip steps that are already done\n progress = [step for step in progress if step > start_step]\n if self.is_main:\n progress = tqdm(progress, desc=\"Training\", unit=\"step\")\n \n for step in progress:\n # Get batch from queue or generate more\n batch = self.get_batch()\n while batch is None:\n self.generate_mode()\n batch = self.get_batch()\n \n # GRPO update\n with self.accelerator.accumulate(self.model):\n loss = self.GRPO_step(batch)\n # for param in self.model.parameters():\n # custom_loss = loss + 0 * param.sum() \n self.accelerator.backward(loss)\n \n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n total_norm = self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n else:\n total_norm = torch.norm(\n torch.stack([\n torch.norm(p.grad.detach(), 2)\n for p in self.model.parameters()\n if p.grad is not None\n ]),\n 2\n )\n \n self.accelerator.log({\n \"grad_norm\": total_norm.item()\n }, step=self.global_step)\n\n self.optimizer.step()\n self.scheduler.step()\n self.optimizer.zero_grad()\n \n self.global_step += 1\n \n # Log metrics\n if self.is_main:\n self.accelerator.log({\n \"loss\": loss.item(),\n \"lr\": self.scheduler.get_last_lr()[0],\n # \"avg_reward\": batch[\"rewards\"].mean().item(),\n # \"max_reward\": batch[\"rewards\"].max().item(),\n # \"min_reward\": batch[\"rewards\"].min().item(),\n }, step=self.global_step)\n progress.set_postfix(\n loss=f\"{loss.item():.4f}\",\n lr=f\"{self.scheduler.get_last_lr()[0]:.8f}\"\n )\n \n # Save checkpoint\n if self.global_step % self.save_per_updates == 0:\n self.save_checkpoint(self.global_step)\n \n # Optional validation logic could be added here\n \n # Save final checkpoint\n self.save_checkpoint(self.global_step, last=True)\n self.accelerator.end_training()","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer","uri":"program://DMOSpeech2/module/src.infer#L1-L559","kind":"module","name":"src.infer","path":"src/infer.py","language":"python","start_line":1,"end_line":559,"context_start_line":1,"context_end_line":559,"code":"import os\nimport torch\nimport torchaudio\nimport torch.nn.functional as F\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torchdiffeq import odeint\nfrom safetensors.torch import load_file\nimport IPython.display as ipd\n\n# Import F5-TTS modules\nfrom f5_tts.model import CFM, UNetT, DiT\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import (\n default, exists, list_str_to_idx, list_str_to_tensor,\n lens_to_mask, mask_from_frac_lengths, get_tokenizer\n)\nfrom f5_tts.infer.utils_infer import (\n load_vocoder, preprocess_ref_audio_text, chunk_text,\n convert_char_to_pinyin, transcribe, target_rms,\n target_sample_rate, hop_length, speed\n)\n\n# Import custom modules\nfrom unimodel import UniModel\nfrom duration_predictor import SpeechLengthPredictor\n\n\nclass DMOInference:\n \"\"\"F5-TTS Inference wrapper class for easy text-to-speech generation.\"\"\"\n \n def __init__(\n self,\n student_checkpoint_path=\"\",\n duration_predictor_path=\"\",\n device=\"cuda\",\n model_type=\"F5TTS_Base\", # \"F5TTS_Base\" or \"E2TTS_Base\"\n tokenizer=\"pinyin\",\n dataset_name=\"Emilia_ZH_EN\",\n ):\n \"\"\"\n Initialize F5-TTS inference model.\n \n Args:\n student_checkpoint_path: Path to student model checkpoint\n duration_predictor_path: Path to duration predictor checkpoint\n device: Device to run inference on\n model_type: Model architecture type\n tokenizer: Tokenizer type (\"pinyin\", \"char\", or \"custom\")\n dataset_name: Dataset name for tokenizer\n cuda_device_id: CUDA device ID to use\n \"\"\"\n \n self.device = device\n self.model_type = model_type\n self.tokenizer = tokenizer\n self.dataset_name = dataset_name\n \n # Model parameters\n self.target_sample_rate = 24000\n self.n_mel_channels = 100\n self.hop_length = 256\n self.real_guidance_scale = 2\n self.fake_guidance_scale = 0\n self.gen_cls_loss = False\n self.num_student_step = 4\n \n # Initialize components\n self._setup_tokenizer()\n self._setup_models(student_checkpoint_path)\n self._setup_mel_spec()\n self._setup_vocoder()\n self._setup_duration_predictor(duration_predictor_path)\n \n def _setup_tokenizer(self):\n \"\"\"Setup tokenizer and vocabulary.\"\"\"\n if self.tokenizer == \"custom\":\n tokenizer_path = self.tokenizer_path\n else:\n tokenizer_path = self.dataset_name\n \n self.vocab_char_map, self.vocab_size = get_tokenizer(tokenizer_path, self.tokenizer)\n \n def _setup_models(self, student_checkpoint_path):\n \"\"\"Initialize teacher and student models.\"\"\"\n # Model configuration\n if self.model_type == \"F5TTS_Base\":\n model_cls = DiT\n model_cfg = dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)\n elif self.model_type == \"E2TTS_Base\":\n model_cls = UNetT\n model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4)\n else:\n raise ValueError(f\"Unknown model type: {self.model_type}\")\n \n # Initialize UniModel (student)\n self.model = UniModel(\n model_cls(**model_cfg, text_num_embeds=self.vocab_size, mel_dim=self.n_mel_channels, \n second_time=self.num_student_step > 1),\n checkpoint_path=\"\",\n vocab_char_map=self.vocab_char_map,\n frac_lengths_mask=(0.5, 0.9),\n real_guidance_scale=self.real_guidance_scale,\n fake_guidance_scale=self.fake_guidance_scale,\n gen_cls_loss=self.gen_cls_loss,\n sway_coeff=0,\n )\n \n # Load student checkpoint\n checkpoint = torch.load(student_checkpoint_path, map_location='cpu')\n self.model.load_state_dict(checkpoint['model_state_dict'], strict=False)\n \n # Setup generator and teacher\n self.generator = self.model.feedforward_model.to(self.device)\n self.teacher = self.model.guidance_model.real_unet.to(self.device)\n \n self.scale = checkpoint['scale']\n \n def _setup_mel_spec(self):\n \"\"\"Initialize mel spectrogram module.\"\"\"\n mel_spec_kwargs = dict(\n target_sample_rate=self.target_sample_rate,\n n_mel_channels=self.n_mel_channels,\n hop_length=self.hop_length,\n )\n self.mel_spec = MelSpec(**mel_spec_kwargs)\n \n def _setup_vocoder(self):\n \"\"\"Initialize vocoder.\"\"\"\n self.vocos = load_vocoder(is_local=False, local_path=\"\")\n self.vocos = self.vocos.to(self.device)\n \n def _setup_duration_predictor(self, checkpoint_path):\n \"\"\"Initialize duration predictor.\"\"\"\n self.wav2mel = MelSpec(\n target_sample_rate=24000,\n n_mel_channels=100,\n hop_length=256,\n win_length=1024,\n n_fft=1024,\n mel_spec_type='vocos'\n ).to(self.device)\n \n self.SLP = SpeechLengthPredictor(\n vocab_size=2545,\n n_mel=100,\n hidden_dim=512,\n n_text_layer=4,\n n_cross_layer=4,\n n_head=8,\n output_dim=301\n ).to(self.device)\n \n self.SLP.eval()\n self.SLP.load_state_dict(torch.load(checkpoint_path, map_location='cpu')['model_state_dict'])\n \n def predict_duration(self, pmt_wav_path, tar_text, pmt_text, dp_softmax_range=0.7, temperature=0):\n \"\"\"\n Predict duration for target text based on prompt audio.\n \n Args:\n pmt_wav_path: Path to prompt audio\n tar_text: Target text to generate\n pmt_text: Prompt text\n dp_softmax_range: softmax annliation range from rate-based duration\n temperature: temperature for softmax sampling (if 0, will use argmax)\n Returns:\n Estimated duration in frames\n \"\"\"\n pmt_wav, sr = torchaudio.load(pmt_wav_path)\n if sr != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, self.target_sample_rate)\n pmt_wav = resampler(pmt_wav)\n if pmt_wav.size(0) > 1:\n pmt_wav = pmt_wav[0].unsqueeze(0)\n pmt_wav = pmt_wav.to(self.device)\n \n pmt_mel = self.wav2mel(pmt_wav).permute(0, 2, 1)\n tar_tokens = self._convert_to_pinyin(list(tar_text))\n pmt_tokens = self._convert_to_pinyin(list(pmt_text))\n \n # Calculate duration\n ref_text_len = len(pmt_tokens)\n gen_text_len = len(tar_tokens)\n ref_audio_len = pmt_mel.size(1)\n duration = int(ref_audio_len / ref_text_len * gen_text_len / speed)\n duration = duration // 10\n \n min_duration = max(int(duration * dp_softmax_range), 0)\n max_duration = min(int(duration * (1 + dp_softmax_range)), 301)\n \n all_tokens = pmt_tokens + [' '] + tar_tokens\n \n text_ids = list_str_to_idx([all_tokens], self.vocab_char_map).to(self.device)\n text_ids = text_ids.masked_fill(text_ids == -1, self.vocab_size)\n \n with torch.no_grad():\n predictions = self.SLP(text_ids=text_ids, mel=pmt_mel)\n predictions = predictions[:, -1, :]\n predictions[:, :min_duration] = float('-inf')\n predictions[:, max_duration:] = float('-inf')\n \n if temperature == 0:\n est_label = predictions.argmax(-1)[..., -1].item() * 10\n else:\n probs = torch.softmax(predictions / temperature, dim=-1)\n sampled_idx = torch.multinomial(probs.squeeze(0), num_samples=1) # Remove the -1 index\n est_label = sampled_idx.item() * 10\n \n return est_label\n \n def _convert_to_pinyin(self, char_list):\n \"\"\"Convert character list to pinyin.\"\"\"\n result = []\n for x in convert_char_to_pinyin(char_list):\n result = result + x\n while result[0] == ' ' and len(result) > 1:\n result = result[1:]\n return result\n \n def generate(\n self,\n gen_text,\n audio_path,\n prompt_text=None,\n teacher_steps=16,\n teacher_stopping_time=0.07,\n student_start_step=1,\n duration=None,\n dp_softmax_range=0.7,\n temperature=0,\n eta=1.0,\n cfg_strength=2.0,\n sway_coefficient=-1.0,\n verbose=False\n ):\n \"\"\"\n Generate speech from text using teacher-student distillation.\n \n Args:\n gen_text: Text to generate\n audio_path: Path to prompt audio\n prompt_text: Prompt text (if None, will use ASR)\n teacher_steps: Number of teacher guidance steps\n teacher_stopping_time: When to stop teacher sampling\n student_start_step: When to start student sampling\n duration: Total duration (if None, will predict)\n dp_softmax_range: Duration predictor softmax range allowed around rate based duration\n temperature: Temperature for duration predictor sampling (0 means use argmax)\n eta: Stochasticity control (0=DDIM, 1=DDPM)\n cfg_strength: Classifier-free guidance strength\n sway_coefficient: Sway sampling coefficient\n verbose: Output sampling steps\n \n Returns:\n Generated audio waveform\n \"\"\"\n if prompt_text is None:\n prompt_text = transcribe(audio_path)\n \n # Predict duration if not provided\n if duration is None:\n duration = self.predict_duration(audio_path, gen_text, prompt_text, dp_softmax_range, temperature)\n \n # Preprocess audio and text\n ref_audio, ref_text = preprocess_ref_audio_text(audio_path, prompt_text)\n audio, sr = torchaudio.load(ref_audio)\n \n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n \n # Normalize audio\n rms = torch.sqrt(torch.mean(torch.square(audio)))\n if rms < target_rms:\n audio = audio * target_rms / rms\n \n if sr != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, self.target_sample_rate)\n audio = resampler(audio)\n \n audio = audio.to(self.device)\n \n # Prepare text\n text_list = [ref_text + gen_text]\n final_text_list = convert_char_to_pinyin(text_list)\n \n # Calculate durations\n ref_audio_len = audio.shape[-1] // self.hop_length\n if duration is None:\n ref_text_len = len(ref_text.encode(\"utf-8\"))\n gen_text_len = len(gen_text.encode(\"utf-8\"))\n duration = ref_audio_len + int(ref_audio_len / ref_text_len * gen_text_len / speed)\n else:\n duration = ref_audio_len + duration\n \n if verbose:\n print('audio:', audio.shape)\n print('text:', final_text_list)\n print('duration:', duration)\n print('eta (stochasticity):', eta) # Print eta value for debugging\n\n # Run inference\n with torch.inference_mode():\n cond, text, step_cond, cond_mask, max_duration, duration_tensor = self._prepare_inputs(\n audio, final_text_list, duration\n )\n \n # Teacher-student sampling\n if teacher_steps > 0 and student_start_step > 0:\n if verbose:\n print('Start teacher sampling with hybrid DDIM/DDPM (eta={})....'.format(eta))\n x1 = self._teacher_sampling(\n step_cond, text, cond_mask, max_duration, duration_tensor, # Use duration_tensor\n teacher_steps, teacher_stopping_time, eta, cfg_strength, verbose, sway_coefficient\n )\n else:\n x1 = step_cond\n \n if verbose:\n print('Start student sampling...')\n # Student sampling\n x1 = self._student_sampling(x1, cond, text, student_start_step, verbose, sway_coefficient)\n \n # Decode to audio\n mel = x1.permute(0, 2, 1) * self.scale\n generated_wave = self.vocos.decode(mel[..., cond_mask.sum():])\n \n return generated_wave.cpu().numpy().squeeze()\n \n def generate_teacher_only(\n self,\n gen_text,\n audio_path,\n prompt_text=None,\n teacher_steps=32,\n duration=None,\n eta=1.0,\n cfg_strength=2.0,\n sway_coefficient=-1.0\n ):\n \"\"\"\n Generate speech using teacher model only (no student distillation).\n \n Args:\n gen_text: Text to generate\n audio_path: Path to prompt audio\n prompt_text: Prompt text (if None, will use ASR)\n teacher_steps: Number of sampling steps\n duration: Total duration (if None, will predict)\n eta: Stochasticity control (0=DDIM, 1=DDPM)\n cfg_strength: Classifier-free guidance strength\n sway_coefficient: Sway sampling coefficient\n \n Returns:\n Generated audio waveform\n \"\"\"\n if prompt_text is None:\n prompt_text = transcribe(audio_path)\n \n # Predict duration if not provided\n if duration is None:\n duration = self.predict_duration(audio_path, gen_text, prompt_text)\n \n # Preprocess audio and text\n ref_audio, ref_text = preprocess_ref_audio_text(audio_path, prompt_text)\n audio, sr = torchaudio.load(ref_audio)\n \n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n \n # Normalize audio\n rms = torch.sqrt(torch.mean(torch.square(audio)))\n if rms < target_rms:\n audio = audio * target_rms / rms\n \n if sr != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, self.target_sample_rate)\n audio = resampler(audio)\n \n audio = audio.to(self.device)\n \n # Prepare text\n text_list = [ref_text + gen_text]\n final_text_list = convert_char_to_pinyin(text_list)\n \n # Calculate durations\n ref_audio_len = audio.shape[-1] // self.hop_length\n if duration is None:\n ref_text_len = len(ref_text.encode(\"utf-8\"))\n gen_text_len = len(gen_text.encode(\"utf-8\"))\n duration = ref_audio_len + int(ref_audio_len / ref_text_len * gen_text_len / speed)\n else:\n duration = ref_audio_len + duration\n \n # Run inference\n with torch.inference_mode():\n cond, text, step_cond, cond_mask, max_duration = self._prepare_inputs(\n audio, final_text_list, duration\n )\n \n # Teacher-only sampling\n x1 = self._teacher_sampling(\n step_cond, text, cond_mask, max_duration, duration,\n teacher_steps, 1.0, eta, cfg_strength, sway_coefficient # stopping_time=1.0 for full sampling\n )\n \n # Decode to audio\n mel = x1.permute(0, 2, 1) * self.scale\n generated_wave = self.vocos.decode(mel[..., cond_mask.sum():])\n \n return generated_wave\n \n def _prepare_inputs(self, audio, text_list, duration):\n \"\"\"Prepare inputs for generation.\"\"\"\n lens = None\n max_duration_limit = 4096\n \n cond = audio\n text = text_list\n \n if cond.ndim == 2:\n cond = self.mel_spec(cond)\n cond = cond.permute(0, 2, 1)\n assert cond.shape[-1] == 100\n \n cond = cond / self.scale\n batch, cond_seq_len, device = *cond.shape[:2], cond.device\n \n if not exists(lens):\n lens = torch.full((batch,), cond_seq_len, device=device, dtype=torch.long)\n \n # Process text\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n \n if exists(text):\n text_lens = (text != -1).sum(dim=-1)\n lens = torch.maximum(text_lens, lens)\n \n # Process duration\n cond_mask = lens_to_mask(lens)\n \n if isinstance(duration, int):\n duration = torch.full((batch,), duration, device=device, dtype=torch.long)\n \n duration = torch.maximum(lens + 1, duration)\n duration = duration.clamp(max=max_duration_limit)\n max_duration = duration.amax()\n \n # Pad conditioning\n cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value=0.0)\n cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value=False)\n cond_mask = cond_mask.unsqueeze(-1)\n step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))\n \n return cond, text, step_cond, cond_mask, max_duration, duration\n \n def _teacher_sampling(self, step_cond, text, cond_mask, max_duration, duration,\n teacher_steps, teacher_stopping_time, eta, cfg_strength, verbose, sway_sampling_coef = -1):\n \"\"\"Perform teacher model sampling.\"\"\"\n device = step_cond.device\n \n # Pre-generate noise sequence for stochastic sampling\n noise_seq = None\n if eta > 0:\n noise_seq = [torch.randn(1, max_duration, 100, device=device) \n for _ in range(teacher_steps)]\n \n def fn(t, x):\n with torch.inference_mode():\n with torch.autocast(device_type=\"cuda\", dtype=torch.float16):\n if verbose:\n print(f'current t: {t}')\n step_frac = 1.0 - t.item()\n step_idx = min(int(step_frac * len(noise_seq)), len(noise_seq) - 1) if noise_seq else 0\n \n # Predict flow\n pred = self.teacher(\n x=x, cond=step_cond, text=text, time=t, mask=None,\n drop_audio_cond=False, drop_text=False\n )\n \n if cfg_strength > 1e-5:\n null_pred = self.teacher(\n x=x, cond=step_cond, text=text, time=t, mask=None,\n drop_audio_cond=True, drop_text=True\n )\n pred = pred + (pred - null_pred) * cfg_strength\n \n # Add stochasticity if eta > 0\n if eta > 0 and noise_seq is not None:\n alpha_t = 1.0 - t.item()\n sigma_t = t.item()\n noise_scale = torch.sqrt(torch.tensor(\n (sigma_t**2) / (alpha_t**2 + sigma_t**2) * eta,\n device=device\n ))\n return pred + noise_scale * noise_seq[step_idx]\n else:\n return pred\n \n # Initialize noise\n y0 = []\n for dur in duration:\n y0.append(torch.randn(dur, 100, device=device, dtype=step_cond.dtype))\n y0 = pad_sequence(y0, padding_value=0, batch_first=True)\n \n # Setup time steps\n t = torch.linspace(0, 1, teacher_steps + 1, device=device, dtype=step_cond.dtype)\n if sway_sampling_coef is not None:\n t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)\n t = t[:(t > teacher_stoppin\n# ... truncated ...","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer.DMOInference","uri":"program://DMOSpeech2/class/src.infer.DMOInference#L28-L559","kind":"class","name":"DMOInference","path":"src/infer.py","language":"python","start_line":28,"end_line":559,"context_start_line":8,"context_end_line":559,"code":"import IPython.display as ipd\n\n# Import F5-TTS modules\nfrom f5_tts.model import CFM, UNetT, DiT\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import (\n default, exists, list_str_to_idx, list_str_to_tensor,\n lens_to_mask, mask_from_frac_lengths, get_tokenizer\n)\nfrom f5_tts.infer.utils_infer import (\n load_vocoder, preprocess_ref_audio_text, chunk_text,\n convert_char_to_pinyin, transcribe, target_rms,\n target_sample_rate, hop_length, speed\n)\n\n# Import custom modules\nfrom unimodel import UniModel\nfrom duration_predictor import SpeechLengthPredictor\n\n\nclass DMOInference:\n \"\"\"F5-TTS Inference wrapper class for easy text-to-speech generation.\"\"\"\n \n def __init__(\n self,\n student_checkpoint_path=\"\",\n duration_predictor_path=\"\",\n device=\"cuda\",\n model_type=\"F5TTS_Base\", # \"F5TTS_Base\" or \"E2TTS_Base\"\n tokenizer=\"pinyin\",\n dataset_name=\"Emilia_ZH_EN\",\n ):\n \"\"\"\n Initialize F5-TTS inference model.\n \n Args:\n student_checkpoint_path: Path to student model checkpoint\n duration_predictor_path: Path to duration predictor checkpoint\n device: Device to run inference on\n model_type: Model architecture type\n tokenizer: Tokenizer type (\"pinyin\", \"char\", or \"custom\")\n dataset_name: Dataset name for tokenizer\n cuda_device_id: CUDA device ID to use\n \"\"\"\n \n self.device = device\n self.model_type = model_type\n self.tokenizer = tokenizer\n self.dataset_name = dataset_name\n \n # Model parameters\n self.target_sample_rate = 24000\n self.n_mel_channels = 100\n self.hop_length = 256\n self.real_guidance_scale = 2\n self.fake_guidance_scale = 0\n self.gen_cls_loss = False\n self.num_student_step = 4\n \n # Initialize components\n self._setup_tokenizer()\n self._setup_models(student_checkpoint_path)\n self._setup_mel_spec()\n self._setup_vocoder()\n self._setup_duration_predictor(duration_predictor_path)\n \n def _setup_tokenizer(self):\n \"\"\"Setup tokenizer and vocabulary.\"\"\"\n if self.tokenizer == \"custom\":\n tokenizer_path = self.tokenizer_path\n else:\n tokenizer_path = self.dataset_name\n \n self.vocab_char_map, self.vocab_size = get_tokenizer(tokenizer_path, self.tokenizer)\n \n def _setup_models(self, student_checkpoint_path):\n \"\"\"Initialize teacher and student models.\"\"\"\n # Model configuration\n if self.model_type == \"F5TTS_Base\":\n model_cls = DiT\n model_cfg = dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)\n elif self.model_type == \"E2TTS_Base\":\n model_cls = UNetT\n model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4)\n else:\n raise ValueError(f\"Unknown model type: {self.model_type}\")\n \n # Initialize UniModel (student)\n self.model = UniModel(\n model_cls(**model_cfg, text_num_embeds=self.vocab_size, mel_dim=self.n_mel_channels, \n second_time=self.num_student_step > 1),\n checkpoint_path=\"\",\n vocab_char_map=self.vocab_char_map,\n frac_lengths_mask=(0.5, 0.9),\n real_guidance_scale=self.real_guidance_scale,\n fake_guidance_scale=self.fake_guidance_scale,\n gen_cls_loss=self.gen_cls_loss,\n sway_coeff=0,\n )\n \n # Load student checkpoint\n checkpoint = torch.load(student_checkpoint_path, map_location='cpu')\n self.model.load_state_dict(checkpoint['model_state_dict'], strict=False)\n \n # Setup generator and teacher\n self.generator = self.model.feedforward_model.to(self.device)\n self.teacher = self.model.guidance_model.real_unet.to(self.device)\n \n self.scale = checkpoint['scale']\n \n def _setup_mel_spec(self):\n \"\"\"Initialize mel spectrogram module.\"\"\"\n mel_spec_kwargs = dict(\n target_sample_rate=self.target_sample_rate,\n n_mel_channels=self.n_mel_channels,\n hop_length=self.hop_length,\n )\n self.mel_spec = MelSpec(**mel_spec_kwargs)\n \n def _setup_vocoder(self):\n \"\"\"Initialize vocoder.\"\"\"\n self.vocos = load_vocoder(is_local=False, local_path=\"\")\n self.vocos = self.vocos.to(self.device)\n \n def _setup_duration_predictor(self, checkpoint_path):\n \"\"\"Initialize duration predictor.\"\"\"\n self.wav2mel = MelSpec(\n target_sample_rate=24000,\n n_mel_channels=100,\n hop_length=256,\n win_length=1024,\n n_fft=1024,\n mel_spec_type='vocos'\n ).to(self.device)\n \n self.SLP = SpeechLengthPredictor(\n vocab_size=2545,\n n_mel=100,\n hidden_dim=512,\n n_text_layer=4,\n n_cross_layer=4,\n n_head=8,\n output_dim=301\n ).to(self.device)\n \n self.SLP.eval()\n self.SLP.load_state_dict(torch.load(checkpoint_path, map_location='cpu')['model_state_dict'])\n \n def predict_duration(self, pmt_wav_path, tar_text, pmt_text, dp_softmax_range=0.7, temperature=0):\n \"\"\"\n Predict duration for target text based on prompt audio.\n \n Args:\n pmt_wav_path: Path to prompt audio\n tar_text: Target text to generate\n pmt_text: Prompt text\n dp_softmax_range: softmax annliation range from rate-based duration\n temperature: temperature for softmax sampling (if 0, will use argmax)\n Returns:\n Estimated duration in frames\n \"\"\"\n pmt_wav, sr = torchaudio.load(pmt_wav_path)\n if sr != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, self.target_sample_rate)\n pmt_wav = resampler(pmt_wav)\n if pmt_wav.size(0) > 1:\n pmt_wav = pmt_wav[0].unsqueeze(0)\n pmt_wav = pmt_wav.to(self.device)\n \n pmt_mel = self.wav2mel(pmt_wav).permute(0, 2, 1)\n tar_tokens = self._convert_to_pinyin(list(tar_text))\n pmt_tokens = self._convert_to_pinyin(list(pmt_text))\n \n # Calculate duration\n ref_text_len = len(pmt_tokens)\n gen_text_len = len(tar_tokens)\n ref_audio_len = pmt_mel.size(1)\n duration = int(ref_audio_len / ref_text_len * gen_text_len / speed)\n duration = duration // 10\n \n min_duration = max(int(duration * dp_softmax_range), 0)\n max_duration = min(int(duration * (1 + dp_softmax_range)), 301)\n \n all_tokens = pmt_tokens + [' '] + tar_tokens\n \n text_ids = list_str_to_idx([all_tokens], self.vocab_char_map).to(self.device)\n text_ids = text_ids.masked_fill(text_ids == -1, self.vocab_size)\n \n with torch.no_grad():\n predictions = self.SLP(text_ids=text_ids, mel=pmt_mel)\n predictions = predictions[:, -1, :]\n predictions[:, :min_duration] = float('-inf')\n predictions[:, max_duration:] = float('-inf')\n \n if temperature == 0:\n est_label = predictions.argmax(-1)[..., -1].item() * 10\n else:\n probs = torch.softmax(predictions / temperature, dim=-1)\n sampled_idx = torch.multinomial(probs.squeeze(0), num_samples=1) # Remove the -1 index\n est_label = sampled_idx.item() * 10\n \n return est_label\n \n def _convert_to_pinyin(self, char_list):\n \"\"\"Convert character list to pinyin.\"\"\"\n result = []\n for x in convert_char_to_pinyin(char_list):\n result = result + x\n while result[0] == ' ' and len(result) > 1:\n result = result[1:]\n return result\n \n def generate(\n self,\n gen_text,\n audio_path,\n prompt_text=None,\n teacher_steps=16,\n teacher_stopping_time=0.07,\n student_start_step=1,\n duration=None,\n dp_softmax_range=0.7,\n temperature=0,\n eta=1.0,\n cfg_strength=2.0,\n sway_coefficient=-1.0,\n verbose=False\n ):\n \"\"\"\n Generate speech from text using teacher-student distillation.\n \n Args:\n gen_text: Text to generate\n audio_path: Path to prompt audio\n prompt_text: Prompt text (if None, will use ASR)\n teacher_steps: Number of teacher guidance steps\n teacher_stopping_time: When to stop teacher sampling\n student_start_step: When to start student sampling\n duration: Total duration (if None, will predict)\n dp_softmax_range: Duration predictor softmax range allowed around rate based duration\n temperature: Temperature for duration predictor sampling (0 means use argmax)\n eta: Stochasticity control (0=DDIM, 1=DDPM)\n cfg_strength: Classifier-free guidance strength\n sway_coefficient: Sway sampling coefficient\n verbose: Output sampling steps\n \n Returns:\n Generated audio waveform\n \"\"\"\n if prompt_text is None:\n prompt_text = transcribe(audio_path)\n \n # Predict duration if not provided\n if duration is None:\n duration = self.predict_duration(audio_path, gen_text, prompt_text, dp_softmax_range, temperature)\n \n # Preprocess audio and text\n ref_audio, ref_text = preprocess_ref_audio_text(audio_path, prompt_text)\n audio, sr = torchaudio.load(ref_audio)\n \n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n \n # Normalize audio\n rms = torch.sqrt(torch.mean(torch.square(audio)))\n if rms < target_rms:\n audio = audio * target_rms / rms\n \n if sr != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, self.target_sample_rate)\n audio = resampler(audio)\n \n audio = audio.to(self.device)\n \n # Prepare text\n text_list = [ref_text + gen_text]\n final_text_list = convert_char_to_pinyin(text_list)\n \n # Calculate durations\n ref_audio_len = audio.shape[-1] // self.hop_length\n if duration is None:\n ref_text_len = len(ref_text.encode(\"utf-8\"))\n gen_text_len = len(gen_text.encode(\"utf-8\"))\n duration = ref_audio_len + int(ref_audio_len / ref_text_len * gen_text_len / speed)\n else:\n duration = ref_audio_len + duration\n \n if verbose:\n print('audio:', audio.shape)\n print('text:', final_text_list)\n print('duration:', duration)\n print('eta (stochasticity):', eta) # Print eta value for debugging\n\n # Run inference\n with torch.inference_mode():\n cond, text, step_cond, cond_mask, max_duration, duration_tensor = self._prepare_inputs(\n audio, final_text_list, duration\n )\n \n # Teacher-student sampling\n if teacher_steps > 0 and student_start_step > 0:\n if verbose:\n print('Start teacher sampling with hybrid DDIM/DDPM (eta={})....'.format(eta))\n x1 = self._teacher_sampling(\n step_cond, text, cond_mask, max_duration, duration_tensor, # Use duration_tensor\n teacher_steps, teacher_stopping_time, eta, cfg_strength, verbose, sway_coefficient\n )\n else:\n x1 = step_cond\n \n if verbose:\n print('Start student sampling...')\n # Student sampling\n x1 = self._student_sampling(x1, cond, text, student_start_step, verbose, sway_coefficient)\n \n # Decode to audio\n mel = x1.permute(0, 2, 1) * self.scale\n generated_wave = self.vocos.decode(mel[..., cond_mask.sum():])\n \n return generated_wave.cpu().numpy().squeeze()\n \n def generate_teacher_only(\n self,\n gen_text,\n audio_path,\n prompt_text=None,\n teacher_steps=32,\n duration=None,\n eta=1.0,\n cfg_strength=2.0,\n sway_coefficient=-1.0\n ):\n \"\"\"\n Generate speech using teacher model only (no student distillation).\n \n Args:\n gen_text: Text to generate\n audio_path: Path to prompt audio\n prompt_text: Prompt text (if None, will use ASR)\n teacher_steps: Number of sampling steps\n duration: Total duration (if None, will predict)\n eta: Stochasticity control (0=DDIM, 1=DDPM)\n cfg_strength: Classifier-free guidance strength\n sway_coefficient: Sway sampling coefficient\n \n Returns:\n Generated audio waveform\n \"\"\"\n if prompt_text is None:\n prompt_text = transcribe(audio_path)\n \n # Predict duration if not provided\n if duration is None:\n duration = self.predict_duration(audio_path, gen_text, prompt_text)\n \n # Preprocess audio and text\n ref_audio, ref_text = preprocess_ref_audio_text(audio_path, prompt_text)\n audio, sr = torchaudio.load(ref_audio)\n \n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n \n # Normalize audio\n rms = torch.sqrt(torch.mean(torch.square(audio)))\n if rms < target_rms:\n audio = audio * target_rms / rms\n \n if sr != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, self.target_sample_rate)\n audio = resampler(audio)\n \n audio = audio.to(self.device)\n \n # Prepare text\n text_list = [ref_text + gen_text]\n final_text_list = convert_char_to_pinyin(text_list)\n \n # Calculate durations\n ref_audio_len = audio.shape[-1] // self.hop_length\n if duration is None:\n ref_text_len = len(ref_text.encode(\"utf-8\"))\n gen_text_len = len(gen_text.encode(\"utf-8\"))\n duration = ref_audio_len + int(ref_audio_len / ref_text_len * gen_text_len / speed)\n else:\n duration = ref_audio_len + duration\n \n # Run inference\n with torch.inference_mode():\n cond, text, step_cond, cond_mask, max_duration = self._prepare_inputs(\n audio, final_text_list, duration\n )\n \n # Teacher-only sampling\n x1 = self._teacher_sampling(\n step_cond, text, cond_mask, max_duration, duration,\n teacher_steps, 1.0, eta, cfg_strength, sway_coefficient # stopping_time=1.0 for full sampling\n )\n \n # Decode to audio\n mel = x1.permute(0, 2, 1) * self.scale\n generated_wave = self.vocos.decode(mel[..., cond_mask.sum():])\n \n return generated_wave\n \n def _prepare_inputs(self, audio, text_list, duration):\n \"\"\"Prepare inputs for generation.\"\"\"\n lens = None\n max_duration_limit = 4096\n \n cond = audio\n text = text_list\n \n if cond.ndim == 2:\n cond = self.mel_spec(cond)\n cond = cond.permute(0, 2, 1)\n assert cond.shape[-1] == 100\n \n cond = cond / self.scale\n batch, cond_seq_len, device = *cond.shape[:2], cond.device\n \n if not exists(lens):\n lens = torch.full((batch,), cond_seq_len, device=device, dtype=torch.long)\n \n # Process text\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n \n if exists(text):\n text_lens = (text != -1).sum(dim=-1)\n lens = torch.maximum(text_lens, lens)\n \n # Process duration\n cond_mask = lens_to_mask(lens)\n \n if isinstance(duration, int):\n duration = torch.full((batch,), duration, device=device, dtype=torch.long)\n \n duration = torch.maximum(lens + 1, duration)\n duration = duration.clamp(max=max_duration_limit)\n max_duration = duration.amax()\n \n # Pad conditioning\n cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value=0.0)\n cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value=False)\n cond_mask = cond_mask.unsqueeze(-1)\n step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))\n \n return cond, text, step_cond, cond_mask, max_duration, duration\n \n def _teacher_sampling(self, step_cond, text, cond_mask, max_duration, duration,\n teacher_steps, teacher_stopping_time, eta, cfg_strength, verbose, sway_sampling_coef = -1):\n \"\"\"Perform teacher model sampling.\"\"\"\n device = step_cond.device\n \n # Pre-generate noise sequence for stochastic sampling\n noise_seq = None\n if eta > 0:\n noise_seq = [torch.randn(1, max_duration, 100, device=device) \n for _ in range(teacher_steps)]\n \n def fn(t, x):\n with torch.inference_mode():\n with torch.autocast(device_type=\"cuda\", dtype=torch.float16):\n if verbose:\n print(f'current t: {t}')\n step_frac = 1.0 - t.item()\n step_idx = min(int(step_frac * len(noise_seq)), len(noise_seq) - 1) if noise_seq else 0\n \n # Predict flow\n pred = self.teacher(\n x=x, cond=step_cond, text=text, time=t, mask=None,\n drop_audio_cond=False, drop_text=False\n )\n \n if cfg_strength > 1e-5:\n null_pred = self.teacher(\n x=x, cond=step_cond, text=text, time=t, mask=None,\n drop_audio_cond=True, drop_text=True\n )\n pred = pred + (pred - null_pred) * cfg_strength\n \n # Add stochasticity if eta > 0\n if eta > 0 and noise_seq is not None:\n alpha_t = 1.0 - t.item()\n sigma_t = t.item()\n noise_scale = torch.sqrt(torch.tensor(\n (sigma_t**2) / (alpha_t**2 + sigma_t**2) * eta,\n device=device\n ))\n return pred + noise_scale * noise_seq[step_idx]\n else:\n return pred\n \n # Initialize noise\n y0 = []\n for dur in duration:\n y0.append(torch.randn(dur, 100, device=device, dtype=step_cond.dtype))\n y0 = pad_sequence(y0, padding_value=0, batch_first=True)\n \n # Setup time steps\n t = torch.linspace(0, 1, teacher_steps + 1, device=device, dtype=step_cond.dtype)\n if sway_sampling_coef is not None:\n t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)\n t = t[:(t > teacher_stopping_time).float().argmax() + 2]\n t = t[:-1]\n \n # Solve ODE\n trajectory = odeint(fn, y0, t, method=\"euler\")\n \n if teacher_stopping_time < 1.0:\n# ... truncated ...","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer.__init__","uri":"program://DMOSpeech2/function/src.infer.__init__#L31-L72","kind":"function","name":"__init__","path":"src/infer.py","language":"python","start_line":31,"end_line":72,"context_start_line":11,"context_end_line":92,"code":"from f5_tts.model import CFM, UNetT, DiT\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import (\n default, exists, list_str_to_idx, list_str_to_tensor,\n lens_to_mask, mask_from_frac_lengths, get_tokenizer\n)\nfrom f5_tts.infer.utils_infer import (\n load_vocoder, preprocess_ref_audio_text, chunk_text,\n convert_char_to_pinyin, transcribe, target_rms,\n target_sample_rate, hop_length, speed\n)\n\n# Import custom modules\nfrom unimodel import UniModel\nfrom duration_predictor import SpeechLengthPredictor\n\n\nclass DMOInference:\n \"\"\"F5-TTS Inference wrapper class for easy text-to-speech generation.\"\"\"\n \n def __init__(\n self,\n student_checkpoint_path=\"\",\n duration_predictor_path=\"\",\n device=\"cuda\",\n model_type=\"F5TTS_Base\", # \"F5TTS_Base\" or \"E2TTS_Base\"\n tokenizer=\"pinyin\",\n dataset_name=\"Emilia_ZH_EN\",\n ):\n \"\"\"\n Initialize F5-TTS inference model.\n \n Args:\n student_checkpoint_path: Path to student model checkpoint\n duration_predictor_path: Path to duration predictor checkpoint\n device: Device to run inference on\n model_type: Model architecture type\n tokenizer: Tokenizer type (\"pinyin\", \"char\", or \"custom\")\n dataset_name: Dataset name for tokenizer\n cuda_device_id: CUDA device ID to use\n \"\"\"\n \n self.device = device\n self.model_type = model_type\n self.tokenizer = tokenizer\n self.dataset_name = dataset_name\n \n # Model parameters\n self.target_sample_rate = 24000\n self.n_mel_channels = 100\n self.hop_length = 256\n self.real_guidance_scale = 2\n self.fake_guidance_scale = 0\n self.gen_cls_loss = False\n self.num_student_step = 4\n \n # Initialize components\n self._setup_tokenizer()\n self._setup_models(student_checkpoint_path)\n self._setup_mel_spec()\n self._setup_vocoder()\n self._setup_duration_predictor(duration_predictor_path)\n \n def _setup_tokenizer(self):\n \"\"\"Setup tokenizer and vocabulary.\"\"\"\n if self.tokenizer == \"custom\":\n tokenizer_path = self.tokenizer_path\n else:\n tokenizer_path = self.dataset_name\n \n self.vocab_char_map, self.vocab_size = get_tokenizer(tokenizer_path, self.tokenizer)\n \n def _setup_models(self, student_checkpoint_path):\n \"\"\"Initialize teacher and student models.\"\"\"\n # Model configuration\n if self.model_type == \"F5TTS_Base\":\n model_cls = DiT\n model_cfg = dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)\n elif self.model_type == \"E2TTS_Base\":\n model_cls = UNetT\n model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4)\n else:","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer._setup_tokenizer","uri":"program://DMOSpeech2/function/src.infer._setup_tokenizer#L74-L81","kind":"function","name":"_setup_tokenizer","path":"src/infer.py","language":"python","start_line":74,"end_line":81,"context_start_line":54,"context_end_line":101,"code":" self.model_type = model_type\n self.tokenizer = tokenizer\n self.dataset_name = dataset_name\n \n # Model parameters\n self.target_sample_rate = 24000\n self.n_mel_channels = 100\n self.hop_length = 256\n self.real_guidance_scale = 2\n self.fake_guidance_scale = 0\n self.gen_cls_loss = False\n self.num_student_step = 4\n \n # Initialize components\n self._setup_tokenizer()\n self._setup_models(student_checkpoint_path)\n self._setup_mel_spec()\n self._setup_vocoder()\n self._setup_duration_predictor(duration_predictor_path)\n \n def _setup_tokenizer(self):\n \"\"\"Setup tokenizer and vocabulary.\"\"\"\n if self.tokenizer == \"custom\":\n tokenizer_path = self.tokenizer_path\n else:\n tokenizer_path = self.dataset_name\n \n self.vocab_char_map, self.vocab_size = get_tokenizer(tokenizer_path, self.tokenizer)\n \n def _setup_models(self, student_checkpoint_path):\n \"\"\"Initialize teacher and student models.\"\"\"\n # Model configuration\n if self.model_type == \"F5TTS_Base\":\n model_cls = DiT\n model_cfg = dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)\n elif self.model_type == \"E2TTS_Base\":\n model_cls = UNetT\n model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4)\n else:\n raise ValueError(f\"Unknown model type: {self.model_type}\")\n \n # Initialize UniModel (student)\n self.model = UniModel(\n model_cls(**model_cfg, text_num_embeds=self.vocab_size, mel_dim=self.n_mel_channels, \n second_time=self.num_student_step > 1),\n checkpoint_path=\"\",\n vocab_char_map=self.vocab_char_map,\n frac_lengths_mask=(0.5, 0.9),","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer._setup_models","uri":"program://DMOSpeech2/function/src.infer._setup_models#L83-L116","kind":"function","name":"_setup_models","path":"src/infer.py","language":"python","start_line":83,"end_line":116,"context_start_line":63,"context_end_line":136,"code":" self.fake_guidance_scale = 0\n self.gen_cls_loss = False\n self.num_student_step = 4\n \n # Initialize components\n self._setup_tokenizer()\n self._setup_models(student_checkpoint_path)\n self._setup_mel_spec()\n self._setup_vocoder()\n self._setup_duration_predictor(duration_predictor_path)\n \n def _setup_tokenizer(self):\n \"\"\"Setup tokenizer and vocabulary.\"\"\"\n if self.tokenizer == \"custom\":\n tokenizer_path = self.tokenizer_path\n else:\n tokenizer_path = self.dataset_name\n \n self.vocab_char_map, self.vocab_size = get_tokenizer(tokenizer_path, self.tokenizer)\n \n def _setup_models(self, student_checkpoint_path):\n \"\"\"Initialize teacher and student models.\"\"\"\n # Model configuration\n if self.model_type == \"F5TTS_Base\":\n model_cls = DiT\n model_cfg = dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)\n elif self.model_type == \"E2TTS_Base\":\n model_cls = UNetT\n model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4)\n else:\n raise ValueError(f\"Unknown model type: {self.model_type}\")\n \n # Initialize UniModel (student)\n self.model = UniModel(\n model_cls(**model_cfg, text_num_embeds=self.vocab_size, mel_dim=self.n_mel_channels, \n second_time=self.num_student_step > 1),\n checkpoint_path=\"\",\n vocab_char_map=self.vocab_char_map,\n frac_lengths_mask=(0.5, 0.9),\n real_guidance_scale=self.real_guidance_scale,\n fake_guidance_scale=self.fake_guidance_scale,\n gen_cls_loss=self.gen_cls_loss,\n sway_coeff=0,\n )\n \n # Load student checkpoint\n checkpoint = torch.load(student_checkpoint_path, map_location='cpu')\n self.model.load_state_dict(checkpoint['model_state_dict'], strict=False)\n \n # Setup generator and teacher\n self.generator = self.model.feedforward_model.to(self.device)\n self.teacher = self.model.guidance_model.real_unet.to(self.device)\n \n self.scale = checkpoint['scale']\n \n def _setup_mel_spec(self):\n \"\"\"Initialize mel spectrogram module.\"\"\"\n mel_spec_kwargs = dict(\n target_sample_rate=self.target_sample_rate,\n n_mel_channels=self.n_mel_channels,\n hop_length=self.hop_length,\n )\n self.mel_spec = MelSpec(**mel_spec_kwargs)\n \n def _setup_vocoder(self):\n \"\"\"Initialize vocoder.\"\"\"\n self.vocos = load_vocoder(is_local=False, local_path=\"\")\n self.vocos = self.vocos.to(self.device)\n \n def _setup_duration_predictor(self, checkpoint_path):\n \"\"\"Initialize duration predictor.\"\"\"\n self.wav2mel = MelSpec(\n target_sample_rate=24000,\n n_mel_channels=100,","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer._setup_mel_spec","uri":"program://DMOSpeech2/function/src.infer._setup_mel_spec#L118-L125","kind":"function","name":"_setup_mel_spec","path":"src/infer.py","language":"python","start_line":118,"end_line":125,"context_start_line":98,"context_end_line":145,"code":" second_time=self.num_student_step > 1),\n checkpoint_path=\"\",\n vocab_char_map=self.vocab_char_map,\n frac_lengths_mask=(0.5, 0.9),\n real_guidance_scale=self.real_guidance_scale,\n fake_guidance_scale=self.fake_guidance_scale,\n gen_cls_loss=self.gen_cls_loss,\n sway_coeff=0,\n )\n \n # Load student checkpoint\n checkpoint = torch.load(student_checkpoint_path, map_location='cpu')\n self.model.load_state_dict(checkpoint['model_state_dict'], strict=False)\n \n # Setup generator and teacher\n self.generator = self.model.feedforward_model.to(self.device)\n self.teacher = self.model.guidance_model.real_unet.to(self.device)\n \n self.scale = checkpoint['scale']\n \n def _setup_mel_spec(self):\n \"\"\"Initialize mel spectrogram module.\"\"\"\n mel_spec_kwargs = dict(\n target_sample_rate=self.target_sample_rate,\n n_mel_channels=self.n_mel_channels,\n hop_length=self.hop_length,\n )\n self.mel_spec = MelSpec(**mel_spec_kwargs)\n \n def _setup_vocoder(self):\n \"\"\"Initialize vocoder.\"\"\"\n self.vocos = load_vocoder(is_local=False, local_path=\"\")\n self.vocos = self.vocos.to(self.device)\n \n def _setup_duration_predictor(self, checkpoint_path):\n \"\"\"Initialize duration predictor.\"\"\"\n self.wav2mel = MelSpec(\n target_sample_rate=24000,\n n_mel_channels=100,\n hop_length=256,\n win_length=1024,\n n_fft=1024,\n mel_spec_type='vocos'\n ).to(self.device)\n \n self.SLP = SpeechLengthPredictor(\n vocab_size=2545,\n n_mel=100,","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer._setup_vocoder","uri":"program://DMOSpeech2/function/src.infer._setup_vocoder#L127-L130","kind":"function","name":"_setup_vocoder","path":"src/infer.py","language":"python","start_line":127,"end_line":130,"context_start_line":107,"context_end_line":150,"code":" \n # Load student checkpoint\n checkpoint = torch.load(student_checkpoint_path, map_location='cpu')\n self.model.load_state_dict(checkpoint['model_state_dict'], strict=False)\n \n # Setup generator and teacher\n self.generator = self.model.feedforward_model.to(self.device)\n self.teacher = self.model.guidance_model.real_unet.to(self.device)\n \n self.scale = checkpoint['scale']\n \n def _setup_mel_spec(self):\n \"\"\"Initialize mel spectrogram module.\"\"\"\n mel_spec_kwargs = dict(\n target_sample_rate=self.target_sample_rate,\n n_mel_channels=self.n_mel_channels,\n hop_length=self.hop_length,\n )\n self.mel_spec = MelSpec(**mel_spec_kwargs)\n \n def _setup_vocoder(self):\n \"\"\"Initialize vocoder.\"\"\"\n self.vocos = load_vocoder(is_local=False, local_path=\"\")\n self.vocos = self.vocos.to(self.device)\n \n def _setup_duration_predictor(self, checkpoint_path):\n \"\"\"Initialize duration predictor.\"\"\"\n self.wav2mel = MelSpec(\n target_sample_rate=24000,\n n_mel_channels=100,\n hop_length=256,\n win_length=1024,\n n_fft=1024,\n mel_spec_type='vocos'\n ).to(self.device)\n \n self.SLP = SpeechLengthPredictor(\n vocab_size=2545,\n n_mel=100,\n hidden_dim=512,\n n_text_layer=4,\n n_cross_layer=4,\n n_head=8,\n output_dim=301","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer._setup_duration_predictor","uri":"program://DMOSpeech2/function/src.infer._setup_duration_predictor#L132-L154","kind":"function","name":"_setup_duration_predictor","path":"src/infer.py","language":"python","start_line":132,"end_line":154,"context_start_line":112,"context_end_line":174,"code":" # Setup generator and teacher\n self.generator = self.model.feedforward_model.to(self.device)\n self.teacher = self.model.guidance_model.real_unet.to(self.device)\n \n self.scale = checkpoint['scale']\n \n def _setup_mel_spec(self):\n \"\"\"Initialize mel spectrogram module.\"\"\"\n mel_spec_kwargs = dict(\n target_sample_rate=self.target_sample_rate,\n n_mel_channels=self.n_mel_channels,\n hop_length=self.hop_length,\n )\n self.mel_spec = MelSpec(**mel_spec_kwargs)\n \n def _setup_vocoder(self):\n \"\"\"Initialize vocoder.\"\"\"\n self.vocos = load_vocoder(is_local=False, local_path=\"\")\n self.vocos = self.vocos.to(self.device)\n \n def _setup_duration_predictor(self, checkpoint_path):\n \"\"\"Initialize duration predictor.\"\"\"\n self.wav2mel = MelSpec(\n target_sample_rate=24000,\n n_mel_channels=100,\n hop_length=256,\n win_length=1024,\n n_fft=1024,\n mel_spec_type='vocos'\n ).to(self.device)\n \n self.SLP = SpeechLengthPredictor(\n vocab_size=2545,\n n_mel=100,\n hidden_dim=512,\n n_text_layer=4,\n n_cross_layer=4,\n n_head=8,\n output_dim=301\n ).to(self.device)\n \n self.SLP.eval()\n self.SLP.load_state_dict(torch.load(checkpoint_path, map_location='cpu')['model_state_dict'])\n \n def predict_duration(self, pmt_wav_path, tar_text, pmt_text, dp_softmax_range=0.7, temperature=0):\n \"\"\"\n Predict duration for target text based on prompt audio.\n \n Args:\n pmt_wav_path: Path to prompt audio\n tar_text: Target text to generate\n pmt_text: Prompt text\n dp_softmax_range: softmax annliation range from rate-based duration\n temperature: temperature for softmax sampling (if 0, will use argmax)\n Returns:\n Estimated duration in frames\n \"\"\"\n pmt_wav, sr = torchaudio.load(pmt_wav_path)\n if sr != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, self.target_sample_rate)\n pmt_wav = resampler(pmt_wav)\n if pmt_wav.size(0) > 1:\n pmt_wav = pmt_wav[0].unsqueeze(0)","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer.predict_duration","uri":"program://DMOSpeech2/function/src.infer.predict_duration#L156-L209","kind":"function","name":"predict_duration","path":"src/infer.py","language":"python","start_line":156,"end_line":209,"context_start_line":136,"context_end_line":229,"code":" n_mel_channels=100,\n hop_length=256,\n win_length=1024,\n n_fft=1024,\n mel_spec_type='vocos'\n ).to(self.device)\n \n self.SLP = SpeechLengthPredictor(\n vocab_size=2545,\n n_mel=100,\n hidden_dim=512,\n n_text_layer=4,\n n_cross_layer=4,\n n_head=8,\n output_dim=301\n ).to(self.device)\n \n self.SLP.eval()\n self.SLP.load_state_dict(torch.load(checkpoint_path, map_location='cpu')['model_state_dict'])\n \n def predict_duration(self, pmt_wav_path, tar_text, pmt_text, dp_softmax_range=0.7, temperature=0):\n \"\"\"\n Predict duration for target text based on prompt audio.\n \n Args:\n pmt_wav_path: Path to prompt audio\n tar_text: Target text to generate\n pmt_text: Prompt text\n dp_softmax_range: softmax annliation range from rate-based duration\n temperature: temperature for softmax sampling (if 0, will use argmax)\n Returns:\n Estimated duration in frames\n \"\"\"\n pmt_wav, sr = torchaudio.load(pmt_wav_path)\n if sr != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, self.target_sample_rate)\n pmt_wav = resampler(pmt_wav)\n if pmt_wav.size(0) > 1:\n pmt_wav = pmt_wav[0].unsqueeze(0)\n pmt_wav = pmt_wav.to(self.device)\n \n pmt_mel = self.wav2mel(pmt_wav).permute(0, 2, 1)\n tar_tokens = self._convert_to_pinyin(list(tar_text))\n pmt_tokens = self._convert_to_pinyin(list(pmt_text))\n \n # Calculate duration\n ref_text_len = len(pmt_tokens)\n gen_text_len = len(tar_tokens)\n ref_audio_len = pmt_mel.size(1)\n duration = int(ref_audio_len / ref_text_len * gen_text_len / speed)\n duration = duration // 10\n \n min_duration = max(int(duration * dp_softmax_range), 0)\n max_duration = min(int(duration * (1 + dp_softmax_range)), 301)\n \n all_tokens = pmt_tokens + [' '] + tar_tokens\n \n text_ids = list_str_to_idx([all_tokens], self.vocab_char_map).to(self.device)\n text_ids = text_ids.masked_fill(text_ids == -1, self.vocab_size)\n \n with torch.no_grad():\n predictions = self.SLP(text_ids=text_ids, mel=pmt_mel)\n predictions = predictions[:, -1, :]\n predictions[:, :min_duration] = float('-inf')\n predictions[:, max_duration:] = float('-inf')\n \n if temperature == 0:\n est_label = predictions.argmax(-1)[..., -1].item() * 10\n else:\n probs = torch.softmax(predictions / temperature, dim=-1)\n sampled_idx = torch.multinomial(probs.squeeze(0), num_samples=1) # Remove the -1 index\n est_label = sampled_idx.item() * 10\n \n return est_label\n \n def _convert_to_pinyin(self, char_list):\n \"\"\"Convert character list to pinyin.\"\"\"\n result = []\n for x in convert_char_to_pinyin(char_list):\n result = result + x\n while result[0] == ' ' and len(result) > 1:\n result = result[1:]\n return result\n \n def generate(\n self,\n gen_text,\n audio_path,\n prompt_text=None,\n teacher_steps=16,\n teacher_stopping_time=0.07,\n student_start_step=1,\n duration=None,\n dp_softmax_range=0.7,","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer._convert_to_pinyin","uri":"program://DMOSpeech2/function/src.infer._convert_to_pinyin#L211-L218","kind":"function","name":"_convert_to_pinyin","path":"src/infer.py","language":"python","start_line":211,"end_line":218,"context_start_line":191,"context_end_line":238,"code":" all_tokens = pmt_tokens + [' '] + tar_tokens\n \n text_ids = list_str_to_idx([all_tokens], self.vocab_char_map).to(self.device)\n text_ids = text_ids.masked_fill(text_ids == -1, self.vocab_size)\n \n with torch.no_grad():\n predictions = self.SLP(text_ids=text_ids, mel=pmt_mel)\n predictions = predictions[:, -1, :]\n predictions[:, :min_duration] = float('-inf')\n predictions[:, max_duration:] = float('-inf')\n \n if temperature == 0:\n est_label = predictions.argmax(-1)[..., -1].item() * 10\n else:\n probs = torch.softmax(predictions / temperature, dim=-1)\n sampled_idx = torch.multinomial(probs.squeeze(0), num_samples=1) # Remove the -1 index\n est_label = sampled_idx.item() * 10\n \n return est_label\n \n def _convert_to_pinyin(self, char_list):\n \"\"\"Convert character list to pinyin.\"\"\"\n result = []\n for x in convert_char_to_pinyin(char_list):\n result = result + x\n while result[0] == ' ' and len(result) > 1:\n result = result[1:]\n return result\n \n def generate(\n self,\n gen_text,\n audio_path,\n prompt_text=None,\n teacher_steps=16,\n teacher_stopping_time=0.07,\n student_start_step=1,\n duration=None,\n dp_softmax_range=0.7,\n temperature=0,\n eta=1.0,\n cfg_strength=2.0,\n sway_coefficient=-1.0,\n verbose=False\n ):\n \"\"\"\n Generate speech from text using teacher-student distillation.\n ","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer.generate","uri":"program://DMOSpeech2/function/src.infer.generate#L220-L327","kind":"function","name":"generate","path":"src/infer.py","language":"python","start_line":220,"end_line":327,"context_start_line":200,"context_end_line":347,"code":" predictions[:, max_duration:] = float('-inf')\n \n if temperature == 0:\n est_label = predictions.argmax(-1)[..., -1].item() * 10\n else:\n probs = torch.softmax(predictions / temperature, dim=-1)\n sampled_idx = torch.multinomial(probs.squeeze(0), num_samples=1) # Remove the -1 index\n est_label = sampled_idx.item() * 10\n \n return est_label\n \n def _convert_to_pinyin(self, char_list):\n \"\"\"Convert character list to pinyin.\"\"\"\n result = []\n for x in convert_char_to_pinyin(char_list):\n result = result + x\n while result[0] == ' ' and len(result) > 1:\n result = result[1:]\n return result\n \n def generate(\n self,\n gen_text,\n audio_path,\n prompt_text=None,\n teacher_steps=16,\n teacher_stopping_time=0.07,\n student_start_step=1,\n duration=None,\n dp_softmax_range=0.7,\n temperature=0,\n eta=1.0,\n cfg_strength=2.0,\n sway_coefficient=-1.0,\n verbose=False\n ):\n \"\"\"\n Generate speech from text using teacher-student distillation.\n \n Args:\n gen_text: Text to generate\n audio_path: Path to prompt audio\n prompt_text: Prompt text (if None, will use ASR)\n teacher_steps: Number of teacher guidance steps\n teacher_stopping_time: When to stop teacher sampling\n student_start_step: When to start student sampling\n duration: Total duration (if None, will predict)\n dp_softmax_range: Duration predictor softmax range allowed around rate based duration\n temperature: Temperature for duration predictor sampling (0 means use argmax)\n eta: Stochasticity control (0=DDIM, 1=DDPM)\n cfg_strength: Classifier-free guidance strength\n sway_coefficient: Sway sampling coefficient\n verbose: Output sampling steps\n \n Returns:\n Generated audio waveform\n \"\"\"\n if prompt_text is None:\n prompt_text = transcribe(audio_path)\n \n # Predict duration if not provided\n if duration is None:\n duration = self.predict_duration(audio_path, gen_text, prompt_text, dp_softmax_range, temperature)\n \n # Preprocess audio and text\n ref_audio, ref_text = preprocess_ref_audio_text(audio_path, prompt_text)\n audio, sr = torchaudio.load(ref_audio)\n \n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n \n # Normalize audio\n rms = torch.sqrt(torch.mean(torch.square(audio)))\n if rms < target_rms:\n audio = audio * target_rms / rms\n \n if sr != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, self.target_sample_rate)\n audio = resampler(audio)\n \n audio = audio.to(self.device)\n \n # Prepare text\n text_list = [ref_text + gen_text]\n final_text_list = convert_char_to_pinyin(text_list)\n \n # Calculate durations\n ref_audio_len = audio.shape[-1] // self.hop_length\n if duration is None:\n ref_text_len = len(ref_text.encode(\"utf-8\"))\n gen_text_len = len(gen_text.encode(\"utf-8\"))\n duration = ref_audio_len + int(ref_audio_len / ref_text_len * gen_text_len / speed)\n else:\n duration = ref_audio_len + duration\n \n if verbose:\n print('audio:', audio.shape)\n print('text:', final_text_list)\n print('duration:', duration)\n print('eta (stochasticity):', eta) # Print eta value for debugging\n\n # Run inference\n with torch.inference_mode():\n cond, text, step_cond, cond_mask, max_duration, duration_tensor = self._prepare_inputs(\n audio, final_text_list, duration\n )\n \n # Teacher-student sampling\n if teacher_steps > 0 and student_start_step > 0:\n if verbose:\n print('Start teacher sampling with hybrid DDIM/DDPM (eta={})....'.format(eta))\n x1 = self._teacher_sampling(\n step_cond, text, cond_mask, max_duration, duration_tensor, # Use duration_tensor\n teacher_steps, teacher_stopping_time, eta, cfg_strength, verbose, sway_coefficient\n )\n else:\n x1 = step_cond\n \n if verbose:\n print('Start student sampling...')\n # Student sampling\n x1 = self._student_sampling(x1, cond, text, student_start_step, verbose, sway_coefficient)\n \n # Decode to audio\n mel = x1.permute(0, 2, 1) * self.scale\n generated_wave = self.vocos.decode(mel[..., cond_mask.sum():])\n \n return generated_wave.cpu().numpy().squeeze()\n \n def generate_teacher_only(\n self,\n gen_text,\n audio_path,\n prompt_text=None,\n teacher_steps=32,\n duration=None,\n eta=1.0,\n cfg_strength=2.0,\n sway_coefficient=-1.0\n ):\n \"\"\"\n Generate speech using teacher model only (no student distillation).\n \n Args:\n gen_text: Text to generate\n audio_path: Path to prompt audio\n prompt_text: Prompt text (if None, will use ASR)\n teacher_steps: Number of sampling steps","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer.generate_teacher_only","uri":"program://DMOSpeech2/function/src.infer.generate_teacher_only#L329-L410","kind":"function","name":"generate_teacher_only","path":"src/infer.py","language":"python","start_line":329,"end_line":410,"context_start_line":309,"context_end_line":430,"code":" if verbose:\n print('Start teacher sampling with hybrid DDIM/DDPM (eta={})....'.format(eta))\n x1 = self._teacher_sampling(\n step_cond, text, cond_mask, max_duration, duration_tensor, # Use duration_tensor\n teacher_steps, teacher_stopping_time, eta, cfg_strength, verbose, sway_coefficient\n )\n else:\n x1 = step_cond\n \n if verbose:\n print('Start student sampling...')\n # Student sampling\n x1 = self._student_sampling(x1, cond, text, student_start_step, verbose, sway_coefficient)\n \n # Decode to audio\n mel = x1.permute(0, 2, 1) * self.scale\n generated_wave = self.vocos.decode(mel[..., cond_mask.sum():])\n \n return generated_wave.cpu().numpy().squeeze()\n \n def generate_teacher_only(\n self,\n gen_text,\n audio_path,\n prompt_text=None,\n teacher_steps=32,\n duration=None,\n eta=1.0,\n cfg_strength=2.0,\n sway_coefficient=-1.0\n ):\n \"\"\"\n Generate speech using teacher model only (no student distillation).\n \n Args:\n gen_text: Text to generate\n audio_path: Path to prompt audio\n prompt_text: Prompt text (if None, will use ASR)\n teacher_steps: Number of sampling steps\n duration: Total duration (if None, will predict)\n eta: Stochasticity control (0=DDIM, 1=DDPM)\n cfg_strength: Classifier-free guidance strength\n sway_coefficient: Sway sampling coefficient\n \n Returns:\n Generated audio waveform\n \"\"\"\n if prompt_text is None:\n prompt_text = transcribe(audio_path)\n \n # Predict duration if not provided\n if duration is None:\n duration = self.predict_duration(audio_path, gen_text, prompt_text)\n \n # Preprocess audio and text\n ref_audio, ref_text = preprocess_ref_audio_text(audio_path, prompt_text)\n audio, sr = torchaudio.load(ref_audio)\n \n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n \n # Normalize audio\n rms = torch.sqrt(torch.mean(torch.square(audio)))\n if rms < target_rms:\n audio = audio * target_rms / rms\n \n if sr != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, self.target_sample_rate)\n audio = resampler(audio)\n \n audio = audio.to(self.device)\n \n # Prepare text\n text_list = [ref_text + gen_text]\n final_text_list = convert_char_to_pinyin(text_list)\n \n # Calculate durations\n ref_audio_len = audio.shape[-1] // self.hop_length\n if duration is None:\n ref_text_len = len(ref_text.encode(\"utf-8\"))\n gen_text_len = len(gen_text.encode(\"utf-8\"))\n duration = ref_audio_len + int(ref_audio_len / ref_text_len * gen_text_len / speed)\n else:\n duration = ref_audio_len + duration\n \n # Run inference\n with torch.inference_mode():\n cond, text, step_cond, cond_mask, max_duration = self._prepare_inputs(\n audio, final_text_list, duration\n )\n \n # Teacher-only sampling\n x1 = self._teacher_sampling(\n step_cond, text, cond_mask, max_duration, duration,\n teacher_steps, 1.0, eta, cfg_strength, sway_coefficient # stopping_time=1.0 for full sampling\n )\n \n # Decode to audio\n mel = x1.permute(0, 2, 1) * self.scale\n generated_wave = self.vocos.decode(mel[..., cond_mask.sum():])\n \n return generated_wave\n \n def _prepare_inputs(self, audio, text_list, duration):\n \"\"\"Prepare inputs for generation.\"\"\"\n lens = None\n max_duration_limit = 4096\n \n cond = audio\n text = text_list\n \n if cond.ndim == 2:\n cond = self.mel_spec(cond)\n cond = cond.permute(0, 2, 1)\n assert cond.shape[-1] == 100\n \n cond = cond / self.scale\n batch, cond_seq_len, device = *cond.shape[:2], cond.device\n \n if not exists(lens):\n lens = torch.full((batch,), cond_seq_len, device=device, dtype=torch.long)\n ","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer._prepare_inputs","uri":"program://DMOSpeech2/function/src.infer._prepare_inputs#L412-L459","kind":"function","name":"_prepare_inputs","path":"src/infer.py","language":"python","start_line":412,"end_line":459,"context_start_line":392,"context_end_line":479,"code":" duration = ref_audio_len + duration\n \n # Run inference\n with torch.inference_mode():\n cond, text, step_cond, cond_mask, max_duration = self._prepare_inputs(\n audio, final_text_list, duration\n )\n \n # Teacher-only sampling\n x1 = self._teacher_sampling(\n step_cond, text, cond_mask, max_duration, duration,\n teacher_steps, 1.0, eta, cfg_strength, sway_coefficient # stopping_time=1.0 for full sampling\n )\n \n # Decode to audio\n mel = x1.permute(0, 2, 1) * self.scale\n generated_wave = self.vocos.decode(mel[..., cond_mask.sum():])\n \n return generated_wave\n \n def _prepare_inputs(self, audio, text_list, duration):\n \"\"\"Prepare inputs for generation.\"\"\"\n lens = None\n max_duration_limit = 4096\n \n cond = audio\n text = text_list\n \n if cond.ndim == 2:\n cond = self.mel_spec(cond)\n cond = cond.permute(0, 2, 1)\n assert cond.shape[-1] == 100\n \n cond = cond / self.scale\n batch, cond_seq_len, device = *cond.shape[:2], cond.device\n \n if not exists(lens):\n lens = torch.full((batch,), cond_seq_len, device=device, dtype=torch.long)\n \n # Process text\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n \n if exists(text):\n text_lens = (text != -1).sum(dim=-1)\n lens = torch.maximum(text_lens, lens)\n \n # Process duration\n cond_mask = lens_to_mask(lens)\n \n if isinstance(duration, int):\n duration = torch.full((batch,), duration, device=device, dtype=torch.long)\n \n duration = torch.maximum(lens + 1, duration)\n duration = duration.clamp(max=max_duration_limit)\n max_duration = duration.amax()\n \n # Pad conditioning\n cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value=0.0)\n cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value=False)\n cond_mask = cond_mask.unsqueeze(-1)\n step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))\n \n return cond, text, step_cond, cond_mask, max_duration, duration\n \n def _teacher_sampling(self, step_cond, text, cond_mask, max_duration, duration,\n teacher_steps, teacher_stopping_time, eta, cfg_strength, verbose, sway_sampling_coef = -1):\n \"\"\"Perform teacher model sampling.\"\"\"\n device = step_cond.device\n \n # Pre-generate noise sequence for stochastic sampling\n noise_seq = None\n if eta > 0:\n noise_seq = [torch.randn(1, max_duration, 100, device=device) \n for _ in range(teacher_steps)]\n \n def fn(t, x):\n with torch.inference_mode():\n with torch.autocast(device_type=\"cuda\", dtype=torch.float16):\n if verbose:\n print(f'current t: {t}')\n step_frac = 1.0 - t.item()\n step_idx = min(int(step_frac * len(noise_seq)), len(noise_seq) - 1) if noise_seq else 0\n ","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer._teacher_sampling","uri":"program://DMOSpeech2/function/src.infer._teacher_sampling#L461-L527","kind":"function","name":"_teacher_sampling","path":"src/infer.py","language":"python","start_line":461,"end_line":527,"context_start_line":441,"context_end_line":547,"code":" lens = torch.maximum(text_lens, lens)\n \n # Process duration\n cond_mask = lens_to_mask(lens)\n \n if isinstance(duration, int):\n duration = torch.full((batch,), duration, device=device, dtype=torch.long)\n \n duration = torch.maximum(lens + 1, duration)\n duration = duration.clamp(max=max_duration_limit)\n max_duration = duration.amax()\n \n # Pad conditioning\n cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value=0.0)\n cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value=False)\n cond_mask = cond_mask.unsqueeze(-1)\n step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))\n \n return cond, text, step_cond, cond_mask, max_duration, duration\n \n def _teacher_sampling(self, step_cond, text, cond_mask, max_duration, duration,\n teacher_steps, teacher_stopping_time, eta, cfg_strength, verbose, sway_sampling_coef = -1):\n \"\"\"Perform teacher model sampling.\"\"\"\n device = step_cond.device\n \n # Pre-generate noise sequence for stochastic sampling\n noise_seq = None\n if eta > 0:\n noise_seq = [torch.randn(1, max_duration, 100, device=device) \n for _ in range(teacher_steps)]\n \n def fn(t, x):\n with torch.inference_mode():\n with torch.autocast(device_type=\"cuda\", dtype=torch.float16):\n if verbose:\n print(f'current t: {t}')\n step_frac = 1.0 - t.item()\n step_idx = min(int(step_frac * len(noise_seq)), len(noise_seq) - 1) if noise_seq else 0\n \n # Predict flow\n pred = self.teacher(\n x=x, cond=step_cond, text=text, time=t, mask=None,\n drop_audio_cond=False, drop_text=False\n )\n \n if cfg_strength > 1e-5:\n null_pred = self.teacher(\n x=x, cond=step_cond, text=text, time=t, mask=None,\n drop_audio_cond=True, drop_text=True\n )\n pred = pred + (pred - null_pred) * cfg_strength\n \n # Add stochasticity if eta > 0\n if eta > 0 and noise_seq is not None:\n alpha_t = 1.0 - t.item()\n sigma_t = t.item()\n noise_scale = torch.sqrt(torch.tensor(\n (sigma_t**2) / (alpha_t**2 + sigma_t**2) * eta,\n device=device\n ))\n return pred + noise_scale * noise_seq[step_idx]\n else:\n return pred\n \n # Initialize noise\n y0 = []\n for dur in duration:\n y0.append(torch.randn(dur, 100, device=device, dtype=step_cond.dtype))\n y0 = pad_sequence(y0, padding_value=0, batch_first=True)\n \n # Setup time steps\n t = torch.linspace(0, 1, teacher_steps + 1, device=device, dtype=step_cond.dtype)\n if sway_sampling_coef is not None:\n t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)\n t = t[:(t > teacher_stopping_time).float().argmax() + 2]\n t = t[:-1]\n \n # Solve ODE\n trajectory = odeint(fn, y0, t, method=\"euler\")\n \n if teacher_stopping_time < 1.0:\n # If early stopping, compute final step\n pred = fn(t[-1], trajectory[-1])\n test_out = trajectory[-1] + (1 - t[-1]) * pred\n return test_out\n else:\n return trajectory[-1]\n \n def _student_sampling(self, x1, cond, text, student_start_step, verbose, sway_coeff = -1):\n \"\"\"Perform student model sampling.\"\"\"\n steps = torch.Tensor([0, 0.25, 0.5, 0.75])\n steps = steps + sway_coeff * (torch.cos(torch.pi / 2 * steps) - 1 + steps)\n steps = steps[student_start_step:]\n \n for step in steps:\n time = torch.Tensor([step]).to(x1.device)\n \n x0 = torch.randn_like(x1)\n t = time.unsqueeze(-1).unsqueeze(-1)\n phi = (1 - t) * x0 + t * x1\n \n if verbose:\n print(f'current step: {step}')\n with torch.no_grad():\n pred = self.generator(\n x=phi,\n cond=cond,","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer._student_sampling","uri":"program://DMOSpeech2/function/src.infer._student_sampling#L529-L559","kind":"function","name":"_student_sampling","path":"src/infer.py","language":"python","start_line":529,"end_line":559,"context_start_line":509,"context_end_line":559,"code":" y0 = pad_sequence(y0, padding_value=0, batch_first=True)\n \n # Setup time steps\n t = torch.linspace(0, 1, teacher_steps + 1, device=device, dtype=step_cond.dtype)\n if sway_sampling_coef is not None:\n t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)\n t = t[:(t > teacher_stopping_time).float().argmax() + 2]\n t = t[:-1]\n \n # Solve ODE\n trajectory = odeint(fn, y0, t, method=\"euler\")\n \n if teacher_stopping_time < 1.0:\n # If early stopping, compute final step\n pred = fn(t[-1], trajectory[-1])\n test_out = trajectory[-1] + (1 - t[-1]) * pred\n return test_out\n else:\n return trajectory[-1]\n \n def _student_sampling(self, x1, cond, text, student_start_step, verbose, sway_coeff = -1):\n \"\"\"Perform student model sampling.\"\"\"\n steps = torch.Tensor([0, 0.25, 0.5, 0.75])\n steps = steps + sway_coeff * (torch.cos(torch.pi / 2 * steps) - 1 + steps)\n steps = steps[student_start_step:]\n \n for step in steps:\n time = torch.Tensor([step]).to(x1.device)\n \n x0 = torch.randn_like(x1)\n t = time.unsqueeze(-1).unsqueeze(-1)\n phi = (1 - t) * x0 + t * x1\n \n if verbose:\n print(f'current step: {step}')\n with torch.no_grad():\n pred = self.generator(\n x=phi,\n cond=cond,\n text=text,\n time=time,\n drop_audio_cond=False,\n drop_text=False\n )\n \n # Predicted mel spectrogram\n output = phi + (1 - t) * pred\n \n x1 = output\n \n return x1","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.infer.fn","uri":"program://DMOSpeech2/function/src.infer.fn#L472-L503","kind":"function","name":"fn","path":"src/infer.py","language":"python","start_line":472,"end_line":503,"context_start_line":452,"context_end_line":523,"code":" \n # Pad conditioning\n cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value=0.0)\n cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value=False)\n cond_mask = cond_mask.unsqueeze(-1)\n step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))\n \n return cond, text, step_cond, cond_mask, max_duration, duration\n \n def _teacher_sampling(self, step_cond, text, cond_mask, max_duration, duration,\n teacher_steps, teacher_stopping_time, eta, cfg_strength, verbose, sway_sampling_coef = -1):\n \"\"\"Perform teacher model sampling.\"\"\"\n device = step_cond.device\n \n # Pre-generate noise sequence for stochastic sampling\n noise_seq = None\n if eta > 0:\n noise_seq = [torch.randn(1, max_duration, 100, device=device) \n for _ in range(teacher_steps)]\n \n def fn(t, x):\n with torch.inference_mode():\n with torch.autocast(device_type=\"cuda\", dtype=torch.float16):\n if verbose:\n print(f'current t: {t}')\n step_frac = 1.0 - t.item()\n step_idx = min(int(step_frac * len(noise_seq)), len(noise_seq) - 1) if noise_seq else 0\n \n # Predict flow\n pred = self.teacher(\n x=x, cond=step_cond, text=text, time=t, mask=None,\n drop_audio_cond=False, drop_text=False\n )\n \n if cfg_strength > 1e-5:\n null_pred = self.teacher(\n x=x, cond=step_cond, text=text, time=t, mask=None,\n drop_audio_cond=True, drop_text=True\n )\n pred = pred + (pred - null_pred) * cfg_strength\n \n # Add stochasticity if eta > 0\n if eta > 0 and noise_seq is not None:\n alpha_t = 1.0 - t.item()\n sigma_t = t.item()\n noise_scale = torch.sqrt(torch.tensor(\n (sigma_t**2) / (alpha_t**2 + sigma_t**2) * eta,\n device=device\n ))\n return pred + noise_scale * noise_seq[step_idx]\n else:\n return pred\n \n # Initialize noise\n y0 = []\n for dur in duration:\n y0.append(torch.randn(dur, 100, device=device, dtype=step_cond.dtype))\n y0 = pad_sequence(y0, padding_value=0, batch_first=True)\n \n # Setup time steps\n t = torch.linspace(0, 1, teacher_steps + 1, device=device, dtype=step_cond.dtype)\n if sway_sampling_coef is not None:\n t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)\n t = t[:(t > teacher_stopping_time).float().argmax() + 2]\n t = t[:-1]\n \n # Solve ODE\n trajectory = odeint(fn, y0, t, method=\"euler\")\n \n if teacher_stopping_time < 1.0:\n # If early stopping, compute final step\n pred = fn(t[-1], trajectory[-1])","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.discriminator_conformer","uri":"program://DMOSpeech2/module/src.discriminator_conformer#L1-L202","kind":"module","name":"src.discriminator_conformer","path":"src/discriminator_conformer.py","language":"python","start_line":1,"end_line":202,"context_start_line":1,"context_end_line":202,"code":"# part of the code is borrowed from https://github.com/lawlict/ECAPA-TDNN\n\nfrom __future__ import annotations\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torchaudio.transforms as trans\nfrom pathlib import Path\nfrom torchaudio.models import Conformer\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nclass ResBlock(nn.Module):\n def __init__(self, hidden_dim, n_conv=3, dropout_p=0.2):\n super().__init__()\n self._n_groups = 8\n self.blocks = nn.ModuleList([\n self._get_conv(hidden_dim, dilation=3**i, dropout_p=dropout_p)\n for i in range(n_conv)])\n\n\n def forward(self, x):\n for block in self.blocks:\n res = x\n x = block(x)\n x += res\n return x\n\n def _get_conv(self, hidden_dim, dilation, dropout_p=0.2):\n layers = [\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=dilation, dilation=dilation),\n nn.ReLU(),\n nn.GroupNorm(num_groups=self._n_groups, num_channels=hidden_dim),\n nn.Dropout(p=dropout_p),\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=1, dilation=1),\n nn.ReLU(),\n nn.Dropout(p=dropout_p)\n ]\n return nn.Sequential(*layers)\n\nclass ConformerDiscirminator(nn.Module):\n def __init__(self, input_dim, channels=512, num_layers=3, num_heads=8, depthwise_conv_kernel_size=15, use_group_norm=True):\n super().__init__()\n \n self.input_layer = nn.Conv1d(input_dim, channels, kernel_size=3, padding=1)\n\n self.resblock1 = nn.Sequential(\n ResBlock(channels),\n nn.GroupNorm(num_groups=1, num_channels=channels)\n )\n \n self.resblock2 = nn.Sequential(\n ResBlock(channels),\n nn.GroupNorm(num_groups=1, num_channels=channels)\n )\n\n self.conformer1 = Conformer(**{\"input_dim\": channels, \n \"num_heads\": num_heads, \n \"ffn_dim\": channels * 2, \n \"num_layers\": 1, \n \"depthwise_conv_kernel_size\": depthwise_conv_kernel_size // 2,\n \"use_group_norm\": use_group_norm})\n\n self.conformer2 = Conformer(**{\"input_dim\": channels, \n \"num_heads\": num_heads, \n \"ffn_dim\": channels * 2, \n \"num_layers\": num_layers - 1, \n \"depthwise_conv_kernel_size\": depthwise_conv_kernel_size,\n \"use_group_norm\": use_group_norm})\n \n self.linear = nn.Conv1d(channels, 1, kernel_size=1)\n\n def forward(self, x):\n # x = torch.stack(x, dim=1).transpose(-1, -2).flatten(start_dim=1, end_dim=2)\n x = torch.cat(x, dim=-1)\n x = x.transpose(1, 2)\n\n x = self.input_layer(x)\n\n x = self.resblock1(x)\n x = nn.functional.avg_pool1d(x, 2)\n x = self.resblock2(x)\n x = nn.functional.avg_pool1d(x, 2)\n \n # Transpose to (B, T, C) for the conformer.\n x = x.transpose(1, 2)\n batch_size, time_steps, _ = x.shape\n # Create a dummy lengths tensor (all sequences are assumed to be full length).\n lengths = torch.full((batch_size,), time_steps, device=x.device, dtype=torch.int64)\n # The built-in Conformer returns (output, output_lengths); we discard lengths.\n\n x, _ = self.conformer1(x, lengths)\n x, _ = self.conformer2(x, lengths)\n # Transpose back to (B, C, T).\n x = x.transpose(1, 2)\n\n # out = self.bn(self.pooling(out))\n out = self.linear(x).squeeze(1)\n\n return out\n\nif __name__ == \"__main__\":\n from f5_tts.model.utils import get_tokenizer\n from f5_tts.model import DiT\n\n bsz = 2\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n \n \n fake_unet = DiT(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4, text_num_embeds=vocab_size, mel_dim=80)\n\n fake_unet = fake_unet.cuda()\n\n text = [\"hello world\"] * bsz\n lens = torch.randint(1, 1000, (bsz,)).cuda()\n inp = torch.randn(bsz, lens.max(), 80).cuda()\n \n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n\n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n \n # handle text as string\n if isinstance(text, list):\n if exists(vocab_char_map):\n text = list_str_to_idx(text, vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # lens and mask\n if not exists(lens):\n lens = torch.full((batch,), seq_len, device=device)\n\n mask = lens_to_mask(lens, length=seq_len) # useless here, as collate_fn will pad to max length in batch\n frac_lengths_mask = (0.7, 1.0)\n \n # get a random span to mask out for training conditionally\n frac_lengths = torch.zeros((batch,), device=device).float().uniform_(*frac_lengths_mask)\n rand_span_mask = mask_from_frac_lengths(lens, frac_lengths)\n \n if exists(mask):\n rand_span_mask &= mask\n\n # Sample a time\n time = torch.rand((batch,), dtype=dtype, device=device)\n\n x1 = inp\n x0 = torch.randn_like(x1)\n t = time.unsqueeze(-1).unsqueeze(-1)\n \n phi = (1 - t) * x0 + t * x1\n flow = x1 - x0\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n\n layers = fake_unet(\n x=phi, \n cond=cond,\n text=text, \n time=time, \n drop_audio_cond=False,\n drop_text=False,\n classify_mode=True\n )\n\n # layers = torch.stack(layers, dim=1).transpose(-1, -2).flatten(start_dim=1, end_dim=2)\n # print(layers.shape)\n\n from ctcmodel import ConformerCTC\n ctcmodel = ConformerCTC(vocab_size=vocab_size, mel_dim=80, num_heads=8, d_hid=512, nlayers=6).cuda()\n real_out, layer = ctcmodel(inp)\n layer = layer[-3:] # only use the last 3 layers\n layer = [F.interpolate(l, mode='nearest', scale_factor=4).transpose(-1, -2) for l in layer]\n if layer[0].size(1) < layers[0].size(1):\n layer = [F.pad(l, (0, 0, 0, layers[0].size(1) - l.size(1))) for l in layer]\n \n layers = layer + layers\n\n model = ConformerDiscirminator(input_dim=23 * 1024 + 3 * 512, \n channels=512\n )\n \n\n model = model.cuda()\n print(model)\n out = model(layers)\n print(out.shape)","source_hash":"25cc145b2a3e168fd92617837b289ebc02a86aa041d3ffbd1b4e303892986baf","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.discriminator_conformer.ResBlock","uri":"program://DMOSpeech2/class/src.discriminator_conformer.ResBlock#L21-L47","kind":"class","name":"ResBlock","path":"src/discriminator_conformer.py","language":"python","start_line":21,"end_line":47,"context_start_line":1,"context_end_line":67,"code":"# part of the code is borrowed from https://github.com/lawlict/ECAPA-TDNN\n\nfrom __future__ import annotations\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torchaudio.transforms as trans\nfrom pathlib import Path\nfrom torchaudio.models import Conformer\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nclass ResBlock(nn.Module):\n def __init__(self, hidden_dim, n_conv=3, dropout_p=0.2):\n super().__init__()\n self._n_groups = 8\n self.blocks = nn.ModuleList([\n self._get_conv(hidden_dim, dilation=3**i, dropout_p=dropout_p)\n for i in range(n_conv)])\n\n\n def forward(self, x):\n for block in self.blocks:\n res = x\n x = block(x)\n x += res\n return x\n\n def _get_conv(self, hidden_dim, dilation, dropout_p=0.2):\n layers = [\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=dilation, dilation=dilation),\n nn.ReLU(),\n nn.GroupNorm(num_groups=self._n_groups, num_channels=hidden_dim),\n nn.Dropout(p=dropout_p),\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=1, dilation=1),\n nn.ReLU(),\n nn.Dropout(p=dropout_p)\n ]\n return nn.Sequential(*layers)\n\nclass ConformerDiscirminator(nn.Module):\n def __init__(self, input_dim, channels=512, num_layers=3, num_heads=8, depthwise_conv_kernel_size=15, use_group_norm=True):\n super().__init__()\n \n self.input_layer = nn.Conv1d(input_dim, channels, kernel_size=3, padding=1)\n\n self.resblock1 = nn.Sequential(\n ResBlock(channels),\n nn.GroupNorm(num_groups=1, num_channels=channels)\n )\n \n self.resblock2 = nn.Sequential(\n ResBlock(channels),\n nn.GroupNorm(num_groups=1, num_channels=channels)\n )\n\n self.conformer1 = Conformer(**{\"input_dim\": channels, \n \"num_heads\": num_heads, \n \"ffn_dim\": channels * 2, ","source_hash":"25cc145b2a3e168fd92617837b289ebc02a86aa041d3ffbd1b4e303892986baf","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.discriminator_conformer.ConformerDiscirminator","uri":"program://DMOSpeech2/class/src.discriminator_conformer.ConformerDiscirminator#L49-L108","kind":"class","name":"ConformerDiscirminator","path":"src/discriminator_conformer.py","language":"python","start_line":49,"end_line":108,"context_start_line":29,"context_end_line":128,"code":"\n def forward(self, x):\n for block in self.blocks:\n res = x\n x = block(x)\n x += res\n return x\n\n def _get_conv(self, hidden_dim, dilation, dropout_p=0.2):\n layers = [\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=dilation, dilation=dilation),\n nn.ReLU(),\n nn.GroupNorm(num_groups=self._n_groups, num_channels=hidden_dim),\n nn.Dropout(p=dropout_p),\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=1, dilation=1),\n nn.ReLU(),\n nn.Dropout(p=dropout_p)\n ]\n return nn.Sequential(*layers)\n\nclass ConformerDiscirminator(nn.Module):\n def __init__(self, input_dim, channels=512, num_layers=3, num_heads=8, depthwise_conv_kernel_size=15, use_group_norm=True):\n super().__init__()\n \n self.input_layer = nn.Conv1d(input_dim, channels, kernel_size=3, padding=1)\n\n self.resblock1 = nn.Sequential(\n ResBlock(channels),\n nn.GroupNorm(num_groups=1, num_channels=channels)\n )\n \n self.resblock2 = nn.Sequential(\n ResBlock(channels),\n nn.GroupNorm(num_groups=1, num_channels=channels)\n )\n\n self.conformer1 = Conformer(**{\"input_dim\": channels, \n \"num_heads\": num_heads, \n \"ffn_dim\": channels * 2, \n \"num_layers\": 1, \n \"depthwise_conv_kernel_size\": depthwise_conv_kernel_size // 2,\n \"use_group_norm\": use_group_norm})\n\n self.conformer2 = Conformer(**{\"input_dim\": channels, \n \"num_heads\": num_heads, \n \"ffn_dim\": channels * 2, \n \"num_layers\": num_layers - 1, \n \"depthwise_conv_kernel_size\": depthwise_conv_kernel_size,\n \"use_group_norm\": use_group_norm})\n \n self.linear = nn.Conv1d(channels, 1, kernel_size=1)\n\n def forward(self, x):\n # x = torch.stack(x, dim=1).transpose(-1, -2).flatten(start_dim=1, end_dim=2)\n x = torch.cat(x, dim=-1)\n x = x.transpose(1, 2)\n\n x = self.input_layer(x)\n\n x = self.resblock1(x)\n x = nn.functional.avg_pool1d(x, 2)\n x = self.resblock2(x)\n x = nn.functional.avg_pool1d(x, 2)\n \n # Transpose to (B, T, C) for the conformer.\n x = x.transpose(1, 2)\n batch_size, time_steps, _ = x.shape\n # Create a dummy lengths tensor (all sequences are assumed to be full length).\n lengths = torch.full((batch_size,), time_steps, device=x.device, dtype=torch.int64)\n # The built-in Conformer returns (output, output_lengths); we discard lengths.\n\n x, _ = self.conformer1(x, lengths)\n x, _ = self.conformer2(x, lengths)\n # Transpose back to (B, C, T).\n x = x.transpose(1, 2)\n\n # out = self.bn(self.pooling(out))\n out = self.linear(x).squeeze(1)\n\n return out\n\nif __name__ == \"__main__\":\n from f5_tts.model.utils import get_tokenizer\n from f5_tts.model import DiT\n\n bsz = 2\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n \n \n fake_unet = DiT(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4, text_num_embeds=vocab_size, mel_dim=80)\n\n fake_unet = fake_unet.cuda()","source_hash":"25cc145b2a3e168fd92617837b289ebc02a86aa041d3ffbd1b4e303892986baf","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.discriminator_conformer.__init__","uri":"program://DMOSpeech2/function/src.discriminator_conformer.__init__#L50-L79","kind":"function","name":"__init__","path":"src/discriminator_conformer.py","language":"python","start_line":50,"end_line":79,"context_start_line":30,"context_end_line":99,"code":" def forward(self, x):\n for block in self.blocks:\n res = x\n x = block(x)\n x += res\n return x\n\n def _get_conv(self, hidden_dim, dilation, dropout_p=0.2):\n layers = [\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=dilation, dilation=dilation),\n nn.ReLU(),\n nn.GroupNorm(num_groups=self._n_groups, num_channels=hidden_dim),\n nn.Dropout(p=dropout_p),\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=1, dilation=1),\n nn.ReLU(),\n nn.Dropout(p=dropout_p)\n ]\n return nn.Sequential(*layers)\n\nclass ConformerDiscirminator(nn.Module):\n def __init__(self, input_dim, channels=512, num_layers=3, num_heads=8, depthwise_conv_kernel_size=15, use_group_norm=True):\n super().__init__()\n \n self.input_layer = nn.Conv1d(input_dim, channels, kernel_size=3, padding=1)\n\n self.resblock1 = nn.Sequential(\n ResBlock(channels),\n nn.GroupNorm(num_groups=1, num_channels=channels)\n )\n \n self.resblock2 = nn.Sequential(\n ResBlock(channels),\n nn.GroupNorm(num_groups=1, num_channels=channels)\n )\n\n self.conformer1 = Conformer(**{\"input_dim\": channels, \n \"num_heads\": num_heads, \n \"ffn_dim\": channels * 2, \n \"num_layers\": 1, \n \"depthwise_conv_kernel_size\": depthwise_conv_kernel_size // 2,\n \"use_group_norm\": use_group_norm})\n\n self.conformer2 = Conformer(**{\"input_dim\": channels, \n \"num_heads\": num_heads, \n \"ffn_dim\": channels * 2, \n \"num_layers\": num_layers - 1, \n \"depthwise_conv_kernel_size\": depthwise_conv_kernel_size,\n \"use_group_norm\": use_group_norm})\n \n self.linear = nn.Conv1d(channels, 1, kernel_size=1)\n\n def forward(self, x):\n # x = torch.stack(x, dim=1).transpose(-1, -2).flatten(start_dim=1, end_dim=2)\n x = torch.cat(x, dim=-1)\n x = x.transpose(1, 2)\n\n x = self.input_layer(x)\n\n x = self.resblock1(x)\n x = nn.functional.avg_pool1d(x, 2)\n x = self.resblock2(x)\n x = nn.functional.avg_pool1d(x, 2)\n \n # Transpose to (B, T, C) for the conformer.\n x = x.transpose(1, 2)\n batch_size, time_steps, _ = x.shape\n # Create a dummy lengths tensor (all sequences are assumed to be full length).\n lengths = torch.full((batch_size,), time_steps, device=x.device, dtype=torch.int64)\n # The built-in Conformer returns (output, output_lengths); we discard lengths.\n","source_hash":"25cc145b2a3e168fd92617837b289ebc02a86aa041d3ffbd1b4e303892986baf","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.discriminator_conformer.forward","uri":"program://DMOSpeech2/function/src.discriminator_conformer.forward#L81-L108","kind":"function","name":"forward","path":"src/discriminator_conformer.py","language":"python","start_line":81,"end_line":108,"context_start_line":61,"context_end_line":128,"code":" ResBlock(channels),\n nn.GroupNorm(num_groups=1, num_channels=channels)\n )\n\n self.conformer1 = Conformer(**{\"input_dim\": channels, \n \"num_heads\": num_heads, \n \"ffn_dim\": channels * 2, \n \"num_layers\": 1, \n \"depthwise_conv_kernel_size\": depthwise_conv_kernel_size // 2,\n \"use_group_norm\": use_group_norm})\n\n self.conformer2 = Conformer(**{\"input_dim\": channels, \n \"num_heads\": num_heads, \n \"ffn_dim\": channels * 2, \n \"num_layers\": num_layers - 1, \n \"depthwise_conv_kernel_size\": depthwise_conv_kernel_size,\n \"use_group_norm\": use_group_norm})\n \n self.linear = nn.Conv1d(channels, 1, kernel_size=1)\n\n def forward(self, x):\n # x = torch.stack(x, dim=1).transpose(-1, -2).flatten(start_dim=1, end_dim=2)\n x = torch.cat(x, dim=-1)\n x = x.transpose(1, 2)\n\n x = self.input_layer(x)\n\n x = self.resblock1(x)\n x = nn.functional.avg_pool1d(x, 2)\n x = self.resblock2(x)\n x = nn.functional.avg_pool1d(x, 2)\n \n # Transpose to (B, T, C) for the conformer.\n x = x.transpose(1, 2)\n batch_size, time_steps, _ = x.shape\n # Create a dummy lengths tensor (all sequences are assumed to be full length).\n lengths = torch.full((batch_size,), time_steps, device=x.device, dtype=torch.int64)\n # The built-in Conformer returns (output, output_lengths); we discard lengths.\n\n x, _ = self.conformer1(x, lengths)\n x, _ = self.conformer2(x, lengths)\n # Transpose back to (B, C, T).\n x = x.transpose(1, 2)\n\n # out = self.bn(self.pooling(out))\n out = self.linear(x).squeeze(1)\n\n return out\n\nif __name__ == \"__main__\":\n from f5_tts.model.utils import get_tokenizer\n from f5_tts.model import DiT\n\n bsz = 2\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n \n \n fake_unet = DiT(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4, text_num_embeds=vocab_size, mel_dim=80)\n\n fake_unet = fake_unet.cuda()","source_hash":"25cc145b2a3e168fd92617837b289ebc02a86aa041d3ffbd1b4e303892986baf","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.discriminator_conformer._get_conv","uri":"program://DMOSpeech2/function/src.discriminator_conformer._get_conv#L37-L47","kind":"function","name":"_get_conv","path":"src/discriminator_conformer.py","language":"python","start_line":37,"end_line":47,"context_start_line":17,"context_end_line":67,"code":" lens_to_mask,\n mask_from_frac_lengths,\n)\n\nclass ResBlock(nn.Module):\n def __init__(self, hidden_dim, n_conv=3, dropout_p=0.2):\n super().__init__()\n self._n_groups = 8\n self.blocks = nn.ModuleList([\n self._get_conv(hidden_dim, dilation=3**i, dropout_p=dropout_p)\n for i in range(n_conv)])\n\n\n def forward(self, x):\n for block in self.blocks:\n res = x\n x = block(x)\n x += res\n return x\n\n def _get_conv(self, hidden_dim, dilation, dropout_p=0.2):\n layers = [\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=dilation, dilation=dilation),\n nn.ReLU(),\n nn.GroupNorm(num_groups=self._n_groups, num_channels=hidden_dim),\n nn.Dropout(p=dropout_p),\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=1, dilation=1),\n nn.ReLU(),\n nn.Dropout(p=dropout_p)\n ]\n return nn.Sequential(*layers)\n\nclass ConformerDiscirminator(nn.Module):\n def __init__(self, input_dim, channels=512, num_layers=3, num_heads=8, depthwise_conv_kernel_size=15, use_group_norm=True):\n super().__init__()\n \n self.input_layer = nn.Conv1d(input_dim, channels, kernel_size=3, padding=1)\n\n self.resblock1 = nn.Sequential(\n ResBlock(channels),\n nn.GroupNorm(num_groups=1, num_channels=channels)\n )\n \n self.resblock2 = nn.Sequential(\n ResBlock(channels),\n nn.GroupNorm(num_groups=1, num_channels=channels)\n )\n\n self.conformer1 = Conformer(**{\"input_dim\": channels, \n \"num_heads\": num_heads, \n \"ffn_dim\": channels * 2, ","source_hash":"25cc145b2a3e168fd92617837b289ebc02a86aa041d3ffbd1b4e303892986baf","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ctcmodel","uri":"program://DMOSpeech2/module/src.ctcmodel#L1-L221","kind":"module","name":"src.ctcmodel","path":"src/ctcmodel.py","language":"python","start_line":1,"end_line":221,"context_start_line":1,"context_end_line":221,"code":"from torch import nn\nimport torch \nimport copy\n\nfrom pathlib import Path\nfrom torchaudio.models import Conformer\n\n\nfrom f5_tts.model.utils import default\nfrom f5_tts.model.utils import exists\nfrom f5_tts.model.utils import list_str_to_idx\nfrom f5_tts.model.utils import list_str_to_tensor\nfrom f5_tts.model.utils import lens_to_mask\nfrom f5_tts.model.utils import mask_from_frac_lengths\n\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nclass ResBlock(nn.Module):\n def __init__(self, hidden_dim, n_conv=3, dropout_p=0.2):\n super().__init__()\n self._n_groups = 8\n self.blocks = nn.ModuleList([\n self._get_conv(hidden_dim, dilation=3**i, dropout_p=dropout_p)\n for i in range(n_conv)])\n\n\n def forward(self, x):\n for block in self.blocks:\n res = x\n x = block(x)\n x += res\n return x\n\n def _get_conv(self, hidden_dim, dilation, dropout_p=0.2):\n layers = [\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=dilation, dilation=dilation),\n nn.ReLU(),\n nn.GroupNorm(num_groups=self._n_groups, num_channels=hidden_dim),\n nn.Dropout(p=dropout_p),\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=1, dilation=1),\n nn.ReLU(),\n nn.Dropout(p=dropout_p)\n ]\n return nn.Sequential(*layers)\n\n\nclass ConformerCTC(nn.Module):\n def __init__(self,\n vocab_size,\n mel_dim=100, \n num_heads=8, \n d_hid=512, \n nlayers=6):\n super().__init__()\n \n self.mel_proj = nn.Conv1d(mel_dim, d_hid, kernel_size=3, padding=1)\n \n self.d_hid = d_hid\n \n self.resblock1 = nn.Sequential(\n ResBlock(d_hid),\n nn.GroupNorm(num_groups=1, num_channels=d_hid)\n )\n \n self.resblock2 = nn.Sequential(\n ResBlock(d_hid),\n nn.GroupNorm(num_groups=1, num_channels=d_hid)\n )\n \n\n self.conf_pre = torch.nn.ModuleList(\n [Conformer(\n input_dim=d_hid,\n num_heads=num_heads,\n ffn_dim=d_hid * 2,\n num_layers=1,\n depthwise_conv_kernel_size=15,\n use_group_norm=True,)\n for _ in range(nlayers // 2)\n ]\n )\n \n self.conf_after = torch.nn.ModuleList(\n [Conformer(\n input_dim=d_hid,\n num_heads=num_heads,\n ffn_dim=d_hid * 2,\n num_layers=1,\n depthwise_conv_kernel_size=7,\n use_group_norm=True,)\n for _ in range(nlayers // 2)\n ]\n )\n\n self.out = nn.Linear(d_hid, 1 + vocab_size) # 1 for blank\n\n self.ctc_loss = nn.CTCLoss(blank=vocab_size, zero_infinity=True).cuda()\n\n \n def forward(self, latent, text=None, text_lens=None):\n layers = []\n\n x = self.mel_proj(latent.transpose(-1, -2)).transpose(-1, -2)\n\n x = x.transpose(1, 2)\n layers.append(nn.functional.avg_pool1d(x, 4))\n # x = x.transpose(1, 2)\n\n x = self.resblock1(x)\n x = nn.functional.avg_pool1d(x, 2)\n layers.append(nn.functional.avg_pool1d(x, 2))\n x = self.resblock2(x)\n x = nn.functional.avg_pool1d(x, 2)\n layers.append(x)\n\n x = x.transpose(1, 2)\n\n batch_size, time_steps, _ = x.shape\n # Create a dummy lengths tensor (all sequences are assumed to be full length).\n input_lengths = torch.full((batch_size,), time_steps, device=x.device, dtype=torch.int64)\n\n for layer in (self.conf_pre):\n x, _ = layer(x, input_lengths)\n layers.append(x.transpose(1, 2))\n\n for layer in (self.conf_after):\n x, _ = layer(x, input_lengths)\n layers.append(x.transpose(1, 2))\n\n x = self.out(x)\n\n if text_lens is not None and text is not None:\n loss = self.ctc_loss(x.log_softmax(dim=2).transpose(0, 1), text, input_lengths, text_lens)\n return x, layers, loss\n else:\n return x, layers\n\n\nif __name__ == \"__main__\":\n from f5_tts.model.utils import get_tokenizer\n\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n \n model = ConformerCTC(vocab_size, mel_dim=80, num_heads=8, d_hid=512, nlayers=6).cuda()\n \n text = [\"hello world\"] * bsz\n lens = torch.randint(1, 1000, (bsz,)).cuda()\n inp = torch.randn(bsz, lens.max(), 80).cuda()\n \n batch, seq_len, dtype, device = *inp.shape[:2], inp.dtype, inp.device\n \n # handle text as string\n text_lens = torch.tensor([len(t) for t in text], device=device)\n if isinstance(text, list):\n if exists(vocab_char_map):\n text = list_str_to_idx(text, vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # lens and mask\n if not exists(lens):\n lens = torch.full((batch,), seq_len, device=device)\n\n out, layers, loss = model(inp, text_lens)\n\n print(out.shape)\n print(out)\n print(len(layers))\n print(torch.stack(layers, axis=1).shape)\n print(loss)\n\n probs = out.softmax(dim=2) # Convert logits to probabilities\n\n # Greedy decoding\n best_path = torch.argmax(probs, dim=2)\n\n decoded_sequences = []\n blank_idx = vocab_size\n\n char_vocab_map = list(vocab_char_map.keys())\n\n\n for batch in best_path:\n decoded_sequence = []\n previous_token = None\n\n for token in batch:\n if token != previous_token: # Collapse repeated tokens\n if token != blank_idx: # Ignore blank tokens\n decoded_sequence.append(token.item())\n previous_token = token\n\n decoded_sequences.append(decoded_sequence)\n\n # Convert token indices to characters\n decoded_texts = [''.join([char_vocab_map[token] for token in sequence]) for sequence in decoded_sequences]\n gt_texts = []\n for i in range(text_lens.size(0)):\n gt_texts.append(''.join([char_vocab_map[token] for token in text[i, :text_lens[i]]]))\n \n print(decoded_texts)\n print(gt_texts)","source_hash":"61bd6ed4c15b5affe57e2e7cf82b89734754e5cb0c0f639947fb1b307278ba06","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ctcmodel.ResBlock","uri":"program://DMOSpeech2/class/src.ctcmodel.ResBlock#L26-L52","kind":"class","name":"ResBlock","path":"src/ctcmodel.py","language":"python","start_line":26,"end_line":52,"context_start_line":6,"context_end_line":72,"code":"from torchaudio.models import Conformer\n\n\nfrom f5_tts.model.utils import default\nfrom f5_tts.model.utils import exists\nfrom f5_tts.model.utils import list_str_to_idx\nfrom f5_tts.model.utils import list_str_to_tensor\nfrom f5_tts.model.utils import lens_to_mask\nfrom f5_tts.model.utils import mask_from_frac_lengths\n\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nclass ResBlock(nn.Module):\n def __init__(self, hidden_dim, n_conv=3, dropout_p=0.2):\n super().__init__()\n self._n_groups = 8\n self.blocks = nn.ModuleList([\n self._get_conv(hidden_dim, dilation=3**i, dropout_p=dropout_p)\n for i in range(n_conv)])\n\n\n def forward(self, x):\n for block in self.blocks:\n res = x\n x = block(x)\n x += res\n return x\n\n def _get_conv(self, hidden_dim, dilation, dropout_p=0.2):\n layers = [\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=dilation, dilation=dilation),\n nn.ReLU(),\n nn.GroupNorm(num_groups=self._n_groups, num_channels=hidden_dim),\n nn.Dropout(p=dropout_p),\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=1, dilation=1),\n nn.ReLU(),\n nn.Dropout(p=dropout_p)\n ]\n return nn.Sequential(*layers)\n\n\nclass ConformerCTC(nn.Module):\n def __init__(self,\n vocab_size,\n mel_dim=100, \n num_heads=8, \n d_hid=512, \n nlayers=6):\n super().__init__()\n \n self.mel_proj = nn.Conv1d(mel_dim, d_hid, kernel_size=3, padding=1)\n \n self.d_hid = d_hid\n \n self.resblock1 = nn.Sequential(\n ResBlock(d_hid),\n nn.GroupNorm(num_groups=1, num_channels=d_hid)\n )\n ","source_hash":"61bd6ed4c15b5affe57e2e7cf82b89734754e5cb0c0f639947fb1b307278ba06","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ctcmodel.ConformerCTC","uri":"program://DMOSpeech2/class/src.ctcmodel.ConformerCTC#L55-L144","kind":"class","name":"ConformerCTC","path":"src/ctcmodel.py","language":"python","start_line":55,"end_line":144,"context_start_line":35,"context_end_line":164,"code":" def forward(self, x):\n for block in self.blocks:\n res = x\n x = block(x)\n x += res\n return x\n\n def _get_conv(self, hidden_dim, dilation, dropout_p=0.2):\n layers = [\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=dilation, dilation=dilation),\n nn.ReLU(),\n nn.GroupNorm(num_groups=self._n_groups, num_channels=hidden_dim),\n nn.Dropout(p=dropout_p),\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=1, dilation=1),\n nn.ReLU(),\n nn.Dropout(p=dropout_p)\n ]\n return nn.Sequential(*layers)\n\n\nclass ConformerCTC(nn.Module):\n def __init__(self,\n vocab_size,\n mel_dim=100, \n num_heads=8, \n d_hid=512, \n nlayers=6):\n super().__init__()\n \n self.mel_proj = nn.Conv1d(mel_dim, d_hid, kernel_size=3, padding=1)\n \n self.d_hid = d_hid\n \n self.resblock1 = nn.Sequential(\n ResBlock(d_hid),\n nn.GroupNorm(num_groups=1, num_channels=d_hid)\n )\n \n self.resblock2 = nn.Sequential(\n ResBlock(d_hid),\n nn.GroupNorm(num_groups=1, num_channels=d_hid)\n )\n \n\n self.conf_pre = torch.nn.ModuleList(\n [Conformer(\n input_dim=d_hid,\n num_heads=num_heads,\n ffn_dim=d_hid * 2,\n num_layers=1,\n depthwise_conv_kernel_size=15,\n use_group_norm=True,)\n for _ in range(nlayers // 2)\n ]\n )\n \n self.conf_after = torch.nn.ModuleList(\n [Conformer(\n input_dim=d_hid,\n num_heads=num_heads,\n ffn_dim=d_hid * 2,\n num_layers=1,\n depthwise_conv_kernel_size=7,\n use_group_norm=True,)\n for _ in range(nlayers // 2)\n ]\n )\n\n self.out = nn.Linear(d_hid, 1 + vocab_size) # 1 for blank\n\n self.ctc_loss = nn.CTCLoss(blank=vocab_size, zero_infinity=True).cuda()\n\n \n def forward(self, latent, text=None, text_lens=None):\n layers = []\n\n x = self.mel_proj(latent.transpose(-1, -2)).transpose(-1, -2)\n\n x = x.transpose(1, 2)\n layers.append(nn.functional.avg_pool1d(x, 4))\n # x = x.transpose(1, 2)\n\n x = self.resblock1(x)\n x = nn.functional.avg_pool1d(x, 2)\n layers.append(nn.functional.avg_pool1d(x, 2))\n x = self.resblock2(x)\n x = nn.functional.avg_pool1d(x, 2)\n layers.append(x)\n\n x = x.transpose(1, 2)\n\n batch_size, time_steps, _ = x.shape\n # Create a dummy lengths tensor (all sequences are assumed to be full length).\n input_lengths = torch.full((batch_size,), time_steps, device=x.device, dtype=torch.int64)\n\n for layer in (self.conf_pre):\n x, _ = layer(x, input_lengths)\n layers.append(x.transpose(1, 2))\n\n for layer in (self.conf_after):\n x, _ = layer(x, input_lengths)\n layers.append(x.transpose(1, 2))\n\n x = self.out(x)\n\n if text_lens is not None and text is not None:\n loss = self.ctc_loss(x.log_softmax(dim=2).transpose(0, 1), text, input_lengths, text_lens)\n return x, layers, loss\n else:\n return x, layers\n\n\nif __name__ == \"__main__\":\n from f5_tts.model.utils import get_tokenizer\n\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n \n model = ConformerCTC(vocab_size, mel_dim=80, num_heads=8, d_hid=512, nlayers=6).cuda()\n \n text = [\"hello world\"] * bsz","source_hash":"61bd6ed4c15b5affe57e2e7cf82b89734754e5cb0c0f639947fb1b307278ba06","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ctcmodel.__init__","uri":"program://DMOSpeech2/function/src.ctcmodel.__init__#L56-L105","kind":"function","name":"__init__","path":"src/ctcmodel.py","language":"python","start_line":56,"end_line":105,"context_start_line":36,"context_end_line":125,"code":" for block in self.blocks:\n res = x\n x = block(x)\n x += res\n return x\n\n def _get_conv(self, hidden_dim, dilation, dropout_p=0.2):\n layers = [\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=dilation, dilation=dilation),\n nn.ReLU(),\n nn.GroupNorm(num_groups=self._n_groups, num_channels=hidden_dim),\n nn.Dropout(p=dropout_p),\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=1, dilation=1),\n nn.ReLU(),\n nn.Dropout(p=dropout_p)\n ]\n return nn.Sequential(*layers)\n\n\nclass ConformerCTC(nn.Module):\n def __init__(self,\n vocab_size,\n mel_dim=100, \n num_heads=8, \n d_hid=512, \n nlayers=6):\n super().__init__()\n \n self.mel_proj = nn.Conv1d(mel_dim, d_hid, kernel_size=3, padding=1)\n \n self.d_hid = d_hid\n \n self.resblock1 = nn.Sequential(\n ResBlock(d_hid),\n nn.GroupNorm(num_groups=1, num_channels=d_hid)\n )\n \n self.resblock2 = nn.Sequential(\n ResBlock(d_hid),\n nn.GroupNorm(num_groups=1, num_channels=d_hid)\n )\n \n\n self.conf_pre = torch.nn.ModuleList(\n [Conformer(\n input_dim=d_hid,\n num_heads=num_heads,\n ffn_dim=d_hid * 2,\n num_layers=1,\n depthwise_conv_kernel_size=15,\n use_group_norm=True,)\n for _ in range(nlayers // 2)\n ]\n )\n \n self.conf_after = torch.nn.ModuleList(\n [Conformer(\n input_dim=d_hid,\n num_heads=num_heads,\n ffn_dim=d_hid * 2,\n num_layers=1,\n depthwise_conv_kernel_size=7,\n use_group_norm=True,)\n for _ in range(nlayers // 2)\n ]\n )\n\n self.out = nn.Linear(d_hid, 1 + vocab_size) # 1 for blank\n\n self.ctc_loss = nn.CTCLoss(blank=vocab_size, zero_infinity=True).cuda()\n\n \n def forward(self, latent, text=None, text_lens=None):\n layers = []\n\n x = self.mel_proj(latent.transpose(-1, -2)).transpose(-1, -2)\n\n x = x.transpose(1, 2)\n layers.append(nn.functional.avg_pool1d(x, 4))\n # x = x.transpose(1, 2)\n\n x = self.resblock1(x)\n x = nn.functional.avg_pool1d(x, 2)\n layers.append(nn.functional.avg_pool1d(x, 2))\n x = self.resblock2(x)\n x = nn.functional.avg_pool1d(x, 2)\n layers.append(x)\n\n x = x.transpose(1, 2)\n","source_hash":"61bd6ed4c15b5affe57e2e7cf82b89734754e5cb0c0f639947fb1b307278ba06","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ctcmodel.forward","uri":"program://DMOSpeech2/function/src.ctcmodel.forward#L108-L144","kind":"function","name":"forward","path":"src/ctcmodel.py","language":"python","start_line":108,"end_line":144,"context_start_line":88,"context_end_line":164,"code":" ]\n )\n \n self.conf_after = torch.nn.ModuleList(\n [Conformer(\n input_dim=d_hid,\n num_heads=num_heads,\n ffn_dim=d_hid * 2,\n num_layers=1,\n depthwise_conv_kernel_size=7,\n use_group_norm=True,)\n for _ in range(nlayers // 2)\n ]\n )\n\n self.out = nn.Linear(d_hid, 1 + vocab_size) # 1 for blank\n\n self.ctc_loss = nn.CTCLoss(blank=vocab_size, zero_infinity=True).cuda()\n\n \n def forward(self, latent, text=None, text_lens=None):\n layers = []\n\n x = self.mel_proj(latent.transpose(-1, -2)).transpose(-1, -2)\n\n x = x.transpose(1, 2)\n layers.append(nn.functional.avg_pool1d(x, 4))\n # x = x.transpose(1, 2)\n\n x = self.resblock1(x)\n x = nn.functional.avg_pool1d(x, 2)\n layers.append(nn.functional.avg_pool1d(x, 2))\n x = self.resblock2(x)\n x = nn.functional.avg_pool1d(x, 2)\n layers.append(x)\n\n x = x.transpose(1, 2)\n\n batch_size, time_steps, _ = x.shape\n # Create a dummy lengths tensor (all sequences are assumed to be full length).\n input_lengths = torch.full((batch_size,), time_steps, device=x.device, dtype=torch.int64)\n\n for layer in (self.conf_pre):\n x, _ = layer(x, input_lengths)\n layers.append(x.transpose(1, 2))\n\n for layer in (self.conf_after):\n x, _ = layer(x, input_lengths)\n layers.append(x.transpose(1, 2))\n\n x = self.out(x)\n\n if text_lens is not None and text is not None:\n loss = self.ctc_loss(x.log_softmax(dim=2).transpose(0, 1), text, input_lengths, text_lens)\n return x, layers, loss\n else:\n return x, layers\n\n\nif __name__ == \"__main__\":\n from f5_tts.model.utils import get_tokenizer\n\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n \n model = ConformerCTC(vocab_size, mel_dim=80, num_heads=8, d_hid=512, nlayers=6).cuda()\n \n text = [\"hello world\"] * bsz","source_hash":"61bd6ed4c15b5affe57e2e7cf82b89734754e5cb0c0f639947fb1b307278ba06","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.ctcmodel._get_conv","uri":"program://DMOSpeech2/function/src.ctcmodel._get_conv#L42-L52","kind":"function","name":"_get_conv","path":"src/ctcmodel.py","language":"python","start_line":42,"end_line":52,"context_start_line":22,"context_end_line":72,"code":" lens_to_mask,\n mask_from_frac_lengths,\n)\n\nclass ResBlock(nn.Module):\n def __init__(self, hidden_dim, n_conv=3, dropout_p=0.2):\n super().__init__()\n self._n_groups = 8\n self.blocks = nn.ModuleList([\n self._get_conv(hidden_dim, dilation=3**i, dropout_p=dropout_p)\n for i in range(n_conv)])\n\n\n def forward(self, x):\n for block in self.blocks:\n res = x\n x = block(x)\n x += res\n return x\n\n def _get_conv(self, hidden_dim, dilation, dropout_p=0.2):\n layers = [\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=dilation, dilation=dilation),\n nn.ReLU(),\n nn.GroupNorm(num_groups=self._n_groups, num_channels=hidden_dim),\n nn.Dropout(p=dropout_p),\n nn.Conv1d(hidden_dim, hidden_dim, kernel_size=3, padding=1, dilation=1),\n nn.ReLU(),\n nn.Dropout(p=dropout_p)\n ]\n return nn.Sequential(*layers)\n\n\nclass ConformerCTC(nn.Module):\n def __init__(self,\n vocab_size,\n mel_dim=100, \n num_heads=8, \n d_hid=512, \n nlayers=6):\n super().__init__()\n \n self.mel_proj = nn.Conv1d(mel_dim, d_hid, kernel_size=3, padding=1)\n \n self.d_hid = d_hid\n \n self.resblock1 = nn.Sequential(\n ResBlock(d_hid),\n nn.GroupNorm(num_groups=1, num_channels=d_hid)\n )\n ","source_hash":"61bd6ed4c15b5affe57e2e7cf82b89734754e5cb0c0f639947fb1b307278ba06","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.api","uri":"program://DMOSpeech2/module/src.f5_tts.api#L1-L164","kind":"module","name":"src.f5_tts.api","path":"src/f5_tts/api.py","language":"python","start_line":1,"end_line":164,"context_start_line":1,"context_end_line":164,"code":"import random\nimport sys\nfrom importlib.resources import files\n\nimport soundfile as sf\nimport tqdm\nfrom cached_path import cached_path\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\n\nfrom f5_tts.infer.utils_infer import (\n infer_process,\n load_model,\n load_vocoder,\n preprocess_ref_audio_text,\n remove_silence_for_generated_wav,\n save_spectrogram,\n transcribe,\n)\nfrom f5_tts.model.utils import seed_everything\n\n\nclass F5TTS:\n def __init__(\n self,\n model=\"F5TTS_v1_Base\",\n ckpt_file=\"\",\n vocab_file=\"\",\n ode_method=\"euler\",\n use_ema=True,\n vocoder_local_path=None,\n device=None,\n hf_cache_dir=None,\n ):\n model_cfg = OmegaConf.load(str(files(\"f5_tts\").joinpath(f\"configs/{model}.yaml\")))\n model_cls = get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n model_arc = model_cfg.model.arch\n\n self.mel_spec_type = model_cfg.model.mel_spec.mel_spec_type\n self.target_sample_rate = model_cfg.model.mel_spec.target_sample_rate\n\n self.ode_method = ode_method\n self.use_ema = use_ema\n\n if device is not None:\n self.device = device\n else:\n import torch\n\n self.device = (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n )\n\n # Load models\n self.vocoder = load_vocoder(\n self.mel_spec_type, vocoder_local_path is not None, vocoder_local_path, self.device, hf_cache_dir\n )\n\n repo_name, ckpt_step, ckpt_type = \"F5-TTS\", 1250000, \"safetensors\"\n\n # override for previous models\n if model == \"F5TTS_Base\":\n if self.mel_spec_type == \"vocos\":\n ckpt_step = 1200000\n elif self.mel_spec_type == \"bigvgan\":\n model = \"F5TTS_Base_bigvgan\"\n ckpt_type = \"pt\"\n elif model == \"E2TTS_Base\":\n repo_name = \"E2-TTS\"\n ckpt_step = 1200000\n\n if not ckpt_file:\n ckpt_file = str(\n cached_path(f\"hf://SWivid/{repo_name}/{model}/model_{ckpt_step}.{ckpt_type}\", cache_dir=hf_cache_dir)\n )\n self.ema_model = load_model(\n model_cls, model_arc, ckpt_file, self.mel_spec_type, vocab_file, self.ode_method, self.use_ema, self.device\n )\n\n def transcribe(self, ref_audio, language=None):\n return transcribe(ref_audio, language)\n\n def export_wav(self, wav, file_wave, remove_silence=False):\n sf.write(file_wave, wav, self.target_sample_rate)\n\n if remove_silence:\n remove_silence_for_generated_wav(file_wave)\n\n def export_spectrogram(self, spec, file_spec):\n save_spectrogram(spec, file_spec)\n\n def infer(\n self,\n ref_file,\n ref_text,\n gen_text,\n show_info=print,\n progress=tqdm,\n target_rms=0.1,\n cross_fade_duration=0.15,\n sway_sampling_coef=-1,\n cfg_strength=2,\n nfe_step=32,\n speed=1.0,\n fix_duration=None,\n remove_silence=False,\n file_wave=None,\n file_spec=None,\n seed=None,\n ):\n if seed is None:\n seed = random.randint(0, sys.maxsize)\n seed_everything(seed)\n self.seed = seed\n\n ref_file, ref_text = preprocess_ref_audio_text(ref_file, ref_text)\n\n wav, sr, spec = infer_process(\n ref_file,\n ref_text,\n gen_text,\n self.ema_model,\n self.vocoder,\n self.mel_spec_type,\n show_info=show_info,\n progress=progress,\n target_rms=target_rms,\n cross_fade_duration=cross_fade_duration,\n nfe_step=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n speed=speed,\n fix_duration=fix_duration,\n device=self.device,\n )\n\n if file_wave is not None:\n self.export_wav(wav, file_wave, remove_silence)\n\n if file_spec is not None:\n self.export_spectrogram(spec, file_spec)\n\n return wav, sr, spec\n\n\nif __name__ == \"__main__\":\n f5tts = F5TTS()\n\n wav, sr, spec = f5tts.infer(\n ref_file=str(files(\"f5_tts\").joinpath(\"infer/examples/basic/basic_ref_en.wav\")),\n ref_text=\"some call me nature, others call me mother nature.\",\n gen_text=\"\"\"I don't really care what you call me. I've been a silent spectator, watching species evolve, empires rise and fall. But always remember, I am mighty and enduring. Respect me and I'll nurture you; ignore me and you shall face the consequences.\"\"\",\n file_wave=str(files(\"f5_tts\").joinpath(\"../../tests/api_out.wav\")),\n file_spec=str(files(\"f5_tts\").joinpath(\"../../tests/api_out.png\")),\n seed=None,\n )\n\n print(\"seed :\", f5tts.seed)","source_hash":"520e3eafa052475f30e038f2a37addeaa4aafe3b17cf9da172d67f1037b47a53","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.api.F5TTS","uri":"program://DMOSpeech2/class/src.f5_tts.api.F5TTS#L23-L149","kind":"class","name":"F5TTS","path":"src/f5_tts/api.py","language":"python","start_line":23,"end_line":149,"context_start_line":3,"context_end_line":164,"code":"from importlib.resources import files\n\nimport soundfile as sf\nimport tqdm\nfrom cached_path import cached_path\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\n\nfrom f5_tts.infer.utils_infer import (\n infer_process,\n load_model,\n load_vocoder,\n preprocess_ref_audio_text,\n remove_silence_for_generated_wav,\n save_spectrogram,\n transcribe,\n)\nfrom f5_tts.model.utils import seed_everything\n\n\nclass F5TTS:\n def __init__(\n self,\n model=\"F5TTS_v1_Base\",\n ckpt_file=\"\",\n vocab_file=\"\",\n ode_method=\"euler\",\n use_ema=True,\n vocoder_local_path=None,\n device=None,\n hf_cache_dir=None,\n ):\n model_cfg = OmegaConf.load(str(files(\"f5_tts\").joinpath(f\"configs/{model}.yaml\")))\n model_cls = get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n model_arc = model_cfg.model.arch\n\n self.mel_spec_type = model_cfg.model.mel_spec.mel_spec_type\n self.target_sample_rate = model_cfg.model.mel_spec.target_sample_rate\n\n self.ode_method = ode_method\n self.use_ema = use_ema\n\n if device is not None:\n self.device = device\n else:\n import torch\n\n self.device = (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n )\n\n # Load models\n self.vocoder = load_vocoder(\n self.mel_spec_type, vocoder_local_path is not None, vocoder_local_path, self.device, hf_cache_dir\n )\n\n repo_name, ckpt_step, ckpt_type = \"F5-TTS\", 1250000, \"safetensors\"\n\n # override for previous models\n if model == \"F5TTS_Base\":\n if self.mel_spec_type == \"vocos\":\n ckpt_step = 1200000\n elif self.mel_spec_type == \"bigvgan\":\n model = \"F5TTS_Base_bigvgan\"\n ckpt_type = \"pt\"\n elif model == \"E2TTS_Base\":\n repo_name = \"E2-TTS\"\n ckpt_step = 1200000\n\n if not ckpt_file:\n ckpt_file = str(\n cached_path(f\"hf://SWivid/{repo_name}/{model}/model_{ckpt_step}.{ckpt_type}\", cache_dir=hf_cache_dir)\n )\n self.ema_model = load_model(\n model_cls, model_arc, ckpt_file, self.mel_spec_type, vocab_file, self.ode_method, self.use_ema, self.device\n )\n\n def transcribe(self, ref_audio, language=None):\n return transcribe(ref_audio, language)\n\n def export_wav(self, wav, file_wave, remove_silence=False):\n sf.write(file_wave, wav, self.target_sample_rate)\n\n if remove_silence:\n remove_silence_for_generated_wav(file_wave)\n\n def export_spectrogram(self, spec, file_spec):\n save_spectrogram(spec, file_spec)\n\n def infer(\n self,\n ref_file,\n ref_text,\n gen_text,\n show_info=print,\n progress=tqdm,\n target_rms=0.1,\n cross_fade_duration=0.15,\n sway_sampling_coef=-1,\n cfg_strength=2,\n nfe_step=32,\n speed=1.0,\n fix_duration=None,\n remove_silence=False,\n file_wave=None,\n file_spec=None,\n seed=None,\n ):\n if seed is None:\n seed = random.randint(0, sys.maxsize)\n seed_everything(seed)\n self.seed = seed\n\n ref_file, ref_text = preprocess_ref_audio_text(ref_file, ref_text)\n\n wav, sr, spec = infer_process(\n ref_file,\n ref_text,\n gen_text,\n self.ema_model,\n self.vocoder,\n self.mel_spec_type,\n show_info=show_info,\n progress=progress,\n target_rms=target_rms,\n cross_fade_duration=cross_fade_duration,\n nfe_step=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n speed=speed,\n fix_duration=fix_duration,\n device=self.device,\n )\n\n if file_wave is not None:\n self.export_wav(wav, file_wave, remove_silence)\n\n if file_spec is not None:\n self.export_spectrogram(spec, file_spec)\n\n return wav, sr, spec\n\n\nif __name__ == \"__main__\":\n f5tts = F5TTS()\n\n wav, sr, spec = f5tts.infer(\n ref_file=str(files(\"f5_tts\").joinpath(\"infer/examples/basic/basic_ref_en.wav\")),\n ref_text=\"some call me nature, others call me mother nature.\",\n gen_text=\"\"\"I don't really care what you call me. I've been a silent spectator, watching species evolve, empires rise and fall. But always remember, I am mighty and enduring. Respect me and I'll nurture you; ignore me and you shall face the consequences.\"\"\",\n file_wave=str(files(\"f5_tts\").joinpath(\"../../tests/api_out.wav\")),\n file_spec=str(files(\"f5_tts\").joinpath(\"../../tests/api_out.png\")),\n seed=None,\n )\n\n print(\"seed :\", f5tts.seed)","source_hash":"520e3eafa052475f30e038f2a37addeaa4aafe3b17cf9da172d67f1037b47a53","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.api.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.api.__init__#L24-L84","kind":"function","name":"__init__","path":"src/f5_tts/api.py","language":"python","start_line":24,"end_line":84,"context_start_line":4,"context_end_line":104,"code":"\nimport soundfile as sf\nimport tqdm\nfrom cached_path import cached_path\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\n\nfrom f5_tts.infer.utils_infer import (\n infer_process,\n load_model,\n load_vocoder,\n preprocess_ref_audio_text,\n remove_silence_for_generated_wav,\n save_spectrogram,\n transcribe,\n)\nfrom f5_tts.model.utils import seed_everything\n\n\nclass F5TTS:\n def __init__(\n self,\n model=\"F5TTS_v1_Base\",\n ckpt_file=\"\",\n vocab_file=\"\",\n ode_method=\"euler\",\n use_ema=True,\n vocoder_local_path=None,\n device=None,\n hf_cache_dir=None,\n ):\n model_cfg = OmegaConf.load(str(files(\"f5_tts\").joinpath(f\"configs/{model}.yaml\")))\n model_cls = get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n model_arc = model_cfg.model.arch\n\n self.mel_spec_type = model_cfg.model.mel_spec.mel_spec_type\n self.target_sample_rate = model_cfg.model.mel_spec.target_sample_rate\n\n self.ode_method = ode_method\n self.use_ema = use_ema\n\n if device is not None:\n self.device = device\n else:\n import torch\n\n self.device = (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n )\n\n # Load models\n self.vocoder = load_vocoder(\n self.mel_spec_type, vocoder_local_path is not None, vocoder_local_path, self.device, hf_cache_dir\n )\n\n repo_name, ckpt_step, ckpt_type = \"F5-TTS\", 1250000, \"safetensors\"\n\n # override for previous models\n if model == \"F5TTS_Base\":\n if self.mel_spec_type == \"vocos\":\n ckpt_step = 1200000\n elif self.mel_spec_type == \"bigvgan\":\n model = \"F5TTS_Base_bigvgan\"\n ckpt_type = \"pt\"\n elif model == \"E2TTS_Base\":\n repo_name = \"E2-TTS\"\n ckpt_step = 1200000\n\n if not ckpt_file:\n ckpt_file = str(\n cached_path(f\"hf://SWivid/{repo_name}/{model}/model_{ckpt_step}.{ckpt_type}\", cache_dir=hf_cache_dir)\n )\n self.ema_model = load_model(\n model_cls, model_arc, ckpt_file, self.mel_spec_type, vocab_file, self.ode_method, self.use_ema, self.device\n )\n\n def transcribe(self, ref_audio, language=None):\n return transcribe(ref_audio, language)\n\n def export_wav(self, wav, file_wave, remove_silence=False):\n sf.write(file_wave, wav, self.target_sample_rate)\n\n if remove_silence:\n remove_silence_for_generated_wav(file_wave)\n\n def export_spectrogram(self, spec, file_spec):\n save_spectrogram(spec, file_spec)\n\n def infer(\n self,\n ref_file,\n ref_text,\n gen_text,\n show_info=print,\n progress=tqdm,","source_hash":"520e3eafa052475f30e038f2a37addeaa4aafe3b17cf9da172d67f1037b47a53","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.api.transcribe","uri":"program://DMOSpeech2/function/src.f5_tts.api.transcribe#L86-L87","kind":"function","name":"transcribe","path":"src/f5_tts/api.py","language":"python","start_line":86,"end_line":87,"context_start_line":66,"context_end_line":107,"code":"\n # override for previous models\n if model == \"F5TTS_Base\":\n if self.mel_spec_type == \"vocos\":\n ckpt_step = 1200000\n elif self.mel_spec_type == \"bigvgan\":\n model = \"F5TTS_Base_bigvgan\"\n ckpt_type = \"pt\"\n elif model == \"E2TTS_Base\":\n repo_name = \"E2-TTS\"\n ckpt_step = 1200000\n\n if not ckpt_file:\n ckpt_file = str(\n cached_path(f\"hf://SWivid/{repo_name}/{model}/model_{ckpt_step}.{ckpt_type}\", cache_dir=hf_cache_dir)\n )\n self.ema_model = load_model(\n model_cls, model_arc, ckpt_file, self.mel_spec_type, vocab_file, self.ode_method, self.use_ema, self.device\n )\n\n def transcribe(self, ref_audio, language=None):\n return transcribe(ref_audio, language)\n\n def export_wav(self, wav, file_wave, remove_silence=False):\n sf.write(file_wave, wav, self.target_sample_rate)\n\n if remove_silence:\n remove_silence_for_generated_wav(file_wave)\n\n def export_spectrogram(self, spec, file_spec):\n save_spectrogram(spec, file_spec)\n\n def infer(\n self,\n ref_file,\n ref_text,\n gen_text,\n show_info=print,\n progress=tqdm,\n target_rms=0.1,\n cross_fade_duration=0.15,\n sway_sampling_coef=-1,","source_hash":"520e3eafa052475f30e038f2a37addeaa4aafe3b17cf9da172d67f1037b47a53","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.api.export_wav","uri":"program://DMOSpeech2/function/src.f5_tts.api.export_wav#L89-L93","kind":"function","name":"export_wav","path":"src/f5_tts/api.py","language":"python","start_line":89,"end_line":93,"context_start_line":69,"context_end_line":113,"code":" if self.mel_spec_type == \"vocos\":\n ckpt_step = 1200000\n elif self.mel_spec_type == \"bigvgan\":\n model = \"F5TTS_Base_bigvgan\"\n ckpt_type = \"pt\"\n elif model == \"E2TTS_Base\":\n repo_name = \"E2-TTS\"\n ckpt_step = 1200000\n\n if not ckpt_file:\n ckpt_file = str(\n cached_path(f\"hf://SWivid/{repo_name}/{model}/model_{ckpt_step}.{ckpt_type}\", cache_dir=hf_cache_dir)\n )\n self.ema_model = load_model(\n model_cls, model_arc, ckpt_file, self.mel_spec_type, vocab_file, self.ode_method, self.use_ema, self.device\n )\n\n def transcribe(self, ref_audio, language=None):\n return transcribe(ref_audio, language)\n\n def export_wav(self, wav, file_wave, remove_silence=False):\n sf.write(file_wave, wav, self.target_sample_rate)\n\n if remove_silence:\n remove_silence_for_generated_wav(file_wave)\n\n def export_spectrogram(self, spec, file_spec):\n save_spectrogram(spec, file_spec)\n\n def infer(\n self,\n ref_file,\n ref_text,\n gen_text,\n show_info=print,\n progress=tqdm,\n target_rms=0.1,\n cross_fade_duration=0.15,\n sway_sampling_coef=-1,\n cfg_strength=2,\n nfe_step=32,\n speed=1.0,\n fix_duration=None,\n remove_silence=False,\n file_wave=None,","source_hash":"520e3eafa052475f30e038f2a37addeaa4aafe3b17cf9da172d67f1037b47a53","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.api.export_spectrogram","uri":"program://DMOSpeech2/function/src.f5_tts.api.export_spectrogram#L95-L96","kind":"function","name":"export_spectrogram","path":"src/f5_tts/api.py","language":"python","start_line":95,"end_line":96,"context_start_line":75,"context_end_line":116,"code":" repo_name = \"E2-TTS\"\n ckpt_step = 1200000\n\n if not ckpt_file:\n ckpt_file = str(\n cached_path(f\"hf://SWivid/{repo_name}/{model}/model_{ckpt_step}.{ckpt_type}\", cache_dir=hf_cache_dir)\n )\n self.ema_model = load_model(\n model_cls, model_arc, ckpt_file, self.mel_spec_type, vocab_file, self.ode_method, self.use_ema, self.device\n )\n\n def transcribe(self, ref_audio, language=None):\n return transcribe(ref_audio, language)\n\n def export_wav(self, wav, file_wave, remove_silence=False):\n sf.write(file_wave, wav, self.target_sample_rate)\n\n if remove_silence:\n remove_silence_for_generated_wav(file_wave)\n\n def export_spectrogram(self, spec, file_spec):\n save_spectrogram(spec, file_spec)\n\n def infer(\n self,\n ref_file,\n ref_text,\n gen_text,\n show_info=print,\n progress=tqdm,\n target_rms=0.1,\n cross_fade_duration=0.15,\n sway_sampling_coef=-1,\n cfg_strength=2,\n nfe_step=32,\n speed=1.0,\n fix_duration=None,\n remove_silence=False,\n file_wave=None,\n file_spec=None,\n seed=None,\n ):","source_hash":"520e3eafa052475f30e038f2a37addeaa4aafe3b17cf9da172d67f1037b47a53","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.api.infer","uri":"program://DMOSpeech2/function/src.f5_tts.api.infer#L98-L149","kind":"function","name":"infer","path":"src/f5_tts/api.py","language":"python","start_line":98,"end_line":149,"context_start_line":78,"context_end_line":164,"code":" if not ckpt_file:\n ckpt_file = str(\n cached_path(f\"hf://SWivid/{repo_name}/{model}/model_{ckpt_step}.{ckpt_type}\", cache_dir=hf_cache_dir)\n )\n self.ema_model = load_model(\n model_cls, model_arc, ckpt_file, self.mel_spec_type, vocab_file, self.ode_method, self.use_ema, self.device\n )\n\n def transcribe(self, ref_audio, language=None):\n return transcribe(ref_audio, language)\n\n def export_wav(self, wav, file_wave, remove_silence=False):\n sf.write(file_wave, wav, self.target_sample_rate)\n\n if remove_silence:\n remove_silence_for_generated_wav(file_wave)\n\n def export_spectrogram(self, spec, file_spec):\n save_spectrogram(spec, file_spec)\n\n def infer(\n self,\n ref_file,\n ref_text,\n gen_text,\n show_info=print,\n progress=tqdm,\n target_rms=0.1,\n cross_fade_duration=0.15,\n sway_sampling_coef=-1,\n cfg_strength=2,\n nfe_step=32,\n speed=1.0,\n fix_duration=None,\n remove_silence=False,\n file_wave=None,\n file_spec=None,\n seed=None,\n ):\n if seed is None:\n seed = random.randint(0, sys.maxsize)\n seed_everything(seed)\n self.seed = seed\n\n ref_file, ref_text = preprocess_ref_audio_text(ref_file, ref_text)\n\n wav, sr, spec = infer_process(\n ref_file,\n ref_text,\n gen_text,\n self.ema_model,\n self.vocoder,\n self.mel_spec_type,\n show_info=show_info,\n progress=progress,\n target_rms=target_rms,\n cross_fade_duration=cross_fade_duration,\n nfe_step=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n speed=speed,\n fix_duration=fix_duration,\n device=self.device,\n )\n\n if file_wave is not None:\n self.export_wav(wav, file_wave, remove_silence)\n\n if file_spec is not None:\n self.export_spectrogram(spec, file_spec)\n\n return wav, sr, spec\n\n\nif __name__ == \"__main__\":\n f5tts = F5TTS()\n\n wav, sr, spec = f5tts.infer(\n ref_file=str(files(\"f5_tts\").joinpath(\"infer/examples/basic/basic_ref_en.wav\")),\n ref_text=\"some call me nature, others call me mother nature.\",\n gen_text=\"\"\"I don't really care what you call me. I've been a silent spectator, watching species evolve, empires rise and fall. But always remember, I am mighty and enduring. Respect me and I'll nurture you; ignore me and you shall face the consequences.\"\"\",\n file_wave=str(files(\"f5_tts\").joinpath(\"../../tests/api_out.wav\")),\n file_spec=str(files(\"f5_tts\").joinpath(\"../../tests/api_out.png\")),\n seed=None,\n )\n\n print(\"seed :\", f5tts.seed)","source_hash":"520e3eafa052475f30e038f2a37addeaa4aafe3b17cf9da172d67f1037b47a53","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_client","uri":"program://DMOSpeech2/module/src.f5_tts.socket_client#L1-L63","kind":"module","name":"src.f5_tts.socket_client","path":"src/f5_tts/socket_client.py","language":"python","start_line":1,"end_line":63,"context_start_line":1,"context_end_line":63,"code":"import asyncio\nimport logging\nimport socket\nimport time\n\nimport numpy as np\nimport pyaudio\n\n\nlogging.basicConfig(level=logging.INFO)\nlogger = logging.getLogger(__name__)\n\n\nasync def listen_to_F5TTS(text, server_ip=\"localhost\", server_port=9998):\n client_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)\n await asyncio.get_event_loop().run_in_executor(None, client_socket.connect, (server_ip, int(server_port)))\n\n start_time = time.time()\n first_chunk_time = None\n\n async def play_audio_stream():\n nonlocal first_chunk_time\n p = pyaudio.PyAudio()\n stream = p.open(format=pyaudio.paFloat32, channels=1, rate=24000, output=True, frames_per_buffer=2048)\n\n try:\n while True:\n data = await asyncio.get_event_loop().run_in_executor(None, client_socket.recv, 8192)\n if not data:\n break\n if data == b\"END\":\n logger.info(\"End of audio received.\")\n break\n\n audio_array = np.frombuffer(data, dtype=np.float32)\n stream.write(audio_array.tobytes())\n\n if first_chunk_time is None:\n first_chunk_time = time.time()\n\n finally:\n stream.stop_stream()\n stream.close()\n p.terminate()\n\n logger.info(f\"Total time taken: {time.time() - start_time:.4f} seconds\")\n\n try:\n data_to_send = f\"{text}\".encode(\"utf-8\")\n await asyncio.get_event_loop().run_in_executor(None, client_socket.sendall, data_to_send)\n await play_audio_stream()\n\n except Exception as e:\n logger.error(f\"Error in listen_to_F5TTS: {e}\")\n\n finally:\n client_socket.close()\n\n\nif __name__ == \"__main__\":\n text_to_send = \"As a Reader assistant, I'm familiar with new technology. which are key to its improved performance in terms of both training speed and inference efficiency. Let's break down the components\"\n\n asyncio.run(listen_to_F5TTS(text_to_send))","source_hash":"301265e66a6fca7849462861c53751be848df709d1b9b755d098798e81cf25e6","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_client.listen_to_F5TTS","uri":"program://DMOSpeech2/function/src.f5_tts.socket_client.listen_to_F5TTS#L14-L57","kind":"function","name":"listen_to_F5TTS","path":"src/f5_tts/socket_client.py","language":"python","start_line":14,"end_line":57,"context_start_line":1,"context_end_line":63,"code":"import asyncio\nimport logging\nimport socket\nimport time\n\nimport numpy as np\nimport pyaudio\n\n\nlogging.basicConfig(level=logging.INFO)\nlogger = logging.getLogger(__name__)\n\n\nasync def listen_to_F5TTS(text, server_ip=\"localhost\", server_port=9998):\n client_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)\n await asyncio.get_event_loop().run_in_executor(None, client_socket.connect, (server_ip, int(server_port)))\n\n start_time = time.time()\n first_chunk_time = None\n\n async def play_audio_stream():\n nonlocal first_chunk_time\n p = pyaudio.PyAudio()\n stream = p.open(format=pyaudio.paFloat32, channels=1, rate=24000, output=True, frames_per_buffer=2048)\n\n try:\n while True:\n data = await asyncio.get_event_loop().run_in_executor(None, client_socket.recv, 8192)\n if not data:\n break\n if data == b\"END\":\n logger.info(\"End of audio received.\")\n break\n\n audio_array = np.frombuffer(data, dtype=np.float32)\n stream.write(audio_array.tobytes())\n\n if first_chunk_time is None:\n first_chunk_time = time.time()\n\n finally:\n stream.stop_stream()\n stream.close()\n p.terminate()\n\n logger.info(f\"Total time taken: {time.time() - start_time:.4f} seconds\")\n\n try:\n data_to_send = f\"{text}\".encode(\"utf-8\")\n await asyncio.get_event_loop().run_in_executor(None, client_socket.sendall, data_to_send)\n await play_audio_stream()\n\n except Exception as e:\n logger.error(f\"Error in listen_to_F5TTS: {e}\")\n\n finally:\n client_socket.close()\n\n\nif __name__ == \"__main__\":\n text_to_send = \"As a Reader assistant, I'm familiar with new technology. which are key to its improved performance in terms of both training speed and inference efficiency. Let's break down the components\"\n\n asyncio.run(listen_to_F5TTS(text_to_send))","source_hash":"301265e66a6fca7849462861c53751be848df709d1b9b755d098798e81cf25e6","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_client.play_audio_stream","uri":"program://DMOSpeech2/function/src.f5_tts.socket_client.play_audio_stream#L21-L46","kind":"function","name":"play_audio_stream","path":"src/f5_tts/socket_client.py","language":"python","start_line":21,"end_line":46,"context_start_line":1,"context_end_line":63,"code":"import asyncio\nimport logging\nimport socket\nimport time\n\nimport numpy as np\nimport pyaudio\n\n\nlogging.basicConfig(level=logging.INFO)\nlogger = logging.getLogger(__name__)\n\n\nasync def listen_to_F5TTS(text, server_ip=\"localhost\", server_port=9998):\n client_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)\n await asyncio.get_event_loop().run_in_executor(None, client_socket.connect, (server_ip, int(server_port)))\n\n start_time = time.time()\n first_chunk_time = None\n\n async def play_audio_stream():\n nonlocal first_chunk_time\n p = pyaudio.PyAudio()\n stream = p.open(format=pyaudio.paFloat32, channels=1, rate=24000, output=True, frames_per_buffer=2048)\n\n try:\n while True:\n data = await asyncio.get_event_loop().run_in_executor(None, client_socket.recv, 8192)\n if not data:\n break\n if data == b\"END\":\n logger.info(\"End of audio received.\")\n break\n\n audio_array = np.frombuffer(data, dtype=np.float32)\n stream.write(audio_array.tobytes())\n\n if first_chunk_time is None:\n first_chunk_time = time.time()\n\n finally:\n stream.stop_stream()\n stream.close()\n p.terminate()\n\n logger.info(f\"Total time taken: {time.time() - start_time:.4f} seconds\")\n\n try:\n data_to_send = f\"{text}\".encode(\"utf-8\")\n await asyncio.get_event_loop().run_in_executor(None, client_socket.sendall, data_to_send)\n await play_audio_stream()\n\n except Exception as e:\n logger.error(f\"Error in listen_to_F5TTS: {e}\")\n\n finally:\n client_socket.close()\n\n\nif __name__ == \"__main__\":\n text_to_send = \"As a Reader assistant, I'm familiar with new technology. which are key to its improved performance in terms of both training speed and inference efficiency. Let's break down the components\"\n\n asyncio.run(listen_to_F5TTS(text_to_send))","source_hash":"301265e66a6fca7849462861c53751be848df709d1b9b755d098798e81cf25e6","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server","uri":"program://DMOSpeech2/module/src.f5_tts.socket_server#L1-L268","kind":"module","name":"src.f5_tts.socket_server","path":"src/f5_tts/socket_server.py","language":"python","start_line":1,"end_line":268,"context_start_line":1,"context_end_line":268,"code":"import argparse\nimport gc\nimport logging\nimport queue\nimport socket\nimport struct\nimport threading\nimport traceback\nimport wave\nfrom importlib.resources import files\n\nimport numpy as np\nimport torch\nimport torchaudio\nfrom huggingface_hub import hf_hub_download\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\n\nfrom f5_tts.infer.utils_infer import (\n chunk_text,\n infer_batch_process,\n load_model,\n load_vocoder,\n preprocess_ref_audio_text,\n)\n\n\nlogging.basicConfig(level=logging.INFO)\nlogger = logging.getLogger(__name__)\n\n\nclass AudioFileWriterThread(threading.Thread):\n \"\"\"Threaded file writer to avoid blocking the TTS streaming process.\"\"\"\n\n def __init__(self, output_file, sampling_rate):\n super().__init__()\n self.output_file = output_file\n self.sampling_rate = sampling_rate\n self.queue = queue.Queue()\n self.stop_event = threading.Event()\n self.audio_data = []\n\n def run(self):\n \"\"\"Process queued audio data and write it to a file.\"\"\"\n logger.info(\"AudioFileWriterThread started.\")\n with wave.open(self.output_file, \"wb\") as wf:\n wf.setnchannels(1)\n wf.setsampwidth(2)\n wf.setframerate(self.sampling_rate)\n\n while not self.stop_event.is_set() or not self.queue.empty():\n try:\n chunk = self.queue.get(timeout=0.1)\n if chunk is not None:\n chunk = np.int16(chunk * 32767)\n self.audio_data.append(chunk)\n wf.writeframes(chunk.tobytes())\n except queue.Empty:\n continue\n\n def add_chunk(self, chunk):\n \"\"\"Add a new chunk to the queue.\"\"\"\n self.queue.put(chunk)\n\n def stop(self):\n \"\"\"Stop writing and ensure all queued data is written.\"\"\"\n self.stop_event.set()\n self.join()\n logger.info(\"Audio writing completed.\")\n\n\nclass TTSStreamingProcessor:\n def __init__(self, model, ckpt_file, vocab_file, ref_audio, ref_text, device=None, dtype=torch.float32):\n self.device = device or (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n )\n model_cfg = OmegaConf.load(str(files(\"f5_tts\").joinpath(f\"configs/{model}.yaml\")))\n self.model_cls = get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n self.model_arc = model_cfg.model.arch\n self.mel_spec_type = model_cfg.model.mel_spec.mel_spec_type\n self.sampling_rate = model_cfg.model.mel_spec.target_sample_rate\n\n self.model = self.load_ema_model(ckpt_file, vocab_file, dtype)\n self.vocoder = self.load_vocoder_model()\n\n self.update_reference(ref_audio, ref_text)\n self._warm_up()\n self.file_writer_thread = None\n self.first_package = True\n\n def load_ema_model(self, ckpt_file, vocab_file, dtype):\n return load_model(\n self.model_cls,\n self.model_arc,\n ckpt_path=ckpt_file,\n mel_spec_type=self.mel_spec_type,\n vocab_file=vocab_file,\n ode_method=\"euler\",\n use_ema=True,\n device=self.device,\n ).to(self.device, dtype=dtype)\n\n def load_vocoder_model(self):\n return load_vocoder(vocoder_name=self.mel_spec_type, is_local=False, local_path=None, device=self.device)\n\n def update_reference(self, ref_audio, ref_text):\n self.ref_audio, self.ref_text = preprocess_ref_audio_text(ref_audio, ref_text)\n self.audio, self.sr = torchaudio.load(self.ref_audio)\n\n ref_audio_duration = self.audio.shape[-1] / self.sr\n ref_text_byte_len = len(self.ref_text.encode(\"utf-8\"))\n self.max_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration))\n self.few_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 2)\n self.min_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 4)\n\n def _warm_up(self):\n logger.info(\"Warming up the model...\")\n gen_text = \"Warm-up text for the model.\"\n for _ in infer_batch_process(\n (self.audio, self.sr),\n self.ref_text,\n [gen_text],\n self.model,\n self.vocoder,\n progress=None,\n device=self.device,\n streaming=True,\n ):\n pass\n logger.info(\"Warm-up completed.\")\n\n def generate_stream(self, text, conn):\n text_batches = chunk_text(text, max_chars=self.max_chars)\n if self.first_package:\n text_batches = chunk_text(text_batches[0], max_chars=self.few_chars) + text_batches[1:]\n text_batches = chunk_text(text_batches[0], max_chars=self.min_chars) + text_batches[1:]\n self.first_package = False\n\n audio_stream = infer_batch_process(\n (self.audio, self.sr),\n self.ref_text,\n text_batches,\n self.model,\n self.vocoder,\n progress=None,\n device=self.device,\n streaming=True,\n chunk_size=2048,\n )\n\n # Reset the file writer thread\n if self.file_writer_thread is not None:\n self.file_writer_thread.stop()\n self.file_writer_thread = AudioFileWriterThread(\"output.wav\", self.sampling_rate)\n self.file_writer_thread.start()\n\n for audio_chunk, _ in audio_stream:\n if len(audio_chunk) > 0:\n logger.info(f\"Generated audio chunk of size: {len(audio_chunk)}\")\n\n # Send audio chunk via socket\n conn.sendall(struct.pack(f\"{len(audio_chunk)}f\", *audio_chunk))\n\n # Write to file asynchronously\n self.file_writer_thread.add_chunk(audio_chunk)\n\n logger.info(\"Finished sending audio stream.\")\n conn.sendall(b\"END\") # Send end signal\n\n # Ensure all audio data is written before exiting\n self.file_writer_thread.stop()\n\n\ndef handle_client(conn, processor):\n try:\n with conn:\n conn.setsockopt(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)\n while True:\n data = conn.recv(1024)\n if not data:\n processor.first_package = True\n break\n data_str = data.decode(\"utf-8\").strip()\n logger.info(f\"Received text: {data_str}\")\n\n try:\n processor.generate_stream(data_str, conn)\n except Exception as inner_e:\n logger.error(f\"Error during processing: {inner_e}\")\n traceback.print_exc()\n break\n except Exception as e:\n logger.error(f\"Error handling client: {e}\")\n traceback.print_exc()\n\n\ndef start_server(host, port, processor):\n with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:\n s.bind((host, port))\n s.listen()\n logger.info(f\"Server started on {host}:{port}\")\n while True:\n conn, addr = s.accept()\n logger.info(f\"Connected by {addr}\")\n handle_client(conn, processor)\n\n\nif __name__ == \"__main__\":\n parser = argparse.ArgumentParser()\n\n parser.add_argument(\"--host\", default=\"0.0.0.0\")\n parser.add_argument(\"--port\", default=9998)\n\n parser.add_argument(\n \"--model\",\n default=\"F5TTS_v1_Base\",\n help=\"The model name, e.g. F5TTS_v1_Base\",\n )\n parser.add_argument(\n \"--ckpt_file\",\n default=str(hf_hub_download(repo_id=\"SWivid/F5-TTS\", filename=\"F5TTS_v1_Base/model_1250000.safetensors\")),\n help=\"Path to the model checkpoint file\",\n )\n parser.add_argument(\n \"--vocab_file\",\n default=\"\",\n help=\"Path to the vocab file if customized\",\n )\n\n parser.add_argument(\n \"--ref_audio\",\n default=str(files(\"f5_tts\").joinpath(\"infer/examples/basic/basic_ref_en.wav\")),\n help=\"Reference audio to provide model with speaker characteristics\",\n )\n parser.add_argument(\n \"--ref_text\",\n default=\"\",\n help=\"Reference audio subtitle, leave empty to auto-transcribe\",\n )\n\n parser.add_argument(\"--device\", default=None, help=\"Device to run the model on\")\n parser.add_argument(\"--dtype\", default=torch.float32, help=\"Data type to use for model inference\")\n\n args = parser.parse_args()\n\n try:\n # Initialize the processor with the model and vocoder\n processor = TTSStreamingProcessor(\n model=args.model,\n ckpt_file=args.ckpt_file,\n vocab_file=args.vocab_file,\n ref_audio=args.ref_audio,\n ref_text=args.ref_text,\n device=args.device,\n dtype=args.dtype,\n )\n\n # Start the server\n start_server(args.host, args.port, processor)\n\n except KeyboardInterrupt:\n gc.collect()","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server.AudioFileWriterThread","uri":"program://DMOSpeech2/class/src.f5_tts.socket_server.AudioFileWriterThread#L32-L69","kind":"class","name":"AudioFileWriterThread","path":"src/f5_tts/socket_server.py","language":"python","start_line":32,"end_line":69,"context_start_line":12,"context_end_line":89,"code":"import numpy as np\nimport torch\nimport torchaudio\nfrom huggingface_hub import hf_hub_download\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\n\nfrom f5_tts.infer.utils_infer import (\n chunk_text,\n infer_batch_process,\n load_model,\n load_vocoder,\n preprocess_ref_audio_text,\n)\n\n\nlogging.basicConfig(level=logging.INFO)\nlogger = logging.getLogger(__name__)\n\n\nclass AudioFileWriterThread(threading.Thread):\n \"\"\"Threaded file writer to avoid blocking the TTS streaming process.\"\"\"\n\n def __init__(self, output_file, sampling_rate):\n super().__init__()\n self.output_file = output_file\n self.sampling_rate = sampling_rate\n self.queue = queue.Queue()\n self.stop_event = threading.Event()\n self.audio_data = []\n\n def run(self):\n \"\"\"Process queued audio data and write it to a file.\"\"\"\n logger.info(\"AudioFileWriterThread started.\")\n with wave.open(self.output_file, \"wb\") as wf:\n wf.setnchannels(1)\n wf.setsampwidth(2)\n wf.setframerate(self.sampling_rate)\n\n while not self.stop_event.is_set() or not self.queue.empty():\n try:\n chunk = self.queue.get(timeout=0.1)\n if chunk is not None:\n chunk = np.int16(chunk * 32767)\n self.audio_data.append(chunk)\n wf.writeframes(chunk.tobytes())\n except queue.Empty:\n continue\n\n def add_chunk(self, chunk):\n \"\"\"Add a new chunk to the queue.\"\"\"\n self.queue.put(chunk)\n\n def stop(self):\n \"\"\"Stop writing and ensure all queued data is written.\"\"\"\n self.stop_event.set()\n self.join()\n logger.info(\"Audio writing completed.\")\n\n\nclass TTSStreamingProcessor:\n def __init__(self, model, ckpt_file, vocab_file, ref_audio, ref_text, device=None, dtype=torch.float32):\n self.device = device or (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n )\n model_cfg = OmegaConf.load(str(files(\"f5_tts\").joinpath(f\"configs/{model}.yaml\")))\n self.model_cls = get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n self.model_arc = model_cfg.model.arch\n self.mel_spec_type = model_cfg.model.mel_spec.mel_spec_type\n self.sampling_rate = model_cfg.model.mel_spec.target_sample_rate\n\n self.model = self.load_ema_model(ckpt_file, vocab_file, dtype)","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server.TTSStreamingProcessor","uri":"program://DMOSpeech2/class/src.f5_tts.socket_server.TTSStreamingProcessor#L72-L177","kind":"class","name":"TTSStreamingProcessor","path":"src/f5_tts/socket_server.py","language":"python","start_line":72,"end_line":177,"context_start_line":52,"context_end_line":197,"code":" try:\n chunk = self.queue.get(timeout=0.1)\n if chunk is not None:\n chunk = np.int16(chunk * 32767)\n self.audio_data.append(chunk)\n wf.writeframes(chunk.tobytes())\n except queue.Empty:\n continue\n\n def add_chunk(self, chunk):\n \"\"\"Add a new chunk to the queue.\"\"\"\n self.queue.put(chunk)\n\n def stop(self):\n \"\"\"Stop writing and ensure all queued data is written.\"\"\"\n self.stop_event.set()\n self.join()\n logger.info(\"Audio writing completed.\")\n\n\nclass TTSStreamingProcessor:\n def __init__(self, model, ckpt_file, vocab_file, ref_audio, ref_text, device=None, dtype=torch.float32):\n self.device = device or (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n )\n model_cfg = OmegaConf.load(str(files(\"f5_tts\").joinpath(f\"configs/{model}.yaml\")))\n self.model_cls = get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n self.model_arc = model_cfg.model.arch\n self.mel_spec_type = model_cfg.model.mel_spec.mel_spec_type\n self.sampling_rate = model_cfg.model.mel_spec.target_sample_rate\n\n self.model = self.load_ema_model(ckpt_file, vocab_file, dtype)\n self.vocoder = self.load_vocoder_model()\n\n self.update_reference(ref_audio, ref_text)\n self._warm_up()\n self.file_writer_thread = None\n self.first_package = True\n\n def load_ema_model(self, ckpt_file, vocab_file, dtype):\n return load_model(\n self.model_cls,\n self.model_arc,\n ckpt_path=ckpt_file,\n mel_spec_type=self.mel_spec_type,\n vocab_file=vocab_file,\n ode_method=\"euler\",\n use_ema=True,\n device=self.device,\n ).to(self.device, dtype=dtype)\n\n def load_vocoder_model(self):\n return load_vocoder(vocoder_name=self.mel_spec_type, is_local=False, local_path=None, device=self.device)\n\n def update_reference(self, ref_audio, ref_text):\n self.ref_audio, self.ref_text = preprocess_ref_audio_text(ref_audio, ref_text)\n self.audio, self.sr = torchaudio.load(self.ref_audio)\n\n ref_audio_duration = self.audio.shape[-1] / self.sr\n ref_text_byte_len = len(self.ref_text.encode(\"utf-8\"))\n self.max_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration))\n self.few_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 2)\n self.min_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 4)\n\n def _warm_up(self):\n logger.info(\"Warming up the model...\")\n gen_text = \"Warm-up text for the model.\"\n for _ in infer_batch_process(\n (self.audio, self.sr),\n self.ref_text,\n [gen_text],\n self.model,\n self.vocoder,\n progress=None,\n device=self.device,\n streaming=True,\n ):\n pass\n logger.info(\"Warm-up completed.\")\n\n def generate_stream(self, text, conn):\n text_batches = chunk_text(text, max_chars=self.max_chars)\n if self.first_package:\n text_batches = chunk_text(text_batches[0], max_chars=self.few_chars) + text_batches[1:]\n text_batches = chunk_text(text_batches[0], max_chars=self.min_chars) + text_batches[1:]\n self.first_package = False\n\n audio_stream = infer_batch_process(\n (self.audio, self.sr),\n self.ref_text,\n text_batches,\n self.model,\n self.vocoder,\n progress=None,\n device=self.device,\n streaming=True,\n chunk_size=2048,\n )\n\n # Reset the file writer thread\n if self.file_writer_thread is not None:\n self.file_writer_thread.stop()\n self.file_writer_thread = AudioFileWriterThread(\"output.wav\", self.sampling_rate)\n self.file_writer_thread.start()\n\n for audio_chunk, _ in audio_stream:\n if len(audio_chunk) > 0:\n logger.info(f\"Generated audio chunk of size: {len(audio_chunk)}\")\n\n # Send audio chunk via socket\n conn.sendall(struct.pack(f\"{len(audio_chunk)}f\", *audio_chunk))\n\n # Write to file asynchronously\n self.file_writer_thread.add_chunk(audio_chunk)\n\n logger.info(\"Finished sending audio stream.\")\n conn.sendall(b\"END\") # Send end signal\n\n # Ensure all audio data is written before exiting\n self.file_writer_thread.stop()\n\n\ndef handle_client(conn, processor):\n try:\n with conn:\n conn.setsockopt(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)\n while True:\n data = conn.recv(1024)\n if not data:\n processor.first_package = True\n break\n data_str = data.decode(\"utf-8\").strip()\n logger.info(f\"Received text: {data_str}\")\n\n try:\n processor.generate_stream(data_str, conn)\n except Exception as inner_e:\n logger.error(f\"Error during processing: {inner_e}\")\n traceback.print_exc()\n break","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server.handle_client","uri":"program://DMOSpeech2/function/src.f5_tts.socket_server.handle_client#L180-L200","kind":"function","name":"handle_client","path":"src/f5_tts/socket_server.py","language":"python","start_line":180,"end_line":200,"context_start_line":160,"context_end_line":220,"code":" self.file_writer_thread = AudioFileWriterThread(\"output.wav\", self.sampling_rate)\n self.file_writer_thread.start()\n\n for audio_chunk, _ in audio_stream:\n if len(audio_chunk) > 0:\n logger.info(f\"Generated audio chunk of size: {len(audio_chunk)}\")\n\n # Send audio chunk via socket\n conn.sendall(struct.pack(f\"{len(audio_chunk)}f\", *audio_chunk))\n\n # Write to file asynchronously\n self.file_writer_thread.add_chunk(audio_chunk)\n\n logger.info(\"Finished sending audio stream.\")\n conn.sendall(b\"END\") # Send end signal\n\n # Ensure all audio data is written before exiting\n self.file_writer_thread.stop()\n\n\ndef handle_client(conn, processor):\n try:\n with conn:\n conn.setsockopt(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)\n while True:\n data = conn.recv(1024)\n if not data:\n processor.first_package = True\n break\n data_str = data.decode(\"utf-8\").strip()\n logger.info(f\"Received text: {data_str}\")\n\n try:\n processor.generate_stream(data_str, conn)\n except Exception as inner_e:\n logger.error(f\"Error during processing: {inner_e}\")\n traceback.print_exc()\n break\n except Exception as e:\n logger.error(f\"Error handling client: {e}\")\n traceback.print_exc()\n\n\ndef start_server(host, port, processor):\n with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:\n s.bind((host, port))\n s.listen()\n logger.info(f\"Server started on {host}:{port}\")\n while True:\n conn, addr = s.accept()\n logger.info(f\"Connected by {addr}\")\n handle_client(conn, processor)\n\n\nif __name__ == \"__main__\":\n parser = argparse.ArgumentParser()\n\n parser.add_argument(\"--host\", default=\"0.0.0.0\")\n parser.add_argument(\"--port\", default=9998)\n\n parser.add_argument(","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server.start_server","uri":"program://DMOSpeech2/function/src.f5_tts.socket_server.start_server#L203-L211","kind":"function","name":"start_server","path":"src/f5_tts/socket_server.py","language":"python","start_line":203,"end_line":211,"context_start_line":183,"context_end_line":231,"code":" conn.setsockopt(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)\n while True:\n data = conn.recv(1024)\n if not data:\n processor.first_package = True\n break\n data_str = data.decode(\"utf-8\").strip()\n logger.info(f\"Received text: {data_str}\")\n\n try:\n processor.generate_stream(data_str, conn)\n except Exception as inner_e:\n logger.error(f\"Error during processing: {inner_e}\")\n traceback.print_exc()\n break\n except Exception as e:\n logger.error(f\"Error handling client: {e}\")\n traceback.print_exc()\n\n\ndef start_server(host, port, processor):\n with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:\n s.bind((host, port))\n s.listen()\n logger.info(f\"Server started on {host}:{port}\")\n while True:\n conn, addr = s.accept()\n logger.info(f\"Connected by {addr}\")\n handle_client(conn, processor)\n\n\nif __name__ == \"__main__\":\n parser = argparse.ArgumentParser()\n\n parser.add_argument(\"--host\", default=\"0.0.0.0\")\n parser.add_argument(\"--port\", default=9998)\n\n parser.add_argument(\n \"--model\",\n default=\"F5TTS_v1_Base\",\n help=\"The model name, e.g. F5TTS_v1_Base\",\n )\n parser.add_argument(\n \"--ckpt_file\",\n default=str(hf_hub_download(repo_id=\"SWivid/F5-TTS\", filename=\"F5TTS_v1_Base/model_1250000.safetensors\")),\n help=\"Path to the model checkpoint file\",\n )\n parser.add_argument(\n \"--vocab_file\",","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.socket_server.__init__#L73-L95","kind":"function","name":"__init__","path":"src/f5_tts/socket_server.py","language":"python","start_line":73,"end_line":95,"context_start_line":53,"context_end_line":115,"code":" chunk = self.queue.get(timeout=0.1)\n if chunk is not None:\n chunk = np.int16(chunk * 32767)\n self.audio_data.append(chunk)\n wf.writeframes(chunk.tobytes())\n except queue.Empty:\n continue\n\n def add_chunk(self, chunk):\n \"\"\"Add a new chunk to the queue.\"\"\"\n self.queue.put(chunk)\n\n def stop(self):\n \"\"\"Stop writing and ensure all queued data is written.\"\"\"\n self.stop_event.set()\n self.join()\n logger.info(\"Audio writing completed.\")\n\n\nclass TTSStreamingProcessor:\n def __init__(self, model, ckpt_file, vocab_file, ref_audio, ref_text, device=None, dtype=torch.float32):\n self.device = device or (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n )\n model_cfg = OmegaConf.load(str(files(\"f5_tts\").joinpath(f\"configs/{model}.yaml\")))\n self.model_cls = get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n self.model_arc = model_cfg.model.arch\n self.mel_spec_type = model_cfg.model.mel_spec.mel_spec_type\n self.sampling_rate = model_cfg.model.mel_spec.target_sample_rate\n\n self.model = self.load_ema_model(ckpt_file, vocab_file, dtype)\n self.vocoder = self.load_vocoder_model()\n\n self.update_reference(ref_audio, ref_text)\n self._warm_up()\n self.file_writer_thread = None\n self.first_package = True\n\n def load_ema_model(self, ckpt_file, vocab_file, dtype):\n return load_model(\n self.model_cls,\n self.model_arc,\n ckpt_path=ckpt_file,\n mel_spec_type=self.mel_spec_type,\n vocab_file=vocab_file,\n ode_method=\"euler\",\n use_ema=True,\n device=self.device,\n ).to(self.device, dtype=dtype)\n\n def load_vocoder_model(self):\n return load_vocoder(vocoder_name=self.mel_spec_type, is_local=False, local_path=None, device=self.device)\n\n def update_reference(self, ref_audio, ref_text):\n self.ref_audio, self.ref_text = preprocess_ref_audio_text(ref_audio, ref_text)\n self.audio, self.sr = torchaudio.load(self.ref_audio)\n","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server.run","uri":"program://DMOSpeech2/function/src.f5_tts.socket_server.run#L43-L59","kind":"function","name":"run","path":"src/f5_tts/socket_server.py","language":"python","start_line":43,"end_line":59,"context_start_line":23,"context_end_line":79,"code":" load_vocoder,\n preprocess_ref_audio_text,\n)\n\n\nlogging.basicConfig(level=logging.INFO)\nlogger = logging.getLogger(__name__)\n\n\nclass AudioFileWriterThread(threading.Thread):\n \"\"\"Threaded file writer to avoid blocking the TTS streaming process.\"\"\"\n\n def __init__(self, output_file, sampling_rate):\n super().__init__()\n self.output_file = output_file\n self.sampling_rate = sampling_rate\n self.queue = queue.Queue()\n self.stop_event = threading.Event()\n self.audio_data = []\n\n def run(self):\n \"\"\"Process queued audio data and write it to a file.\"\"\"\n logger.info(\"AudioFileWriterThread started.\")\n with wave.open(self.output_file, \"wb\") as wf:\n wf.setnchannels(1)\n wf.setsampwidth(2)\n wf.setframerate(self.sampling_rate)\n\n while not self.stop_event.is_set() or not self.queue.empty():\n try:\n chunk = self.queue.get(timeout=0.1)\n if chunk is not None:\n chunk = np.int16(chunk * 32767)\n self.audio_data.append(chunk)\n wf.writeframes(chunk.tobytes())\n except queue.Empty:\n continue\n\n def add_chunk(self, chunk):\n \"\"\"Add a new chunk to the queue.\"\"\"\n self.queue.put(chunk)\n\n def stop(self):\n \"\"\"Stop writing and ensure all queued data is written.\"\"\"\n self.stop_event.set()\n self.join()\n logger.info(\"Audio writing completed.\")\n\n\nclass TTSStreamingProcessor:\n def __init__(self, model, ckpt_file, vocab_file, ref_audio, ref_text, device=None, dtype=torch.float32):\n self.device = device or (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server.add_chunk","uri":"program://DMOSpeech2/function/src.f5_tts.socket_server.add_chunk#L61-L63","kind":"function","name":"add_chunk","path":"src/f5_tts/socket_server.py","language":"python","start_line":61,"end_line":63,"context_start_line":41,"context_end_line":83,"code":" self.audio_data = []\n\n def run(self):\n \"\"\"Process queued audio data and write it to a file.\"\"\"\n logger.info(\"AudioFileWriterThread started.\")\n with wave.open(self.output_file, \"wb\") as wf:\n wf.setnchannels(1)\n wf.setsampwidth(2)\n wf.setframerate(self.sampling_rate)\n\n while not self.stop_event.is_set() or not self.queue.empty():\n try:\n chunk = self.queue.get(timeout=0.1)\n if chunk is not None:\n chunk = np.int16(chunk * 32767)\n self.audio_data.append(chunk)\n wf.writeframes(chunk.tobytes())\n except queue.Empty:\n continue\n\n def add_chunk(self, chunk):\n \"\"\"Add a new chunk to the queue.\"\"\"\n self.queue.put(chunk)\n\n def stop(self):\n \"\"\"Stop writing and ensure all queued data is written.\"\"\"\n self.stop_event.set()\n self.join()\n logger.info(\"Audio writing completed.\")\n\n\nclass TTSStreamingProcessor:\n def __init__(self, model, ckpt_file, vocab_file, ref_audio, ref_text, device=None, dtype=torch.float32):\n self.device = device or (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n )\n model_cfg = OmegaConf.load(str(files(\"f5_tts\").joinpath(f\"configs/{model}.yaml\")))","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server.stop","uri":"program://DMOSpeech2/function/src.f5_tts.socket_server.stop#L65-L69","kind":"function","name":"stop","path":"src/f5_tts/socket_server.py","language":"python","start_line":65,"end_line":69,"context_start_line":45,"context_end_line":89,"code":" logger.info(\"AudioFileWriterThread started.\")\n with wave.open(self.output_file, \"wb\") as wf:\n wf.setnchannels(1)\n wf.setsampwidth(2)\n wf.setframerate(self.sampling_rate)\n\n while not self.stop_event.is_set() or not self.queue.empty():\n try:\n chunk = self.queue.get(timeout=0.1)\n if chunk is not None:\n chunk = np.int16(chunk * 32767)\n self.audio_data.append(chunk)\n wf.writeframes(chunk.tobytes())\n except queue.Empty:\n continue\n\n def add_chunk(self, chunk):\n \"\"\"Add a new chunk to the queue.\"\"\"\n self.queue.put(chunk)\n\n def stop(self):\n \"\"\"Stop writing and ensure all queued data is written.\"\"\"\n self.stop_event.set()\n self.join()\n logger.info(\"Audio writing completed.\")\n\n\nclass TTSStreamingProcessor:\n def __init__(self, model, ckpt_file, vocab_file, ref_audio, ref_text, device=None, dtype=torch.float32):\n self.device = device or (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n )\n model_cfg = OmegaConf.load(str(files(\"f5_tts\").joinpath(f\"configs/{model}.yaml\")))\n self.model_cls = get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n self.model_arc = model_cfg.model.arch\n self.mel_spec_type = model_cfg.model.mel_spec.mel_spec_type\n self.sampling_rate = model_cfg.model.mel_spec.target_sample_rate\n\n self.model = self.load_ema_model(ckpt_file, vocab_file, dtype)","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server.load_ema_model","uri":"program://DMOSpeech2/function/src.f5_tts.socket_server.load_ema_model#L97-L107","kind":"function","name":"load_ema_model","path":"src/f5_tts/socket_server.py","language":"python","start_line":97,"end_line":107,"context_start_line":77,"context_end_line":127,"code":" else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n )\n model_cfg = OmegaConf.load(str(files(\"f5_tts\").joinpath(f\"configs/{model}.yaml\")))\n self.model_cls = get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n self.model_arc = model_cfg.model.arch\n self.mel_spec_type = model_cfg.model.mel_spec.mel_spec_type\n self.sampling_rate = model_cfg.model.mel_spec.target_sample_rate\n\n self.model = self.load_ema_model(ckpt_file, vocab_file, dtype)\n self.vocoder = self.load_vocoder_model()\n\n self.update_reference(ref_audio, ref_text)\n self._warm_up()\n self.file_writer_thread = None\n self.first_package = True\n\n def load_ema_model(self, ckpt_file, vocab_file, dtype):\n return load_model(\n self.model_cls,\n self.model_arc,\n ckpt_path=ckpt_file,\n mel_spec_type=self.mel_spec_type,\n vocab_file=vocab_file,\n ode_method=\"euler\",\n use_ema=True,\n device=self.device,\n ).to(self.device, dtype=dtype)\n\n def load_vocoder_model(self):\n return load_vocoder(vocoder_name=self.mel_spec_type, is_local=False, local_path=None, device=self.device)\n\n def update_reference(self, ref_audio, ref_text):\n self.ref_audio, self.ref_text = preprocess_ref_audio_text(ref_audio, ref_text)\n self.audio, self.sr = torchaudio.load(self.ref_audio)\n\n ref_audio_duration = self.audio.shape[-1] / self.sr\n ref_text_byte_len = len(self.ref_text.encode(\"utf-8\"))\n self.max_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration))\n self.few_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 2)\n self.min_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 4)\n\n def _warm_up(self):\n logger.info(\"Warming up the model...\")\n gen_text = \"Warm-up text for the model.\"\n for _ in infer_batch_process(\n (self.audio, self.sr),\n self.ref_text,","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server.load_vocoder_model","uri":"program://DMOSpeech2/function/src.f5_tts.socket_server.load_vocoder_model#L109-L110","kind":"function","name":"load_vocoder_model","path":"src/f5_tts/socket_server.py","language":"python","start_line":109,"end_line":110,"context_start_line":89,"context_end_line":130,"code":" self.model = self.load_ema_model(ckpt_file, vocab_file, dtype)\n self.vocoder = self.load_vocoder_model()\n\n self.update_reference(ref_audio, ref_text)\n self._warm_up()\n self.file_writer_thread = None\n self.first_package = True\n\n def load_ema_model(self, ckpt_file, vocab_file, dtype):\n return load_model(\n self.model_cls,\n self.model_arc,\n ckpt_path=ckpt_file,\n mel_spec_type=self.mel_spec_type,\n vocab_file=vocab_file,\n ode_method=\"euler\",\n use_ema=True,\n device=self.device,\n ).to(self.device, dtype=dtype)\n\n def load_vocoder_model(self):\n return load_vocoder(vocoder_name=self.mel_spec_type, is_local=False, local_path=None, device=self.device)\n\n def update_reference(self, ref_audio, ref_text):\n self.ref_audio, self.ref_text = preprocess_ref_audio_text(ref_audio, ref_text)\n self.audio, self.sr = torchaudio.load(self.ref_audio)\n\n ref_audio_duration = self.audio.shape[-1] / self.sr\n ref_text_byte_len = len(self.ref_text.encode(\"utf-8\"))\n self.max_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration))\n self.few_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 2)\n self.min_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 4)\n\n def _warm_up(self):\n logger.info(\"Warming up the model...\")\n gen_text = \"Warm-up text for the model.\"\n for _ in infer_batch_process(\n (self.audio, self.sr),\n self.ref_text,\n [gen_text],\n self.model,\n self.vocoder,","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server.update_reference","uri":"program://DMOSpeech2/function/src.f5_tts.socket_server.update_reference#L112-L120","kind":"function","name":"update_reference","path":"src/f5_tts/socket_server.py","language":"python","start_line":112,"end_line":120,"context_start_line":92,"context_end_line":140,"code":" self.update_reference(ref_audio, ref_text)\n self._warm_up()\n self.file_writer_thread = None\n self.first_package = True\n\n def load_ema_model(self, ckpt_file, vocab_file, dtype):\n return load_model(\n self.model_cls,\n self.model_arc,\n ckpt_path=ckpt_file,\n mel_spec_type=self.mel_spec_type,\n vocab_file=vocab_file,\n ode_method=\"euler\",\n use_ema=True,\n device=self.device,\n ).to(self.device, dtype=dtype)\n\n def load_vocoder_model(self):\n return load_vocoder(vocoder_name=self.mel_spec_type, is_local=False, local_path=None, device=self.device)\n\n def update_reference(self, ref_audio, ref_text):\n self.ref_audio, self.ref_text = preprocess_ref_audio_text(ref_audio, ref_text)\n self.audio, self.sr = torchaudio.load(self.ref_audio)\n\n ref_audio_duration = self.audio.shape[-1] / self.sr\n ref_text_byte_len = len(self.ref_text.encode(\"utf-8\"))\n self.max_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration))\n self.few_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 2)\n self.min_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 4)\n\n def _warm_up(self):\n logger.info(\"Warming up the model...\")\n gen_text = \"Warm-up text for the model.\"\n for _ in infer_batch_process(\n (self.audio, self.sr),\n self.ref_text,\n [gen_text],\n self.model,\n self.vocoder,\n progress=None,\n device=self.device,\n streaming=True,\n ):\n pass\n logger.info(\"Warm-up completed.\")\n\n def generate_stream(self, text, conn):\n text_batches = chunk_text(text, max_chars=self.max_chars)\n if self.first_package:","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server._warm_up","uri":"program://DMOSpeech2/function/src.f5_tts.socket_server._warm_up#L122-L136","kind":"function","name":"_warm_up","path":"src/f5_tts/socket_server.py","language":"python","start_line":122,"end_line":136,"context_start_line":102,"context_end_line":156,"code":" mel_spec_type=self.mel_spec_type,\n vocab_file=vocab_file,\n ode_method=\"euler\",\n use_ema=True,\n device=self.device,\n ).to(self.device, dtype=dtype)\n\n def load_vocoder_model(self):\n return load_vocoder(vocoder_name=self.mel_spec_type, is_local=False, local_path=None, device=self.device)\n\n def update_reference(self, ref_audio, ref_text):\n self.ref_audio, self.ref_text = preprocess_ref_audio_text(ref_audio, ref_text)\n self.audio, self.sr = torchaudio.load(self.ref_audio)\n\n ref_audio_duration = self.audio.shape[-1] / self.sr\n ref_text_byte_len = len(self.ref_text.encode(\"utf-8\"))\n self.max_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration))\n self.few_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 2)\n self.min_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 4)\n\n def _warm_up(self):\n logger.info(\"Warming up the model...\")\n gen_text = \"Warm-up text for the model.\"\n for _ in infer_batch_process(\n (self.audio, self.sr),\n self.ref_text,\n [gen_text],\n self.model,\n self.vocoder,\n progress=None,\n device=self.device,\n streaming=True,\n ):\n pass\n logger.info(\"Warm-up completed.\")\n\n def generate_stream(self, text, conn):\n text_batches = chunk_text(text, max_chars=self.max_chars)\n if self.first_package:\n text_batches = chunk_text(text_batches[0], max_chars=self.few_chars) + text_batches[1:]\n text_batches = chunk_text(text_batches[0], max_chars=self.min_chars) + text_batches[1:]\n self.first_package = False\n\n audio_stream = infer_batch_process(\n (self.audio, self.sr),\n self.ref_text,\n text_batches,\n self.model,\n self.vocoder,\n progress=None,\n device=self.device,\n streaming=True,\n chunk_size=2048,\n )\n","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.socket_server.generate_stream","uri":"program://DMOSpeech2/function/src.f5_tts.socket_server.generate_stream#L138-L177","kind":"function","name":"generate_stream","path":"src/f5_tts/socket_server.py","language":"python","start_line":138,"end_line":177,"context_start_line":118,"context_end_line":197,"code":" self.max_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration))\n self.few_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 2)\n self.min_chars = int(ref_text_byte_len / (ref_audio_duration) * (25 - ref_audio_duration) / 4)\n\n def _warm_up(self):\n logger.info(\"Warming up the model...\")\n gen_text = \"Warm-up text for the model.\"\n for _ in infer_batch_process(\n (self.audio, self.sr),\n self.ref_text,\n [gen_text],\n self.model,\n self.vocoder,\n progress=None,\n device=self.device,\n streaming=True,\n ):\n pass\n logger.info(\"Warm-up completed.\")\n\n def generate_stream(self, text, conn):\n text_batches = chunk_text(text, max_chars=self.max_chars)\n if self.first_package:\n text_batches = chunk_text(text_batches[0], max_chars=self.few_chars) + text_batches[1:]\n text_batches = chunk_text(text_batches[0], max_chars=self.min_chars) + text_batches[1:]\n self.first_package = False\n\n audio_stream = infer_batch_process(\n (self.audio, self.sr),\n self.ref_text,\n text_batches,\n self.model,\n self.vocoder,\n progress=None,\n device=self.device,\n streaming=True,\n chunk_size=2048,\n )\n\n # Reset the file writer thread\n if self.file_writer_thread is not None:\n self.file_writer_thread.stop()\n self.file_writer_thread = AudioFileWriterThread(\"output.wav\", self.sampling_rate)\n self.file_writer_thread.start()\n\n for audio_chunk, _ in audio_stream:\n if len(audio_chunk) > 0:\n logger.info(f\"Generated audio chunk of size: {len(audio_chunk)}\")\n\n # Send audio chunk via socket\n conn.sendall(struct.pack(f\"{len(audio_chunk)}f\", *audio_chunk))\n\n # Write to file asynchronously\n self.file_writer_thread.add_chunk(audio_chunk)\n\n logger.info(\"Finished sending audio stream.\")\n conn.sendall(b\"END\") # Send end signal\n\n # Ensure all audio data is written before exiting\n self.file_writer_thread.stop()\n\n\ndef handle_client(conn, processor):\n try:\n with conn:\n conn.setsockopt(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)\n while True:\n data = conn.recv(1024)\n if not data:\n processor.first_package = True\n break\n data_str = data.decode(\"utf-8\").strip()\n logger.info(f\"Received text: {data_str}\")\n\n try:\n processor.generate_stream(data_str, conn)\n except Exception as inner_e:\n logger.error(f\"Error during processing: {inner_e}\")\n traceback.print_exc()\n break","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.train","uri":"program://DMOSpeech2/module/src.f5_tts.train.train#L1-L77","kind":"module","name":"src.f5_tts.train.train","path":"src/f5_tts/train/train.py","language":"python","start_line":1,"end_line":77,"context_start_line":1,"context_end_line":77,"code":"# training script.\n\nimport os\nfrom importlib.resources import files\n\nimport hydra\nfrom omegaconf import OmegaConf\n\nfrom f5_tts.model import CFM, Trainer\nfrom f5_tts.model.dataset import load_dataset\nfrom f5_tts.model.utils import get_tokenizer\n\n\nos.chdir(str(files(\"f5_tts\").joinpath(\"../..\"))) # change working directory to root of project (local editable)\n\n\n@hydra.main(version_base=\"1.3\", config_path=str(files(\"f5_tts\").joinpath(\"configs\")), config_name=None)\ndef main(model_cfg):\n model_cls = hydra.utils.get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n model_arc = model_cfg.model.arch\n tokenizer = model_cfg.model.tokenizer\n mel_spec_type = model_cfg.model.mel_spec.mel_spec_type\n\n exp_name = f\"{model_cfg.model.name}_{mel_spec_type}_{model_cfg.model.tokenizer}_{model_cfg.datasets.name}\"\n wandb_resume_id = None\n\n # set text tokenizer\n if tokenizer != \"custom\":\n tokenizer_path = model_cfg.datasets.name\n else:\n tokenizer_path = model_cfg.model.tokenizer_path\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n\n # set model\n model = CFM(\n transformer=model_cls(**model_arc, text_num_embeds=vocab_size, mel_dim=model_cfg.model.mel_spec.n_mel_channels),\n mel_spec_kwargs=model_cfg.model.mel_spec,\n vocab_char_map=vocab_char_map,\n )\n\n # init trainer\n trainer = Trainer(\n model,\n epochs=model_cfg.optim.epochs,\n learning_rate=model_cfg.optim.learning_rate,\n num_warmup_updates=model_cfg.optim.num_warmup_updates,\n save_per_updates=model_cfg.ckpts.save_per_updates,\n keep_last_n_checkpoints=model_cfg.ckpts.keep_last_n_checkpoints,\n checkpoint_path=str(files(\"f5_tts\").joinpath(f\"../../{model_cfg.ckpts.save_dir}\")),\n batch_size_per_gpu=model_cfg.datasets.batch_size_per_gpu,\n batch_size_type=model_cfg.datasets.batch_size_type,\n max_samples=model_cfg.datasets.max_samples,\n grad_accumulation_steps=model_cfg.optim.grad_accumulation_steps,\n max_grad_norm=model_cfg.optim.max_grad_norm,\n logger=model_cfg.ckpts.logger,\n wandb_project=\"CFM-TTS\",\n wandb_run_name=exp_name,\n wandb_resume_id=wandb_resume_id,\n last_per_updates=model_cfg.ckpts.last_per_updates,\n log_samples=model_cfg.ckpts.log_samples,\n bnb_optimizer=model_cfg.optim.bnb_optimizer,\n mel_spec_type=mel_spec_type,\n is_local_vocoder=model_cfg.model.vocoder.is_local,\n local_vocoder_path=model_cfg.model.vocoder.local_path,\n model_cfg_dict=OmegaConf.to_container(model_cfg, resolve=True),\n )\n\n train_dataset = load_dataset(model_cfg.datasets.name, tokenizer, mel_spec_kwargs=model_cfg.model.mel_spec)\n trainer.train(\n train_dataset,\n num_workers=model_cfg.datasets.num_workers,\n resumable_with_seed=666, # seed for shuffling dataset\n )\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"0b23913aed7f806c7697e3bf8d390c96a5b868c8c05158129999b98c70a50ca4","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.train.main","uri":"program://DMOSpeech2/function/src.f5_tts.train.train.main#L18-L73","kind":"function","name":"main","path":"src/f5_tts/train/train.py","language":"python","start_line":18,"end_line":73,"context_start_line":1,"context_end_line":77,"code":"# training script.\n\nimport os\nfrom importlib.resources import files\n\nimport hydra\nfrom omegaconf import OmegaConf\n\nfrom f5_tts.model import CFM, Trainer\nfrom f5_tts.model.dataset import load_dataset\nfrom f5_tts.model.utils import get_tokenizer\n\n\nos.chdir(str(files(\"f5_tts\").joinpath(\"../..\"))) # change working directory to root of project (local editable)\n\n\n@hydra.main(version_base=\"1.3\", config_path=str(files(\"f5_tts\").joinpath(\"configs\")), config_name=None)\ndef main(model_cfg):\n model_cls = hydra.utils.get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n model_arc = model_cfg.model.arch\n tokenizer = model_cfg.model.tokenizer\n mel_spec_type = model_cfg.model.mel_spec.mel_spec_type\n\n exp_name = f\"{model_cfg.model.name}_{mel_spec_type}_{model_cfg.model.tokenizer}_{model_cfg.datasets.name}\"\n wandb_resume_id = None\n\n # set text tokenizer\n if tokenizer != \"custom\":\n tokenizer_path = model_cfg.datasets.name\n else:\n tokenizer_path = model_cfg.model.tokenizer_path\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n\n # set model\n model = CFM(\n transformer=model_cls(**model_arc, text_num_embeds=vocab_size, mel_dim=model_cfg.model.mel_spec.n_mel_channels),\n mel_spec_kwargs=model_cfg.model.mel_spec,\n vocab_char_map=vocab_char_map,\n )\n\n # init trainer\n trainer = Trainer(\n model,\n epochs=model_cfg.optim.epochs,\n learning_rate=model_cfg.optim.learning_rate,\n num_warmup_updates=model_cfg.optim.num_warmup_updates,\n save_per_updates=model_cfg.ckpts.save_per_updates,\n keep_last_n_checkpoints=model_cfg.ckpts.keep_last_n_checkpoints,\n checkpoint_path=str(files(\"f5_tts\").joinpath(f\"../../{model_cfg.ckpts.save_dir}\")),\n batch_size_per_gpu=model_cfg.datasets.batch_size_per_gpu,\n batch_size_type=model_cfg.datasets.batch_size_type,\n max_samples=model_cfg.datasets.max_samples,\n grad_accumulation_steps=model_cfg.optim.grad_accumulation_steps,\n max_grad_norm=model_cfg.optim.max_grad_norm,\n logger=model_cfg.ckpts.logger,\n wandb_project=\"CFM-TTS\",\n wandb_run_name=exp_name,\n wandb_resume_id=wandb_resume_id,\n last_per_updates=model_cfg.ckpts.last_per_updates,\n log_samples=model_cfg.ckpts.log_samples,\n bnb_optimizer=model_cfg.optim.bnb_optimizer,\n mel_spec_type=mel_spec_type,\n is_local_vocoder=model_cfg.model.vocoder.is_local,\n local_vocoder_path=model_cfg.model.vocoder.local_path,\n model_cfg_dict=OmegaConf.to_container(model_cfg, resolve=True),\n )\n\n train_dataset = load_dataset(model_cfg.datasets.name, tokenizer, mel_spec_kwargs=model_cfg.model.mel_spec)\n trainer.train(\n train_dataset,\n num_workers=model_cfg.datasets.num_workers,\n resumable_with_seed=666, # seed for shuffling dataset\n )\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"0b23913aed7f806c7697e3bf8d390c96a5b868c8c05158129999b98c70a50ca4","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_cli","uri":"program://DMOSpeech2/module/src.f5_tts.train.finetune_cli#L1-L214","kind":"module","name":"src.f5_tts.train.finetune_cli","path":"src/f5_tts/train/finetune_cli.py","language":"python","start_line":1,"end_line":214,"context_start_line":1,"context_end_line":214,"code":"import argparse\nimport os\nimport shutil\nfrom importlib.resources import files\n\nfrom cached_path import cached_path\n\nfrom f5_tts.model import CFM, DiT, Trainer, UNetT\nfrom f5_tts.model.dataset import load_dataset\nfrom f5_tts.model.utils import get_tokenizer\n\n\n# -------------------------- Dataset Settings --------------------------- #\ntarget_sample_rate = 24000\nn_mel_channels = 100\nhop_length = 256\nwin_length = 1024\nn_fft = 1024\nmel_spec_type = \"vocos\" # 'vocos' or 'bigvgan'\n\n\n# -------------------------- Argument Parsing --------------------------- #\ndef parse_args():\n parser = argparse.ArgumentParser(description=\"Train CFM Model\")\n\n parser.add_argument(\n \"--exp_name\",\n type=str,\n default=\"F5TTS_v1_Base\",\n choices=[\"F5TTS_v1_Base\", \"F5TTS_Base\", \"E2TTS_Base\"],\n help=\"Experiment name\",\n )\n parser.add_argument(\"--dataset_name\", type=str, default=\"Emilia_ZH_EN\", help=\"Name of the dataset to use\")\n parser.add_argument(\"--learning_rate\", type=float, default=1e-5, help=\"Learning rate for training\")\n parser.add_argument(\"--batch_size_per_gpu\", type=int, default=3200, help=\"Batch size per GPU\")\n parser.add_argument(\n \"--batch_size_type\", type=str, default=\"frame\", choices=[\"frame\", \"sample\"], help=\"Batch size type\"\n )\n parser.add_argument(\"--max_samples\", type=int, default=64, help=\"Max sequences per batch\")\n parser.add_argument(\"--grad_accumulation_steps\", type=int, default=1, help=\"Gradient accumulation steps\")\n parser.add_argument(\"--max_grad_norm\", type=float, default=1.0, help=\"Max gradient norm for clipping\")\n parser.add_argument(\"--epochs\", type=int, default=100, help=\"Number of training epochs\")\n parser.add_argument(\"--num_warmup_updates\", type=int, default=20000, help=\"Warmup updates\")\n parser.add_argument(\"--save_per_updates\", type=int, default=50000, help=\"Save checkpoint every N updates\")\n parser.add_argument(\n \"--keep_last_n_checkpoints\",\n type=int,\n default=-1,\n help=\"-1 to keep all, 0 to not save intermediate, > 0 to keep last N checkpoints\",\n )\n parser.add_argument(\"--last_per_updates\", type=int, default=5000, help=\"Save last checkpoint every N updates\")\n parser.add_argument(\"--finetune\", action=\"store_true\", help=\"Use Finetune\")\n parser.add_argument(\"--pretrain\", type=str, default=None, help=\"the path to the checkpoint\")\n parser.add_argument(\n \"--tokenizer\", type=str, default=\"pinyin\", choices=[\"pinyin\", \"char\", \"custom\"], help=\"Tokenizer type\"\n )\n parser.add_argument(\n \"--tokenizer_path\",\n type=str,\n default=None,\n help=\"Path to custom tokenizer vocab file (only used if tokenizer = 'custom')\",\n )\n parser.add_argument(\n \"--log_samples\",\n action=\"store_true\",\n help=\"Log inferenced samples per ckpt save updates\",\n )\n parser.add_argument(\"--logger\", type=str, default=None, choices=[None, \"wandb\", \"tensorboard\"], help=\"logger\")\n parser.add_argument(\n \"--bnb_optimizer\",\n action=\"store_true\",\n help=\"Use 8-bit Adam optimizer from bitsandbytes\",\n )\n\n return parser.parse_args()\n\n\n# -------------------------- Training Settings -------------------------- #\n\n\ndef main():\n args = parse_args()\n\n checkpoint_path = str(files(\"f5_tts\").joinpath(f\"../../ckpts/{args.dataset_name}\"))\n\n # Model parameters based on experiment name\n\n if args.exp_name == \"F5TTS_v1_Base\":\n wandb_resume_id = None\n model_cls = DiT\n model_cfg = dict(\n dim=1024,\n depth=22,\n heads=16,\n ff_mult=2,\n text_dim=512,\n conv_layers=4,\n )\n if args.finetune:\n if args.pretrain is None:\n ckpt_path = str(cached_path(\"hf://SWivid/F5-TTS/F5TTS_v1_Base/model_1250000.safetensors\"))\n else:\n ckpt_path = args.pretrain\n\n elif args.exp_name == \"F5TTS_Base\":\n wandb_resume_id = None\n model_cls = DiT\n model_cfg = dict(\n dim=1024,\n depth=22,\n heads=16,\n ff_mult=2,\n text_dim=512,\n text_mask_padding=False,\n conv_layers=4,\n pe_attn_head=1,\n )\n if args.finetune:\n if args.pretrain is None:\n ckpt_path = str(cached_path(\"hf://SWivid/F5-TTS/F5TTS_Base/model_1200000.pt\"))\n else:\n ckpt_path = args.pretrain\n\n elif args.exp_name == \"E2TTS_Base\":\n wandb_resume_id = None\n model_cls = UNetT\n model_cfg = dict(\n dim=1024,\n depth=24,\n heads=16,\n ff_mult=4,\n text_mask_padding=False,\n pe_attn_head=1,\n )\n if args.finetune:\n if args.pretrain is None:\n ckpt_path = str(cached_path(\"hf://SWivid/E2-TTS/E2TTS_Base/model_1200000.pt\"))\n else:\n ckpt_path = args.pretrain\n\n if args.finetune:\n if not os.path.isdir(checkpoint_path):\n os.makedirs(checkpoint_path, exist_ok=True)\n\n file_checkpoint = os.path.basename(ckpt_path)\n if not file_checkpoint.startswith(\"pretrained_\"): # Change: Add 'pretrained_' prefix to copied model\n file_checkpoint = \"pretrained_\" + file_checkpoint\n file_checkpoint = os.path.join(checkpoint_path, file_checkpoint)\n if not os.path.isfile(file_checkpoint):\n shutil.copy2(ckpt_path, file_checkpoint)\n print(\"copy checkpoint for finetune\")\n\n # Use the tokenizer and tokenizer_path provided in the command line arguments\n\n tokenizer = args.tokenizer\n if tokenizer == \"custom\":\n if not args.tokenizer_path:\n raise ValueError(\"Custom tokenizer selected, but no tokenizer_path provided.\")\n tokenizer_path = args.tokenizer_path\n else:\n tokenizer_path = args.dataset_name\n\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n\n print(\"\\nvocab : \", vocab_size)\n print(\"\\nvocoder : \", mel_spec_type)\n\n mel_spec_kwargs = dict(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n )\n\n model = CFM(\n transformer=model_cls(**model_cfg, text_num_embeds=vocab_size, mel_dim=n_mel_channels),\n mel_spec_kwargs=mel_spec_kwargs,\n vocab_char_map=vocab_char_map,\n )\n\n trainer = Trainer(\n model,\n args.epochs,\n args.learning_rate,\n num_warmup_updates=args.num_warmup_updates,\n save_per_updates=args.save_per_updates,\n keep_last_n_checkpoints=args.keep_last_n_checkpoints,\n checkpoint_path=checkpoint_path,\n batch_size_per_gpu=args.batch_size_per_gpu,\n batch_size_type=args.batch_size_type,\n max_samples=args.max_samples,\n grad_accumulation_steps=args.grad_accumulation_steps,\n max_grad_norm=args.max_grad_norm,\n logger=args.logger,\n wandb_project=args.dataset_name,\n wandb_run_name=args.exp_name,\n wandb_resume_id=wandb_resume_id,\n log_samples=args.log_samples,\n last_per_updates=args.last_per_updates,\n bnb_optimizer=args.bnb_optimizer,\n )\n\n train_dataset = load_dataset(args.dataset_name, tokenizer, mel_spec_kwargs=mel_spec_kwargs)\n\n trainer.train(\n train_dataset,\n resumable_with_seed=666, # seed for shuffling dataset\n )\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"c04786bf9631406fbea26b1b51f469fa6e7ddb9a3607090baa217ef45c23c4ae","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_cli.parse_args","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_cli.parse_args#L23-L75","kind":"function","name":"parse_args","path":"src/f5_tts/train/finetune_cli.py","language":"python","start_line":23,"end_line":75,"context_start_line":3,"context_end_line":95,"code":"import shutil\nfrom importlib.resources import files\n\nfrom cached_path import cached_path\n\nfrom f5_tts.model import CFM, DiT, Trainer, UNetT\nfrom f5_tts.model.dataset import load_dataset\nfrom f5_tts.model.utils import get_tokenizer\n\n\n# -------------------------- Dataset Settings --------------------------- #\ntarget_sample_rate = 24000\nn_mel_channels = 100\nhop_length = 256\nwin_length = 1024\nn_fft = 1024\nmel_spec_type = \"vocos\" # 'vocos' or 'bigvgan'\n\n\n# -------------------------- Argument Parsing --------------------------- #\ndef parse_args():\n parser = argparse.ArgumentParser(description=\"Train CFM Model\")\n\n parser.add_argument(\n \"--exp_name\",\n type=str,\n default=\"F5TTS_v1_Base\",\n choices=[\"F5TTS_v1_Base\", \"F5TTS_Base\", \"E2TTS_Base\"],\n help=\"Experiment name\",\n )\n parser.add_argument(\"--dataset_name\", type=str, default=\"Emilia_ZH_EN\", help=\"Name of the dataset to use\")\n parser.add_argument(\"--learning_rate\", type=float, default=1e-5, help=\"Learning rate for training\")\n parser.add_argument(\"--batch_size_per_gpu\", type=int, default=3200, help=\"Batch size per GPU\")\n parser.add_argument(\n \"--batch_size_type\", type=str, default=\"frame\", choices=[\"frame\", \"sample\"], help=\"Batch size type\"\n )\n parser.add_argument(\"--max_samples\", type=int, default=64, help=\"Max sequences per batch\")\n parser.add_argument(\"--grad_accumulation_steps\", type=int, default=1, help=\"Gradient accumulation steps\")\n parser.add_argument(\"--max_grad_norm\", type=float, default=1.0, help=\"Max gradient norm for clipping\")\n parser.add_argument(\"--epochs\", type=int, default=100, help=\"Number of training epochs\")\n parser.add_argument(\"--num_warmup_updates\", type=int, default=20000, help=\"Warmup updates\")\n parser.add_argument(\"--save_per_updates\", type=int, default=50000, help=\"Save checkpoint every N updates\")\n parser.add_argument(\n \"--keep_last_n_checkpoints\",\n type=int,\n default=-1,\n help=\"-1 to keep all, 0 to not save intermediate, > 0 to keep last N checkpoints\",\n )\n parser.add_argument(\"--last_per_updates\", type=int, default=5000, help=\"Save last checkpoint every N updates\")\n parser.add_argument(\"--finetune\", action=\"store_true\", help=\"Use Finetune\")\n parser.add_argument(\"--pretrain\", type=str, default=None, help=\"the path to the checkpoint\")\n parser.add_argument(\n \"--tokenizer\", type=str, default=\"pinyin\", choices=[\"pinyin\", \"char\", \"custom\"], help=\"Tokenizer type\"\n )\n parser.add_argument(\n \"--tokenizer_path\",\n type=str,\n default=None,\n help=\"Path to custom tokenizer vocab file (only used if tokenizer = 'custom')\",\n )\n parser.add_argument(\n \"--log_samples\",\n action=\"store_true\",\n help=\"Log inferenced samples per ckpt save updates\",\n )\n parser.add_argument(\"--logger\", type=str, default=None, choices=[None, \"wandb\", \"tensorboard\"], help=\"logger\")\n parser.add_argument(\n \"--bnb_optimizer\",\n action=\"store_true\",\n help=\"Use 8-bit Adam optimizer from bitsandbytes\",\n )\n\n return parser.parse_args()\n\n\n# -------------------------- Training Settings -------------------------- #\n\n\ndef main():\n args = parse_args()\n\n checkpoint_path = str(files(\"f5_tts\").joinpath(f\"../../ckpts/{args.dataset_name}\"))\n\n # Model parameters based on experiment name\n\n if args.exp_name == \"F5TTS_v1_Base\":\n wandb_resume_id = None\n model_cls = DiT\n model_cfg = dict(\n dim=1024,\n depth=22,\n heads=16,\n ff_mult=2,","source_hash":"c04786bf9631406fbea26b1b51f469fa6e7ddb9a3607090baa217ef45c23c4ae","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_cli.main","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_cli.main#L81-L210","kind":"function","name":"main","path":"src/f5_tts/train/finetune_cli.py","language":"python","start_line":81,"end_line":210,"context_start_line":61,"context_end_line":214,"code":" help=\"Path to custom tokenizer vocab file (only used if tokenizer = 'custom')\",\n )\n parser.add_argument(\n \"--log_samples\",\n action=\"store_true\",\n help=\"Log inferenced samples per ckpt save updates\",\n )\n parser.add_argument(\"--logger\", type=str, default=None, choices=[None, \"wandb\", \"tensorboard\"], help=\"logger\")\n parser.add_argument(\n \"--bnb_optimizer\",\n action=\"store_true\",\n help=\"Use 8-bit Adam optimizer from bitsandbytes\",\n )\n\n return parser.parse_args()\n\n\n# -------------------------- Training Settings -------------------------- #\n\n\ndef main():\n args = parse_args()\n\n checkpoint_path = str(files(\"f5_tts\").joinpath(f\"../../ckpts/{args.dataset_name}\"))\n\n # Model parameters based on experiment name\n\n if args.exp_name == \"F5TTS_v1_Base\":\n wandb_resume_id = None\n model_cls = DiT\n model_cfg = dict(\n dim=1024,\n depth=22,\n heads=16,\n ff_mult=2,\n text_dim=512,\n conv_layers=4,\n )\n if args.finetune:\n if args.pretrain is None:\n ckpt_path = str(cached_path(\"hf://SWivid/F5-TTS/F5TTS_v1_Base/model_1250000.safetensors\"))\n else:\n ckpt_path = args.pretrain\n\n elif args.exp_name == \"F5TTS_Base\":\n wandb_resume_id = None\n model_cls = DiT\n model_cfg = dict(\n dim=1024,\n depth=22,\n heads=16,\n ff_mult=2,\n text_dim=512,\n text_mask_padding=False,\n conv_layers=4,\n pe_attn_head=1,\n )\n if args.finetune:\n if args.pretrain is None:\n ckpt_path = str(cached_path(\"hf://SWivid/F5-TTS/F5TTS_Base/model_1200000.pt\"))\n else:\n ckpt_path = args.pretrain\n\n elif args.exp_name == \"E2TTS_Base\":\n wandb_resume_id = None\n model_cls = UNetT\n model_cfg = dict(\n dim=1024,\n depth=24,\n heads=16,\n ff_mult=4,\n text_mask_padding=False,\n pe_attn_head=1,\n )\n if args.finetune:\n if args.pretrain is None:\n ckpt_path = str(cached_path(\"hf://SWivid/E2-TTS/E2TTS_Base/model_1200000.pt\"))\n else:\n ckpt_path = args.pretrain\n\n if args.finetune:\n if not os.path.isdir(checkpoint_path):\n os.makedirs(checkpoint_path, exist_ok=True)\n\n file_checkpoint = os.path.basename(ckpt_path)\n if not file_checkpoint.startswith(\"pretrained_\"): # Change: Add 'pretrained_' prefix to copied model\n file_checkpoint = \"pretrained_\" + file_checkpoint\n file_checkpoint = os.path.join(checkpoint_path, file_checkpoint)\n if not os.path.isfile(file_checkpoint):\n shutil.copy2(ckpt_path, file_checkpoint)\n print(\"copy checkpoint for finetune\")\n\n # Use the tokenizer and tokenizer_path provided in the command line arguments\n\n tokenizer = args.tokenizer\n if tokenizer == \"custom\":\n if not args.tokenizer_path:\n raise ValueError(\"Custom tokenizer selected, but no tokenizer_path provided.\")\n tokenizer_path = args.tokenizer_path\n else:\n tokenizer_path = args.dataset_name\n\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n\n print(\"\\nvocab : \", vocab_size)\n print(\"\\nvocoder : \", mel_spec_type)\n\n mel_spec_kwargs = dict(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n )\n\n model = CFM(\n transformer=model_cls(**model_cfg, text_num_embeds=vocab_size, mel_dim=n_mel_channels),\n mel_spec_kwargs=mel_spec_kwargs,\n vocab_char_map=vocab_char_map,\n )\n\n trainer = Trainer(\n model,\n args.epochs,\n args.learning_rate,\n num_warmup_updates=args.num_warmup_updates,\n save_per_updates=args.save_per_updates,\n keep_last_n_checkpoints=args.keep_last_n_checkpoints,\n checkpoint_path=checkpoint_path,\n batch_size_per_gpu=args.batch_size_per_gpu,\n batch_size_type=args.batch_size_type,\n max_samples=args.max_samples,\n grad_accumulation_steps=args.grad_accumulation_steps,\n max_grad_norm=args.max_grad_norm,\n logger=args.logger,\n wandb_project=args.dataset_name,\n wandb_run_name=args.exp_name,\n wandb_resume_id=wandb_resume_id,\n log_samples=args.log_samples,\n last_per_updates=args.last_per_updates,\n bnb_optimizer=args.bnb_optimizer,\n )\n\n train_dataset = load_dataset(args.dataset_name, tokenizer, mel_spec_kwargs=mel_spec_kwargs)\n\n trainer.train(\n train_dataset,\n resumable_with_seed=666, # seed for shuffling dataset\n )\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"c04786bf9631406fbea26b1b51f469fa6e7ddb9a3607090baa217ef45c23c4ae","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio","uri":"program://DMOSpeech2/module/src.f5_tts.train.finetune_gradio#L1-L1865","kind":"module","name":"src.f5_tts.train.finetune_gradio","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1,"end_line":1865,"context_start_line":1,"context_end_line":1865,"code":"import gc\nimport json\nimport os\nimport platform\nimport queue\nimport random\nimport re\nimport shutil\nimport signal\nimport subprocess\nimport sys\nimport tempfile\nimport threading\nimport time\nfrom glob import glob\nfrom importlib.resources import files\n\nimport click\nimport gradio as gr\nimport librosa\nimport numpy as np\nimport psutil\nimport torch\nimport torchaudio\nfrom cached_path import cached_path\nfrom datasets import Dataset as Dataset_\nfrom datasets.arrow_writer import ArrowWriter\nfrom safetensors.torch import load_file, save_file\nfrom scipy.io import wavfile\n\nfrom f5_tts.api import F5TTS\nfrom f5_tts.infer.utils_infer import transcribe\nfrom f5_tts.model.utils import convert_char_to_pinyin\n\n\ntraining_process = None\nsystem = platform.system()\npython_executable = sys.executable or \"python\"\ntts_api = None\nlast_checkpoint = \"\"\nlast_device = \"\"\nlast_ema = None\n\n\npath_data = str(files(\"f5_tts\").joinpath(\"../../data\"))\npath_project_ckpts = str(files(\"f5_tts\").joinpath(\"../../ckpts\"))\nfile_train = str(files(\"f5_tts\").joinpath(\"train/finetune_cli.py\"))\n\ndevice = (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n)\n\n\n# Save settings from a JSON file\ndef save_settings(\n project_name,\n exp_name,\n learning_rate,\n batch_size_per_gpu,\n batch_size_type,\n max_samples,\n grad_accumulation_steps,\n max_grad_norm,\n epochs,\n num_warmup_updates,\n save_per_updates,\n keep_last_n_checkpoints,\n last_per_updates,\n finetune,\n file_checkpoint_train,\n tokenizer_type,\n tokenizer_file,\n mixed_precision,\n logger,\n ch_8bit_adam,\n):\n path_project = os.path.join(path_project_ckpts, project_name)\n os.makedirs(path_project, exist_ok=True)\n file_setting = os.path.join(path_project, \"setting.json\")\n\n settings = {\n \"exp_name\": exp_name,\n \"learning_rate\": learning_rate,\n \"batch_size_per_gpu\": batch_size_per_gpu,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"epochs\": epochs,\n \"num_warmup_updates\": num_warmup_updates,\n \"save_per_updates\": save_per_updates,\n \"keep_last_n_checkpoints\": keep_last_n_checkpoints,\n \"last_per_updates\": last_per_updates,\n \"finetune\": finetune,\n \"file_checkpoint_train\": file_checkpoint_train,\n \"tokenizer_type\": tokenizer_type,\n \"tokenizer_file\": tokenizer_file,\n \"mixed_precision\": mixed_precision,\n \"logger\": logger,\n \"bnb_optimizer\": ch_8bit_adam,\n }\n with open(file_setting, \"w\") as f:\n json.dump(settings, f, indent=4)\n return \"Settings saved!\"\n\n\n# Load settings from a JSON file\ndef load_settings(project_name):\n project_name = project_name.replace(\"_pinyin\", \"\").replace(\"_char\", \"\")\n path_project = os.path.join(path_project_ckpts, project_name)\n file_setting = os.path.join(path_project, \"setting.json\")\n\n # Default settings\n default_settings = {\n \"exp_name\": \"F5TTS_v1_Base\",\n \"learning_rate\": 1e-5,\n \"batch_size_per_gpu\": 3200,\n \"batch_size_type\": \"frame\",\n \"max_samples\": 64,\n \"grad_accumulation_steps\": 1,\n \"max_grad_norm\": 1.0,\n \"epochs\": 100,\n \"num_warmup_updates\": 100,\n \"save_per_updates\": 500,\n \"keep_last_n_checkpoints\": -1,\n \"last_per_updates\": 100,\n \"finetune\": True,\n \"file_checkpoint_train\": \"\",\n \"tokenizer_type\": \"pinyin\",\n \"tokenizer_file\": \"\",\n \"mixed_precision\": \"fp16\",\n \"logger\": \"none\",\n \"bnb_optimizer\": False,\n }\n if device == \"mps\":\n default_settings[\"mixed_precision\"] = \"none\"\n\n # Load settings from file if it exists\n if os.path.isfile(file_setting):\n with open(file_setting, \"r\") as f:\n file_settings = json.load(f)\n default_settings.update(file_settings)\n\n # Return as a tuple in the correct order\n return (\n default_settings[\"exp_name\"],\n default_settings[\"learning_rate\"],\n default_settings[\"batch_size_per_gpu\"],\n default_settings[\"batch_size_type\"],\n default_settings[\"max_samples\"],\n default_settings[\"grad_accumulation_steps\"],\n default_settings[\"max_grad_norm\"],\n default_settings[\"epochs\"],\n default_settings[\"num_warmup_updates\"],\n default_settings[\"save_per_updates\"],\n default_settings[\"keep_last_n_checkpoints\"],\n default_settings[\"last_per_updates\"],\n default_settings[\"finetune\"],\n default_settings[\"file_checkpoint_train\"],\n default_settings[\"tokenizer_type\"],\n default_settings[\"tokenizer_file\"],\n default_settings[\"mixed_precision\"],\n default_settings[\"logger\"],\n default_settings[\"bnb_optimizer\"],\n )\n\n\n# Load metadata\ndef get_audio_duration(audio_path):\n \"\"\"Calculate the duration mono of an audio file.\"\"\"\n audio, sample_rate = torchaudio.load(audio_path)\n return audio.shape[1] / sample_rate\n\n\nclass Slicer: # https://github.com/RVC-Boss/GPT-SoVITS/blob/main/tools/slicer2.py\n def __init__(\n self,\n sr: int,\n threshold: float = -40.0,\n min_length: int = 20000, # 20 seconds\n min_interval: int = 300,\n hop_size: int = 20,\n max_sil_kept: int = 2000,\n ):\n if not min_length >= min_interval >= hop_size:\n raise ValueError(\"The following condition must be satisfied: min_length >= min_interval >= hop_size\")\n if not max_sil_kept >= hop_size:\n raise ValueError(\"The following condition must be satisfied: max_sil_kept >= hop_size\")\n min_interval = sr * min_interval / 1000\n self.threshold = 10 ** (threshold / 20.0)\n self.hop_size = round(sr * hop_size / 1000)\n self.win_size = min(round(min_interval), 4 * self.hop_size)\n self.min_length = round(sr * min_length / 1000 / self.hop_size)\n self.min_interval = round(min_interval / self.hop_size)\n self.max_sil_kept = round(sr * max_sil_kept / 1000 / self.hop_size)\n\n def _apply_slice(self, waveform, begin, end):\n if len(waveform.shape) > 1:\n return waveform[:, begin * self.hop_size : min(waveform.shape[1], end * self.hop_size)]\n else:\n return waveform[begin * self.hop_size : min(waveform.shape[0], end * self.hop_size)]\n\n # @timeit\n def slice(self, waveform):\n if len(waveform.shape) > 1:\n samples = waveform.mean(axis=0)\n else:\n samples = waveform\n if samples.shape[0] <= self.min_length:\n return [waveform]\n rms_list = librosa.feature.rms(y=samples, frame_length=self.win_size, hop_length=self.hop_size).squeeze(0)\n sil_tags = []\n silence_start = None\n clip_start = 0\n for i, rms in enumerate(rms_list):\n # Keep looping while frame is silent.\n if rms < self.threshold:\n # Record start of silent frames.\n if silence_start is None:\n silence_start = i\n continue\n # Keep looping while frame is not silent and silence start has not been recorded.\n if silence_start is None:\n continue\n # Clear recorded silence start if interval is not enough or clip is too short\n is_leading_silence = silence_start == 0 and i > self.max_sil_kept\n need_slice_middle = i - silence_start >= self.min_interval and i - clip_start >= self.min_length\n if not is_leading_silence and not need_slice_middle:\n silence_start = None\n continue\n # Need slicing. Record the range of silent frames to be removed.\n if i - silence_start <= self.max_sil_kept:\n pos = rms_list[silence_start : i + 1].argmin() + silence_start\n if silence_start == 0:\n sil_tags.append((0, pos))\n else:\n sil_tags.append((pos, pos))\n clip_start = pos\n elif i - silence_start <= self.max_sil_kept * 2:\n pos = rms_list[i - self.max_sil_kept : silence_start + self.max_sil_kept + 1].argmin()\n pos += i - self.max_sil_kept\n pos_l = rms_list[silence_start : silence_start + self.max_sil_kept + 1].argmin() + silence_start\n pos_r = rms_list[i - self.max_sil_kept : i + 1].argmin() + i - self.max_sil_kept\n if silence_start == 0:\n sil_tags.append((0, pos_r))\n clip_start = pos_r\n else:\n sil_tags.append((min(pos_l, pos), max(pos_r, pos)))\n clip_start = max(pos_r, pos)\n else:\n pos_l = rms_list[silence_start : silence_start + self.max_sil_kept + 1].argmin() + silence_start\n pos_r = rms_list[i - self.max_sil_kept : i + 1].argmin() + i - self.max_sil_kept\n if silence_start == 0:\n sil_tags.append((0, pos_r))\n else:\n sil_tags.append((pos_l, pos_r))\n clip_start = pos_r\n silence_start = None\n # Deal with trailing silence.\n total_frames = rms_list.shape[0]\n if silence_start is not None and total_frames - silence_start >= self.min_interval:\n silence_end = min(total_frames, silence_start + self.max_sil_kept)\n pos = rms_list[silence_start : silence_end + 1].argmin() + silence_start\n sil_tags.append((pos, total_frames + 1))\n # Apply and return slices: [chunk, start, end]\n if len(sil_tags) == 0:\n return [[waveform, 0, int(total_frames * self.hop_size)]]\n else:\n chunks = []\n if sil_tags[0][0] > 0:\n chunks.append([self._apply_slice(waveform, 0, sil_tags[0][0]), 0, int(sil_tags[0][0] * self.hop_size)])\n for i in range(len(sil_tags) - 1):\n chunks.append(\n [\n self._apply_slice(waveform, sil_tags[i][1], sil_tags[i + 1][0]),\n int(sil_tags[i][1] * self.hop_size),\n int(sil_tags[i + 1][0] * self.hop_size),\n ]\n )\n if sil_tags[-1][1] < total_frames:\n chunks.append(\n [\n self._apply_slice(waveform, sil_tags[-1][1], total_frames),\n int(sil_tags[-1][1] * self.hop_size),\n int(total_frames * self.hop_size),\n ]\n )\n return chunks\n\n\n# terminal\ndef terminate_process_tree(pid, including_parent=True):\n try:\n parent = psutil.Process(pid)\n except psutil.NoSuchProcess:\n # Process already terminated\n return\n\n children = parent.children(recursive=True)\n for child in children:\n try:\n os.kill(child.pid, signal.SIGTERM) # or signal.SIGKILL\n except OSError:\n pass\n if including_parent:\n try:\n os.kill(parent.pid, signal.SIGTERM) # or signal.SIGKILL\n except OSError:\n pass\n\n\ndef terminate_process(pid):\n if system == \"Windows\":\n cmd = f\"taskkill /t /f /pid {pid}\"\n os.system(cmd)\n else:\n terminate_process_tree(pid)\n\n\ndef start_training(\n dataset_name,\n exp_name,\n learning_rate,\n batch_size_per_gpu,\n batch_size_type,\n max_samples,\n grad_accumulation_steps,\n max_grad_norm,\n epochs,\n num_warmup_updates,\n save_per_updates,\n keep_last_n_checkpoints,\n last_per_updates,\n finetune,\n file_checkpoint_train,\n tokenizer_type,\n tokenizer_file,\n mixed_precision,\n stream,\n logger,\n ch_8bit_adam,\n):\n global training_process, tts_api, stop_signal\n\n if tts_api is not None:\n if tts_api is not None:\n del tts_api\n\n gc.collect()\n torch.cuda.empty_cache()\n tts_api = None\n\n path_project = os.path.join(path_data, dataset_name)\n\n if not os.path.isdir(path_project):\n yield (\n f\"There is not project with name {dataset_name}\",\n gr.update(interactive=True),\n gr.update(interactive=False),\n )\n return\n\n file_raw = os.path.join(path_project, \"raw.arrow\")\n if not os.path.isfile(file_raw):\n yield f\"There is no file {file_raw}\", gr.update(interactive=True), gr.update(interactive=False)\n return\n\n # Check if a training process is already running\n if training_process is not None:\n return \"Train run already!\", gr.update(interactive=False), gr.update(interactive=True)\n\n yield \"start train\", gr.update(interactive=False), gr.update(interactive=False)\n\n # Command to run the training script with the specified arguments\n\n if tokenizer_file == \"\":\n if dataset_name.endswith(\"_pinyin\"):\n tokenizer_type = \"pinyin\"\n elif dataset_name.endswith(\"_char\"):\n tokenizer_type = \"char\"\n else:\n tokenizer_type = \"custom\"\n\n dataset_name = dataset_name.replace(\"_pinyin\", \"\").replace(\"_char\", \"\")\n\n if mixed_precision != \"none\":\n fp16 = f\"--mixed_precision={mixed_precision}\"\n else:\n fp16 = \"\"\n\n cmd = (\n f'accelerate launch {fp16} \"{file_train}\" --exp_name {exp_name}'\n f\" --learning_rate {learning_rate}\"\n f\" --batch_size_per_gpu {batch_size_per_gpu}\"\n f\" --batch_size_type {batch_size_type}\"\n f\" --max_samples {max_samples}\"\n f\" --grad_accumulation_steps {grad_accumulation_steps}\"\n f\" --max_grad_norm {max_grad_norm}\"\n f\" --epochs {epochs}\"\n f\" --num_warmup_updates {num_warmup_updates}\"\n f\" --save_per_updates {save_per_updates}\"\n f\" --keep_last_n_checkpoints {keep_last_n_checkpoints}\"\n f\" --last_per_updates {last_per_updates}\"\n f\" --dataset_name {dataset_name}\"\n )\n\n if finetune:\n cmd += \" --finetune\"\n\n if file_checkpoint_train != \"\":\n cmd += f' --pretrain \"{file_checkpoint_train}\"'\n\n if tokenizer_file != \"\":\n cmd += f\" --tokenizer_path {tokenizer_file}\"\n\n cmd += f\" --tokenizer {tokenizer_type}\"\n\n if logger != \"none\":\n cmd += f\" --logger {logger}\"\n\n cmd += \" --log_samples\"\n\n if ch_8bit_adam:\n cmd += \" --bnb_optimizer\"\n\n print(\"run command : \\n\" + cmd + \"\\n\")\n\n save_settings(\n dataset_name,\n exp_name,\n learning_rate,\n batch_size_per_gpu,\n batch_size_type,\n max_samples,\n grad_accumulation_steps,\n max_grad_norm,\n epochs,\n num_warmup_updates,\n save_per_updates,\n keep_last_n_checkpoints,\n last_per_updates,\n finetune,\n file_checkpoint_train,\n tokenizer_type,\n tokenizer_file,\n mixed_precision,\n logger,\n ch_8bit_adam,\n )\n\n try:\n if not stream:\n # Start the training process\n training_process = subprocess.Popen(cmd, shell=True)\n\n time.sleep(5)\n yield \"train start\", gr.update(interactive=False), gr.update(interactive=True)\n\n # Wait for the training process to finish\n training_process.wait()\n else:\n\n def stream_output(pipe, output_queue):\n try:\n for line in iter(pipe.readline, \"\"):\n output_queue.put(line)\n except Exception as e:\n output_queue.put(f\"Error reading pipe: {str(e)}\")\n finally:\n pipe.close()\n\n env = os.environ.copy()\n env[\"PYTHONUNBUFFERED\"] = \"1\"\n\n training_process = subprocess.Popen(\n cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, bufsize=1, env=env\n )\n yield \"Training started ...\", gr.update(interactive=False), gr.update(interactive=True)\n\n stdout_queue = queue.Queue()\n stderr_queue = queue.Queue()\n\n stdout_thread = threading.Thread(target=stream_output, args=(training_process.stdout, stdout_queue))\n stderr_thread = threading.Thread(target=stream_output, args=(training_process.stderr, stderr_queue))\n stdout_thread.daemon = True\n stderr_thread.daemon = True\n stdout_thread.start()\n stderr_thread.start()\n stop_signal = False\n while True:\n if stop_signal:\n training_process.terminate()\n time.sleep(0.5)\n if training_process.poll() is None:\n training_process.kill()\n yield \"Training stopped by user.\", gr.update(interactive=True), gr.update(interactive=False)\n break\n\n process_status = training_process.poll()\n\n # Handle stdout\n try:\n while True:\n output = stdout_queue.get_nowait()\n print(output, end=\"\")\n match = re.search(\n r\"Epoch (\\d+)/(\\d+):\\s+(\\d+)%\\|.*\\[(\\d+:\\d+)<.*?loss=(\\d+\\.\\d+), update=(\\d+)\", output\n )\n if match:\n current_epoch = match.group(1)\n total_epochs = match.group(2)\n percent_complete = match.group(3)\n elapsed_time = match.group(4)\n loss = match.group(5)\n current_update = match.group(6)\n message = (\n f\"Epoch: {current_epoch}/{total_epochs}, \"\n f\"Progress: {percent_complete}%, \"\n f\"Elapsed Time: {elapsed_time}, \"\n f\"Loss: {loss}, \"\n f\"Update: {current_update}\"\n )\n yield message, gr.update(interactive=False), gr.update(interactive=True)\n elif output.strip():\n yield output, gr.update(interactive=False), gr.update(interactive=True)\n except queue.Empty:\n pass\n\n # Handle stderr\n try:\n while True:\n error_output = stderr_queue.get_nowait()\n print(error_output, end=\"\")\n if error_output.strip():\n yield f\"{error_output.strip()}\", gr.update(interactive=False), gr.update(interactive=True)\n except queue.Empty:\n pass\n\n if process_status is not None and stdout_queue.empty() and stderr_queue.empty():\n if process_status != 0:\n yield (\n f\"Process crashed with exit code {process_status}!\",\n gr.update(interactive=False),\n gr.update(interactive=True),\n )\n else:\n yield (\n \"Training complete or paused ...\",\n gr.update(interactive=False),\n gr.update(interactive=True),\n )\n break\n\n # Small sleep to prevent CPU thrashing\n time.sleep(0.1)\n\n # Clean up\n training_process.stdout.close()\n training_process.stderr.close()\n training_process.wait()\n\n time.sleep(1)\n\n if training_process is None:\n text_info = \"Train stopped !\"\n else:\n# ... truncated ...","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.save_settings","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.save_settings#L61-L110","kind":"function","name":"save_settings","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":61,"end_line":110,"context_start_line":41,"context_end_line":130,"code":"last_device = \"\"\nlast_ema = None\n\n\npath_data = str(files(\"f5_tts\").joinpath(\"../../data\"))\npath_project_ckpts = str(files(\"f5_tts\").joinpath(\"../../ckpts\"))\nfile_train = str(files(\"f5_tts\").joinpath(\"train/finetune_cli.py\"))\n\ndevice = (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n)\n\n\n# Save settings from a JSON file\ndef save_settings(\n project_name,\n exp_name,\n learning_rate,\n batch_size_per_gpu,\n batch_size_type,\n max_samples,\n grad_accumulation_steps,\n max_grad_norm,\n epochs,\n num_warmup_updates,\n save_per_updates,\n keep_last_n_checkpoints,\n last_per_updates,\n finetune,\n file_checkpoint_train,\n tokenizer_type,\n tokenizer_file,\n mixed_precision,\n logger,\n ch_8bit_adam,\n):\n path_project = os.path.join(path_project_ckpts, project_name)\n os.makedirs(path_project, exist_ok=True)\n file_setting = os.path.join(path_project, \"setting.json\")\n\n settings = {\n \"exp_name\": exp_name,\n \"learning_rate\": learning_rate,\n \"batch_size_per_gpu\": batch_size_per_gpu,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"epochs\": epochs,\n \"num_warmup_updates\": num_warmup_updates,\n \"save_per_updates\": save_per_updates,\n \"keep_last_n_checkpoints\": keep_last_n_checkpoints,\n \"last_per_updates\": last_per_updates,\n \"finetune\": finetune,\n \"file_checkpoint_train\": file_checkpoint_train,\n \"tokenizer_type\": tokenizer_type,\n \"tokenizer_file\": tokenizer_file,\n \"mixed_precision\": mixed_precision,\n \"logger\": logger,\n \"bnb_optimizer\": ch_8bit_adam,\n }\n with open(file_setting, \"w\") as f:\n json.dump(settings, f, indent=4)\n return \"Settings saved!\"\n\n\n# Load settings from a JSON file\ndef load_settings(project_name):\n project_name = project_name.replace(\"_pinyin\", \"\").replace(\"_char\", \"\")\n path_project = os.path.join(path_project_ckpts, project_name)\n file_setting = os.path.join(path_project, \"setting.json\")\n\n # Default settings\n default_settings = {\n \"exp_name\": \"F5TTS_v1_Base\",\n \"learning_rate\": 1e-5,\n \"batch_size_per_gpu\": 3200,\n \"batch_size_type\": \"frame\",\n \"max_samples\": 64,\n \"grad_accumulation_steps\": 1,\n \"max_grad_norm\": 1.0,\n \"epochs\": 100,\n \"num_warmup_updates\": 100,\n \"save_per_updates\": 500,","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.load_settings","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.load_settings#L114-L171","kind":"function","name":"load_settings","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":114,"end_line":171,"context_start_line":94,"context_end_line":191,"code":" \"max_grad_norm\": max_grad_norm,\n \"epochs\": epochs,\n \"num_warmup_updates\": num_warmup_updates,\n \"save_per_updates\": save_per_updates,\n \"keep_last_n_checkpoints\": keep_last_n_checkpoints,\n \"last_per_updates\": last_per_updates,\n \"finetune\": finetune,\n \"file_checkpoint_train\": file_checkpoint_train,\n \"tokenizer_type\": tokenizer_type,\n \"tokenizer_file\": tokenizer_file,\n \"mixed_precision\": mixed_precision,\n \"logger\": logger,\n \"bnb_optimizer\": ch_8bit_adam,\n }\n with open(file_setting, \"w\") as f:\n json.dump(settings, f, indent=4)\n return \"Settings saved!\"\n\n\n# Load settings from a JSON file\ndef load_settings(project_name):\n project_name = project_name.replace(\"_pinyin\", \"\").replace(\"_char\", \"\")\n path_project = os.path.join(path_project_ckpts, project_name)\n file_setting = os.path.join(path_project, \"setting.json\")\n\n # Default settings\n default_settings = {\n \"exp_name\": \"F5TTS_v1_Base\",\n \"learning_rate\": 1e-5,\n \"batch_size_per_gpu\": 3200,\n \"batch_size_type\": \"frame\",\n \"max_samples\": 64,\n \"grad_accumulation_steps\": 1,\n \"max_grad_norm\": 1.0,\n \"epochs\": 100,\n \"num_warmup_updates\": 100,\n \"save_per_updates\": 500,\n \"keep_last_n_checkpoints\": -1,\n \"last_per_updates\": 100,\n \"finetune\": True,\n \"file_checkpoint_train\": \"\",\n \"tokenizer_type\": \"pinyin\",\n \"tokenizer_file\": \"\",\n \"mixed_precision\": \"fp16\",\n \"logger\": \"none\",\n \"bnb_optimizer\": False,\n }\n if device == \"mps\":\n default_settings[\"mixed_precision\"] = \"none\"\n\n # Load settings from file if it exists\n if os.path.isfile(file_setting):\n with open(file_setting, \"r\") as f:\n file_settings = json.load(f)\n default_settings.update(file_settings)\n\n # Return as a tuple in the correct order\n return (\n default_settings[\"exp_name\"],\n default_settings[\"learning_rate\"],\n default_settings[\"batch_size_per_gpu\"],\n default_settings[\"batch_size_type\"],\n default_settings[\"max_samples\"],\n default_settings[\"grad_accumulation_steps\"],\n default_settings[\"max_grad_norm\"],\n default_settings[\"epochs\"],\n default_settings[\"num_warmup_updates\"],\n default_settings[\"save_per_updates\"],\n default_settings[\"keep_last_n_checkpoints\"],\n default_settings[\"last_per_updates\"],\n default_settings[\"finetune\"],\n default_settings[\"file_checkpoint_train\"],\n default_settings[\"tokenizer_type\"],\n default_settings[\"tokenizer_file\"],\n default_settings[\"mixed_precision\"],\n default_settings[\"logger\"],\n default_settings[\"bnb_optimizer\"],\n )\n\n\n# Load metadata\ndef get_audio_duration(audio_path):\n \"\"\"Calculate the duration mono of an audio file.\"\"\"\n audio, sample_rate = torchaudio.load(audio_path)\n return audio.shape[1] / sample_rate\n\n\nclass Slicer: # https://github.com/RVC-Boss/GPT-SoVITS/blob/main/tools/slicer2.py\n def __init__(\n self,\n sr: int,\n threshold: float = -40.0,\n min_length: int = 20000, # 20 seconds\n min_interval: int = 300,\n hop_size: int = 20,\n max_sil_kept: int = 2000,\n ):\n if not min_length >= min_interval >= hop_size:","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.get_audio_duration","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.get_audio_duration#L175-L178","kind":"function","name":"get_audio_duration","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":175,"end_line":178,"context_start_line":155,"context_end_line":198,"code":" default_settings[\"batch_size_type\"],\n default_settings[\"max_samples\"],\n default_settings[\"grad_accumulation_steps\"],\n default_settings[\"max_grad_norm\"],\n default_settings[\"epochs\"],\n default_settings[\"num_warmup_updates\"],\n default_settings[\"save_per_updates\"],\n default_settings[\"keep_last_n_checkpoints\"],\n default_settings[\"last_per_updates\"],\n default_settings[\"finetune\"],\n default_settings[\"file_checkpoint_train\"],\n default_settings[\"tokenizer_type\"],\n default_settings[\"tokenizer_file\"],\n default_settings[\"mixed_precision\"],\n default_settings[\"logger\"],\n default_settings[\"bnb_optimizer\"],\n )\n\n\n# Load metadata\ndef get_audio_duration(audio_path):\n \"\"\"Calculate the duration mono of an audio file.\"\"\"\n audio, sample_rate = torchaudio.load(audio_path)\n return audio.shape[1] / sample_rate\n\n\nclass Slicer: # https://github.com/RVC-Boss/GPT-SoVITS/blob/main/tools/slicer2.py\n def __init__(\n self,\n sr: int,\n threshold: float = -40.0,\n min_length: int = 20000, # 20 seconds\n min_interval: int = 300,\n hop_size: int = 20,\n max_sil_kept: int = 2000,\n ):\n if not min_length >= min_interval >= hop_size:\n raise ValueError(\"The following condition must be satisfied: min_length >= min_interval >= hop_size\")\n if not max_sil_kept >= hop_size:\n raise ValueError(\"The following condition must be satisfied: max_sil_kept >= hop_size\")\n min_interval = sr * min_interval / 1000\n self.threshold = 10 ** (threshold / 20.0)\n self.hop_size = round(sr * hop_size / 1000)\n self.win_size = min(round(min_interval), 4 * self.hop_size)","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.Slicer","uri":"program://DMOSpeech2/class/src.f5_tts.train.finetune_gradio.Slicer#L181-L294","kind":"class","name":"Slicer","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":181,"end_line":294,"context_start_line":161,"context_end_line":314,"code":" default_settings[\"save_per_updates\"],\n default_settings[\"keep_last_n_checkpoints\"],\n default_settings[\"last_per_updates\"],\n default_settings[\"finetune\"],\n default_settings[\"file_checkpoint_train\"],\n default_settings[\"tokenizer_type\"],\n default_settings[\"tokenizer_file\"],\n default_settings[\"mixed_precision\"],\n default_settings[\"logger\"],\n default_settings[\"bnb_optimizer\"],\n )\n\n\n# Load metadata\ndef get_audio_duration(audio_path):\n \"\"\"Calculate the duration mono of an audio file.\"\"\"\n audio, sample_rate = torchaudio.load(audio_path)\n return audio.shape[1] / sample_rate\n\n\nclass Slicer: # https://github.com/RVC-Boss/GPT-SoVITS/blob/main/tools/slicer2.py\n def __init__(\n self,\n sr: int,\n threshold: float = -40.0,\n min_length: int = 20000, # 20 seconds\n min_interval: int = 300,\n hop_size: int = 20,\n max_sil_kept: int = 2000,\n ):\n if not min_length >= min_interval >= hop_size:\n raise ValueError(\"The following condition must be satisfied: min_length >= min_interval >= hop_size\")\n if not max_sil_kept >= hop_size:\n raise ValueError(\"The following condition must be satisfied: max_sil_kept >= hop_size\")\n min_interval = sr * min_interval / 1000\n self.threshold = 10 ** (threshold / 20.0)\n self.hop_size = round(sr * hop_size / 1000)\n self.win_size = min(round(min_interval), 4 * self.hop_size)\n self.min_length = round(sr * min_length / 1000 / self.hop_size)\n self.min_interval = round(min_interval / self.hop_size)\n self.max_sil_kept = round(sr * max_sil_kept / 1000 / self.hop_size)\n\n def _apply_slice(self, waveform, begin, end):\n if len(waveform.shape) > 1:\n return waveform[:, begin * self.hop_size : min(waveform.shape[1], end * self.hop_size)]\n else:\n return waveform[begin * self.hop_size : min(waveform.shape[0], end * self.hop_size)]\n\n # @timeit\n def slice(self, waveform):\n if len(waveform.shape) > 1:\n samples = waveform.mean(axis=0)\n else:\n samples = waveform\n if samples.shape[0] <= self.min_length:\n return [waveform]\n rms_list = librosa.feature.rms(y=samples, frame_length=self.win_size, hop_length=self.hop_size).squeeze(0)\n sil_tags = []\n silence_start = None\n clip_start = 0\n for i, rms in enumerate(rms_list):\n # Keep looping while frame is silent.\n if rms < self.threshold:\n # Record start of silent frames.\n if silence_start is None:\n silence_start = i\n continue\n # Keep looping while frame is not silent and silence start has not been recorded.\n if silence_start is None:\n continue\n # Clear recorded silence start if interval is not enough or clip is too short\n is_leading_silence = silence_start == 0 and i > self.max_sil_kept\n need_slice_middle = i - silence_start >= self.min_interval and i - clip_start >= self.min_length\n if not is_leading_silence and not need_slice_middle:\n silence_start = None\n continue\n # Need slicing. Record the range of silent frames to be removed.\n if i - silence_start <= self.max_sil_kept:\n pos = rms_list[silence_start : i + 1].argmin() + silence_start\n if silence_start == 0:\n sil_tags.append((0, pos))\n else:\n sil_tags.append((pos, pos))\n clip_start = pos\n elif i - silence_start <= self.max_sil_kept * 2:\n pos = rms_list[i - self.max_sil_kept : silence_start + self.max_sil_kept + 1].argmin()\n pos += i - self.max_sil_kept\n pos_l = rms_list[silence_start : silence_start + self.max_sil_kept + 1].argmin() + silence_start\n pos_r = rms_list[i - self.max_sil_kept : i + 1].argmin() + i - self.max_sil_kept\n if silence_start == 0:\n sil_tags.append((0, pos_r))\n clip_start = pos_r\n else:\n sil_tags.append((min(pos_l, pos), max(pos_r, pos)))\n clip_start = max(pos_r, pos)\n else:\n pos_l = rms_list[silence_start : silence_start + self.max_sil_kept + 1].argmin() + silence_start\n pos_r = rms_list[i - self.max_sil_kept : i + 1].argmin() + i - self.max_sil_kept\n if silence_start == 0:\n sil_tags.append((0, pos_r))\n else:\n sil_tags.append((pos_l, pos_r))\n clip_start = pos_r\n silence_start = None\n # Deal with trailing silence.\n total_frames = rms_list.shape[0]\n if silence_start is not None and total_frames - silence_start >= self.min_interval:\n silence_end = min(total_frames, silence_start + self.max_sil_kept)\n pos = rms_list[silence_start : silence_end + 1].argmin() + silence_start\n sil_tags.append((pos, total_frames + 1))\n # Apply and return slices: [chunk, start, end]\n if len(sil_tags) == 0:\n return [[waveform, 0, int(total_frames * self.hop_size)]]\n else:\n chunks = []\n if sil_tags[0][0] > 0:\n chunks.append([self._apply_slice(waveform, 0, sil_tags[0][0]), 0, int(sil_tags[0][0] * self.hop_size)])\n for i in range(len(sil_tags) - 1):\n chunks.append(\n [\n self._apply_slice(waveform, sil_tags[i][1], sil_tags[i + 1][0]),\n int(sil_tags[i][1] * self.hop_size),\n int(sil_tags[i + 1][0] * self.hop_size),\n ]\n )\n if sil_tags[-1][1] < total_frames:\n chunks.append(\n [\n self._apply_slice(waveform, sil_tags[-1][1], total_frames),\n int(sil_tags[-1][1] * self.hop_size),\n int(total_frames * self.hop_size),\n ]\n )\n return chunks\n\n\n# terminal\ndef terminate_process_tree(pid, including_parent=True):\n try:\n parent = psutil.Process(pid)\n except psutil.NoSuchProcess:\n # Process already terminated\n return\n\n children = parent.children(recursive=True)\n for child in children:\n try:\n os.kill(child.pid, signal.SIGTERM) # or signal.SIGKILL\n except OSError:\n pass\n if including_parent:\n try:\n os.kill(parent.pid, signal.SIGTERM) # or signal.SIGKILL\n except OSError:","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.terminate_process_tree","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.terminate_process_tree#L298-L315","kind":"function","name":"terminate_process_tree","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":298,"end_line":315,"context_start_line":278,"context_end_line":335,"code":" for i in range(len(sil_tags) - 1):\n chunks.append(\n [\n self._apply_slice(waveform, sil_tags[i][1], sil_tags[i + 1][0]),\n int(sil_tags[i][1] * self.hop_size),\n int(sil_tags[i + 1][0] * self.hop_size),\n ]\n )\n if sil_tags[-1][1] < total_frames:\n chunks.append(\n [\n self._apply_slice(waveform, sil_tags[-1][1], total_frames),\n int(sil_tags[-1][1] * self.hop_size),\n int(total_frames * self.hop_size),\n ]\n )\n return chunks\n\n\n# terminal\ndef terminate_process_tree(pid, including_parent=True):\n try:\n parent = psutil.Process(pid)\n except psutil.NoSuchProcess:\n # Process already terminated\n return\n\n children = parent.children(recursive=True)\n for child in children:\n try:\n os.kill(child.pid, signal.SIGTERM) # or signal.SIGKILL\n except OSError:\n pass\n if including_parent:\n try:\n os.kill(parent.pid, signal.SIGTERM) # or signal.SIGKILL\n except OSError:\n pass\n\n\ndef terminate_process(pid):\n if system == \"Windows\":\n cmd = f\"taskkill /t /f /pid {pid}\"\n os.system(cmd)\n else:\n terminate_process_tree(pid)\n\n\ndef start_training(\n dataset_name,\n exp_name,\n learning_rate,\n batch_size_per_gpu,\n batch_size_type,\n max_samples,\n grad_accumulation_steps,\n max_grad_norm,\n epochs,","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.terminate_process","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.terminate_process#L318-L323","kind":"function","name":"terminate_process","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":318,"end_line":323,"context_start_line":298,"context_end_line":343,"code":"def terminate_process_tree(pid, including_parent=True):\n try:\n parent = psutil.Process(pid)\n except psutil.NoSuchProcess:\n # Process already terminated\n return\n\n children = parent.children(recursive=True)\n for child in children:\n try:\n os.kill(child.pid, signal.SIGTERM) # or signal.SIGKILL\n except OSError:\n pass\n if including_parent:\n try:\n os.kill(parent.pid, signal.SIGTERM) # or signal.SIGKILL\n except OSError:\n pass\n\n\ndef terminate_process(pid):\n if system == \"Windows\":\n cmd = f\"taskkill /t /f /pid {pid}\"\n os.system(cmd)\n else:\n terminate_process_tree(pid)\n\n\ndef start_training(\n dataset_name,\n exp_name,\n learning_rate,\n batch_size_per_gpu,\n batch_size_type,\n max_samples,\n grad_accumulation_steps,\n max_grad_norm,\n epochs,\n num_warmup_updates,\n save_per_updates,\n keep_last_n_checkpoints,\n last_per_updates,\n finetune,\n file_checkpoint_train,\n tokenizer_type,\n tokenizer_file,","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.start_training","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.start_training#L326-L581","kind":"function","name":"start_training","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":326,"end_line":581,"context_start_line":306,"context_end_line":601,"code":" for child in children:\n try:\n os.kill(child.pid, signal.SIGTERM) # or signal.SIGKILL\n except OSError:\n pass\n if including_parent:\n try:\n os.kill(parent.pid, signal.SIGTERM) # or signal.SIGKILL\n except OSError:\n pass\n\n\ndef terminate_process(pid):\n if system == \"Windows\":\n cmd = f\"taskkill /t /f /pid {pid}\"\n os.system(cmd)\n else:\n terminate_process_tree(pid)\n\n\ndef start_training(\n dataset_name,\n exp_name,\n learning_rate,\n batch_size_per_gpu,\n batch_size_type,\n max_samples,\n grad_accumulation_steps,\n max_grad_norm,\n epochs,\n num_warmup_updates,\n save_per_updates,\n keep_last_n_checkpoints,\n last_per_updates,\n finetune,\n file_checkpoint_train,\n tokenizer_type,\n tokenizer_file,\n mixed_precision,\n stream,\n logger,\n ch_8bit_adam,\n):\n global training_process, tts_api, stop_signal\n\n if tts_api is not None:\n if tts_api is not None:\n del tts_api\n\n gc.collect()\n torch.cuda.empty_cache()\n tts_api = None\n\n path_project = os.path.join(path_data, dataset_name)\n\n if not os.path.isdir(path_project):\n yield (\n f\"There is not project with name {dataset_name}\",\n gr.update(interactive=True),\n gr.update(interactive=False),\n )\n return\n\n file_raw = os.path.join(path_project, \"raw.arrow\")\n if not os.path.isfile(file_raw):\n yield f\"There is no file {file_raw}\", gr.update(interactive=True), gr.update(interactive=False)\n return\n\n # Check if a training process is already running\n if training_process is not None:\n return \"Train run already!\", gr.update(interactive=False), gr.update(interactive=True)\n\n yield \"start train\", gr.update(interactive=False), gr.update(interactive=False)\n\n # Command to run the training script with the specified arguments\n\n if tokenizer_file == \"\":\n if dataset_name.endswith(\"_pinyin\"):\n tokenizer_type = \"pinyin\"\n elif dataset_name.endswith(\"_char\"):\n tokenizer_type = \"char\"\n else:\n tokenizer_type = \"custom\"\n\n dataset_name = dataset_name.replace(\"_pinyin\", \"\").replace(\"_char\", \"\")\n\n if mixed_precision != \"none\":\n fp16 = f\"--mixed_precision={mixed_precision}\"\n else:\n fp16 = \"\"\n\n cmd = (\n f'accelerate launch {fp16} \"{file_train}\" --exp_name {exp_name}'\n f\" --learning_rate {learning_rate}\"\n f\" --batch_size_per_gpu {batch_size_per_gpu}\"\n f\" --batch_size_type {batch_size_type}\"\n f\" --max_samples {max_samples}\"\n f\" --grad_accumulation_steps {grad_accumulation_steps}\"\n f\" --max_grad_norm {max_grad_norm}\"\n f\" --epochs {epochs}\"\n f\" --num_warmup_updates {num_warmup_updates}\"\n f\" --save_per_updates {save_per_updates}\"\n f\" --keep_last_n_checkpoints {keep_last_n_checkpoints}\"\n f\" --last_per_updates {last_per_updates}\"\n f\" --dataset_name {dataset_name}\"\n )\n\n if finetune:\n cmd += \" --finetune\"\n\n if file_checkpoint_train != \"\":\n cmd += f' --pretrain \"{file_checkpoint_train}\"'\n\n if tokenizer_file != \"\":\n cmd += f\" --tokenizer_path {tokenizer_file}\"\n\n cmd += f\" --tokenizer {tokenizer_type}\"\n\n if logger != \"none\":\n cmd += f\" --logger {logger}\"\n\n cmd += \" --log_samples\"\n\n if ch_8bit_adam:\n cmd += \" --bnb_optimizer\"\n\n print(\"run command : \\n\" + cmd + \"\\n\")\n\n save_settings(\n dataset_name,\n exp_name,\n learning_rate,\n batch_size_per_gpu,\n batch_size_type,\n max_samples,\n grad_accumulation_steps,\n max_grad_norm,\n epochs,\n num_warmup_updates,\n save_per_updates,\n keep_last_n_checkpoints,\n last_per_updates,\n finetune,\n file_checkpoint_train,\n tokenizer_type,\n tokenizer_file,\n mixed_precision,\n logger,\n ch_8bit_adam,\n )\n\n try:\n if not stream:\n # Start the training process\n training_process = subprocess.Popen(cmd, shell=True)\n\n time.sleep(5)\n yield \"train start\", gr.update(interactive=False), gr.update(interactive=True)\n\n # Wait for the training process to finish\n training_process.wait()\n else:\n\n def stream_output(pipe, output_queue):\n try:\n for line in iter(pipe.readline, \"\"):\n output_queue.put(line)\n except Exception as e:\n output_queue.put(f\"Error reading pipe: {str(e)}\")\n finally:\n pipe.close()\n\n env = os.environ.copy()\n env[\"PYTHONUNBUFFERED\"] = \"1\"\n\n training_process = subprocess.Popen(\n cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, bufsize=1, env=env\n )\n yield \"Training started ...\", gr.update(interactive=False), gr.update(interactive=True)\n\n stdout_queue = queue.Queue()\n stderr_queue = queue.Queue()\n\n stdout_thread = threading.Thread(target=stream_output, args=(training_process.stdout, stdout_queue))\n stderr_thread = threading.Thread(target=stream_output, args=(training_process.stderr, stderr_queue))\n stdout_thread.daemon = True\n stderr_thread.daemon = True\n stdout_thread.start()\n stderr_thread.start()\n stop_signal = False\n while True:\n if stop_signal:\n training_process.terminate()\n time.sleep(0.5)\n if training_process.poll() is None:\n training_process.kill()\n yield \"Training stopped by user.\", gr.update(interactive=True), gr.update(interactive=False)\n break\n\n process_status = training_process.poll()\n\n # Handle stdout\n try:\n while True:\n output = stdout_queue.get_nowait()\n print(output, end=\"\")\n match = re.search(\n r\"Epoch (\\d+)/(\\d+):\\s+(\\d+)%\\|.*\\[(\\d+:\\d+)<.*?loss=(\\d+\\.\\d+), update=(\\d+)\", output\n )\n if match:\n current_epoch = match.group(1)\n total_epochs = match.group(2)\n percent_complete = match.group(3)\n elapsed_time = match.group(4)\n loss = match.group(5)\n current_update = match.group(6)\n message = (\n f\"Epoch: {current_epoch}/{total_epochs}, \"\n f\"Progress: {percent_complete}%, \"\n f\"Elapsed Time: {elapsed_time}, \"\n f\"Loss: {loss}, \"\n f\"Update: {current_update}\"\n )\n yield message, gr.update(interactive=False), gr.update(interactive=True)\n elif output.strip():\n yield output, gr.update(interactive=False), gr.update(interactive=True)\n except queue.Empty:\n pass\n\n # Handle stderr\n try:\n while True:\n error_output = stderr_queue.get_nowait()\n print(error_output, end=\"\")\n if error_output.strip():\n yield f\"{error_output.strip()}\", gr.update(interactive=False), gr.update(interactive=True)\n except queue.Empty:\n pass\n\n if process_status is not None and stdout_queue.empty() and stderr_queue.empty():\n if process_status != 0:\n yield (\n f\"Process crashed with exit code {process_status}!\",\n gr.update(interactive=False),\n gr.update(interactive=True),\n )\n else:\n yield (\n \"Training complete or paused ...\",\n gr.update(interactive=False),\n gr.update(interactive=True),\n )\n break\n\n # Small sleep to prevent CPU thrashing\n time.sleep(0.1)\n\n # Clean up\n training_process.stdout.close()\n training_process.stderr.close()\n training_process.wait()\n\n time.sleep(1)\n\n if training_process is None:\n text_info = \"Train stopped !\"\n else:\n text_info = \"Train complete at end !\"\n\n except Exception as e: # Catch all exceptions\n # Ensure that we reset the training process variable in case of an error\n text_info = f\"An error occurred: {str(e)}\"\n\n training_process = None\n\n yield text_info, gr.update(interactive=True), gr.update(interactive=False)\n\n\ndef stop_training():\n global training_process, stop_signal\n\n if training_process is None:\n return \"Train not running !\", gr.update(interactive=True), gr.update(interactive=False)\n terminate_process_tree(training_process.pid)\n # training_process = None\n stop_signal = True\n return \"Train stopped !\", gr.update(interactive=True), gr.update(interactive=False)\n\n\ndef get_list_projects():\n project_list = []\n for folder in os.listdir(path_data):\n path_folder = os.path.join(path_data, folder)\n if not os.path.isdir(path_folder):\n continue\n folder = folder.lower()","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.stop_training","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.stop_training#L584-L592","kind":"function","name":"stop_training","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":584,"end_line":592,"context_start_line":564,"context_end_line":612,"code":" training_process.stdout.close()\n training_process.stderr.close()\n training_process.wait()\n\n time.sleep(1)\n\n if training_process is None:\n text_info = \"Train stopped !\"\n else:\n text_info = \"Train complete at end !\"\n\n except Exception as e: # Catch all exceptions\n # Ensure that we reset the training process variable in case of an error\n text_info = f\"An error occurred: {str(e)}\"\n\n training_process = None\n\n yield text_info, gr.update(interactive=True), gr.update(interactive=False)\n\n\ndef stop_training():\n global training_process, stop_signal\n\n if training_process is None:\n return \"Train not running !\", gr.update(interactive=True), gr.update(interactive=False)\n terminate_process_tree(training_process.pid)\n # training_process = None\n stop_signal = True\n return \"Train stopped !\", gr.update(interactive=True), gr.update(interactive=False)\n\n\ndef get_list_projects():\n project_list = []\n for folder in os.listdir(path_data):\n path_folder = os.path.join(path_data, folder)\n if not os.path.isdir(path_folder):\n continue\n folder = folder.lower()\n if folder == \"emilia_zh_en_pinyin\":\n continue\n project_list.append(folder)\n\n projects_selelect = None if not project_list else project_list[-1]\n\n return project_list, projects_selelect\n\n\ndef create_data_project(name, tokenizer_type):\n name += \"_\" + tokenizer_type","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.get_list_projects","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.get_list_projects#L595-L608","kind":"function","name":"get_list_projects","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":595,"end_line":608,"context_start_line":575,"context_end_line":628,"code":" except Exception as e: # Catch all exceptions\n # Ensure that we reset the training process variable in case of an error\n text_info = f\"An error occurred: {str(e)}\"\n\n training_process = None\n\n yield text_info, gr.update(interactive=True), gr.update(interactive=False)\n\n\ndef stop_training():\n global training_process, stop_signal\n\n if training_process is None:\n return \"Train not running !\", gr.update(interactive=True), gr.update(interactive=False)\n terminate_process_tree(training_process.pid)\n # training_process = None\n stop_signal = True\n return \"Train stopped !\", gr.update(interactive=True), gr.update(interactive=False)\n\n\ndef get_list_projects():\n project_list = []\n for folder in os.listdir(path_data):\n path_folder = os.path.join(path_data, folder)\n if not os.path.isdir(path_folder):\n continue\n folder = folder.lower()\n if folder == \"emilia_zh_en_pinyin\":\n continue\n project_list.append(folder)\n\n projects_selelect = None if not project_list else project_list[-1]\n\n return project_list, projects_selelect\n\n\ndef create_data_project(name, tokenizer_type):\n name += \"_\" + tokenizer_type\n os.makedirs(os.path.join(path_data, name), exist_ok=True)\n os.makedirs(os.path.join(path_data, name, \"dataset\"), exist_ok=True)\n project_list, projects_selelect = get_list_projects()\n return gr.update(choices=project_list, value=name)\n\n\ndef transcribe_all(name_project, audio_files, language, user=False, progress=gr.Progress()):\n path_project = os.path.join(path_data, name_project)\n path_dataset = os.path.join(path_project, \"dataset\")\n path_project_wavs = os.path.join(path_project, \"wavs\")\n file_metadata = os.path.join(path_project, \"metadata.csv\")\n\n if not user:\n if audio_files is None:\n return \"You need to load an audio file.\"\n","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.create_data_project","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.create_data_project#L611-L616","kind":"function","name":"create_data_project","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":611,"end_line":616,"context_start_line":591,"context_end_line":636,"code":" stop_signal = True\n return \"Train stopped !\", gr.update(interactive=True), gr.update(interactive=False)\n\n\ndef get_list_projects():\n project_list = []\n for folder in os.listdir(path_data):\n path_folder = os.path.join(path_data, folder)\n if not os.path.isdir(path_folder):\n continue\n folder = folder.lower()\n if folder == \"emilia_zh_en_pinyin\":\n continue\n project_list.append(folder)\n\n projects_selelect = None if not project_list else project_list[-1]\n\n return project_list, projects_selelect\n\n\ndef create_data_project(name, tokenizer_type):\n name += \"_\" + tokenizer_type\n os.makedirs(os.path.join(path_data, name), exist_ok=True)\n os.makedirs(os.path.join(path_data, name, \"dataset\"), exist_ok=True)\n project_list, projects_selelect = get_list_projects()\n return gr.update(choices=project_list, value=name)\n\n\ndef transcribe_all(name_project, audio_files, language, user=False, progress=gr.Progress()):\n path_project = os.path.join(path_data, name_project)\n path_dataset = os.path.join(path_project, \"dataset\")\n path_project_wavs = os.path.join(path_project, \"wavs\")\n file_metadata = os.path.join(path_project, \"metadata.csv\")\n\n if not user:\n if audio_files is None:\n return \"You need to load an audio file.\"\n\n if os.path.isdir(path_project_wavs):\n shutil.rmtree(path_project_wavs)\n\n if os.path.isfile(file_metadata):\n os.remove(file_metadata)\n\n os.makedirs(path_project_wavs, exist_ok=True)\n","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.transcribe_all","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.transcribe_all#L619-L687","kind":"function","name":"transcribe_all","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":619,"end_line":687,"context_start_line":599,"context_end_line":707,"code":" if not os.path.isdir(path_folder):\n continue\n folder = folder.lower()\n if folder == \"emilia_zh_en_pinyin\":\n continue\n project_list.append(folder)\n\n projects_selelect = None if not project_list else project_list[-1]\n\n return project_list, projects_selelect\n\n\ndef create_data_project(name, tokenizer_type):\n name += \"_\" + tokenizer_type\n os.makedirs(os.path.join(path_data, name), exist_ok=True)\n os.makedirs(os.path.join(path_data, name, \"dataset\"), exist_ok=True)\n project_list, projects_selelect = get_list_projects()\n return gr.update(choices=project_list, value=name)\n\n\ndef transcribe_all(name_project, audio_files, language, user=False, progress=gr.Progress()):\n path_project = os.path.join(path_data, name_project)\n path_dataset = os.path.join(path_project, \"dataset\")\n path_project_wavs = os.path.join(path_project, \"wavs\")\n file_metadata = os.path.join(path_project, \"metadata.csv\")\n\n if not user:\n if audio_files is None:\n return \"You need to load an audio file.\"\n\n if os.path.isdir(path_project_wavs):\n shutil.rmtree(path_project_wavs)\n\n if os.path.isfile(file_metadata):\n os.remove(file_metadata)\n\n os.makedirs(path_project_wavs, exist_ok=True)\n\n if user:\n file_audios = [\n file\n for format in (\"*.wav\", \"*.ogg\", \"*.opus\", \"*.mp3\", \"*.flac\")\n for file in glob(os.path.join(path_dataset, format))\n ]\n if file_audios == []:\n return \"No audio file was found in the dataset.\"\n else:\n file_audios = audio_files\n\n alpha = 0.5\n _max = 1.0\n slicer = Slicer(24000)\n\n num = 0\n error_num = 0\n data = \"\"\n for file_audio in progress.tqdm(file_audios, desc=\"transcribe files\", total=len((file_audios))):\n audio, _ = librosa.load(file_audio, sr=24000, mono=True)\n\n list_slicer = slicer.slice(audio)\n for chunk, start, end in progress.tqdm(list_slicer, total=len(list_slicer), desc=\"slicer files\"):\n name_segment = os.path.join(f\"segment_{num}\")\n file_segment = os.path.join(path_project_wavs, f\"{name_segment}.wav\")\n\n tmp_max = np.abs(chunk).max()\n if tmp_max > 1:\n chunk /= tmp_max\n chunk = (chunk / tmp_max * (_max * alpha)) + (1 - alpha) * chunk\n wavfile.write(file_segment, 24000, (chunk * 32767).astype(np.int16))\n\n try:\n text = transcribe(file_segment, language)\n text = text.strip()\n\n data += f\"{name_segment}|{text}\\n\"\n\n num += 1\n except: # noqa: E722\n error_num += 1\n\n with open(file_metadata, \"w\", encoding=\"utf-8-sig\") as f:\n f.write(data)\n\n if error_num != []:\n error_text = f\"\\nerror files : {error_num}\"\n else:\n error_text = \"\"\n\n return f\"transcribe complete samples : {num}\\npath : {path_project_wavs}{error_text}\"\n\n\ndef format_seconds_to_hms(seconds):\n hours = int(seconds / 3600)\n minutes = int((seconds % 3600) / 60)\n seconds = seconds % 60\n return \"{:02d}:{:02d}:{:02d}\".format(hours, minutes, int(seconds))\n\n\ndef get_correct_audio_path(\n audio_input,\n base_path=\"wavs\",\n supported_formats=(\"wav\", \"mp3\", \"aac\", \"flac\", \"m4a\", \"alac\", \"ogg\", \"aiff\", \"wma\", \"amr\"),\n):\n file_audio = None\n\n # Helper function to check if file has a supported extension\n def has_supported_extension(file_name):\n return any(file_name.endswith(f\".{ext}\") for ext in supported_formats)\n","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.format_seconds_to_hms","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.format_seconds_to_hms#L690-L694","kind":"function","name":"format_seconds_to_hms","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":690,"end_line":694,"context_start_line":670,"context_end_line":714,"code":" text = transcribe(file_segment, language)\n text = text.strip()\n\n data += f\"{name_segment}|{text}\\n\"\n\n num += 1\n except: # noqa: E722\n error_num += 1\n\n with open(file_metadata, \"w\", encoding=\"utf-8-sig\") as f:\n f.write(data)\n\n if error_num != []:\n error_text = f\"\\nerror files : {error_num}\"\n else:\n error_text = \"\"\n\n return f\"transcribe complete samples : {num}\\npath : {path_project_wavs}{error_text}\"\n\n\ndef format_seconds_to_hms(seconds):\n hours = int(seconds / 3600)\n minutes = int((seconds % 3600) / 60)\n seconds = seconds % 60\n return \"{:02d}:{:02d}:{:02d}\".format(hours, minutes, int(seconds))\n\n\ndef get_correct_audio_path(\n audio_input,\n base_path=\"wavs\",\n supported_formats=(\"wav\", \"mp3\", \"aac\", \"flac\", \"m4a\", \"alac\", \"ogg\", \"aiff\", \"wma\", \"amr\"),\n):\n file_audio = None\n\n # Helper function to check if file has a supported extension\n def has_supported_extension(file_name):\n return any(file_name.endswith(f\".{ext}\") for ext in supported_formats)\n\n # Case 1: If it's a full path with a valid extension, use it directly\n if os.path.isabs(audio_input) and has_supported_extension(audio_input):\n file_audio = audio_input\n\n # Case 2: If it has a supported extension but is not a full path\n elif has_supported_extension(audio_input) and not os.path.isabs(audio_input):\n file_audio = os.path.join(base_path, audio_input)","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.get_correct_audio_path","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.get_correct_audio_path#L697-L725","kind":"function","name":"get_correct_audio_path","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":697,"end_line":725,"context_start_line":677,"context_end_line":745,"code":" error_num += 1\n\n with open(file_metadata, \"w\", encoding=\"utf-8-sig\") as f:\n f.write(data)\n\n if error_num != []:\n error_text = f\"\\nerror files : {error_num}\"\n else:\n error_text = \"\"\n\n return f\"transcribe complete samples : {num}\\npath : {path_project_wavs}{error_text}\"\n\n\ndef format_seconds_to_hms(seconds):\n hours = int(seconds / 3600)\n minutes = int((seconds % 3600) / 60)\n seconds = seconds % 60\n return \"{:02d}:{:02d}:{:02d}\".format(hours, minutes, int(seconds))\n\n\ndef get_correct_audio_path(\n audio_input,\n base_path=\"wavs\",\n supported_formats=(\"wav\", \"mp3\", \"aac\", \"flac\", \"m4a\", \"alac\", \"ogg\", \"aiff\", \"wma\", \"amr\"),\n):\n file_audio = None\n\n # Helper function to check if file has a supported extension\n def has_supported_extension(file_name):\n return any(file_name.endswith(f\".{ext}\") for ext in supported_formats)\n\n # Case 1: If it's a full path with a valid extension, use it directly\n if os.path.isabs(audio_input) and has_supported_extension(audio_input):\n file_audio = audio_input\n\n # Case 2: If it has a supported extension but is not a full path\n elif has_supported_extension(audio_input) and not os.path.isabs(audio_input):\n file_audio = os.path.join(base_path, audio_input)\n\n # Case 3: If only the name is given (no extension and not a full path)\n elif not has_supported_extension(audio_input) and not os.path.isabs(audio_input):\n for ext in supported_formats:\n potential_file = os.path.join(base_path, f\"{audio_input}.{ext}\")\n if os.path.exists(potential_file):\n file_audio = potential_file\n break\n else:\n file_audio = os.path.join(base_path, f\"{audio_input}.{supported_formats[0]}\")\n return file_audio\n\n\ndef create_metadata(name_project, ch_tokenizer, progress=gr.Progress()):\n path_project = os.path.join(path_data, name_project)\n path_project_wavs = os.path.join(path_project, \"wavs\")\n file_metadata = os.path.join(path_project, \"metadata.csv\")\n file_raw = os.path.join(path_project, \"raw.arrow\")\n file_duration = os.path.join(path_project, \"duration.json\")\n file_vocab = os.path.join(path_project, \"vocab.txt\")\n\n if not os.path.isfile(file_metadata):\n return \"The file was not found in \" + file_metadata, \"\"\n\n with open(file_metadata, \"r\", encoding=\"utf-8-sig\") as f:\n data = f.read()\n\n audio_path_list = []\n text_list = []\n duration_list = []\n","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.create_metadata","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.create_metadata#L728-L835","kind":"function","name":"create_metadata","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":728,"end_line":835,"context_start_line":708,"context_end_line":855,"code":" # Case 1: If it's a full path with a valid extension, use it directly\n if os.path.isabs(audio_input) and has_supported_extension(audio_input):\n file_audio = audio_input\n\n # Case 2: If it has a supported extension but is not a full path\n elif has_supported_extension(audio_input) and not os.path.isabs(audio_input):\n file_audio = os.path.join(base_path, audio_input)\n\n # Case 3: If only the name is given (no extension and not a full path)\n elif not has_supported_extension(audio_input) and not os.path.isabs(audio_input):\n for ext in supported_formats:\n potential_file = os.path.join(base_path, f\"{audio_input}.{ext}\")\n if os.path.exists(potential_file):\n file_audio = potential_file\n break\n else:\n file_audio = os.path.join(base_path, f\"{audio_input}.{supported_formats[0]}\")\n return file_audio\n\n\ndef create_metadata(name_project, ch_tokenizer, progress=gr.Progress()):\n path_project = os.path.join(path_data, name_project)\n path_project_wavs = os.path.join(path_project, \"wavs\")\n file_metadata = os.path.join(path_project, \"metadata.csv\")\n file_raw = os.path.join(path_project, \"raw.arrow\")\n file_duration = os.path.join(path_project, \"duration.json\")\n file_vocab = os.path.join(path_project, \"vocab.txt\")\n\n if not os.path.isfile(file_metadata):\n return \"The file was not found in \" + file_metadata, \"\"\n\n with open(file_metadata, \"r\", encoding=\"utf-8-sig\") as f:\n data = f.read()\n\n audio_path_list = []\n text_list = []\n duration_list = []\n\n count = data.split(\"\\n\")\n lenght = 0\n result = []\n error_files = []\n text_vocab_set = set()\n for line in progress.tqdm(data.split(\"\\n\"), total=count):\n sp_line = line.split(\"|\")\n if len(sp_line) != 2:\n continue\n name_audio, text = sp_line[:2]\n\n file_audio = get_correct_audio_path(name_audio, path_project_wavs)\n\n if not os.path.isfile(file_audio):\n error_files.append([file_audio, \"error path\"])\n continue\n\n try:\n duration = get_audio_duration(file_audio)\n except Exception as e:\n error_files.append([file_audio, \"duration\"])\n print(f\"Error processing {file_audio}: {e}\")\n continue\n\n if duration < 1 or duration > 30:\n if duration > 30:\n error_files.append([file_audio, \"duration > 30 sec\"])\n if duration < 1:\n error_files.append([file_audio, \"duration < 1 sec \"])\n continue\n if len(text) < 3:\n error_files.append([file_audio, \"very short text length 3\"])\n continue\n\n text = text.strip()\n text = convert_char_to_pinyin([text], polyphone=True)[0]\n\n audio_path_list.append(file_audio)\n duration_list.append(duration)\n text_list.append(text)\n\n result.append({\"audio_path\": file_audio, \"text\": text, \"duration\": duration})\n if ch_tokenizer:\n text_vocab_set.update(list(text))\n\n lenght += duration\n\n if duration_list == []:\n return f\"Error: No audio files found in the specified path : {path_project_wavs}\", \"\"\n\n min_second = round(min(duration_list), 2)\n max_second = round(max(duration_list), 2)\n\n with ArrowWriter(path=file_raw, writer_batch_size=1) as writer:\n for line in progress.tqdm(result, total=len(result), desc=\"prepare data\"):\n writer.write(line)\n\n with open(file_duration, \"w\") as f:\n json.dump({\"duration\": duration_list}, f, ensure_ascii=False)\n\n new_vocal = \"\"\n if not ch_tokenizer:\n if not os.path.isfile(file_vocab):\n file_vocab_finetune = os.path.join(path_data, \"Emilia_ZH_EN_pinyin/vocab.txt\")\n if not os.path.isfile(file_vocab_finetune):\n return \"Error: Vocabulary file 'Emilia_ZH_EN_pinyin' not found!\", \"\"\n shutil.copy2(file_vocab_finetune, file_vocab)\n\n with open(file_vocab, \"r\", encoding=\"utf-8-sig\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n\n else:\n with open(file_vocab, \"w\", encoding=\"utf-8-sig\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n new_vocal += vocab + \"\\n\"\n vocab_size = len(text_vocab_set)\n\n if error_files != []:\n error_text = \"\\n\".join([\" = \".join(item) for item in error_files])\n else:\n error_text = \"\"\n\n return (\n f\"prepare complete \\nsamples : {len(text_list)}\\ntime data : {format_seconds_to_hms(lenght)}\\nmin sec : {min_second}\\nmax sec : {max_second}\\nfile_arrow : {file_raw}\\nvocab : {vocab_size}\\n{error_text}\",\n new_vocal,\n )\n\n\ndef check_user(value):\n return gr.update(visible=not value), gr.update(visible=value)\n\n\ndef calculate_train(\n name_project,\n epochs,\n learning_rate,\n batch_size_per_gpu,\n batch_size_type,\n max_samples,\n num_warmup_updates,\n finetune,\n):\n path_project = os.path.join(path_data, name_project)\n file_duration = os.path.join(path_project, \"duration.json\")\n\n hop_length = 256","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.check_user","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.check_user#L838-L839","kind":"function","name":"check_user","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":838,"end_line":839,"context_start_line":818,"context_end_line":859,"code":" vocab_size = len(vocab_char_map)\n\n else:\n with open(file_vocab, \"w\", encoding=\"utf-8-sig\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n new_vocal += vocab + \"\\n\"\n vocab_size = len(text_vocab_set)\n\n if error_files != []:\n error_text = \"\\n\".join([\" = \".join(item) for item in error_files])\n else:\n error_text = \"\"\n\n return (\n f\"prepare complete \\nsamples : {len(text_list)}\\ntime data : {format_seconds_to_hms(lenght)}\\nmin sec : {min_second}\\nmax sec : {max_second}\\nfile_arrow : {file_raw}\\nvocab : {vocab_size}\\n{error_text}\",\n new_vocal,\n )\n\n\ndef check_user(value):\n return gr.update(visible=not value), gr.update(visible=value)\n\n\ndef calculate_train(\n name_project,\n epochs,\n learning_rate,\n batch_size_per_gpu,\n batch_size_type,\n max_samples,\n num_warmup_updates,\n finetune,\n):\n path_project = os.path.join(path_data, name_project)\n file_duration = os.path.join(path_project, \"duration.json\")\n\n hop_length = 256\n sampling_rate = 24000\n\n if not os.path.isfile(file_duration):\n return (","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.calculate_train","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.calculate_train#L842-L928","kind":"function","name":"calculate_train","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":842,"end_line":928,"context_start_line":822,"context_end_line":948,"code":" for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n new_vocal += vocab + \"\\n\"\n vocab_size = len(text_vocab_set)\n\n if error_files != []:\n error_text = \"\\n\".join([\" = \".join(item) for item in error_files])\n else:\n error_text = \"\"\n\n return (\n f\"prepare complete \\nsamples : {len(text_list)}\\ntime data : {format_seconds_to_hms(lenght)}\\nmin sec : {min_second}\\nmax sec : {max_second}\\nfile_arrow : {file_raw}\\nvocab : {vocab_size}\\n{error_text}\",\n new_vocal,\n )\n\n\ndef check_user(value):\n return gr.update(visible=not value), gr.update(visible=value)\n\n\ndef calculate_train(\n name_project,\n epochs,\n learning_rate,\n batch_size_per_gpu,\n batch_size_type,\n max_samples,\n num_warmup_updates,\n finetune,\n):\n path_project = os.path.join(path_data, name_project)\n file_duration = os.path.join(path_project, \"duration.json\")\n\n hop_length = 256\n sampling_rate = 24000\n\n if not os.path.isfile(file_duration):\n return (\n epochs,\n learning_rate,\n batch_size_per_gpu,\n max_samples,\n num_warmup_updates,\n \"project not found !\",\n )\n\n with open(file_duration, \"r\") as file:\n data = json.load(file)\n\n duration_list = data[\"duration\"]\n max_sample_length = max(duration_list) * sampling_rate / hop_length\n total_samples = len(duration_list)\n total_duration = sum(duration_list)\n\n if torch.cuda.is_available():\n gpu_count = torch.cuda.device_count()\n total_memory = 0\n for i in range(gpu_count):\n gpu_properties = torch.cuda.get_device_properties(i)\n total_memory += gpu_properties.total_memory / (1024**3) # in GB\n elif torch.xpu.is_available():\n gpu_count = torch.xpu.device_count()\n total_memory = 0\n for i in range(gpu_count):\n gpu_properties = torch.xpu.get_device_properties(i)\n total_memory += gpu_properties.total_memory / (1024**3)\n elif torch.backends.mps.is_available():\n gpu_count = 1\n total_memory = psutil.virtual_memory().available / (1024**3)\n\n avg_gpu_memory = total_memory / gpu_count\n\n # rough estimate of batch size\n if batch_size_type == \"frame\":\n batch_size_per_gpu = max(int(38400 * (avg_gpu_memory - 5) / 75), int(max_sample_length))\n elif batch_size_type == \"sample\":\n batch_size_per_gpu = int(200 / (total_duration / total_samples))\n\n if total_samples < 64:\n max_samples = int(total_samples * 0.25)\n\n num_warmup_updates = max(num_warmup_updates, int(total_samples * 0.05))\n\n # take 1.2M updates as the maximum\n max_updates = 1200000\n\n if batch_size_type == \"frame\":\n mini_batch_duration = batch_size_per_gpu * gpu_count * hop_length / sampling_rate\n updates_per_epoch = total_duration / mini_batch_duration\n elif batch_size_type == \"sample\":\n updates_per_epoch = total_samples / batch_size_per_gpu / gpu_count\n\n epochs = int(max_updates / updates_per_epoch)\n\n if finetune:\n learning_rate = 1e-5\n else:\n learning_rate = 7.5e-5\n\n return (\n epochs,\n learning_rate,\n batch_size_per_gpu,\n max_samples,\n num_warmup_updates,\n total_samples,\n )\n\n\ndef prune_checkpoint(checkpoint_path: str, new_checkpoint_path: str, save_ema: bool, safetensors: bool) -> str:\n try:\n checkpoint = torch.load(checkpoint_path, weights_only=True)\n print(\"Original Checkpoint Keys:\", checkpoint.keys())\n\n to_retain = \"ema_model_state_dict\" if save_ema else \"model_state_dict\"\n try:\n model_state_dict_to_retain = checkpoint[to_retain]\n except KeyError:\n return f\"{to_retain} not found in the checkpoint.\"\n\n if safetensors:\n new_checkpoint_path = new_checkpoint_path.replace(\".pt\", \".safetensors\")\n save_file(model_state_dict_to_retain, new_checkpoint_path)\n else:\n new_checkpoint_path = new_checkpoint_path.replace(\".safetensors\", \".pt\")\n new_checkpoint = {\"ema_model_state_dict\": model_state_dict_to_retain}\n torch.save(new_checkpoint, new_checkpoint_path)","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.prune_checkpoint","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.prune_checkpoint#L931-L953","kind":"function","name":"prune_checkpoint","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":931,"end_line":953,"context_start_line":911,"context_end_line":973,"code":" elif batch_size_type == \"sample\":\n updates_per_epoch = total_samples / batch_size_per_gpu / gpu_count\n\n epochs = int(max_updates / updates_per_epoch)\n\n if finetune:\n learning_rate = 1e-5\n else:\n learning_rate = 7.5e-5\n\n return (\n epochs,\n learning_rate,\n batch_size_per_gpu,\n max_samples,\n num_warmup_updates,\n total_samples,\n )\n\n\ndef prune_checkpoint(checkpoint_path: str, new_checkpoint_path: str, save_ema: bool, safetensors: bool) -> str:\n try:\n checkpoint = torch.load(checkpoint_path, weights_only=True)\n print(\"Original Checkpoint Keys:\", checkpoint.keys())\n\n to_retain = \"ema_model_state_dict\" if save_ema else \"model_state_dict\"\n try:\n model_state_dict_to_retain = checkpoint[to_retain]\n except KeyError:\n return f\"{to_retain} not found in the checkpoint.\"\n\n if safetensors:\n new_checkpoint_path = new_checkpoint_path.replace(\".pt\", \".safetensors\")\n save_file(model_state_dict_to_retain, new_checkpoint_path)\n else:\n new_checkpoint_path = new_checkpoint_path.replace(\".safetensors\", \".pt\")\n new_checkpoint = {\"ema_model_state_dict\": model_state_dict_to_retain}\n torch.save(new_checkpoint, new_checkpoint_path)\n\n return f\"New checkpoint saved at: {new_checkpoint_path}\"\n\n except Exception as e:\n return f\"An error occurred: {e}\"\n\n\ndef expand_model_embeddings(ckpt_path, new_ckpt_path, num_new_tokens=42):\n seed = 666\n random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.cuda.manual_seed_all(seed)\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n\n if ckpt_path.endswith(\".safetensors\"):\n ckpt = load_file(ckpt_path, device=\"cpu\")\n ckpt = {\"ema_model_state_dict\": ckpt}\n elif ckpt_path.endswith(\".pt\"):\n ckpt = torch.load(ckpt_path, map_location=\"cpu\")\n\n ema_sd = ckpt.get(\"ema_model_state_dict\", {})\n embed_key_ema = \"ema_model.transformer.text_embed.text_embed.weight\"","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.expand_model_embeddings","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.expand_model_embeddings#L956-L993","kind":"function","name":"expand_model_embeddings","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":956,"end_line":993,"context_start_line":936,"context_end_line":1013,"code":" to_retain = \"ema_model_state_dict\" if save_ema else \"model_state_dict\"\n try:\n model_state_dict_to_retain = checkpoint[to_retain]\n except KeyError:\n return f\"{to_retain} not found in the checkpoint.\"\n\n if safetensors:\n new_checkpoint_path = new_checkpoint_path.replace(\".pt\", \".safetensors\")\n save_file(model_state_dict_to_retain, new_checkpoint_path)\n else:\n new_checkpoint_path = new_checkpoint_path.replace(\".safetensors\", \".pt\")\n new_checkpoint = {\"ema_model_state_dict\": model_state_dict_to_retain}\n torch.save(new_checkpoint, new_checkpoint_path)\n\n return f\"New checkpoint saved at: {new_checkpoint_path}\"\n\n except Exception as e:\n return f\"An error occurred: {e}\"\n\n\ndef expand_model_embeddings(ckpt_path, new_ckpt_path, num_new_tokens=42):\n seed = 666\n random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.cuda.manual_seed_all(seed)\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n\n if ckpt_path.endswith(\".safetensors\"):\n ckpt = load_file(ckpt_path, device=\"cpu\")\n ckpt = {\"ema_model_state_dict\": ckpt}\n elif ckpt_path.endswith(\".pt\"):\n ckpt = torch.load(ckpt_path, map_location=\"cpu\")\n\n ema_sd = ckpt.get(\"ema_model_state_dict\", {})\n embed_key_ema = \"ema_model.transformer.text_embed.text_embed.weight\"\n old_embed_ema = ema_sd[embed_key_ema]\n\n vocab_old = old_embed_ema.size(0)\n embed_dim = old_embed_ema.size(1)\n vocab_new = vocab_old + num_new_tokens\n\n def expand_embeddings(old_embeddings):\n new_embeddings = torch.zeros((vocab_new, embed_dim))\n new_embeddings[:vocab_old] = old_embeddings\n new_embeddings[vocab_old:] = torch.randn((num_new_tokens, embed_dim))\n return new_embeddings\n\n ema_sd[embed_key_ema] = expand_embeddings(ema_sd[embed_key_ema])\n\n if new_ckpt_path.endswith(\".safetensors\"):\n save_file(ema_sd, new_ckpt_path)\n elif new_ckpt_path.endswith(\".pt\"):\n torch.save(ckpt, new_ckpt_path)\n\n return vocab_new\n\n\ndef vocab_count(text):\n return str(len(text.split(\",\")))\n\n\ndef vocab_extend(project_name, symbols, model_type):\n if symbols == \"\":\n return \"Symbols empty!\"\n\n name_project = project_name\n path_project = os.path.join(path_data, name_project)\n file_vocab_project = os.path.join(path_project, \"vocab.txt\")\n\n file_vocab = os.path.join(path_data, \"Emilia_ZH_EN_pinyin/vocab.txt\")\n if not os.path.isfile(file_vocab):\n return f\"the file {file_vocab} not found !\"\n\n symbols = symbols.split(\",\")\n if symbols == []:","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.vocab_count","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.vocab_count#L996-L997","kind":"function","name":"vocab_count","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":996,"end_line":997,"context_start_line":976,"context_end_line":1017,"code":" vocab_old = old_embed_ema.size(0)\n embed_dim = old_embed_ema.size(1)\n vocab_new = vocab_old + num_new_tokens\n\n def expand_embeddings(old_embeddings):\n new_embeddings = torch.zeros((vocab_new, embed_dim))\n new_embeddings[:vocab_old] = old_embeddings\n new_embeddings[vocab_old:] = torch.randn((num_new_tokens, embed_dim))\n return new_embeddings\n\n ema_sd[embed_key_ema] = expand_embeddings(ema_sd[embed_key_ema])\n\n if new_ckpt_path.endswith(\".safetensors\"):\n save_file(ema_sd, new_ckpt_path)\n elif new_ckpt_path.endswith(\".pt\"):\n torch.save(ckpt, new_ckpt_path)\n\n return vocab_new\n\n\ndef vocab_count(text):\n return str(len(text.split(\",\")))\n\n\ndef vocab_extend(project_name, symbols, model_type):\n if symbols == \"\":\n return \"Symbols empty!\"\n\n name_project = project_name\n path_project = os.path.join(path_data, name_project)\n file_vocab_project = os.path.join(path_project, \"vocab.txt\")\n\n file_vocab = os.path.join(path_data, \"Emilia_ZH_EN_pinyin/vocab.txt\")\n if not os.path.isfile(file_vocab):\n return f\"the file {file_vocab} not found !\"\n\n symbols = symbols.split(\",\")\n if symbols == []:\n return \"Symbols to extend not found.\"\n\n with open(file_vocab, \"r\", encoding=\"utf-8-sig\") as f:\n data = f.read()","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.vocab_extend","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.vocab_extend#L1000-L1060","kind":"function","name":"vocab_extend","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1000,"end_line":1060,"context_start_line":980,"context_end_line":1080,"code":" def expand_embeddings(old_embeddings):\n new_embeddings = torch.zeros((vocab_new, embed_dim))\n new_embeddings[:vocab_old] = old_embeddings\n new_embeddings[vocab_old:] = torch.randn((num_new_tokens, embed_dim))\n return new_embeddings\n\n ema_sd[embed_key_ema] = expand_embeddings(ema_sd[embed_key_ema])\n\n if new_ckpt_path.endswith(\".safetensors\"):\n save_file(ema_sd, new_ckpt_path)\n elif new_ckpt_path.endswith(\".pt\"):\n torch.save(ckpt, new_ckpt_path)\n\n return vocab_new\n\n\ndef vocab_count(text):\n return str(len(text.split(\",\")))\n\n\ndef vocab_extend(project_name, symbols, model_type):\n if symbols == \"\":\n return \"Symbols empty!\"\n\n name_project = project_name\n path_project = os.path.join(path_data, name_project)\n file_vocab_project = os.path.join(path_project, \"vocab.txt\")\n\n file_vocab = os.path.join(path_data, \"Emilia_ZH_EN_pinyin/vocab.txt\")\n if not os.path.isfile(file_vocab):\n return f\"the file {file_vocab} not found !\"\n\n symbols = symbols.split(\",\")\n if symbols == []:\n return \"Symbols to extend not found.\"\n\n with open(file_vocab, \"r\", encoding=\"utf-8-sig\") as f:\n data = f.read()\n vocab = data.split(\"\\n\")\n vocab_check = set(vocab)\n\n miss_symbols = []\n for item in symbols:\n item = item.replace(\" \", \"\")\n if item in vocab_check:\n continue\n miss_symbols.append(item)\n\n if miss_symbols == []:\n return \"Symbols are okay no need to extend.\"\n\n size_vocab = len(vocab)\n vocab.pop()\n for item in miss_symbols:\n vocab.append(item)\n\n vocab.append(\"\")\n\n with open(file_vocab_project, \"w\", encoding=\"utf-8\") as f:\n f.write(\"\\n\".join(vocab))\n\n if model_type == \"F5TTS_v1_Base\":\n ckpt_path = str(cached_path(\"hf://SWivid/F5-TTS/F5TTS_v1_Base/model_1250000.safetensors\"))\n elif model_type == \"F5TTS_Base\":\n ckpt_path = str(cached_path(\"hf://SWivid/F5-TTS/F5TTS_Base/model_1200000.pt\"))\n elif model_type == \"E2TTS_Base\":\n ckpt_path = str(cached_path(\"hf://SWivid/E2-TTS/E2TTS_Base/model_1200000.pt\"))\n\n vocab_size_new = len(miss_symbols)\n\n dataset_name = name_project.replace(\"_pinyin\", \"\").replace(\"_char\", \"\")\n new_ckpt_path = os.path.join(path_project_ckpts, dataset_name)\n os.makedirs(new_ckpt_path, exist_ok=True)\n\n # Add pretrained_ prefix to model when copying for consistency with finetune_cli.py\n new_ckpt_file = os.path.join(new_ckpt_path, \"pretrained_\" + os.path.basename(ckpt_path))\n\n size = expand_model_embeddings(ckpt_path, new_ckpt_file, num_new_tokens=vocab_size_new)\n\n vocab_new = \"\\n\".join(miss_symbols)\n return f\"vocab old size : {size_vocab}\\nvocab new size : {size}\\nvocab add : {vocab_size_new}\\nnew symbols :\\n{vocab_new}\"\n\n\ndef vocab_check(project_name, tokenizer_type):\n name_project = project_name\n path_project = os.path.join(path_data, name_project)\n\n file_metadata = os.path.join(path_project, \"metadata.csv\")\n\n file_vocab = os.path.join(path_data, \"Emilia_ZH_EN_pinyin/vocab.txt\")\n if not os.path.isfile(file_vocab):\n return f\"the file {file_vocab} not found !\", \"\"\n\n with open(file_vocab, \"r\", encoding=\"utf-8-sig\") as f:\n data = f.read()\n vocab = data.split(\"\\n\")\n vocab = set(vocab)\n\n if not os.path.isfile(file_metadata):\n return f\"the file {file_metadata} not found !\", \"\"\n","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.vocab_check","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.vocab_check#L1063-L1107","kind":"function","name":"vocab_check","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1063,"end_line":1107,"context_start_line":1043,"context_end_line":1127,"code":" elif model_type == \"F5TTS_Base\":\n ckpt_path = str(cached_path(\"hf://SWivid/F5-TTS/F5TTS_Base/model_1200000.pt\"))\n elif model_type == \"E2TTS_Base\":\n ckpt_path = str(cached_path(\"hf://SWivid/E2-TTS/E2TTS_Base/model_1200000.pt\"))\n\n vocab_size_new = len(miss_symbols)\n\n dataset_name = name_project.replace(\"_pinyin\", \"\").replace(\"_char\", \"\")\n new_ckpt_path = os.path.join(path_project_ckpts, dataset_name)\n os.makedirs(new_ckpt_path, exist_ok=True)\n\n # Add pretrained_ prefix to model when copying for consistency with finetune_cli.py\n new_ckpt_file = os.path.join(new_ckpt_path, \"pretrained_\" + os.path.basename(ckpt_path))\n\n size = expand_model_embeddings(ckpt_path, new_ckpt_file, num_new_tokens=vocab_size_new)\n\n vocab_new = \"\\n\".join(miss_symbols)\n return f\"vocab old size : {size_vocab}\\nvocab new size : {size}\\nvocab add : {vocab_size_new}\\nnew symbols :\\n{vocab_new}\"\n\n\ndef vocab_check(project_name, tokenizer_type):\n name_project = project_name\n path_project = os.path.join(path_data, name_project)\n\n file_metadata = os.path.join(path_project, \"metadata.csv\")\n\n file_vocab = os.path.join(path_data, \"Emilia_ZH_EN_pinyin/vocab.txt\")\n if not os.path.isfile(file_vocab):\n return f\"the file {file_vocab} not found !\", \"\"\n\n with open(file_vocab, \"r\", encoding=\"utf-8-sig\") as f:\n data = f.read()\n vocab = data.split(\"\\n\")\n vocab = set(vocab)\n\n if not os.path.isfile(file_metadata):\n return f\"the file {file_metadata} not found !\", \"\"\n\n with open(file_metadata, \"r\", encoding=\"utf-8-sig\") as f:\n data = f.read()\n\n miss_symbols = []\n miss_symbols_keep = {}\n for item in data.split(\"\\n\"):\n sp = item.split(\"|\")\n if len(sp) != 2:\n continue\n\n text = sp[1].strip()\n if tokenizer_type == \"pinyin\":\n text = convert_char_to_pinyin([text], polyphone=True)[0]\n\n for t in text:\n if t not in vocab and t not in miss_symbols_keep:\n miss_symbols.append(t)\n miss_symbols_keep[t] = t\n\n if miss_symbols == []:\n vocab_miss = \"\"\n info = \"You can train using your language !\"\n else:\n vocab_miss = \",\".join(miss_symbols)\n info = f\"The following {len(miss_symbols)} symbols are missing in your language\\n\\n\"\n\n return info, vocab_miss\n\n\ndef get_random_sample_prepare(project_name):\n name_project = project_name\n path_project = os.path.join(path_data, name_project)\n file_arrow = os.path.join(path_project, \"raw.arrow\")\n if not os.path.isfile(file_arrow):\n return \"\", None\n dataset = Dataset_.from_file(file_arrow)\n random_sample = dataset.shuffle(seed=random.randint(0, 1000)).select([0])\n text = \"[\" + \" , \".join([\"' \" + t + \" '\" for t in random_sample[\"text\"][0]]) + \"]\"\n audio_path = random_sample[\"audio_path\"][0]\n return text, audio_path\n\n\ndef get_random_sample_transcribe(project_name):\n name_project = project_name\n path_project = os.path.join(path_data, name_project)\n file_metadata = os.path.join(path_project, \"metadata.csv\")\n if not os.path.isfile(file_metadata):","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.get_random_sample_prepare","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.get_random_sample_prepare#L1110-L1120","kind":"function","name":"get_random_sample_prepare","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1110,"end_line":1120,"context_start_line":1090,"context_end_line":1140,"code":"\n text = sp[1].strip()\n if tokenizer_type == \"pinyin\":\n text = convert_char_to_pinyin([text], polyphone=True)[0]\n\n for t in text:\n if t not in vocab and t not in miss_symbols_keep:\n miss_symbols.append(t)\n miss_symbols_keep[t] = t\n\n if miss_symbols == []:\n vocab_miss = \"\"\n info = \"You can train using your language !\"\n else:\n vocab_miss = \",\".join(miss_symbols)\n info = f\"The following {len(miss_symbols)} symbols are missing in your language\\n\\n\"\n\n return info, vocab_miss\n\n\ndef get_random_sample_prepare(project_name):\n name_project = project_name\n path_project = os.path.join(path_data, name_project)\n file_arrow = os.path.join(path_project, \"raw.arrow\")\n if not os.path.isfile(file_arrow):\n return \"\", None\n dataset = Dataset_.from_file(file_arrow)\n random_sample = dataset.shuffle(seed=random.randint(0, 1000)).select([0])\n text = \"[\" + \" , \".join([\"' \" + t + \" '\" for t in random_sample[\"text\"][0]]) + \"]\"\n audio_path = random_sample[\"audio_path\"][0]\n return text, audio_path\n\n\ndef get_random_sample_transcribe(project_name):\n name_project = project_name\n path_project = os.path.join(path_data, name_project)\n file_metadata = os.path.join(path_project, \"metadata.csv\")\n if not os.path.isfile(file_metadata):\n return \"\", None\n\n data = \"\"\n with open(file_metadata, \"r\", encoding=\"utf-8-sig\") as f:\n data = f.read()\n\n list_data = []\n for item in data.split(\"\\n\"):\n sp = item.split(\"|\")\n if len(sp) != 2:\n continue\n\n # fixed audio when it is absolute","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.get_random_sample_transcribe","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.get_random_sample_transcribe#L1123-L1149","kind":"function","name":"get_random_sample_transcribe","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1123,"end_line":1149,"context_start_line":1103,"context_end_line":1169,"code":" else:\n vocab_miss = \",\".join(miss_symbols)\n info = f\"The following {len(miss_symbols)} symbols are missing in your language\\n\\n\"\n\n return info, vocab_miss\n\n\ndef get_random_sample_prepare(project_name):\n name_project = project_name\n path_project = os.path.join(path_data, name_project)\n file_arrow = os.path.join(path_project, \"raw.arrow\")\n if not os.path.isfile(file_arrow):\n return \"\", None\n dataset = Dataset_.from_file(file_arrow)\n random_sample = dataset.shuffle(seed=random.randint(0, 1000)).select([0])\n text = \"[\" + \" , \".join([\"' \" + t + \" '\" for t in random_sample[\"text\"][0]]) + \"]\"\n audio_path = random_sample[\"audio_path\"][0]\n return text, audio_path\n\n\ndef get_random_sample_transcribe(project_name):\n name_project = project_name\n path_project = os.path.join(path_data, name_project)\n file_metadata = os.path.join(path_project, \"metadata.csv\")\n if not os.path.isfile(file_metadata):\n return \"\", None\n\n data = \"\"\n with open(file_metadata, \"r\", encoding=\"utf-8-sig\") as f:\n data = f.read()\n\n list_data = []\n for item in data.split(\"\\n\"):\n sp = item.split(\"|\")\n if len(sp) != 2:\n continue\n\n # fixed audio when it is absolute\n file_audio = get_correct_audio_path(sp[0], os.path.join(path_project, \"wavs\"))\n list_data.append([file_audio, sp[1]])\n\n if list_data == []:\n return \"\", None\n\n random_item = random.choice(list_data)\n\n return random_item[1], random_item[0]\n\n\ndef get_random_sample_infer(project_name):\n text, audio = get_random_sample_transcribe(project_name)\n return (\n text,\n text,\n audio,\n )\n\n\ndef infer(\n project, file_checkpoint, exp_name, ref_text, ref_audio, gen_text, nfe_step, use_ema, speed, seed, remove_silence\n):\n global last_checkpoint, last_device, tts_api, last_ema\n\n if not os.path.isfile(file_checkpoint):\n return None, \"checkpoint not found!\"\n\n if training_process is not None:","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.get_random_sample_infer","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.get_random_sample_infer#L1152-L1158","kind":"function","name":"get_random_sample_infer","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1152,"end_line":1158,"context_start_line":1132,"context_end_line":1178,"code":" data = f.read()\n\n list_data = []\n for item in data.split(\"\\n\"):\n sp = item.split(\"|\")\n if len(sp) != 2:\n continue\n\n # fixed audio when it is absolute\n file_audio = get_correct_audio_path(sp[0], os.path.join(path_project, \"wavs\"))\n list_data.append([file_audio, sp[1]])\n\n if list_data == []:\n return \"\", None\n\n random_item = random.choice(list_data)\n\n return random_item[1], random_item[0]\n\n\ndef get_random_sample_infer(project_name):\n text, audio = get_random_sample_transcribe(project_name)\n return (\n text,\n text,\n audio,\n )\n\n\ndef infer(\n project, file_checkpoint, exp_name, ref_text, ref_audio, gen_text, nfe_step, use_ema, speed, seed, remove_silence\n):\n global last_checkpoint, last_device, tts_api, last_ema\n\n if not os.path.isfile(file_checkpoint):\n return None, \"checkpoint not found!\"\n\n if training_process is not None:\n device_test = \"cpu\"\n else:\n device_test = None\n\n if last_checkpoint != file_checkpoint or last_device != device_test or last_ema != use_ema or tts_api is None:\n if last_checkpoint != file_checkpoint:\n last_checkpoint = file_checkpoint\n\n if last_device != device_test:","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.infer","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.infer#L1161-L1206","kind":"function","name":"infer","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1161,"end_line":1206,"context_start_line":1141,"context_end_line":1226,"code":" file_audio = get_correct_audio_path(sp[0], os.path.join(path_project, \"wavs\"))\n list_data.append([file_audio, sp[1]])\n\n if list_data == []:\n return \"\", None\n\n random_item = random.choice(list_data)\n\n return random_item[1], random_item[0]\n\n\ndef get_random_sample_infer(project_name):\n text, audio = get_random_sample_transcribe(project_name)\n return (\n text,\n text,\n audio,\n )\n\n\ndef infer(\n project, file_checkpoint, exp_name, ref_text, ref_audio, gen_text, nfe_step, use_ema, speed, seed, remove_silence\n):\n global last_checkpoint, last_device, tts_api, last_ema\n\n if not os.path.isfile(file_checkpoint):\n return None, \"checkpoint not found!\"\n\n if training_process is not None:\n device_test = \"cpu\"\n else:\n device_test = None\n\n if last_checkpoint != file_checkpoint or last_device != device_test or last_ema != use_ema or tts_api is None:\n if last_checkpoint != file_checkpoint:\n last_checkpoint = file_checkpoint\n\n if last_device != device_test:\n last_device = device_test\n\n if last_ema != use_ema:\n last_ema = use_ema\n\n vocab_file = os.path.join(path_data, project, \"vocab.txt\")\n\n tts_api = F5TTS(\n model=exp_name, ckpt_file=file_checkpoint, vocab_file=vocab_file, device=device_test, use_ema=use_ema\n )\n\n print(\"update >> \", device_test, file_checkpoint, use_ema)\n\n if seed == -1: # -1 used for random\n seed = None\n\n with tempfile.NamedTemporaryFile(delete=False, suffix=\".wav\") as f:\n tts_api.infer(\n ref_file=ref_audio,\n ref_text=ref_text.strip(),\n gen_text=gen_text.strip(),\n nfe_step=nfe_step,\n speed=speed,\n remove_silence=remove_silence,\n file_wave=f.name,\n seed=seed,\n )\n return f.name, tts_api.device, str(tts_api.seed)\n\n\ndef check_finetune(finetune):\n return gr.update(interactive=finetune), gr.update(interactive=finetune), gr.update(interactive=finetune)\n\n\ndef get_checkpoints_project(project_name, is_gradio=True):\n if project_name is None:\n return [], \"\"\n project_name = project_name.replace(\"_pinyin\", \"\").replace(\"_char\", \"\")\n\n if os.path.isdir(path_project_ckpts):\n files_checkpoints = glob(os.path.join(path_project_ckpts, project_name, \"*.pt\"))\n # Separate pretrained and regular checkpoints\n pretrained_checkpoints = [f for f in files_checkpoints if \"pretrained_\" in os.path.basename(f)]\n regular_checkpoints = [\n f\n for f in files_checkpoints\n if \"pretrained_\" not in os.path.basename(f) and \"model_last.pt\" not in os.path.basename(f)\n ]","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.check_finetune","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.check_finetune#L1209-L1210","kind":"function","name":"check_finetune","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1209,"end_line":1210,"context_start_line":1189,"context_end_line":1230,"code":"\n print(\"update >> \", device_test, file_checkpoint, use_ema)\n\n if seed == -1: # -1 used for random\n seed = None\n\n with tempfile.NamedTemporaryFile(delete=False, suffix=\".wav\") as f:\n tts_api.infer(\n ref_file=ref_audio,\n ref_text=ref_text.strip(),\n gen_text=gen_text.strip(),\n nfe_step=nfe_step,\n speed=speed,\n remove_silence=remove_silence,\n file_wave=f.name,\n seed=seed,\n )\n return f.name, tts_api.device, str(tts_api.seed)\n\n\ndef check_finetune(finetune):\n return gr.update(interactive=finetune), gr.update(interactive=finetune), gr.update(interactive=finetune)\n\n\ndef get_checkpoints_project(project_name, is_gradio=True):\n if project_name is None:\n return [], \"\"\n project_name = project_name.replace(\"_pinyin\", \"\").replace(\"_char\", \"\")\n\n if os.path.isdir(path_project_ckpts):\n files_checkpoints = glob(os.path.join(path_project_ckpts, project_name, \"*.pt\"))\n # Separate pretrained and regular checkpoints\n pretrained_checkpoints = [f for f in files_checkpoints if \"pretrained_\" in os.path.basename(f)]\n regular_checkpoints = [\n f\n for f in files_checkpoints\n if \"pretrained_\" not in os.path.basename(f) and \"model_last.pt\" not in os.path.basename(f)\n ]\n last_checkpoint = [f for f in files_checkpoints if \"model_last.pt\" in os.path.basename(f)]\n\n # Sort regular checkpoints by number\n regular_checkpoints = sorted(","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.get_checkpoints_project","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.get_checkpoints_project#L1213-L1244","kind":"function","name":"get_checkpoints_project","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1213,"end_line":1244,"context_start_line":1193,"context_end_line":1264,"code":" seed = None\n\n with tempfile.NamedTemporaryFile(delete=False, suffix=\".wav\") as f:\n tts_api.infer(\n ref_file=ref_audio,\n ref_text=ref_text.strip(),\n gen_text=gen_text.strip(),\n nfe_step=nfe_step,\n speed=speed,\n remove_silence=remove_silence,\n file_wave=f.name,\n seed=seed,\n )\n return f.name, tts_api.device, str(tts_api.seed)\n\n\ndef check_finetune(finetune):\n return gr.update(interactive=finetune), gr.update(interactive=finetune), gr.update(interactive=finetune)\n\n\ndef get_checkpoints_project(project_name, is_gradio=True):\n if project_name is None:\n return [], \"\"\n project_name = project_name.replace(\"_pinyin\", \"\").replace(\"_char\", \"\")\n\n if os.path.isdir(path_project_ckpts):\n files_checkpoints = glob(os.path.join(path_project_ckpts, project_name, \"*.pt\"))\n # Separate pretrained and regular checkpoints\n pretrained_checkpoints = [f for f in files_checkpoints if \"pretrained_\" in os.path.basename(f)]\n regular_checkpoints = [\n f\n for f in files_checkpoints\n if \"pretrained_\" not in os.path.basename(f) and \"model_last.pt\" not in os.path.basename(f)\n ]\n last_checkpoint = [f for f in files_checkpoints if \"model_last.pt\" in os.path.basename(f)]\n\n # Sort regular checkpoints by number\n regular_checkpoints = sorted(\n regular_checkpoints, key=lambda x: int(os.path.basename(x).split(\"_\")[1].split(\".\")[0])\n )\n\n # Combine in order: pretrained, regular, last\n files_checkpoints = pretrained_checkpoints + regular_checkpoints + last_checkpoint\n else:\n files_checkpoints = []\n\n selelect_checkpoint = None if not files_checkpoints else files_checkpoints[0]\n\n if is_gradio:\n return gr.update(choices=files_checkpoints, value=selelect_checkpoint)\n\n return files_checkpoints, selelect_checkpoint\n\n\ndef get_audio_project(project_name, is_gradio=True):\n if project_name is None:\n return [], \"\"\n project_name = project_name.replace(\"_pinyin\", \"\").replace(\"_char\", \"\")\n\n if os.path.isdir(path_project_ckpts):\n files_audios = glob(os.path.join(path_project_ckpts, project_name, \"samples\", \"*.wav\"))\n files_audios = sorted(files_audios, key=lambda x: int(os.path.basename(x).split(\"_\")[1].split(\".\")[0]))\n\n files_audios = [item.replace(\"_gen.wav\", \"\") for item in files_audios if item.endswith(\"_gen.wav\")]\n else:\n files_audios = []\n\n selelect_checkpoint = None if not files_audios else files_audios[0]\n\n if is_gradio:\n return gr.update(choices=files_audios, value=selelect_checkpoint)\n","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.get_audio_project","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.get_audio_project#L1247-L1265","kind":"function","name":"get_audio_project","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1247,"end_line":1265,"context_start_line":1227,"context_end_line":1285,"code":" last_checkpoint = [f for f in files_checkpoints if \"model_last.pt\" in os.path.basename(f)]\n\n # Sort regular checkpoints by number\n regular_checkpoints = sorted(\n regular_checkpoints, key=lambda x: int(os.path.basename(x).split(\"_\")[1].split(\".\")[0])\n )\n\n # Combine in order: pretrained, regular, last\n files_checkpoints = pretrained_checkpoints + regular_checkpoints + last_checkpoint\n else:\n files_checkpoints = []\n\n selelect_checkpoint = None if not files_checkpoints else files_checkpoints[0]\n\n if is_gradio:\n return gr.update(choices=files_checkpoints, value=selelect_checkpoint)\n\n return files_checkpoints, selelect_checkpoint\n\n\ndef get_audio_project(project_name, is_gradio=True):\n if project_name is None:\n return [], \"\"\n project_name = project_name.replace(\"_pinyin\", \"\").replace(\"_char\", \"\")\n\n if os.path.isdir(path_project_ckpts):\n files_audios = glob(os.path.join(path_project_ckpts, project_name, \"samples\", \"*.wav\"))\n files_audios = sorted(files_audios, key=lambda x: int(os.path.basename(x).split(\"_\")[1].split(\".\")[0]))\n\n files_audios = [item.replace(\"_gen.wav\", \"\") for item in files_audios if item.endswith(\"_gen.wav\")]\n else:\n files_audios = []\n\n selelect_checkpoint = None if not files_audios else files_audios[0]\n\n if is_gradio:\n return gr.update(choices=files_audios, value=selelect_checkpoint)\n\n return files_audios, selelect_checkpoint\n\n\ndef get_gpu_stats():\n gpu_stats = \"\"\n\n if torch.cuda.is_available():\n gpu_count = torch.cuda.device_count()\n for i in range(gpu_count):\n gpu_name = torch.cuda.get_device_name(i)\n gpu_properties = torch.cuda.get_device_properties(i)\n total_memory = gpu_properties.total_memory / (1024**3) # in GB\n allocated_memory = torch.cuda.memory_allocated(i) / (1024**2) # in MB\n reserved_memory = torch.cuda.memory_reserved(i) / (1024**2) # in MB\n\n gpu_stats += (\n f\"GPU {i} Name: {gpu_name}\\n\"\n f\"Total GPU memory (GPU {i}): {total_memory:.2f} GB\\n\"\n f\"Allocated GPU memory (GPU {i}): {allocated_memory:.2f} MB\\n\"\n f\"Reserved GPU memory (GPU {i}): {reserved_memory:.2f} MB\\n\\n\"\n )","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.get_gpu_stats","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.get_gpu_stats#L1268-L1319","kind":"function","name":"get_gpu_stats","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1268,"end_line":1319,"context_start_line":1248,"context_end_line":1339,"code":" if project_name is None:\n return [], \"\"\n project_name = project_name.replace(\"_pinyin\", \"\").replace(\"_char\", \"\")\n\n if os.path.isdir(path_project_ckpts):\n files_audios = glob(os.path.join(path_project_ckpts, project_name, \"samples\", \"*.wav\"))\n files_audios = sorted(files_audios, key=lambda x: int(os.path.basename(x).split(\"_\")[1].split(\".\")[0]))\n\n files_audios = [item.replace(\"_gen.wav\", \"\") for item in files_audios if item.endswith(\"_gen.wav\")]\n else:\n files_audios = []\n\n selelect_checkpoint = None if not files_audios else files_audios[0]\n\n if is_gradio:\n return gr.update(choices=files_audios, value=selelect_checkpoint)\n\n return files_audios, selelect_checkpoint\n\n\ndef get_gpu_stats():\n gpu_stats = \"\"\n\n if torch.cuda.is_available():\n gpu_count = torch.cuda.device_count()\n for i in range(gpu_count):\n gpu_name = torch.cuda.get_device_name(i)\n gpu_properties = torch.cuda.get_device_properties(i)\n total_memory = gpu_properties.total_memory / (1024**3) # in GB\n allocated_memory = torch.cuda.memory_allocated(i) / (1024**2) # in MB\n reserved_memory = torch.cuda.memory_reserved(i) / (1024**2) # in MB\n\n gpu_stats += (\n f\"GPU {i} Name: {gpu_name}\\n\"\n f\"Total GPU memory (GPU {i}): {total_memory:.2f} GB\\n\"\n f\"Allocated GPU memory (GPU {i}): {allocated_memory:.2f} MB\\n\"\n f\"Reserved GPU memory (GPU {i}): {reserved_memory:.2f} MB\\n\\n\"\n )\n elif torch.xpu.is_available():\n gpu_count = torch.xpu.device_count()\n for i in range(gpu_count):\n gpu_name = torch.xpu.get_device_name(i)\n gpu_properties = torch.xpu.get_device_properties(i)\n total_memory = gpu_properties.total_memory / (1024**3) # in GB\n allocated_memory = torch.xpu.memory_allocated(i) / (1024**2) # in MB\n reserved_memory = torch.xpu.memory_reserved(i) / (1024**2) # in MB\n\n gpu_stats += (\n f\"GPU {i} Name: {gpu_name}\\n\"\n f\"Total GPU memory (GPU {i}): {total_memory:.2f} GB\\n\"\n f\"Allocated GPU memory (GPU {i}): {allocated_memory:.2f} MB\\n\"\n f\"Reserved GPU memory (GPU {i}): {reserved_memory:.2f} MB\\n\\n\"\n )\n elif torch.backends.mps.is_available():\n gpu_count = 1\n gpu_stats += \"MPS GPU\\n\"\n total_memory = psutil.virtual_memory().total / (\n 1024**3\n ) # Total system memory (MPS doesn't have its own memory)\n allocated_memory = 0\n reserved_memory = 0\n\n gpu_stats += (\n f\"Total system memory: {total_memory:.2f} GB\\n\"\n f\"Allocated GPU memory (MPS): {allocated_memory:.2f} MB\\n\"\n f\"Reserved GPU memory (MPS): {reserved_memory:.2f} MB\\n\"\n )\n\n else:\n gpu_stats = \"No GPU available\"\n\n return gpu_stats\n\n\ndef get_cpu_stats():\n cpu_usage = psutil.cpu_percent(interval=1)\n memory_info = psutil.virtual_memory()\n memory_used = memory_info.used / (1024**2)\n memory_total = memory_info.total / (1024**2)\n memory_percent = memory_info.percent\n\n pid = os.getpid()\n process = psutil.Process(pid)\n nice_value = process.nice()\n\n cpu_stats = (\n f\"CPU Usage: {cpu_usage:.2f}%\\n\"\n f\"System Memory: {memory_used:.2f} MB used / {memory_total:.2f} MB total ({memory_percent}% used)\\n\"\n f\"Process Priority (Nice value): {nice_value}\"\n )\n\n return cpu_stats","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.get_cpu_stats","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.get_cpu_stats#L1322-L1339","kind":"function","name":"get_cpu_stats","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1322,"end_line":1339,"context_start_line":1302,"context_end_line":1359,"code":" gpu_count = 1\n gpu_stats += \"MPS GPU\\n\"\n total_memory = psutil.virtual_memory().total / (\n 1024**3\n ) # Total system memory (MPS doesn't have its own memory)\n allocated_memory = 0\n reserved_memory = 0\n\n gpu_stats += (\n f\"Total system memory: {total_memory:.2f} GB\\n\"\n f\"Allocated GPU memory (MPS): {allocated_memory:.2f} MB\\n\"\n f\"Reserved GPU memory (MPS): {reserved_memory:.2f} MB\\n\"\n )\n\n else:\n gpu_stats = \"No GPU available\"\n\n return gpu_stats\n\n\ndef get_cpu_stats():\n cpu_usage = psutil.cpu_percent(interval=1)\n memory_info = psutil.virtual_memory()\n memory_used = memory_info.used / (1024**2)\n memory_total = memory_info.total / (1024**2)\n memory_percent = memory_info.percent\n\n pid = os.getpid()\n process = psutil.Process(pid)\n nice_value = process.nice()\n\n cpu_stats = (\n f\"CPU Usage: {cpu_usage:.2f}%\\n\"\n f\"System Memory: {memory_used:.2f} MB used / {memory_total:.2f} MB total ({memory_percent}% used)\\n\"\n f\"Process Priority (Nice value): {nice_value}\"\n )\n\n return cpu_stats\n\n\ndef get_combined_stats():\n gpu_stats = get_gpu_stats()\n cpu_stats = get_cpu_stats()\n combined_stats = f\"### GPU Stats\\n{gpu_stats}\\n\\n### CPU Stats\\n{cpu_stats}\"\n return combined_stats\n\n\ndef get_audio_select(file_sample):\n select_audio_ref = file_sample\n select_audio_gen = file_sample\n\n if file_sample is not None:\n select_audio_ref += \"_ref.wav\"\n select_audio_gen += \"_gen.wav\"\n\n return select_audio_ref, select_audio_gen\n\n","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.get_combined_stats","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.get_combined_stats#L1342-L1346","kind":"function","name":"get_combined_stats","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1342,"end_line":1346,"context_start_line":1322,"context_end_line":1366,"code":"def get_cpu_stats():\n cpu_usage = psutil.cpu_percent(interval=1)\n memory_info = psutil.virtual_memory()\n memory_used = memory_info.used / (1024**2)\n memory_total = memory_info.total / (1024**2)\n memory_percent = memory_info.percent\n\n pid = os.getpid()\n process = psutil.Process(pid)\n nice_value = process.nice()\n\n cpu_stats = (\n f\"CPU Usage: {cpu_usage:.2f}%\\n\"\n f\"System Memory: {memory_used:.2f} MB used / {memory_total:.2f} MB total ({memory_percent}% used)\\n\"\n f\"Process Priority (Nice value): {nice_value}\"\n )\n\n return cpu_stats\n\n\ndef get_combined_stats():\n gpu_stats = get_gpu_stats()\n cpu_stats = get_cpu_stats()\n combined_stats = f\"### GPU Stats\\n{gpu_stats}\\n\\n### CPU Stats\\n{cpu_stats}\"\n return combined_stats\n\n\ndef get_audio_select(file_sample):\n select_audio_ref = file_sample\n select_audio_gen = file_sample\n\n if file_sample is not None:\n select_audio_ref += \"_ref.wav\"\n select_audio_gen += \"_gen.wav\"\n\n return select_audio_ref, select_audio_gen\n\n\nwith gr.Blocks() as app:\n gr.Markdown(\n \"\"\"\n# F5 TTS Automatic Finetune\n\nThis is a local web UI for F5 TTS finetuning support. This app supports the following TTS models:\n","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.get_audio_select","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.get_audio_select#L1349-L1357","kind":"function","name":"get_audio_select","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1349,"end_line":1357,"context_start_line":1329,"context_end_line":1377,"code":" pid = os.getpid()\n process = psutil.Process(pid)\n nice_value = process.nice()\n\n cpu_stats = (\n f\"CPU Usage: {cpu_usage:.2f}%\\n\"\n f\"System Memory: {memory_used:.2f} MB used / {memory_total:.2f} MB total ({memory_percent}% used)\\n\"\n f\"Process Priority (Nice value): {nice_value}\"\n )\n\n return cpu_stats\n\n\ndef get_combined_stats():\n gpu_stats = get_gpu_stats()\n cpu_stats = get_cpu_stats()\n combined_stats = f\"### GPU Stats\\n{gpu_stats}\\n\\n### CPU Stats\\n{cpu_stats}\"\n return combined_stats\n\n\ndef get_audio_select(file_sample):\n select_audio_ref = file_sample\n select_audio_gen = file_sample\n\n if file_sample is not None:\n select_audio_ref += \"_ref.wav\"\n select_audio_gen += \"_gen.wav\"\n\n return select_audio_ref, select_audio_gen\n\n\nwith gr.Blocks() as app:\n gr.Markdown(\n \"\"\"\n# F5 TTS Automatic Finetune\n\nThis is a local web UI for F5 TTS finetuning support. This app supports the following TTS models:\n\n* [F5-TTS](https://arxiv.org/abs/2410.06885) (A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching)\n* [E2 TTS](https://arxiv.org/abs/2406.18009) (Embarrassingly Easy Fully Non-Autoregressive Zero-Shot TTS)\n\nThe pretrained checkpoints support English and Chinese.\n\nFor tutorial and updates check here (https://github.com/SWivid/F5-TTS/discussions/143)\n\"\"\"\n )\n\n with gr.Row():\n projects, projects_selelect = get_list_projects()","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.main","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.main#L1858-L1861","kind":"function","name":"main","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1858,"end_line":1861,"context_start_line":1838,"context_end_line":1865,"code":" update_button = gr.Button(\"Update Stats\")\n update_button.click(fn=update_stats, outputs=output_box)\n\n def auto_update():\n yield gr.update(value=update_stats())\n\n gr.update(fn=auto_update, inputs=[], outputs=output_box)\n\n\n@click.command()\n@click.option(\"--port\", \"-p\", default=None, type=int, help=\"Port to run the app on\")\n@click.option(\"--host\", \"-H\", default=None, help=\"Host to run the app on\")\n@click.option(\n \"--share\",\n \"-s\",\n default=False,\n is_flag=True,\n help=\"Share the app via Gradio share link\",\n)\n@click.option(\"--api\", \"-a\", default=True, is_flag=True, help=\"Allow API access\")\ndef main(port, host, share, api):\n global app\n print(\"Starting app...\")\n app.queue(api_open=api).launch(server_name=host, server_port=port, share=share, show_api=api)\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.__init__#L182-L201","kind":"function","name":"__init__","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":182,"end_line":201,"context_start_line":162,"context_end_line":221,"code":" default_settings[\"keep_last_n_checkpoints\"],\n default_settings[\"last_per_updates\"],\n default_settings[\"finetune\"],\n default_settings[\"file_checkpoint_train\"],\n default_settings[\"tokenizer_type\"],\n default_settings[\"tokenizer_file\"],\n default_settings[\"mixed_precision\"],\n default_settings[\"logger\"],\n default_settings[\"bnb_optimizer\"],\n )\n\n\n# Load metadata\ndef get_audio_duration(audio_path):\n \"\"\"Calculate the duration mono of an audio file.\"\"\"\n audio, sample_rate = torchaudio.load(audio_path)\n return audio.shape[1] / sample_rate\n\n\nclass Slicer: # https://github.com/RVC-Boss/GPT-SoVITS/blob/main/tools/slicer2.py\n def __init__(\n self,\n sr: int,\n threshold: float = -40.0,\n min_length: int = 20000, # 20 seconds\n min_interval: int = 300,\n hop_size: int = 20,\n max_sil_kept: int = 2000,\n ):\n if not min_length >= min_interval >= hop_size:\n raise ValueError(\"The following condition must be satisfied: min_length >= min_interval >= hop_size\")\n if not max_sil_kept >= hop_size:\n raise ValueError(\"The following condition must be satisfied: max_sil_kept >= hop_size\")\n min_interval = sr * min_interval / 1000\n self.threshold = 10 ** (threshold / 20.0)\n self.hop_size = round(sr * hop_size / 1000)\n self.win_size = min(round(min_interval), 4 * self.hop_size)\n self.min_length = round(sr * min_length / 1000 / self.hop_size)\n self.min_interval = round(min_interval / self.hop_size)\n self.max_sil_kept = round(sr * max_sil_kept / 1000 / self.hop_size)\n\n def _apply_slice(self, waveform, begin, end):\n if len(waveform.shape) > 1:\n return waveform[:, begin * self.hop_size : min(waveform.shape[1], end * self.hop_size)]\n else:\n return waveform[begin * self.hop_size : min(waveform.shape[0], end * self.hop_size)]\n\n # @timeit\n def slice(self, waveform):\n if len(waveform.shape) > 1:\n samples = waveform.mean(axis=0)\n else:\n samples = waveform\n if samples.shape[0] <= self.min_length:\n return [waveform]\n rms_list = librosa.feature.rms(y=samples, frame_length=self.win_size, hop_length=self.hop_size).squeeze(0)\n sil_tags = []\n silence_start = None\n clip_start = 0\n for i, rms in enumerate(rms_list):","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio._apply_slice","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio._apply_slice#L203-L207","kind":"function","name":"_apply_slice","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":203,"end_line":207,"context_start_line":183,"context_end_line":227,"code":" self,\n sr: int,\n threshold: float = -40.0,\n min_length: int = 20000, # 20 seconds\n min_interval: int = 300,\n hop_size: int = 20,\n max_sil_kept: int = 2000,\n ):\n if not min_length >= min_interval >= hop_size:\n raise ValueError(\"The following condition must be satisfied: min_length >= min_interval >= hop_size\")\n if not max_sil_kept >= hop_size:\n raise ValueError(\"The following condition must be satisfied: max_sil_kept >= hop_size\")\n min_interval = sr * min_interval / 1000\n self.threshold = 10 ** (threshold / 20.0)\n self.hop_size = round(sr * hop_size / 1000)\n self.win_size = min(round(min_interval), 4 * self.hop_size)\n self.min_length = round(sr * min_length / 1000 / self.hop_size)\n self.min_interval = round(min_interval / self.hop_size)\n self.max_sil_kept = round(sr * max_sil_kept / 1000 / self.hop_size)\n\n def _apply_slice(self, waveform, begin, end):\n if len(waveform.shape) > 1:\n return waveform[:, begin * self.hop_size : min(waveform.shape[1], end * self.hop_size)]\n else:\n return waveform[begin * self.hop_size : min(waveform.shape[0], end * self.hop_size)]\n\n # @timeit\n def slice(self, waveform):\n if len(waveform.shape) > 1:\n samples = waveform.mean(axis=0)\n else:\n samples = waveform\n if samples.shape[0] <= self.min_length:\n return [waveform]\n rms_list = librosa.feature.rms(y=samples, frame_length=self.win_size, hop_length=self.hop_size).squeeze(0)\n sil_tags = []\n silence_start = None\n clip_start = 0\n for i, rms in enumerate(rms_list):\n # Keep looping while frame is silent.\n if rms < self.threshold:\n # Record start of silent frames.\n if silence_start is None:\n silence_start = i\n continue","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.slice","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.slice#L210-L294","kind":"function","name":"slice","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":210,"end_line":294,"context_start_line":190,"context_end_line":314,"code":" ):\n if not min_length >= min_interval >= hop_size:\n raise ValueError(\"The following condition must be satisfied: min_length >= min_interval >= hop_size\")\n if not max_sil_kept >= hop_size:\n raise ValueError(\"The following condition must be satisfied: max_sil_kept >= hop_size\")\n min_interval = sr * min_interval / 1000\n self.threshold = 10 ** (threshold / 20.0)\n self.hop_size = round(sr * hop_size / 1000)\n self.win_size = min(round(min_interval), 4 * self.hop_size)\n self.min_length = round(sr * min_length / 1000 / self.hop_size)\n self.min_interval = round(min_interval / self.hop_size)\n self.max_sil_kept = round(sr * max_sil_kept / 1000 / self.hop_size)\n\n def _apply_slice(self, waveform, begin, end):\n if len(waveform.shape) > 1:\n return waveform[:, begin * self.hop_size : min(waveform.shape[1], end * self.hop_size)]\n else:\n return waveform[begin * self.hop_size : min(waveform.shape[0], end * self.hop_size)]\n\n # @timeit\n def slice(self, waveform):\n if len(waveform.shape) > 1:\n samples = waveform.mean(axis=0)\n else:\n samples = waveform\n if samples.shape[0] <= self.min_length:\n return [waveform]\n rms_list = librosa.feature.rms(y=samples, frame_length=self.win_size, hop_length=self.hop_size).squeeze(0)\n sil_tags = []\n silence_start = None\n clip_start = 0\n for i, rms in enumerate(rms_list):\n # Keep looping while frame is silent.\n if rms < self.threshold:\n # Record start of silent frames.\n if silence_start is None:\n silence_start = i\n continue\n # Keep looping while frame is not silent and silence start has not been recorded.\n if silence_start is None:\n continue\n # Clear recorded silence start if interval is not enough or clip is too short\n is_leading_silence = silence_start == 0 and i > self.max_sil_kept\n need_slice_middle = i - silence_start >= self.min_interval and i - clip_start >= self.min_length\n if not is_leading_silence and not need_slice_middle:\n silence_start = None\n continue\n # Need slicing. Record the range of silent frames to be removed.\n if i - silence_start <= self.max_sil_kept:\n pos = rms_list[silence_start : i + 1].argmin() + silence_start\n if silence_start == 0:\n sil_tags.append((0, pos))\n else:\n sil_tags.append((pos, pos))\n clip_start = pos\n elif i - silence_start <= self.max_sil_kept * 2:\n pos = rms_list[i - self.max_sil_kept : silence_start + self.max_sil_kept + 1].argmin()\n pos += i - self.max_sil_kept\n pos_l = rms_list[silence_start : silence_start + self.max_sil_kept + 1].argmin() + silence_start\n pos_r = rms_list[i - self.max_sil_kept : i + 1].argmin() + i - self.max_sil_kept\n if silence_start == 0:\n sil_tags.append((0, pos_r))\n clip_start = pos_r\n else:\n sil_tags.append((min(pos_l, pos), max(pos_r, pos)))\n clip_start = max(pos_r, pos)\n else:\n pos_l = rms_list[silence_start : silence_start + self.max_sil_kept + 1].argmin() + silence_start\n pos_r = rms_list[i - self.max_sil_kept : i + 1].argmin() + i - self.max_sil_kept\n if silence_start == 0:\n sil_tags.append((0, pos_r))\n else:\n sil_tags.append((pos_l, pos_r))\n clip_start = pos_r\n silence_start = None\n # Deal with trailing silence.\n total_frames = rms_list.shape[0]\n if silence_start is not None and total_frames - silence_start >= self.min_interval:\n silence_end = min(total_frames, silence_start + self.max_sil_kept)\n pos = rms_list[silence_start : silence_end + 1].argmin() + silence_start\n sil_tags.append((pos, total_frames + 1))\n # Apply and return slices: [chunk, start, end]\n if len(sil_tags) == 0:\n return [[waveform, 0, int(total_frames * self.hop_size)]]\n else:\n chunks = []\n if sil_tags[0][0] > 0:\n chunks.append([self._apply_slice(waveform, 0, sil_tags[0][0]), 0, int(sil_tags[0][0] * self.hop_size)])\n for i in range(len(sil_tags) - 1):\n chunks.append(\n [\n self._apply_slice(waveform, sil_tags[i][1], sil_tags[i + 1][0]),\n int(sil_tags[i][1] * self.hop_size),\n int(sil_tags[i + 1][0] * self.hop_size),\n ]\n )\n if sil_tags[-1][1] < total_frames:\n chunks.append(\n [\n self._apply_slice(waveform, sil_tags[-1][1], total_frames),\n int(sil_tags[-1][1] * self.hop_size),\n int(total_frames * self.hop_size),\n ]\n )\n return chunks\n\n\n# terminal\ndef terminate_process_tree(pid, including_parent=True):\n try:\n parent = psutil.Process(pid)\n except psutil.NoSuchProcess:\n # Process already terminated\n return\n\n children = parent.children(recursive=True)\n for child in children:\n try:\n os.kill(child.pid, signal.SIGTERM) # or signal.SIGKILL\n except OSError:\n pass\n if including_parent:\n try:\n os.kill(parent.pid, signal.SIGTERM) # or signal.SIGKILL\n except OSError:","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.has_supported_extension","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.has_supported_extension#L705-L706","kind":"function","name":"has_supported_extension","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":705,"end_line":706,"context_start_line":685,"context_end_line":726,"code":" error_text = \"\"\n\n return f\"transcribe complete samples : {num}\\npath : {path_project_wavs}{error_text}\"\n\n\ndef format_seconds_to_hms(seconds):\n hours = int(seconds / 3600)\n minutes = int((seconds % 3600) / 60)\n seconds = seconds % 60\n return \"{:02d}:{:02d}:{:02d}\".format(hours, minutes, int(seconds))\n\n\ndef get_correct_audio_path(\n audio_input,\n base_path=\"wavs\",\n supported_formats=(\"wav\", \"mp3\", \"aac\", \"flac\", \"m4a\", \"alac\", \"ogg\", \"aiff\", \"wma\", \"amr\"),\n):\n file_audio = None\n\n # Helper function to check if file has a supported extension\n def has_supported_extension(file_name):\n return any(file_name.endswith(f\".{ext}\") for ext in supported_formats)\n\n # Case 1: If it's a full path with a valid extension, use it directly\n if os.path.isabs(audio_input) and has_supported_extension(audio_input):\n file_audio = audio_input\n\n # Case 2: If it has a supported extension but is not a full path\n elif has_supported_extension(audio_input) and not os.path.isabs(audio_input):\n file_audio = os.path.join(base_path, audio_input)\n\n # Case 3: If only the name is given (no extension and not a full path)\n elif not has_supported_extension(audio_input) and not os.path.isabs(audio_input):\n for ext in supported_formats:\n potential_file = os.path.join(base_path, f\"{audio_input}.{ext}\")\n if os.path.exists(potential_file):\n file_audio = potential_file\n break\n else:\n file_audio = os.path.join(base_path, f\"{audio_input}.{supported_formats[0]}\")\n return file_audio\n","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.expand_embeddings","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.expand_embeddings#L980-L984","kind":"function","name":"expand_embeddings","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":980,"end_line":984,"context_start_line":960,"context_end_line":1004,"code":" torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.cuda.manual_seed_all(seed)\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n\n if ckpt_path.endswith(\".safetensors\"):\n ckpt = load_file(ckpt_path, device=\"cpu\")\n ckpt = {\"ema_model_state_dict\": ckpt}\n elif ckpt_path.endswith(\".pt\"):\n ckpt = torch.load(ckpt_path, map_location=\"cpu\")\n\n ema_sd = ckpt.get(\"ema_model_state_dict\", {})\n embed_key_ema = \"ema_model.transformer.text_embed.text_embed.weight\"\n old_embed_ema = ema_sd[embed_key_ema]\n\n vocab_old = old_embed_ema.size(0)\n embed_dim = old_embed_ema.size(1)\n vocab_new = vocab_old + num_new_tokens\n\n def expand_embeddings(old_embeddings):\n new_embeddings = torch.zeros((vocab_new, embed_dim))\n new_embeddings[:vocab_old] = old_embeddings\n new_embeddings[vocab_old:] = torch.randn((num_new_tokens, embed_dim))\n return new_embeddings\n\n ema_sd[embed_key_ema] = expand_embeddings(ema_sd[embed_key_ema])\n\n if new_ckpt_path.endswith(\".safetensors\"):\n save_file(ema_sd, new_ckpt_path)\n elif new_ckpt_path.endswith(\".pt\"):\n torch.save(ckpt, new_ckpt_path)\n\n return vocab_new\n\n\ndef vocab_count(text):\n return str(len(text.split(\",\")))\n\n\ndef vocab_extend(project_name, symbols, model_type):\n if symbols == \"\":\n return \"Symbols empty!\"\n\n name_project = project_name","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.stream_output","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.stream_output#L469-L476","kind":"function","name":"stream_output","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":469,"end_line":476,"context_start_line":449,"context_end_line":496,"code":" file_checkpoint_train,\n tokenizer_type,\n tokenizer_file,\n mixed_precision,\n logger,\n ch_8bit_adam,\n )\n\n try:\n if not stream:\n # Start the training process\n training_process = subprocess.Popen(cmd, shell=True)\n\n time.sleep(5)\n yield \"train start\", gr.update(interactive=False), gr.update(interactive=True)\n\n # Wait for the training process to finish\n training_process.wait()\n else:\n\n def stream_output(pipe, output_queue):\n try:\n for line in iter(pipe.readline, \"\"):\n output_queue.put(line)\n except Exception as e:\n output_queue.put(f\"Error reading pipe: {str(e)}\")\n finally:\n pipe.close()\n\n env = os.environ.copy()\n env[\"PYTHONUNBUFFERED\"] = \"1\"\n\n training_process = subprocess.Popen(\n cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, bufsize=1, env=env\n )\n yield \"Training started ...\", gr.update(interactive=False), gr.update(interactive=True)\n\n stdout_queue = queue.Queue()\n stderr_queue = queue.Queue()\n\n stdout_thread = threading.Thread(target=stream_output, args=(training_process.stdout, stdout_queue))\n stderr_thread = threading.Thread(target=stream_output, args=(training_process.stderr, stderr_queue))\n stdout_thread.daemon = True\n stderr_thread.daemon = True\n stdout_thread.start()\n stderr_thread.start()\n stop_signal = False\n while True:","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.setup_load_settings","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.setup_load_settings#L1715-L1737","kind":"function","name":"setup_load_settings","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1715,"end_line":1737,"context_start_line":1695,"context_end_line":1757,"code":" batch_size_per_gpu,\n batch_size_type,\n max_samples,\n num_warmup_updates,\n ch_finetune,\n ],\n outputs=[\n epochs,\n learning_rate,\n batch_size_per_gpu,\n max_samples,\n num_warmup_updates,\n lb_samples,\n ],\n )\n\n ch_finetune.change(\n check_finetune, inputs=[ch_finetune], outputs=[file_checkpoint_train, tokenizer_file, tokenizer_type]\n )\n\n def setup_load_settings():\n output_components = [\n exp_name,\n learning_rate,\n batch_size_per_gpu,\n batch_size_type,\n max_samples,\n grad_accumulation_steps,\n max_grad_norm,\n epochs,\n num_warmup_updates,\n save_per_updates,\n keep_last_n_checkpoints,\n last_per_updates,\n ch_finetune,\n file_checkpoint_train,\n tokenizer_type,\n tokenizer_file,\n mixed_precision,\n cd_logger,\n ch_8bit_adam,\n ]\n return output_components\n\n outputs = setup_load_settings()\n\n cm_project.change(\n fn=load_settings,\n inputs=[cm_project],\n outputs=outputs,\n )\n\n ch_refresh_project.click(\n fn=load_settings,\n inputs=[cm_project],\n outputs=outputs,\n )\n\n with gr.TabItem(\"Test Model\"):\n gr.Markdown(\"\"\"```plaintext \nCheck the use_ema setting (True or False) for your model to see what works best for you. Set seed to -1 for random.\n```\"\"\")\n exp_name = gr.Radio(","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.update_stats","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.update_stats#L1835-L1836","kind":"function","name":"update_stats","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1835,"end_line":1836,"context_start_line":1815,"context_end_line":1856,"code":" with gr.TabItem(\"Prune Checkpoint\"):\n gr.Markdown(\"\"\"```plaintext \nReduce the Base model size from 5GB to 1.3GB. The new checkpoint file prunes out optimizer and etc., can be used for inference or finetuning afterward, but not able to resume pretraining.\n```\"\"\")\n txt_path_checkpoint = gr.Textbox(label=\"Path to Checkpoint:\")\n txt_path_checkpoint_small = gr.Textbox(label=\"Path to Output:\")\n with gr.Row():\n ch_save_ema = gr.Checkbox(label=\"Save EMA checkpoint\", value=True)\n ch_safetensors = gr.Checkbox(label=\"Save with safetensors format\", value=True)\n txt_info_reduse = gr.Textbox(label=\"Info\", value=\"\")\n reduse_button = gr.Button(\"Prune\")\n reduse_button.click(\n fn=prune_checkpoint,\n inputs=[txt_path_checkpoint, txt_path_checkpoint_small, ch_save_ema, ch_safetensors],\n outputs=[txt_info_reduse],\n )\n\n with gr.TabItem(\"System Info\"):\n output_box = gr.Textbox(label=\"GPU and CPU Information\", lines=20)\n\n def update_stats():\n return get_combined_stats()\n\n update_button = gr.Button(\"Update Stats\")\n update_button.click(fn=update_stats, outputs=output_box)\n\n def auto_update():\n yield gr.update(value=update_stats())\n\n gr.update(fn=auto_update, inputs=[], outputs=output_box)\n\n\n@click.command()\n@click.option(\"--port\", \"-p\", default=None, type=int, help=\"Port to run the app on\")\n@click.option(\"--host\", \"-H\", default=None, help=\"Host to run the app on\")\n@click.option(\n \"--share\",\n \"-s\",\n default=False,\n is_flag=True,\n help=\"Share the app via Gradio share link\",\n)","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.finetune_gradio.auto_update","uri":"program://DMOSpeech2/function/src.f5_tts.train.finetune_gradio.auto_update#L1841-L1842","kind":"function","name":"auto_update","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1841,"end_line":1842,"context_start_line":1821,"context_end_line":1862,"code":" with gr.Row():\n ch_save_ema = gr.Checkbox(label=\"Save EMA checkpoint\", value=True)\n ch_safetensors = gr.Checkbox(label=\"Save with safetensors format\", value=True)\n txt_info_reduse = gr.Textbox(label=\"Info\", value=\"\")\n reduse_button = gr.Button(\"Prune\")\n reduse_button.click(\n fn=prune_checkpoint,\n inputs=[txt_path_checkpoint, txt_path_checkpoint_small, ch_save_ema, ch_safetensors],\n outputs=[txt_info_reduse],\n )\n\n with gr.TabItem(\"System Info\"):\n output_box = gr.Textbox(label=\"GPU and CPU Information\", lines=20)\n\n def update_stats():\n return get_combined_stats()\n\n update_button = gr.Button(\"Update Stats\")\n update_button.click(fn=update_stats, outputs=output_box)\n\n def auto_update():\n yield gr.update(value=update_stats())\n\n gr.update(fn=auto_update, inputs=[], outputs=output_box)\n\n\n@click.command()\n@click.option(\"--port\", \"-p\", default=None, type=int, help=\"Port to run the app on\")\n@click.option(\"--host\", \"-H\", default=None, help=\"Host to run the app on\")\n@click.option(\n \"--share\",\n \"-s\",\n default=False,\n is_flag=True,\n help=\"Share the app via Gradio share link\",\n)\n@click.option(\"--api\", \"-a\", default=True, is_flag=True, help=\"Allow API access\")\ndef main(port, host, share, api):\n global app\n print(\"Starting app...\")\n app.queue(api_open=api).launch(server_name=host, server_port=port, share=share, show_api=api)\n","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_libritts","uri":"program://DMOSpeech2/module/src.f5_tts.train.datasets.prepare_libritts#L1-L94","kind":"module","name":"src.f5_tts.train.datasets.prepare_libritts","path":"src/f5_tts/train/datasets/prepare_libritts.py","language":"python","start_line":1,"end_line":94,"context_start_line":1,"context_end_line":94,"code":"import os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport json\nfrom concurrent.futures import ProcessPoolExecutor\nfrom importlib.resources import files\nfrom pathlib import Path\n\nimport soundfile as sf\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm\n\n\ndef deal_with_audio_dir(audio_dir):\n sub_result, durations = [], []\n vocab_set = set()\n audio_lists = list(audio_dir.rglob(\"*.wav\"))\n\n for line in audio_lists:\n text_path = line.with_suffix(\".normalized.txt\")\n text = open(text_path, \"r\").read().strip()\n duration = sf.info(line).duration\n if duration < 0.4 or duration > 30:\n continue\n sub_result.append({\"audio_path\": str(line), \"text\": text, \"duration\": duration})\n durations.append(duration)\n vocab_set.update(list(text))\n return sub_result, durations, vocab_set\n\n\ndef main():\n result = []\n duration_list = []\n text_vocab_set = set()\n\n # process raw data\n executor = ProcessPoolExecutor(max_workers=max_workers)\n futures = []\n\n for subset in tqdm(SUB_SET):\n dataset_path = Path(os.path.join(dataset_dir, subset))\n [\n futures.append(executor.submit(deal_with_audio_dir, audio_dir))\n for audio_dir in dataset_path.iterdir()\n if audio_dir.is_dir()\n ]\n for future in tqdm(futures, total=len(futures)):\n sub_result, durations, vocab_set = future.result()\n result.extend(sub_result)\n duration_list.extend(durations)\n text_vocab_set.update(vocab_set)\n executor.shutdown()\n\n # save preprocessed dataset to disk\n if not os.path.exists(f\"{save_dir}\"):\n os.makedirs(f\"{save_dir}\")\n print(f\"\\nSaving to {save_dir} ...\")\n\n with ArrowWriter(path=f\"{save_dir}/raw.arrow\") as writer:\n for line in tqdm(result, desc=\"Writing to raw.arrow ...\"):\n writer.write(line)\n\n # dup a json separately saving duration in case for DynamicBatchSampler ease\n with open(f\"{save_dir}/duration.json\", \"w\", encoding=\"utf-8\") as f:\n json.dump({\"duration\": duration_list}, f, ensure_ascii=False)\n\n # vocab map, i.e. tokenizer\n with open(f\"{save_dir}/vocab.txt\", \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n\n print(f\"\\nFor {dataset_name}, sample count: {len(result)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\")\n print(f\"For {dataset_name}, total {sum(duration_list) / 3600:.2f} hours\")\n\n\nif __name__ == \"__main__\":\n max_workers = 36\n\n tokenizer = \"char\" # \"pinyin\" | \"char\"\n\n SUB_SET = [\"train-clean-100\", \"train-clean-360\", \"train-other-500\"]\n dataset_dir = \"/LibriTTS\"\n dataset_name = f\"LibriTTS_{'_'.join(SUB_SET)}_{tokenizer}\".replace(\"train-clean-\", \"\").replace(\"train-other-\", \"\")\n save_dir = str(files(\"f5_tts\").joinpath(\"../../\")) + f\"/data/{dataset_name}\"\n print(f\"\\nPrepare for {dataset_name}, will save to {save_dir}\\n\")\n main()\n\n # For LibriTTS_100_360_500_char, sample count: 354218\n # For LibriTTS_100_360_500_char, vocab size is: 78\n # For LibriTTS_100_360_500_char, total 554.09 hours","source_hash":"b3d6cc06b84e9c295a75bc558141f87c1328b6d50fac57419a39d4658f382a20","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_libritts.deal_with_audio_dir","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_libritts.deal_with_audio_dir#L17-L31","kind":"function","name":"deal_with_audio_dir","path":"src/f5_tts/train/datasets/prepare_libritts.py","language":"python","start_line":17,"end_line":31,"context_start_line":1,"context_end_line":51,"code":"import os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport json\nfrom concurrent.futures import ProcessPoolExecutor\nfrom importlib.resources import files\nfrom pathlib import Path\n\nimport soundfile as sf\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm\n\n\ndef deal_with_audio_dir(audio_dir):\n sub_result, durations = [], []\n vocab_set = set()\n audio_lists = list(audio_dir.rglob(\"*.wav\"))\n\n for line in audio_lists:\n text_path = line.with_suffix(\".normalized.txt\")\n text = open(text_path, \"r\").read().strip()\n duration = sf.info(line).duration\n if duration < 0.4 or duration > 30:\n continue\n sub_result.append({\"audio_path\": str(line), \"text\": text, \"duration\": duration})\n durations.append(duration)\n vocab_set.update(list(text))\n return sub_result, durations, vocab_set\n\n\ndef main():\n result = []\n duration_list = []\n text_vocab_set = set()\n\n # process raw data\n executor = ProcessPoolExecutor(max_workers=max_workers)\n futures = []\n\n for subset in tqdm(SUB_SET):\n dataset_path = Path(os.path.join(dataset_dir, subset))\n [\n futures.append(executor.submit(deal_with_audio_dir, audio_dir))\n for audio_dir in dataset_path.iterdir()\n if audio_dir.is_dir()\n ]\n for future in tqdm(futures, total=len(futures)):\n sub_result, durations, vocab_set = future.result()","source_hash":"b3d6cc06b84e9c295a75bc558141f87c1328b6d50fac57419a39d4658f382a20","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_libritts.main","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_libritts.main#L34-L77","kind":"function","name":"main","path":"src/f5_tts/train/datasets/prepare_libritts.py","language":"python","start_line":34,"end_line":77,"context_start_line":14,"context_end_line":94,"code":"from tqdm import tqdm\n\n\ndef deal_with_audio_dir(audio_dir):\n sub_result, durations = [], []\n vocab_set = set()\n audio_lists = list(audio_dir.rglob(\"*.wav\"))\n\n for line in audio_lists:\n text_path = line.with_suffix(\".normalized.txt\")\n text = open(text_path, \"r\").read().strip()\n duration = sf.info(line).duration\n if duration < 0.4 or duration > 30:\n continue\n sub_result.append({\"audio_path\": str(line), \"text\": text, \"duration\": duration})\n durations.append(duration)\n vocab_set.update(list(text))\n return sub_result, durations, vocab_set\n\n\ndef main():\n result = []\n duration_list = []\n text_vocab_set = set()\n\n # process raw data\n executor = ProcessPoolExecutor(max_workers=max_workers)\n futures = []\n\n for subset in tqdm(SUB_SET):\n dataset_path = Path(os.path.join(dataset_dir, subset))\n [\n futures.append(executor.submit(deal_with_audio_dir, audio_dir))\n for audio_dir in dataset_path.iterdir()\n if audio_dir.is_dir()\n ]\n for future in tqdm(futures, total=len(futures)):\n sub_result, durations, vocab_set = future.result()\n result.extend(sub_result)\n duration_list.extend(durations)\n text_vocab_set.update(vocab_set)\n executor.shutdown()\n\n # save preprocessed dataset to disk\n if not os.path.exists(f\"{save_dir}\"):\n os.makedirs(f\"{save_dir}\")\n print(f\"\\nSaving to {save_dir} ...\")\n\n with ArrowWriter(path=f\"{save_dir}/raw.arrow\") as writer:\n for line in tqdm(result, desc=\"Writing to raw.arrow ...\"):\n writer.write(line)\n\n # dup a json separately saving duration in case for DynamicBatchSampler ease\n with open(f\"{save_dir}/duration.json\", \"w\", encoding=\"utf-8\") as f:\n json.dump({\"duration\": duration_list}, f, ensure_ascii=False)\n\n # vocab map, i.e. tokenizer\n with open(f\"{save_dir}/vocab.txt\", \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n\n print(f\"\\nFor {dataset_name}, sample count: {len(result)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\")\n print(f\"For {dataset_name}, total {sum(duration_list) / 3600:.2f} hours\")\n\n\nif __name__ == \"__main__\":\n max_workers = 36\n\n tokenizer = \"char\" # \"pinyin\" | \"char\"\n\n SUB_SET = [\"train-clean-100\", \"train-clean-360\", \"train-other-500\"]\n dataset_dir = \"/LibriTTS\"\n dataset_name = f\"LibriTTS_{'_'.join(SUB_SET)}_{tokenizer}\".replace(\"train-clean-\", \"\").replace(\"train-other-\", \"\")\n save_dir = str(files(\"f5_tts\").joinpath(\"../../\")) + f\"/data/{dataset_name}\"\n print(f\"\\nPrepare for {dataset_name}, will save to {save_dir}\\n\")\n main()\n\n # For LibriTTS_100_360_500_char, sample count: 354218\n # For LibriTTS_100_360_500_char, vocab size is: 78\n # For LibriTTS_100_360_500_char, total 554.09 hours","source_hash":"b3d6cc06b84e9c295a75bc558141f87c1328b6d50fac57419a39d4658f382a20","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_ljspeech","uri":"program://DMOSpeech2/module/src.f5_tts.train.datasets.prepare_ljspeech#L1-L67","kind":"module","name":"src.f5_tts.train.datasets.prepare_ljspeech","path":"src/f5_tts/train/datasets/prepare_ljspeech.py","language":"python","start_line":1,"end_line":67,"context_start_line":1,"context_end_line":67,"code":"import os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport json\nfrom importlib.resources import files\nfrom pathlib import Path\n\nimport soundfile as sf\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm\n\n\ndef main():\n result = []\n duration_list = []\n text_vocab_set = set()\n\n with open(meta_info, \"r\") as f:\n lines = f.readlines()\n for line in tqdm(lines):\n uttr, text, norm_text = line.split(\"|\")\n norm_text = norm_text.strip()\n wav_path = Path(dataset_dir) / \"wavs\" / f\"{uttr}.wav\"\n duration = sf.info(wav_path).duration\n if duration < 0.4 or duration > 30:\n continue\n result.append({\"audio_path\": str(wav_path), \"text\": norm_text, \"duration\": duration})\n duration_list.append(duration)\n text_vocab_set.update(list(norm_text))\n\n # save preprocessed dataset to disk\n if not os.path.exists(f\"{save_dir}\"):\n os.makedirs(f\"{save_dir}\")\n print(f\"\\nSaving to {save_dir} ...\")\n\n with ArrowWriter(path=f\"{save_dir}/raw.arrow\") as writer:\n for line in tqdm(result, desc=\"Writing to raw.arrow ...\"):\n writer.write(line)\n\n # dup a json separately saving duration in case for DynamicBatchSampler ease\n with open(f\"{save_dir}/duration.json\", \"w\", encoding=\"utf-8\") as f:\n json.dump({\"duration\": duration_list}, f, ensure_ascii=False)\n\n # vocab map, i.e. tokenizer\n # add alphabets and symbols (optional, if plan to ft on de/fr etc.)\n with open(f\"{save_dir}/vocab.txt\", \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n\n print(f\"\\nFor {dataset_name}, sample count: {len(result)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\")\n print(f\"For {dataset_name}, total {sum(duration_list) / 3600:.2f} hours\")\n\n\nif __name__ == \"__main__\":\n tokenizer = \"char\" # \"pinyin\" | \"char\"\n\n dataset_dir = \"/LJSpeech-1.1\"\n dataset_name = f\"LJSpeech_{tokenizer}\"\n meta_info = os.path.join(dataset_dir, \"metadata.csv\")\n save_dir = str(files(\"f5_tts\").joinpath(\"../../\")) + f\"/data/{dataset_name}\"\n print(f\"\\nPrepare for {dataset_name}, will save to {save_dir}\\n\")\n\n main()","source_hash":"7f03a0819517ade8a7722e4ae5f0378b3b790ea4878e226ab032d9910c608750","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_ljspeech.main","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_ljspeech.main#L16-L55","kind":"function","name":"main","path":"src/f5_tts/train/datasets/prepare_ljspeech.py","language":"python","start_line":16,"end_line":55,"context_start_line":1,"context_end_line":67,"code":"import os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport json\nfrom importlib.resources import files\nfrom pathlib import Path\n\nimport soundfile as sf\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm\n\n\ndef main():\n result = []\n duration_list = []\n text_vocab_set = set()\n\n with open(meta_info, \"r\") as f:\n lines = f.readlines()\n for line in tqdm(lines):\n uttr, text, norm_text = line.split(\"|\")\n norm_text = norm_text.strip()\n wav_path = Path(dataset_dir) / \"wavs\" / f\"{uttr}.wav\"\n duration = sf.info(wav_path).duration\n if duration < 0.4 or duration > 30:\n continue\n result.append({\"audio_path\": str(wav_path), \"text\": norm_text, \"duration\": duration})\n duration_list.append(duration)\n text_vocab_set.update(list(norm_text))\n\n # save preprocessed dataset to disk\n if not os.path.exists(f\"{save_dir}\"):\n os.makedirs(f\"{save_dir}\")\n print(f\"\\nSaving to {save_dir} ...\")\n\n with ArrowWriter(path=f\"{save_dir}/raw.arrow\") as writer:\n for line in tqdm(result, desc=\"Writing to raw.arrow ...\"):\n writer.write(line)\n\n # dup a json separately saving duration in case for DynamicBatchSampler ease\n with open(f\"{save_dir}/duration.json\", \"w\", encoding=\"utf-8\") as f:\n json.dump({\"duration\": duration_list}, f, ensure_ascii=False)\n\n # vocab map, i.e. tokenizer\n # add alphabets and symbols (optional, if plan to ft on de/fr etc.)\n with open(f\"{save_dir}/vocab.txt\", \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n\n print(f\"\\nFor {dataset_name}, sample count: {len(result)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\")\n print(f\"For {dataset_name}, total {sum(duration_list) / 3600:.2f} hours\")\n\n\nif __name__ == \"__main__\":\n tokenizer = \"char\" # \"pinyin\" | \"char\"\n\n dataset_dir = \"/LJSpeech-1.1\"\n dataset_name = f\"LJSpeech_{tokenizer}\"\n meta_info = os.path.join(dataset_dir, \"metadata.csv\")\n save_dir = str(files(\"f5_tts\").joinpath(\"../../\")) + f\"/data/{dataset_name}\"\n print(f\"\\nPrepare for {dataset_name}, will save to {save_dir}\\n\")\n\n main()","source_hash":"7f03a0819517ade8a7722e4ae5f0378b3b790ea4878e226ab032d9910c608750","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_emilia_v2","uri":"program://DMOSpeech2/module/src.f5_tts.train.datasets.prepare_emilia_v2#L1-L94","kind":"module","name":"src.f5_tts.train.datasets.prepare_emilia_v2","path":"src/f5_tts/train/datasets/prepare_emilia_v2.py","language":"python","start_line":1,"end_line":94,"context_start_line":1,"context_end_line":94,"code":"# put in src/f5_tts/train/datasets/prepare_emilia_v2.py\n# prepares Emilia dataset with the new format w/ Emilia-YODAS\n\nimport json\nimport os\nfrom concurrent.futures import ProcessPoolExecutor\nfrom importlib.resources import files\nfrom pathlib import Path\n\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm\n\nfrom f5_tts.model.utils import repetition_found\n\n\n# Define filters for exclusion\nout_en = set()\nen_filters = [\"ا\", \"い\", \"て\"]\n\n\ndef process_audio_directory(audio_dir):\n sub_result, durations, vocab_set = [], [], set()\n bad_case_en = 0\n\n for file in audio_dir.iterdir():\n if file.suffix == \".json\":\n with open(file, \"r\") as f:\n obj = json.load(f)\n text = obj[\"text\"]\n if any(f in text for f in en_filters) or repetition_found(text, length=4):\n bad_case_en += 1\n continue\n\n duration = obj[\"duration\"]\n audio_file = file.with_suffix(\".mp3\")\n if audio_file.exists():\n sub_result.append({\"audio_path\": str(audio_file), \"text\": text, \"duration\": duration})\n durations.append(duration)\n vocab_set.update(list(text))\n\n return sub_result, durations, vocab_set, bad_case_en\n\n\ndef main():\n assert tokenizer in [\"pinyin\", \"char\"]\n result, duration_list, text_vocab_set = [], [], set()\n total_bad_case_en = 0\n\n executor = ProcessPoolExecutor(max_workers=max_workers)\n futures = []\n dataset_path = Path(dataset_dir)\n for sub_dir in dataset_path.iterdir():\n if sub_dir.is_dir():\n futures.append(executor.submit(process_audio_directory, sub_dir))\n\n for future in tqdm(futures, total=len(futures)):\n sub_result, durations, vocab_set, bad_case_en = future.result()\n result.extend(sub_result)\n duration_list.extend(durations)\n text_vocab_set.update(vocab_set)\n total_bad_case_en += bad_case_en\n\n executor.shutdown()\n\n if not os.path.exists(f\"{save_dir}\"):\n os.makedirs(f\"{save_dir}\")\n\n with ArrowWriter(path=f\"{save_dir}/raw.arrow\") as writer:\n for line in tqdm(result, desc=\"Writing to raw.arrow ...\"):\n writer.write(line)\n\n with open(f\"{save_dir}/duration.json\", \"w\", encoding=\"utf-8\") as f:\n json.dump({\"duration\": duration_list}, f, ensure_ascii=False)\n\n with open(f\"{save_dir}/vocab.txt\", \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n\n print(f\"For {dataset_name}, sample count: {len(result)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\")\n print(f\"For {dataset_name}, total {sum(duration_list) / 3600:.2f} hours\")\n print(f\"Bad en transcription case: {total_bad_case_en}\\n\")\n\n\nif __name__ == \"__main__\":\n max_workers = 32\n tokenizer = \"char\"\n dataset_dir = \"/home/ubuntu/emilia-dataset/Emilia-YODAS/EN\"\n dataset_name = f\"Emilia_EN_{tokenizer}\"\n # save_dir = os.path.expanduser(f\"~/F5-TTS/data/{dataset_name}\")\n save_dir = str(files(\"f5_tts\").joinpath(\"../../\")) + f\"/data/{dataset_name}\"\n\n print(f\"Prepare for {dataset_name}, will save to {save_dir}\\n\")\n main()","source_hash":"3757de14aff828d561c375a976c6c9a52ed08492d1e9c99d77df9c5673f50174","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_emilia_v2.process_audio_directory","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_emilia_v2.process_audio_directory#L21-L41","kind":"function","name":"process_audio_directory","path":"src/f5_tts/train/datasets/prepare_emilia_v2.py","language":"python","start_line":21,"end_line":41,"context_start_line":1,"context_end_line":61,"code":"# put in src/f5_tts/train/datasets/prepare_emilia_v2.py\n# prepares Emilia dataset with the new format w/ Emilia-YODAS\n\nimport json\nimport os\nfrom concurrent.futures import ProcessPoolExecutor\nfrom importlib.resources import files\nfrom pathlib import Path\n\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm\n\nfrom f5_tts.model.utils import repetition_found\n\n\n# Define filters for exclusion\nout_en = set()\nen_filters = [\"ا\", \"い\", \"て\"]\n\n\ndef process_audio_directory(audio_dir):\n sub_result, durations, vocab_set = [], [], set()\n bad_case_en = 0\n\n for file in audio_dir.iterdir():\n if file.suffix == \".json\":\n with open(file, \"r\") as f:\n obj = json.load(f)\n text = obj[\"text\"]\n if any(f in text for f in en_filters) or repetition_found(text, length=4):\n bad_case_en += 1\n continue\n\n duration = obj[\"duration\"]\n audio_file = file.with_suffix(\".mp3\")\n if audio_file.exists():\n sub_result.append({\"audio_path\": str(audio_file), \"text\": text, \"duration\": duration})\n durations.append(duration)\n vocab_set.update(list(text))\n\n return sub_result, durations, vocab_set, bad_case_en\n\n\ndef main():\n assert tokenizer in [\"pinyin\", \"char\"]\n result, duration_list, text_vocab_set = [], [], set()\n total_bad_case_en = 0\n\n executor = ProcessPoolExecutor(max_workers=max_workers)\n futures = []\n dataset_path = Path(dataset_dir)\n for sub_dir in dataset_path.iterdir():\n if sub_dir.is_dir():\n futures.append(executor.submit(process_audio_directory, sub_dir))\n\n for future in tqdm(futures, total=len(futures)):\n sub_result, durations, vocab_set, bad_case_en = future.result()\n result.extend(sub_result)\n duration_list.extend(durations)\n text_vocab_set.update(vocab_set)\n total_bad_case_en += bad_case_en","source_hash":"3757de14aff828d561c375a976c6c9a52ed08492d1e9c99d77df9c5673f50174","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_emilia_v2.main","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_emilia_v2.main#L44-L82","kind":"function","name":"main","path":"src/f5_tts/train/datasets/prepare_emilia_v2.py","language":"python","start_line":44,"end_line":82,"context_start_line":24,"context_end_line":94,"code":"\n for file in audio_dir.iterdir():\n if file.suffix == \".json\":\n with open(file, \"r\") as f:\n obj = json.load(f)\n text = obj[\"text\"]\n if any(f in text for f in en_filters) or repetition_found(text, length=4):\n bad_case_en += 1\n continue\n\n duration = obj[\"duration\"]\n audio_file = file.with_suffix(\".mp3\")\n if audio_file.exists():\n sub_result.append({\"audio_path\": str(audio_file), \"text\": text, \"duration\": duration})\n durations.append(duration)\n vocab_set.update(list(text))\n\n return sub_result, durations, vocab_set, bad_case_en\n\n\ndef main():\n assert tokenizer in [\"pinyin\", \"char\"]\n result, duration_list, text_vocab_set = [], [], set()\n total_bad_case_en = 0\n\n executor = ProcessPoolExecutor(max_workers=max_workers)\n futures = []\n dataset_path = Path(dataset_dir)\n for sub_dir in dataset_path.iterdir():\n if sub_dir.is_dir():\n futures.append(executor.submit(process_audio_directory, sub_dir))\n\n for future in tqdm(futures, total=len(futures)):\n sub_result, durations, vocab_set, bad_case_en = future.result()\n result.extend(sub_result)\n duration_list.extend(durations)\n text_vocab_set.update(vocab_set)\n total_bad_case_en += bad_case_en\n\n executor.shutdown()\n\n if not os.path.exists(f\"{save_dir}\"):\n os.makedirs(f\"{save_dir}\")\n\n with ArrowWriter(path=f\"{save_dir}/raw.arrow\") as writer:\n for line in tqdm(result, desc=\"Writing to raw.arrow ...\"):\n writer.write(line)\n\n with open(f\"{save_dir}/duration.json\", \"w\", encoding=\"utf-8\") as f:\n json.dump({\"duration\": duration_list}, f, ensure_ascii=False)\n\n with open(f\"{save_dir}/vocab.txt\", \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n\n print(f\"For {dataset_name}, sample count: {len(result)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\")\n print(f\"For {dataset_name}, total {sum(duration_list) / 3600:.2f} hours\")\n print(f\"Bad en transcription case: {total_bad_case_en}\\n\")\n\n\nif __name__ == \"__main__\":\n max_workers = 32\n tokenizer = \"char\"\n dataset_dir = \"/home/ubuntu/emilia-dataset/Emilia-YODAS/EN\"\n dataset_name = f\"Emilia_EN_{tokenizer}\"\n # save_dir = os.path.expanduser(f\"~/F5-TTS/data/{dataset_name}\")\n save_dir = str(files(\"f5_tts\").joinpath(\"../../\")) + f\"/data/{dataset_name}\"\n\n print(f\"Prepare for {dataset_name}, will save to {save_dir}\\n\")\n main()","source_hash":"3757de14aff828d561c375a976c6c9a52ed08492d1e9c99d77df9c5673f50174","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_wenetspeech4tts","uri":"program://DMOSpeech2/module/src.f5_tts.train.datasets.prepare_wenetspeech4tts#L1-L126","kind":"module","name":"src.f5_tts.train.datasets.prepare_wenetspeech4tts","path":"src/f5_tts/train/datasets/prepare_wenetspeech4tts.py","language":"python","start_line":1,"end_line":126,"context_start_line":1,"context_end_line":126,"code":"# generate audio text map for WenetSpeech4TTS\n# evaluate for vocab size\n\nimport os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport json\nfrom concurrent.futures import ProcessPoolExecutor\nfrom importlib.resources import files\n\nimport torchaudio\nfrom datasets import Dataset\nfrom tqdm import tqdm\n\nfrom f5_tts.model.utils import convert_char_to_pinyin\n\n\ndef deal_with_sub_path_files(dataset_path, sub_path):\n print(f\"Dealing with: {sub_path}\")\n\n text_dir = os.path.join(dataset_path, sub_path, \"txts\")\n audio_dir = os.path.join(dataset_path, sub_path, \"wavs\")\n text_files = os.listdir(text_dir)\n\n audio_paths, texts, durations = [], [], []\n for text_file in tqdm(text_files):\n with open(os.path.join(text_dir, text_file), \"r\", encoding=\"utf-8\") as file:\n first_line = file.readline().split(\"\\t\")\n audio_nm = first_line[0]\n audio_path = os.path.join(audio_dir, audio_nm + \".wav\")\n text = first_line[1].strip()\n\n audio_paths.append(audio_path)\n\n if tokenizer == \"pinyin\":\n texts.extend(convert_char_to_pinyin([text], polyphone=polyphone))\n elif tokenizer == \"char\":\n texts.append(text)\n\n audio, sample_rate = torchaudio.load(audio_path)\n durations.append(audio.shape[-1] / sample_rate)\n\n return audio_paths, texts, durations\n\n\ndef main():\n assert tokenizer in [\"pinyin\", \"char\"]\n\n audio_path_list, text_list, duration_list = [], [], []\n\n executor = ProcessPoolExecutor(max_workers=max_workers)\n futures = []\n for dataset_path in dataset_paths:\n sub_items = os.listdir(dataset_path)\n sub_paths = [item for item in sub_items if os.path.isdir(os.path.join(dataset_path, item))]\n for sub_path in sub_paths:\n futures.append(executor.submit(deal_with_sub_path_files, dataset_path, sub_path))\n for future in tqdm(futures, total=len(futures)):\n audio_paths, texts, durations = future.result()\n audio_path_list.extend(audio_paths)\n text_list.extend(texts)\n duration_list.extend(durations)\n executor.shutdown()\n\n if not os.path.exists(\"data\"):\n os.makedirs(\"data\")\n\n print(f\"\\nSaving to {save_dir} ...\")\n dataset = Dataset.from_dict({\"audio_path\": audio_path_list, \"text\": text_list, \"duration\": duration_list})\n dataset.save_to_disk(f\"{save_dir}/raw\", max_shard_size=\"2GB\") # arrow format\n\n with open(f\"{save_dir}/duration.json\", \"w\", encoding=\"utf-8\") as f:\n json.dump(\n {\"duration\": duration_list}, f, ensure_ascii=False\n ) # dup a json separately saving duration in case for DynamicBatchSampler ease\n\n print(\"\\nEvaluating vocab size (all characters and symbols / all phonemes) ...\")\n text_vocab_set = set()\n for text in tqdm(text_list):\n text_vocab_set.update(list(text))\n\n # add alphabets and symbols (optional, if plan to ft on de/fr etc.)\n if tokenizer == \"pinyin\":\n text_vocab_set.update([chr(i) for i in range(32, 127)] + [chr(i) for i in range(192, 256)])\n\n with open(f\"{save_dir}/vocab.txt\", \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n print(f\"\\nFor {dataset_name}, sample count: {len(text_list)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\\n\")\n\n\nif __name__ == \"__main__\":\n max_workers = 32\n\n tokenizer = \"pinyin\" # \"pinyin\" | \"char\"\n polyphone = True\n dataset_choice = 1 # 1: Premium, 2: Standard, 3: Basic\n\n dataset_name = (\n [\"WenetSpeech4TTS_Premium\", \"WenetSpeech4TTS_Standard\", \"WenetSpeech4TTS_Basic\"][dataset_choice - 1]\n + \"_\"\n + tokenizer\n )\n dataset_paths = [\n \"/WenetSpeech4TTS/Basic\",\n \"/WenetSpeech4TTS/Standard\",\n \"/WenetSpeech4TTS/Premium\",\n ][-dataset_choice:]\n save_dir = str(files(\"f5_tts\").joinpath(\"../../\")) + f\"/data/{dataset_name}\"\n print(f\"\\nChoose Dataset: {dataset_name}, will save to {save_dir}\\n\")\n\n main()\n\n # Results (if adding alphabets with accents and symbols):\n # WenetSpeech4TTS Basic Standard Premium\n # samples count 3932473 1941220 407494\n # pinyin vocab size 1349 1348 1344 (no polyphone)\n # - - 1459 (polyphone)\n # char vocab size 5264 5219 5042\n\n # vocab size may be slightly different due to jieba tokenizer and pypinyin (e.g. way of polyphoneme)\n # please be careful if using pretrained model, make sure the vocab.txt is same","source_hash":"75f27f2a8632386e881a40a58c865011761c8d5f2cdf7d9fe0d2e97e2b896cb2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_wenetspeech4tts.deal_with_sub_path_files","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_wenetspeech4tts.deal_with_sub_path_files#L21-L46","kind":"function","name":"deal_with_sub_path_files","path":"src/f5_tts/train/datasets/prepare_wenetspeech4tts.py","language":"python","start_line":21,"end_line":46,"context_start_line":1,"context_end_line":66,"code":"# generate audio text map for WenetSpeech4TTS\n# evaluate for vocab size\n\nimport os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport json\nfrom concurrent.futures import ProcessPoolExecutor\nfrom importlib.resources import files\n\nimport torchaudio\nfrom datasets import Dataset\nfrom tqdm import tqdm\n\nfrom f5_tts.model.utils import convert_char_to_pinyin\n\n\ndef deal_with_sub_path_files(dataset_path, sub_path):\n print(f\"Dealing with: {sub_path}\")\n\n text_dir = os.path.join(dataset_path, sub_path, \"txts\")\n audio_dir = os.path.join(dataset_path, sub_path, \"wavs\")\n text_files = os.listdir(text_dir)\n\n audio_paths, texts, durations = [], [], []\n for text_file in tqdm(text_files):\n with open(os.path.join(text_dir, text_file), \"r\", encoding=\"utf-8\") as file:\n first_line = file.readline().split(\"\\t\")\n audio_nm = first_line[0]\n audio_path = os.path.join(audio_dir, audio_nm + \".wav\")\n text = first_line[1].strip()\n\n audio_paths.append(audio_path)\n\n if tokenizer == \"pinyin\":\n texts.extend(convert_char_to_pinyin([text], polyphone=polyphone))\n elif tokenizer == \"char\":\n texts.append(text)\n\n audio, sample_rate = torchaudio.load(audio_path)\n durations.append(audio.shape[-1] / sample_rate)\n\n return audio_paths, texts, durations\n\n\ndef main():\n assert tokenizer in [\"pinyin\", \"char\"]\n\n audio_path_list, text_list, duration_list = [], [], []\n\n executor = ProcessPoolExecutor(max_workers=max_workers)\n futures = []\n for dataset_path in dataset_paths:\n sub_items = os.listdir(dataset_path)\n sub_paths = [item for item in sub_items if os.path.isdir(os.path.join(dataset_path, item))]\n for sub_path in sub_paths:\n futures.append(executor.submit(deal_with_sub_path_files, dataset_path, sub_path))\n for future in tqdm(futures, total=len(futures)):\n audio_paths, texts, durations = future.result()\n audio_path_list.extend(audio_paths)\n text_list.extend(texts)\n duration_list.extend(durations)\n executor.shutdown()","source_hash":"75f27f2a8632386e881a40a58c865011761c8d5f2cdf7d9fe0d2e97e2b896cb2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_wenetspeech4tts.main","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_wenetspeech4tts.main#L49-L93","kind":"function","name":"main","path":"src/f5_tts/train/datasets/prepare_wenetspeech4tts.py","language":"python","start_line":49,"end_line":93,"context_start_line":29,"context_end_line":113,"code":" for text_file in tqdm(text_files):\n with open(os.path.join(text_dir, text_file), \"r\", encoding=\"utf-8\") as file:\n first_line = file.readline().split(\"\\t\")\n audio_nm = first_line[0]\n audio_path = os.path.join(audio_dir, audio_nm + \".wav\")\n text = first_line[1].strip()\n\n audio_paths.append(audio_path)\n\n if tokenizer == \"pinyin\":\n texts.extend(convert_char_to_pinyin([text], polyphone=polyphone))\n elif tokenizer == \"char\":\n texts.append(text)\n\n audio, sample_rate = torchaudio.load(audio_path)\n durations.append(audio.shape[-1] / sample_rate)\n\n return audio_paths, texts, durations\n\n\ndef main():\n assert tokenizer in [\"pinyin\", \"char\"]\n\n audio_path_list, text_list, duration_list = [], [], []\n\n executor = ProcessPoolExecutor(max_workers=max_workers)\n futures = []\n for dataset_path in dataset_paths:\n sub_items = os.listdir(dataset_path)\n sub_paths = [item for item in sub_items if os.path.isdir(os.path.join(dataset_path, item))]\n for sub_path in sub_paths:\n futures.append(executor.submit(deal_with_sub_path_files, dataset_path, sub_path))\n for future in tqdm(futures, total=len(futures)):\n audio_paths, texts, durations = future.result()\n audio_path_list.extend(audio_paths)\n text_list.extend(texts)\n duration_list.extend(durations)\n executor.shutdown()\n\n if not os.path.exists(\"data\"):\n os.makedirs(\"data\")\n\n print(f\"\\nSaving to {save_dir} ...\")\n dataset = Dataset.from_dict({\"audio_path\": audio_path_list, \"text\": text_list, \"duration\": duration_list})\n dataset.save_to_disk(f\"{save_dir}/raw\", max_shard_size=\"2GB\") # arrow format\n\n with open(f\"{save_dir}/duration.json\", \"w\", encoding=\"utf-8\") as f:\n json.dump(\n {\"duration\": duration_list}, f, ensure_ascii=False\n ) # dup a json separately saving duration in case for DynamicBatchSampler ease\n\n print(\"\\nEvaluating vocab size (all characters and symbols / all phonemes) ...\")\n text_vocab_set = set()\n for text in tqdm(text_list):\n text_vocab_set.update(list(text))\n\n # add alphabets and symbols (optional, if plan to ft on de/fr etc.)\n if tokenizer == \"pinyin\":\n text_vocab_set.update([chr(i) for i in range(32, 127)] + [chr(i) for i in range(192, 256)])\n\n with open(f\"{save_dir}/vocab.txt\", \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n print(f\"\\nFor {dataset_name}, sample count: {len(text_list)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\\n\")\n\n\nif __name__ == \"__main__\":\n max_workers = 32\n\n tokenizer = \"pinyin\" # \"pinyin\" | \"char\"\n polyphone = True\n dataset_choice = 1 # 1: Premium, 2: Standard, 3: Basic\n\n dataset_name = (\n [\"WenetSpeech4TTS_Premium\", \"WenetSpeech4TTS_Standard\", \"WenetSpeech4TTS_Basic\"][dataset_choice - 1]\n + \"_\"\n + tokenizer\n )\n dataset_paths = [\n \"/WenetSpeech4TTS/Basic\",\n \"/WenetSpeech4TTS/Standard\",\n \"/WenetSpeech4TTS/Premium\",\n ][-dataset_choice:]\n save_dir = str(files(\"f5_tts\").joinpath(\"../../\")) + f\"/data/{dataset_name}\"","source_hash":"75f27f2a8632386e881a40a58c865011761c8d5f2cdf7d9fe0d2e97e2b896cb2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_emilia","uri":"program://DMOSpeech2/module/src.f5_tts.train.datasets.prepare_emilia#L1-L228","kind":"module","name":"src.f5_tts.train.datasets.prepare_emilia","path":"src/f5_tts/train/datasets/prepare_emilia.py","language":"python","start_line":1,"end_line":228,"context_start_line":1,"context_end_line":228,"code":"# Emilia Dataset: https://huggingface.co/datasets/amphion/Emilia-Dataset/tree/fc71e07\n# if use updated new version, i.e. WebDataset, feel free to modify / draft your own script\n\n# generate audio text map for Emilia ZH & EN\n# evaluate for vocab size\n\nimport os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport json\nfrom concurrent.futures import ProcessPoolExecutor\nfrom importlib.resources import files\nfrom pathlib import Path\n\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm\n\nfrom f5_tts.model.utils import convert_char_to_pinyin, repetition_found\n\n\nout_zh = {\n \"ZH_B00041_S06226\",\n \"ZH_B00042_S09204\",\n \"ZH_B00065_S09430\",\n \"ZH_B00065_S09431\",\n \"ZH_B00066_S09327\",\n \"ZH_B00066_S09328\",\n}\nzh_filters = [\"い\", \"て\"]\n# seems synthesized audios, or heavily code-switched\nout_en = {\n \"EN_B00013_S00913\",\n \"EN_B00042_S00120\",\n \"EN_B00055_S04111\",\n \"EN_B00061_S00693\",\n \"EN_B00061_S01494\",\n \"EN_B00061_S03375\",\n \"EN_B00059_S00092\",\n \"EN_B00111_S04300\",\n \"EN_B00100_S03759\",\n \"EN_B00087_S03811\",\n \"EN_B00059_S00950\",\n \"EN_B00089_S00946\",\n \"EN_B00078_S05127\",\n \"EN_B00070_S04089\",\n \"EN_B00074_S09659\",\n \"EN_B00061_S06983\",\n \"EN_B00061_S07060\",\n \"EN_B00059_S08397\",\n \"EN_B00082_S06192\",\n \"EN_B00091_S01238\",\n \"EN_B00089_S07349\",\n \"EN_B00070_S04343\",\n \"EN_B00061_S02400\",\n \"EN_B00076_S01262\",\n \"EN_B00068_S06467\",\n \"EN_B00076_S02943\",\n \"EN_B00064_S05954\",\n \"EN_B00061_S05386\",\n \"EN_B00066_S06544\",\n \"EN_B00076_S06944\",\n \"EN_B00072_S08620\",\n \"EN_B00076_S07135\",\n \"EN_B00076_S09127\",\n \"EN_B00065_S00497\",\n \"EN_B00059_S06227\",\n \"EN_B00063_S02859\",\n \"EN_B00075_S01547\",\n \"EN_B00061_S08286\",\n \"EN_B00079_S02901\",\n \"EN_B00092_S03643\",\n \"EN_B00096_S08653\",\n \"EN_B00063_S04297\",\n \"EN_B00063_S04614\",\n \"EN_B00079_S04698\",\n \"EN_B00104_S01666\",\n \"EN_B00061_S09504\",\n \"EN_B00061_S09694\",\n \"EN_B00065_S05444\",\n \"EN_B00063_S06860\",\n \"EN_B00065_S05725\",\n \"EN_B00069_S07628\",\n \"EN_B00083_S03875\",\n \"EN_B00071_S07665\",\n \"EN_B00071_S07665\",\n \"EN_B00062_S04187\",\n \"EN_B00065_S09873\",\n \"EN_B00065_S09922\",\n \"EN_B00084_S02463\",\n \"EN_B00067_S05066\",\n \"EN_B00106_S08060\",\n \"EN_B00073_S06399\",\n \"EN_B00073_S09236\",\n \"EN_B00087_S00432\",\n \"EN_B00085_S05618\",\n \"EN_B00064_S01262\",\n \"EN_B00072_S01739\",\n \"EN_B00059_S03913\",\n \"EN_B00069_S04036\",\n \"EN_B00067_S05623\",\n \"EN_B00060_S05389\",\n \"EN_B00060_S07290\",\n \"EN_B00062_S08995\",\n}\nen_filters = [\"ا\", \"い\", \"て\"]\n\n\ndef deal_with_audio_dir(audio_dir):\n audio_jsonl = audio_dir.with_suffix(\".jsonl\")\n sub_result, durations = [], []\n vocab_set = set()\n bad_case_zh = 0\n bad_case_en = 0\n with open(audio_jsonl, \"r\") as f:\n lines = f.readlines()\n for line in tqdm(lines, desc=f\"{audio_jsonl.stem}\"):\n obj = json.loads(line)\n text = obj[\"text\"]\n if obj[\"language\"] == \"zh\":\n if obj[\"wav\"].split(\"/\")[1] in out_zh or any(f in text for f in zh_filters) or repetition_found(text):\n bad_case_zh += 1\n continue\n else:\n text = text.translate(\n str.maketrans({\",\": \"，\", \"!\": \"！\", \"?\": \"？\"})\n ) # not \"。\" cuz much code-switched\n if obj[\"language\"] == \"en\":\n if (\n obj[\"wav\"].split(\"/\")[1] in out_en\n or any(f in text for f in en_filters)\n or repetition_found(text, length=4)\n ):\n bad_case_en += 1\n continue\n if tokenizer == \"pinyin\":\n text = convert_char_to_pinyin([text], polyphone=polyphone)[0]\n duration = obj[\"duration\"]\n sub_result.append({\"audio_path\": str(audio_dir.parent / obj[\"wav\"]), \"text\": text, \"duration\": duration})\n durations.append(duration)\n vocab_set.update(list(text))\n return sub_result, durations, vocab_set, bad_case_zh, bad_case_en\n\n\ndef main():\n assert tokenizer in [\"pinyin\", \"char\"]\n result = []\n duration_list = []\n text_vocab_set = set()\n total_bad_case_zh = 0\n total_bad_case_en = 0\n\n # process raw data\n executor = ProcessPoolExecutor(max_workers=max_workers)\n futures = []\n for lang in langs:\n dataset_path = Path(os.path.join(dataset_dir, lang))\n [\n futures.append(executor.submit(deal_with_audio_dir, audio_dir))\n for audio_dir in dataset_path.iterdir()\n if audio_dir.is_dir()\n ]\n for futures in tqdm(futures, total=len(futures)):\n sub_result, durations, vocab_set, bad_case_zh, bad_case_en = futures.result()\n result.extend(sub_result)\n duration_list.extend(durations)\n text_vocab_set.update(vocab_set)\n total_bad_case_zh += bad_case_zh\n total_bad_case_en += bad_case_en\n executor.shutdown()\n\n # save preprocessed dataset to disk\n if not os.path.exists(f\"{save_dir}\"):\n os.makedirs(f\"{save_dir}\")\n print(f\"\\nSaving to {save_dir} ...\")\n\n # dataset = Dataset.from_dict({\"audio_path\": audio_path_list, \"text\": text_list, \"duration\": duration_list}) # oom\n # dataset.save_to_disk(f\"{save_dir}/raw\", max_shard_size=\"2GB\")\n with ArrowWriter(path=f\"{save_dir}/raw.arrow\") as writer:\n for line in tqdm(result, desc=\"Writing to raw.arrow ...\"):\n writer.write(line)\n\n # dup a json separately saving duration in case for DynamicBatchSampler ease\n with open(f\"{save_dir}/duration.json\", \"w\", encoding=\"utf-8\") as f:\n json.dump({\"duration\": duration_list}, f, ensure_ascii=False)\n\n # vocab map, i.e. tokenizer\n # add alphabets and symbols (optional, if plan to ft on de/fr etc.)\n # if tokenizer == \"pinyin\":\n # text_vocab_set.update([chr(i) for i in range(32, 127)] + [chr(i) for i in range(192, 256)])\n with open(f\"{save_dir}/vocab.txt\", \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n\n print(f\"\\nFor {dataset_name}, sample count: {len(result)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\")\n print(f\"For {dataset_name}, total {sum(duration_list) / 3600:.2f} hours\")\n if \"ZH\" in langs:\n print(f\"Bad zh transcription case: {total_bad_case_zh}\")\n if \"EN\" in langs:\n print(f\"Bad en transcription case: {total_bad_case_en}\\n\")\n\n\nif __name__ == \"__main__\":\n max_workers = 32\n\n tokenizer = \"pinyin\" # \"pinyin\" | \"char\"\n polyphone = True\n\n langs = [\"ZH\", \"EN\"]\n dataset_dir = \"/Emilia_Dataset/raw\"\n dataset_name = f\"Emilia_{'_'.join(langs)}_{tokenizer}\"\n save_dir = str(files(\"f5_tts\").joinpath(\"../../\")) + f\"/data/{dataset_name}\"\n print(f\"\\nPrepare for {dataset_name}, will save to {save_dir}\\n\")\n\n main()\n\n # Emilia ZH & EN\n # samples count 37837916 (after removal)\n # pinyin vocab size 2543 (polyphone)\n # total duration 95281.87 (hours)\n # bad zh asr cnt 230435 (samples)\n # bad eh asr cnt 37217 (samples)\n\n # vocab size may be slightly different due to jieba tokenizer and pypinyin (e.g. way of polyphoneme)\n # please be careful if using pretrained model, make sure the vocab.txt is same","source_hash":"418b1ecd9af0922bb02a7bc11617b00dda6b918620148bd8a47df64df3c9ec4e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_emilia.deal_with_audio_dir","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_emilia.deal_with_audio_dir#L111-L144","kind":"function","name":"deal_with_audio_dir","path":"src/f5_tts/train/datasets/prepare_emilia.py","language":"python","start_line":111,"end_line":144,"context_start_line":91,"context_end_line":164,"code":" \"EN_B00065_S09922\",\n \"EN_B00084_S02463\",\n \"EN_B00067_S05066\",\n \"EN_B00106_S08060\",\n \"EN_B00073_S06399\",\n \"EN_B00073_S09236\",\n \"EN_B00087_S00432\",\n \"EN_B00085_S05618\",\n \"EN_B00064_S01262\",\n \"EN_B00072_S01739\",\n \"EN_B00059_S03913\",\n \"EN_B00069_S04036\",\n \"EN_B00067_S05623\",\n \"EN_B00060_S05389\",\n \"EN_B00060_S07290\",\n \"EN_B00062_S08995\",\n}\nen_filters = [\"ا\", \"い\", \"て\"]\n\n\ndef deal_with_audio_dir(audio_dir):\n audio_jsonl = audio_dir.with_suffix(\".jsonl\")\n sub_result, durations = [], []\n vocab_set = set()\n bad_case_zh = 0\n bad_case_en = 0\n with open(audio_jsonl, \"r\") as f:\n lines = f.readlines()\n for line in tqdm(lines, desc=f\"{audio_jsonl.stem}\"):\n obj = json.loads(line)\n text = obj[\"text\"]\n if obj[\"language\"] == \"zh\":\n if obj[\"wav\"].split(\"/\")[1] in out_zh or any(f in text for f in zh_filters) or repetition_found(text):\n bad_case_zh += 1\n continue\n else:\n text = text.translate(\n str.maketrans({\",\": \"，\", \"!\": \"！\", \"?\": \"？\"})\n ) # not \"。\" cuz much code-switched\n if obj[\"language\"] == \"en\":\n if (\n obj[\"wav\"].split(\"/\")[1] in out_en\n or any(f in text for f in en_filters)\n or repetition_found(text, length=4)\n ):\n bad_case_en += 1\n continue\n if tokenizer == \"pinyin\":\n text = convert_char_to_pinyin([text], polyphone=polyphone)[0]\n duration = obj[\"duration\"]\n sub_result.append({\"audio_path\": str(audio_dir.parent / obj[\"wav\"]), \"text\": text, \"duration\": duration})\n durations.append(duration)\n vocab_set.update(list(text))\n return sub_result, durations, vocab_set, bad_case_zh, bad_case_en\n\n\ndef main():\n assert tokenizer in [\"pinyin\", \"char\"]\n result = []\n duration_list = []\n text_vocab_set = set()\n total_bad_case_zh = 0\n total_bad_case_en = 0\n\n # process raw data\n executor = ProcessPoolExecutor(max_workers=max_workers)\n futures = []\n for lang in langs:\n dataset_path = Path(os.path.join(dataset_dir, lang))\n [\n futures.append(executor.submit(deal_with_audio_dir, audio_dir))\n for audio_dir in dataset_path.iterdir()\n if audio_dir.is_dir()\n ]","source_hash":"418b1ecd9af0922bb02a7bc11617b00dda6b918620148bd8a47df64df3c9ec4e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_emilia.main","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_emilia.main#L147-L203","kind":"function","name":"main","path":"src/f5_tts/train/datasets/prepare_emilia.py","language":"python","start_line":147,"end_line":203,"context_start_line":127,"context_end_line":223,"code":" text = text.translate(\n str.maketrans({\",\": \"，\", \"!\": \"！\", \"?\": \"？\"})\n ) # not \"。\" cuz much code-switched\n if obj[\"language\"] == \"en\":\n if (\n obj[\"wav\"].split(\"/\")[1] in out_en\n or any(f in text for f in en_filters)\n or repetition_found(text, length=4)\n ):\n bad_case_en += 1\n continue\n if tokenizer == \"pinyin\":\n text = convert_char_to_pinyin([text], polyphone=polyphone)[0]\n duration = obj[\"duration\"]\n sub_result.append({\"audio_path\": str(audio_dir.parent / obj[\"wav\"]), \"text\": text, \"duration\": duration})\n durations.append(duration)\n vocab_set.update(list(text))\n return sub_result, durations, vocab_set, bad_case_zh, bad_case_en\n\n\ndef main():\n assert tokenizer in [\"pinyin\", \"char\"]\n result = []\n duration_list = []\n text_vocab_set = set()\n total_bad_case_zh = 0\n total_bad_case_en = 0\n\n # process raw data\n executor = ProcessPoolExecutor(max_workers=max_workers)\n futures = []\n for lang in langs:\n dataset_path = Path(os.path.join(dataset_dir, lang))\n [\n futures.append(executor.submit(deal_with_audio_dir, audio_dir))\n for audio_dir in dataset_path.iterdir()\n if audio_dir.is_dir()\n ]\n for futures in tqdm(futures, total=len(futures)):\n sub_result, durations, vocab_set, bad_case_zh, bad_case_en = futures.result()\n result.extend(sub_result)\n duration_list.extend(durations)\n text_vocab_set.update(vocab_set)\n total_bad_case_zh += bad_case_zh\n total_bad_case_en += bad_case_en\n executor.shutdown()\n\n # save preprocessed dataset to disk\n if not os.path.exists(f\"{save_dir}\"):\n os.makedirs(f\"{save_dir}\")\n print(f\"\\nSaving to {save_dir} ...\")\n\n # dataset = Dataset.from_dict({\"audio_path\": audio_path_list, \"text\": text_list, \"duration\": duration_list}) # oom\n # dataset.save_to_disk(f\"{save_dir}/raw\", max_shard_size=\"2GB\")\n with ArrowWriter(path=f\"{save_dir}/raw.arrow\") as writer:\n for line in tqdm(result, desc=\"Writing to raw.arrow ...\"):\n writer.write(line)\n\n # dup a json separately saving duration in case for DynamicBatchSampler ease\n with open(f\"{save_dir}/duration.json\", \"w\", encoding=\"utf-8\") as f:\n json.dump({\"duration\": duration_list}, f, ensure_ascii=False)\n\n # vocab map, i.e. tokenizer\n # add alphabets and symbols (optional, if plan to ft on de/fr etc.)\n # if tokenizer == \"pinyin\":\n # text_vocab_set.update([chr(i) for i in range(32, 127)] + [chr(i) for i in range(192, 256)])\n with open(f\"{save_dir}/vocab.txt\", \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n\n print(f\"\\nFor {dataset_name}, sample count: {len(result)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\")\n print(f\"For {dataset_name}, total {sum(duration_list) / 3600:.2f} hours\")\n if \"ZH\" in langs:\n print(f\"Bad zh transcription case: {total_bad_case_zh}\")\n if \"EN\" in langs:\n print(f\"Bad en transcription case: {total_bad_case_en}\\n\")\n\n\nif __name__ == \"__main__\":\n max_workers = 32\n\n tokenizer = \"pinyin\" # \"pinyin\" | \"char\"\n polyphone = True\n\n langs = [\"ZH\", \"EN\"]\n dataset_dir = \"/Emilia_Dataset/raw\"\n dataset_name = f\"Emilia_{'_'.join(langs)}_{tokenizer}\"\n save_dir = str(files(\"f5_tts\").joinpath(\"../../\")) + f\"/data/{dataset_name}\"\n print(f\"\\nPrepare for {dataset_name}, will save to {save_dir}\\n\")\n\n main()\n\n # Emilia ZH & EN\n # samples count 37837916 (after removal)\n # pinyin vocab size 2543 (polyphone)\n # total duration 95281.87 (hours)","source_hash":"418b1ecd9af0922bb02a7bc11617b00dda6b918620148bd8a47df64df3c9ec4e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_csv_wavs","uri":"program://DMOSpeech2/module/src.f5_tts.train.datasets.prepare_csv_wavs#L1-L283","kind":"module","name":"src.f5_tts.train.datasets.prepare_csv_wavs","path":"src/f5_tts/train/datasets/prepare_csv_wavs.py","language":"python","start_line":1,"end_line":283,"context_start_line":1,"context_end_line":283,"code":"import concurrent.futures\nimport multiprocessing\nimport os\nimport shutil\nimport signal\nimport subprocess # For invoking ffprobe\nimport sys\nfrom contextlib import contextmanager\n\n\nsys.path.append(os.getcwd())\n\nimport argparse\nimport csv\nimport json\nfrom importlib.resources import files\nfrom pathlib import Path\n\nimport torchaudio\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm\n\nfrom f5_tts.model.utils import convert_char_to_pinyin\n\n\nPRETRAINED_VOCAB_PATH = files(\"f5_tts\").joinpath(\"../../data/Emilia_ZH_EN_pinyin/vocab.txt\")\n\n\ndef is_csv_wavs_format(input_dataset_dir):\n fpath = Path(input_dataset_dir)\n metadata = fpath / \"metadata.csv\"\n wavs = fpath / \"wavs\"\n return metadata.exists() and metadata.is_file() and wavs.exists() and wavs.is_dir()\n\n\n# Configuration constants\nBATCH_SIZE = 100 # Batch size for text conversion\nMAX_WORKERS = max(1, multiprocessing.cpu_count() - 1) # Leave one CPU free\nTHREAD_NAME_PREFIX = \"AudioProcessor\"\nCHUNK_SIZE = 100 # Number of files to process per worker batch\n\nexecutor = None # Global executor for cleanup\n\n\n@contextmanager\ndef graceful_exit():\n \"\"\"Context manager for graceful shutdown on signals\"\"\"\n\n def signal_handler(signum, frame):\n print(\"\\nReceived signal to terminate. Cleaning up...\")\n if executor is not None:\n print(\"Shutting down executor...\")\n executor.shutdown(wait=False, cancel_futures=True)\n sys.exit(1)\n\n # Set up signal handlers\n signal.signal(signal.SIGINT, signal_handler)\n signal.signal(signal.SIGTERM, signal_handler)\n\n try:\n yield\n finally:\n if executor is not None:\n executor.shutdown(wait=False)\n\n\ndef process_audio_file(audio_path, text, polyphone):\n \"\"\"Process a single audio file by checking its existence and extracting duration.\"\"\"\n if not Path(audio_path).exists():\n print(f\"audio {audio_path} not found, skipping\")\n return None\n try:\n audio_duration = get_audio_duration(audio_path)\n if audio_duration <= 0:\n raise ValueError(f\"Duration {audio_duration} is non-positive.\")\n return (audio_path, text, audio_duration)\n except Exception as e:\n print(f\"Warning: Failed to process {audio_path} due to error: {e}. Skipping corrupt file.\")\n return None\n\n\ndef batch_convert_texts(texts, polyphone, batch_size=BATCH_SIZE):\n \"\"\"Convert a list of texts to pinyin in batches.\"\"\"\n converted_texts = []\n for i in range(0, len(texts), batch_size):\n batch = texts[i : i + batch_size]\n converted_batch = convert_char_to_pinyin(batch, polyphone=polyphone)\n converted_texts.extend(converted_batch)\n return converted_texts\n\n\ndef prepare_csv_wavs_dir(input_dir, num_workers=None):\n global executor\n assert is_csv_wavs_format(input_dir), f\"not csv_wavs format: {input_dir}\"\n input_dir = Path(input_dir)\n metadata_path = input_dir / \"metadata.csv\"\n audio_path_text_pairs = read_audio_text_pairs(metadata_path.as_posix())\n\n polyphone = True\n total_files = len(audio_path_text_pairs)\n\n # Use provided worker count or calculate optimal number\n worker_count = num_workers if num_workers is not None else min(MAX_WORKERS, total_files)\n print(f\"\\nProcessing {total_files} audio files using {worker_count} workers...\")\n\n with graceful_exit():\n # Initialize thread pool with optimized settings\n with concurrent.futures.ThreadPoolExecutor(\n max_workers=worker_count, thread_name_prefix=THREAD_NAME_PREFIX\n ) as exec:\n executor = exec\n results = []\n\n # Process files in chunks for better efficiency\n for i in range(0, len(audio_path_text_pairs), CHUNK_SIZE):\n chunk = audio_path_text_pairs[i : i + CHUNK_SIZE]\n # Submit futures in order\n chunk_futures = [executor.submit(process_audio_file, pair[0], pair[1], polyphone) for pair in chunk]\n\n # Iterate over futures in the original submission order to preserve ordering\n for future in tqdm(\n chunk_futures,\n total=len(chunk),\n desc=f\"Processing chunk {i // CHUNK_SIZE + 1}/{(total_files + CHUNK_SIZE - 1) // CHUNK_SIZE}\",\n ):\n try:\n result = future.result()\n if result is not None:\n results.append(result)\n except Exception as e:\n print(f\"Error processing file: {e}\")\n\n executor = None\n\n # Filter out failed results\n processed = [res for res in results if res is not None]\n if not processed:\n raise RuntimeError(\"No valid audio files were processed!\")\n\n # Batch process text conversion\n raw_texts = [item[1] for item in processed]\n converted_texts = batch_convert_texts(raw_texts, polyphone, batch_size=BATCH_SIZE)\n\n # Prepare final results\n sub_result = []\n durations = []\n vocab_set = set()\n\n for (audio_path, _, duration), conv_text in zip(processed, converted_texts):\n sub_result.append({\"audio_path\": audio_path, \"text\": conv_text, \"duration\": duration})\n durations.append(duration)\n vocab_set.update(list(conv_text))\n\n return sub_result, durations, vocab_set\n\n\ndef get_audio_duration(audio_path, timeout=5):\n \"\"\"\n Get the duration of an audio file in seconds using ffmpeg's ffprobe.\n Falls back to torchaudio.load() if ffprobe fails.\n \"\"\"\n try:\n cmd = [\n \"ffprobe\",\n \"-v\",\n \"error\",\n \"-show_entries\",\n \"format=duration\",\n \"-of\",\n \"default=noprint_wrappers=1:nokey=1\",\n audio_path,\n ]\n result = subprocess.run(\n cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, check=True, timeout=timeout\n )\n duration_str = result.stdout.strip()\n if duration_str:\n return float(duration_str)\n raise ValueError(\"Empty duration string from ffprobe.\")\n except (subprocess.TimeoutExpired, subprocess.SubprocessError, ValueError) as e:\n print(f\"Warning: ffprobe failed for {audio_path} with error: {e}. Falling back to torchaudio.\")\n try:\n audio, sample_rate = torchaudio.load(audio_path)\n return audio.shape[1] / sample_rate\n except Exception as e:\n raise RuntimeError(f\"Both ffprobe and torchaudio failed for {audio_path}: {e}\")\n\n\ndef read_audio_text_pairs(csv_file_path):\n audio_text_pairs = []\n\n parent = Path(csv_file_path).parent\n with open(csv_file_path, mode=\"r\", newline=\"\", encoding=\"utf-8-sig\") as csvfile:\n reader = csv.reader(csvfile, delimiter=\"|\")\n next(reader) # Skip the header row\n for row in reader:\n if len(row) >= 2:\n audio_file = row[0].strip() # First column: audio file path\n text = row[1].strip() # Second column: text\n audio_file_path = parent / audio_file\n audio_text_pairs.append((audio_file_path.as_posix(), text))\n\n return audio_text_pairs\n\n\ndef save_prepped_dataset(out_dir, result, duration_list, text_vocab_set, is_finetune):\n out_dir = Path(out_dir)\n out_dir.mkdir(exist_ok=True, parents=True)\n print(f\"\\nSaving to {out_dir} ...\")\n\n # Save dataset with improved batch size for better I/O performance\n raw_arrow_path = out_dir / \"raw.arrow\"\n with ArrowWriter(path=raw_arrow_path.as_posix(), writer_batch_size=100) as writer:\n for line in tqdm(result, desc=\"Writing to raw.arrow ...\"):\n writer.write(line)\n\n # Save durations to JSON\n dur_json_path = out_dir / \"duration.json\"\n with open(dur_json_path.as_posix(), \"w\", encoding=\"utf-8\") as f:\n json.dump({\"duration\": duration_list}, f, ensure_ascii=False)\n\n # Handle vocab file - write only once based on finetune flag\n voca_out_path = out_dir / \"vocab.txt\"\n if is_finetune:\n file_vocab_finetune = PRETRAINED_VOCAB_PATH.as_posix()\n shutil.copy2(file_vocab_finetune, voca_out_path)\n else:\n with open(voca_out_path.as_posix(), \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n\n dataset_name = out_dir.stem\n print(f\"\\nFor {dataset_name}, sample count: {len(result)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\")\n print(f\"For {dataset_name}, total {sum(duration_list) / 3600:.2f} hours\")\n\n\ndef prepare_and_save_set(inp_dir, out_dir, is_finetune: bool = True, num_workers: int = None):\n if is_finetune:\n assert PRETRAINED_VOCAB_PATH.exists(), f\"pretrained vocab.txt not found: {PRETRAINED_VOCAB_PATH}\"\n sub_result, durations, vocab_set = prepare_csv_wavs_dir(inp_dir, num_workers=num_workers)\n save_prepped_dataset(out_dir, sub_result, durations, vocab_set, is_finetune)\n\n\ndef cli():\n try:\n # Before processing, check if ffprobe is available.\n if shutil.which(\"ffprobe\") is None:\n print(\n \"Warning: ffprobe is not available. Duration extraction will rely on torchaudio (which may be slower).\"\n )\n\n # Usage examples in help text\n parser = argparse.ArgumentParser(\n description=\"Prepare and save dataset.\",\n epilog=\"\"\"\nExamples:\n # For fine-tuning (default):\n python prepare_csv_wavs.py /input/dataset/path /output/dataset/path\n \n # For pre-training:\n python prepare_csv_wavs.py /input/dataset/path /output/dataset/path --pretrain\n \n # With custom worker count:\n python prepare_csv_wavs.py /input/dataset/path /output/dataset/path --workers 4\n \"\"\",\n )\n parser.add_argument(\"inp_dir\", type=str, help=\"Input directory containing the data.\")\n parser.add_argument(\"out_dir\", type=str, help=\"Output directory to save the prepared data.\")\n parser.add_argument(\"--pretrain\", action=\"store_true\", help=\"Enable for new pretrain, otherwise is a fine-tune\")\n parser.add_argument(\"--workers\", type=int, help=f\"Number of worker threads (default: {MAX_WORKERS})\")\n args = parser.parse_args()\n\n prepare_and_save_set(args.inp_dir, args.out_dir, is_finetune=not args.pretrain, num_workers=args.workers)\n except KeyboardInterrupt:\n print(\"\\nOperation cancelled by user. Cleaning up...\")\n if executor is not None:\n executor.shutdown(wait=False, cancel_futures=True)\n sys.exit(1)\n\n\nif __name__ == \"__main__\":\n cli()","source_hash":"a3a4a9fe211097c49fdbdd87c343f81bb992e127d473b27a46a970336eaff9e0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_csv_wavs.is_csv_wavs_format","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_csv_wavs.is_csv_wavs_format#L29-L33","kind":"function","name":"is_csv_wavs_format","path":"src/f5_tts/train/datasets/prepare_csv_wavs.py","language":"python","start_line":29,"end_line":33,"context_start_line":9,"context_end_line":53,"code":"\n\nsys.path.append(os.getcwd())\n\nimport argparse\nimport csv\nimport json\nfrom importlib.resources import files\nfrom pathlib import Path\n\nimport torchaudio\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm\n\nfrom f5_tts.model.utils import convert_char_to_pinyin\n\n\nPRETRAINED_VOCAB_PATH = files(\"f5_tts\").joinpath(\"../../data/Emilia_ZH_EN_pinyin/vocab.txt\")\n\n\ndef is_csv_wavs_format(input_dataset_dir):\n fpath = Path(input_dataset_dir)\n metadata = fpath / \"metadata.csv\"\n wavs = fpath / \"wavs\"\n return metadata.exists() and metadata.is_file() and wavs.exists() and wavs.is_dir()\n\n\n# Configuration constants\nBATCH_SIZE = 100 # Batch size for text conversion\nMAX_WORKERS = max(1, multiprocessing.cpu_count() - 1) # Leave one CPU free\nTHREAD_NAME_PREFIX = \"AudioProcessor\"\nCHUNK_SIZE = 100 # Number of files to process per worker batch\n\nexecutor = None # Global executor for cleanup\n\n\n@contextmanager\ndef graceful_exit():\n \"\"\"Context manager for graceful shutdown on signals\"\"\"\n\n def signal_handler(signum, frame):\n print(\"\\nReceived signal to terminate. Cleaning up...\")\n if executor is not None:\n print(\"Shutting down executor...\")\n executor.shutdown(wait=False, cancel_futures=True)","source_hash":"a3a4a9fe211097c49fdbdd87c343f81bb992e127d473b27a46a970336eaff9e0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_csv_wavs.graceful_exit","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_csv_wavs.graceful_exit#L46-L64","kind":"function","name":"graceful_exit","path":"src/f5_tts/train/datasets/prepare_csv_wavs.py","language":"python","start_line":46,"end_line":64,"context_start_line":26,"context_end_line":84,"code":"PRETRAINED_VOCAB_PATH = files(\"f5_tts\").joinpath(\"../../data/Emilia_ZH_EN_pinyin/vocab.txt\")\n\n\ndef is_csv_wavs_format(input_dataset_dir):\n fpath = Path(input_dataset_dir)\n metadata = fpath / \"metadata.csv\"\n wavs = fpath / \"wavs\"\n return metadata.exists() and metadata.is_file() and wavs.exists() and wavs.is_dir()\n\n\n# Configuration constants\nBATCH_SIZE = 100 # Batch size for text conversion\nMAX_WORKERS = max(1, multiprocessing.cpu_count() - 1) # Leave one CPU free\nTHREAD_NAME_PREFIX = \"AudioProcessor\"\nCHUNK_SIZE = 100 # Number of files to process per worker batch\n\nexecutor = None # Global executor for cleanup\n\n\n@contextmanager\ndef graceful_exit():\n \"\"\"Context manager for graceful shutdown on signals\"\"\"\n\n def signal_handler(signum, frame):\n print(\"\\nReceived signal to terminate. Cleaning up...\")\n if executor is not None:\n print(\"Shutting down executor...\")\n executor.shutdown(wait=False, cancel_futures=True)\n sys.exit(1)\n\n # Set up signal handlers\n signal.signal(signal.SIGINT, signal_handler)\n signal.signal(signal.SIGTERM, signal_handler)\n\n try:\n yield\n finally:\n if executor is not None:\n executor.shutdown(wait=False)\n\n\ndef process_audio_file(audio_path, text, polyphone):\n \"\"\"Process a single audio file by checking its existence and extracting duration.\"\"\"\n if not Path(audio_path).exists():\n print(f\"audio {audio_path} not found, skipping\")\n return None\n try:\n audio_duration = get_audio_duration(audio_path)\n if audio_duration <= 0:\n raise ValueError(f\"Duration {audio_duration} is non-positive.\")\n return (audio_path, text, audio_duration)\n except Exception as e:\n print(f\"Warning: Failed to process {audio_path} due to error: {e}. Skipping corrupt file.\")\n return None\n\n\ndef batch_convert_texts(texts, polyphone, batch_size=BATCH_SIZE):\n \"\"\"Convert a list of texts to pinyin in batches.\"\"\"\n converted_texts = []","source_hash":"a3a4a9fe211097c49fdbdd87c343f81bb992e127d473b27a46a970336eaff9e0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_csv_wavs.process_audio_file","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_csv_wavs.process_audio_file#L67-L79","kind":"function","name":"process_audio_file","path":"src/f5_tts/train/datasets/prepare_csv_wavs.py","language":"python","start_line":67,"end_line":79,"context_start_line":47,"context_end_line":99,"code":" \"\"\"Context manager for graceful shutdown on signals\"\"\"\n\n def signal_handler(signum, frame):\n print(\"\\nReceived signal to terminate. Cleaning up...\")\n if executor is not None:\n print(\"Shutting down executor...\")\n executor.shutdown(wait=False, cancel_futures=True)\n sys.exit(1)\n\n # Set up signal handlers\n signal.signal(signal.SIGINT, signal_handler)\n signal.signal(signal.SIGTERM, signal_handler)\n\n try:\n yield\n finally:\n if executor is not None:\n executor.shutdown(wait=False)\n\n\ndef process_audio_file(audio_path, text, polyphone):\n \"\"\"Process a single audio file by checking its existence and extracting duration.\"\"\"\n if not Path(audio_path).exists():\n print(f\"audio {audio_path} not found, skipping\")\n return None\n try:\n audio_duration = get_audio_duration(audio_path)\n if audio_duration <= 0:\n raise ValueError(f\"Duration {audio_duration} is non-positive.\")\n return (audio_path, text, audio_duration)\n except Exception as e:\n print(f\"Warning: Failed to process {audio_path} due to error: {e}. Skipping corrupt file.\")\n return None\n\n\ndef batch_convert_texts(texts, polyphone, batch_size=BATCH_SIZE):\n \"\"\"Convert a list of texts to pinyin in batches.\"\"\"\n converted_texts = []\n for i in range(0, len(texts), batch_size):\n batch = texts[i : i + batch_size]\n converted_batch = convert_char_to_pinyin(batch, polyphone=polyphone)\n converted_texts.extend(converted_batch)\n return converted_texts\n\n\ndef prepare_csv_wavs_dir(input_dir, num_workers=None):\n global executor\n assert is_csv_wavs_format(input_dir), f\"not csv_wavs format: {input_dir}\"\n input_dir = Path(input_dir)\n metadata_path = input_dir / \"metadata.csv\"\n audio_path_text_pairs = read_audio_text_pairs(metadata_path.as_posix())\n\n polyphone = True","source_hash":"a3a4a9fe211097c49fdbdd87c343f81bb992e127d473b27a46a970336eaff9e0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_csv_wavs.batch_convert_texts","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_csv_wavs.batch_convert_texts#L82-L89","kind":"function","name":"batch_convert_texts","path":"src/f5_tts/train/datasets/prepare_csv_wavs.py","language":"python","start_line":82,"end_line":89,"context_start_line":62,"context_end_line":109,"code":" finally:\n if executor is not None:\n executor.shutdown(wait=False)\n\n\ndef process_audio_file(audio_path, text, polyphone):\n \"\"\"Process a single audio file by checking its existence and extracting duration.\"\"\"\n if not Path(audio_path).exists():\n print(f\"audio {audio_path} not found, skipping\")\n return None\n try:\n audio_duration = get_audio_duration(audio_path)\n if audio_duration <= 0:\n raise ValueError(f\"Duration {audio_duration} is non-positive.\")\n return (audio_path, text, audio_duration)\n except Exception as e:\n print(f\"Warning: Failed to process {audio_path} due to error: {e}. Skipping corrupt file.\")\n return None\n\n\ndef batch_convert_texts(texts, polyphone, batch_size=BATCH_SIZE):\n \"\"\"Convert a list of texts to pinyin in batches.\"\"\"\n converted_texts = []\n for i in range(0, len(texts), batch_size):\n batch = texts[i : i + batch_size]\n converted_batch = convert_char_to_pinyin(batch, polyphone=polyphone)\n converted_texts.extend(converted_batch)\n return converted_texts\n\n\ndef prepare_csv_wavs_dir(input_dir, num_workers=None):\n global executor\n assert is_csv_wavs_format(input_dir), f\"not csv_wavs format: {input_dir}\"\n input_dir = Path(input_dir)\n metadata_path = input_dir / \"metadata.csv\"\n audio_path_text_pairs = read_audio_text_pairs(metadata_path.as_posix())\n\n polyphone = True\n total_files = len(audio_path_text_pairs)\n\n # Use provided worker count or calculate optimal number\n worker_count = num_workers if num_workers is not None else min(MAX_WORKERS, total_files)\n print(f\"\\nProcessing {total_files} audio files using {worker_count} workers...\")\n\n with graceful_exit():\n # Initialize thread pool with optimized settings\n with concurrent.futures.ThreadPoolExecutor(\n max_workers=worker_count, thread_name_prefix=THREAD_NAME_PREFIX","source_hash":"a3a4a9fe211097c49fdbdd87c343f81bb992e127d473b27a46a970336eaff9e0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_csv_wavs.prepare_csv_wavs_dir","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_csv_wavs.prepare_csv_wavs_dir#L92-L154","kind":"function","name":"prepare_csv_wavs_dir","path":"src/f5_tts/train/datasets/prepare_csv_wavs.py","language":"python","start_line":92,"end_line":154,"context_start_line":72,"context_end_line":174,"code":" try:\n audio_duration = get_audio_duration(audio_path)\n if audio_duration <= 0:\n raise ValueError(f\"Duration {audio_duration} is non-positive.\")\n return (audio_path, text, audio_duration)\n except Exception as e:\n print(f\"Warning: Failed to process {audio_path} due to error: {e}. Skipping corrupt file.\")\n return None\n\n\ndef batch_convert_texts(texts, polyphone, batch_size=BATCH_SIZE):\n \"\"\"Convert a list of texts to pinyin in batches.\"\"\"\n converted_texts = []\n for i in range(0, len(texts), batch_size):\n batch = texts[i : i + batch_size]\n converted_batch = convert_char_to_pinyin(batch, polyphone=polyphone)\n converted_texts.extend(converted_batch)\n return converted_texts\n\n\ndef prepare_csv_wavs_dir(input_dir, num_workers=None):\n global executor\n assert is_csv_wavs_format(input_dir), f\"not csv_wavs format: {input_dir}\"\n input_dir = Path(input_dir)\n metadata_path = input_dir / \"metadata.csv\"\n audio_path_text_pairs = read_audio_text_pairs(metadata_path.as_posix())\n\n polyphone = True\n total_files = len(audio_path_text_pairs)\n\n # Use provided worker count or calculate optimal number\n worker_count = num_workers if num_workers is not None else min(MAX_WORKERS, total_files)\n print(f\"\\nProcessing {total_files} audio files using {worker_count} workers...\")\n\n with graceful_exit():\n # Initialize thread pool with optimized settings\n with concurrent.futures.ThreadPoolExecutor(\n max_workers=worker_count, thread_name_prefix=THREAD_NAME_PREFIX\n ) as exec:\n executor = exec\n results = []\n\n # Process files in chunks for better efficiency\n for i in range(0, len(audio_path_text_pairs), CHUNK_SIZE):\n chunk = audio_path_text_pairs[i : i + CHUNK_SIZE]\n # Submit futures in order\n chunk_futures = [executor.submit(process_audio_file, pair[0], pair[1], polyphone) for pair in chunk]\n\n # Iterate over futures in the original submission order to preserve ordering\n for future in tqdm(\n chunk_futures,\n total=len(chunk),\n desc=f\"Processing chunk {i // CHUNK_SIZE + 1}/{(total_files + CHUNK_SIZE - 1) // CHUNK_SIZE}\",\n ):\n try:\n result = future.result()\n if result is not None:\n results.append(result)\n except Exception as e:\n print(f\"Error processing file: {e}\")\n\n executor = None\n\n # Filter out failed results\n processed = [res for res in results if res is not None]\n if not processed:\n raise RuntimeError(\"No valid audio files were processed!\")\n\n # Batch process text conversion\n raw_texts = [item[1] for item in processed]\n converted_texts = batch_convert_texts(raw_texts, polyphone, batch_size=BATCH_SIZE)\n\n # Prepare final results\n sub_result = []\n durations = []\n vocab_set = set()\n\n for (audio_path, _, duration), conv_text in zip(processed, converted_texts):\n sub_result.append({\"audio_path\": audio_path, \"text\": conv_text, \"duration\": duration})\n durations.append(duration)\n vocab_set.update(list(conv_text))\n\n return sub_result, durations, vocab_set\n\n\ndef get_audio_duration(audio_path, timeout=5):\n \"\"\"\n Get the duration of an audio file in seconds using ffmpeg's ffprobe.\n Falls back to torchaudio.load() if ffprobe fails.\n \"\"\"\n try:\n cmd = [\n \"ffprobe\",\n \"-v\",\n \"error\",\n \"-show_entries\",\n \"format=duration\",\n \"-of\",\n \"default=noprint_wrappers=1:nokey=1\",\n audio_path,\n ]\n result = subprocess.run(\n cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, check=True, timeout=timeout","source_hash":"a3a4a9fe211097c49fdbdd87c343f81bb992e127d473b27a46a970336eaff9e0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_csv_wavs.get_audio_duration","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_csv_wavs.get_audio_duration#L157-L186","kind":"function","name":"get_audio_duration","path":"src/f5_tts/train/datasets/prepare_csv_wavs.py","language":"python","start_line":157,"end_line":186,"context_start_line":137,"context_end_line":206,"code":" if not processed:\n raise RuntimeError(\"No valid audio files were processed!\")\n\n # Batch process text conversion\n raw_texts = [item[1] for item in processed]\n converted_texts = batch_convert_texts(raw_texts, polyphone, batch_size=BATCH_SIZE)\n\n # Prepare final results\n sub_result = []\n durations = []\n vocab_set = set()\n\n for (audio_path, _, duration), conv_text in zip(processed, converted_texts):\n sub_result.append({\"audio_path\": audio_path, \"text\": conv_text, \"duration\": duration})\n durations.append(duration)\n vocab_set.update(list(conv_text))\n\n return sub_result, durations, vocab_set\n\n\ndef get_audio_duration(audio_path, timeout=5):\n \"\"\"\n Get the duration of an audio file in seconds using ffmpeg's ffprobe.\n Falls back to torchaudio.load() if ffprobe fails.\n \"\"\"\n try:\n cmd = [\n \"ffprobe\",\n \"-v\",\n \"error\",\n \"-show_entries\",\n \"format=duration\",\n \"-of\",\n \"default=noprint_wrappers=1:nokey=1\",\n audio_path,\n ]\n result = subprocess.run(\n cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, check=True, timeout=timeout\n )\n duration_str = result.stdout.strip()\n if duration_str:\n return float(duration_str)\n raise ValueError(\"Empty duration string from ffprobe.\")\n except (subprocess.TimeoutExpired, subprocess.SubprocessError, ValueError) as e:\n print(f\"Warning: ffprobe failed for {audio_path} with error: {e}. Falling back to torchaudio.\")\n try:\n audio, sample_rate = torchaudio.load(audio_path)\n return audio.shape[1] / sample_rate\n except Exception as e:\n raise RuntimeError(f\"Both ffprobe and torchaudio failed for {audio_path}: {e}\")\n\n\ndef read_audio_text_pairs(csv_file_path):\n audio_text_pairs = []\n\n parent = Path(csv_file_path).parent\n with open(csv_file_path, mode=\"r\", newline=\"\", encoding=\"utf-8-sig\") as csvfile:\n reader = csv.reader(csvfile, delimiter=\"|\")\n next(reader) # Skip the header row\n for row in reader:\n if len(row) >= 2:\n audio_file = row[0].strip() # First column: audio file path\n text = row[1].strip() # Second column: text\n audio_file_path = parent / audio_file\n audio_text_pairs.append((audio_file_path.as_posix(), text))\n\n return audio_text_pairs\n\n\ndef save_prepped_dataset(out_dir, result, duration_list, text_vocab_set, is_finetune):","source_hash":"a3a4a9fe211097c49fdbdd87c343f81bb992e127d473b27a46a970336eaff9e0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_csv_wavs.read_audio_text_pairs","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_csv_wavs.read_audio_text_pairs#L189-L203","kind":"function","name":"read_audio_text_pairs","path":"src/f5_tts/train/datasets/prepare_csv_wavs.py","language":"python","start_line":189,"end_line":203,"context_start_line":169,"context_end_line":223,"code":" \"-of\",\n \"default=noprint_wrappers=1:nokey=1\",\n audio_path,\n ]\n result = subprocess.run(\n cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, check=True, timeout=timeout\n )\n duration_str = result.stdout.strip()\n if duration_str:\n return float(duration_str)\n raise ValueError(\"Empty duration string from ffprobe.\")\n except (subprocess.TimeoutExpired, subprocess.SubprocessError, ValueError) as e:\n print(f\"Warning: ffprobe failed for {audio_path} with error: {e}. Falling back to torchaudio.\")\n try:\n audio, sample_rate = torchaudio.load(audio_path)\n return audio.shape[1] / sample_rate\n except Exception as e:\n raise RuntimeError(f\"Both ffprobe and torchaudio failed for {audio_path}: {e}\")\n\n\ndef read_audio_text_pairs(csv_file_path):\n audio_text_pairs = []\n\n parent = Path(csv_file_path).parent\n with open(csv_file_path, mode=\"r\", newline=\"\", encoding=\"utf-8-sig\") as csvfile:\n reader = csv.reader(csvfile, delimiter=\"|\")\n next(reader) # Skip the header row\n for row in reader:\n if len(row) >= 2:\n audio_file = row[0].strip() # First column: audio file path\n text = row[1].strip() # Second column: text\n audio_file_path = parent / audio_file\n audio_text_pairs.append((audio_file_path.as_posix(), text))\n\n return audio_text_pairs\n\n\ndef save_prepped_dataset(out_dir, result, duration_list, text_vocab_set, is_finetune):\n out_dir = Path(out_dir)\n out_dir.mkdir(exist_ok=True, parents=True)\n print(f\"\\nSaving to {out_dir} ...\")\n\n # Save dataset with improved batch size for better I/O performance\n raw_arrow_path = out_dir / \"raw.arrow\"\n with ArrowWriter(path=raw_arrow_path.as_posix(), writer_batch_size=100) as writer:\n for line in tqdm(result, desc=\"Writing to raw.arrow ...\"):\n writer.write(line)\n\n # Save durations to JSON\n dur_json_path = out_dir / \"duration.json\"\n with open(dur_json_path.as_posix(), \"w\", encoding=\"utf-8\") as f:\n json.dump({\"duration\": duration_list}, f, ensure_ascii=False)\n\n # Handle vocab file - write only once based on finetune flag\n voca_out_path = out_dir / \"vocab.txt\"","source_hash":"a3a4a9fe211097c49fdbdd87c343f81bb992e127d473b27a46a970336eaff9e0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_csv_wavs.save_prepped_dataset","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_csv_wavs.save_prepped_dataset#L206-L235","kind":"function","name":"save_prepped_dataset","path":"src/f5_tts/train/datasets/prepare_csv_wavs.py","language":"python","start_line":206,"end_line":235,"context_start_line":186,"context_end_line":255,"code":" raise RuntimeError(f\"Both ffprobe and torchaudio failed for {audio_path}: {e}\")\n\n\ndef read_audio_text_pairs(csv_file_path):\n audio_text_pairs = []\n\n parent = Path(csv_file_path).parent\n with open(csv_file_path, mode=\"r\", newline=\"\", encoding=\"utf-8-sig\") as csvfile:\n reader = csv.reader(csvfile, delimiter=\"|\")\n next(reader) # Skip the header row\n for row in reader:\n if len(row) >= 2:\n audio_file = row[0].strip() # First column: audio file path\n text = row[1].strip() # Second column: text\n audio_file_path = parent / audio_file\n audio_text_pairs.append((audio_file_path.as_posix(), text))\n\n return audio_text_pairs\n\n\ndef save_prepped_dataset(out_dir, result, duration_list, text_vocab_set, is_finetune):\n out_dir = Path(out_dir)\n out_dir.mkdir(exist_ok=True, parents=True)\n print(f\"\\nSaving to {out_dir} ...\")\n\n # Save dataset with improved batch size for better I/O performance\n raw_arrow_path = out_dir / \"raw.arrow\"\n with ArrowWriter(path=raw_arrow_path.as_posix(), writer_batch_size=100) as writer:\n for line in tqdm(result, desc=\"Writing to raw.arrow ...\"):\n writer.write(line)\n\n # Save durations to JSON\n dur_json_path = out_dir / \"duration.json\"\n with open(dur_json_path.as_posix(), \"w\", encoding=\"utf-8\") as f:\n json.dump({\"duration\": duration_list}, f, ensure_ascii=False)\n\n # Handle vocab file - write only once based on finetune flag\n voca_out_path = out_dir / \"vocab.txt\"\n if is_finetune:\n file_vocab_finetune = PRETRAINED_VOCAB_PATH.as_posix()\n shutil.copy2(file_vocab_finetune, voca_out_path)\n else:\n with open(voca_out_path.as_posix(), \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n\n dataset_name = out_dir.stem\n print(f\"\\nFor {dataset_name}, sample count: {len(result)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\")\n print(f\"For {dataset_name}, total {sum(duration_list) / 3600:.2f} hours\")\n\n\ndef prepare_and_save_set(inp_dir, out_dir, is_finetune: bool = True, num_workers: int = None):\n if is_finetune:\n assert PRETRAINED_VOCAB_PATH.exists(), f\"pretrained vocab.txt not found: {PRETRAINED_VOCAB_PATH}\"\n sub_result, durations, vocab_set = prepare_csv_wavs_dir(inp_dir, num_workers=num_workers)\n save_prepped_dataset(out_dir, sub_result, durations, vocab_set, is_finetune)\n\n\ndef cli():\n try:\n # Before processing, check if ffprobe is available.\n if shutil.which(\"ffprobe\") is None:\n print(\n \"Warning: ffprobe is not available. Duration extraction will rely on torchaudio (which may be slower).\"\n )\n\n # Usage examples in help text\n parser = argparse.ArgumentParser(\n description=\"Prepare and save dataset.\",","source_hash":"a3a4a9fe211097c49fdbdd87c343f81bb992e127d473b27a46a970336eaff9e0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_csv_wavs.prepare_and_save_set","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_csv_wavs.prepare_and_save_set#L238-L242","kind":"function","name":"prepare_and_save_set","path":"src/f5_tts/train/datasets/prepare_csv_wavs.py","language":"python","start_line":238,"end_line":242,"context_start_line":218,"context_end_line":262,"code":" dur_json_path = out_dir / \"duration.json\"\n with open(dur_json_path.as_posix(), \"w\", encoding=\"utf-8\") as f:\n json.dump({\"duration\": duration_list}, f, ensure_ascii=False)\n\n # Handle vocab file - write only once based on finetune flag\n voca_out_path = out_dir / \"vocab.txt\"\n if is_finetune:\n file_vocab_finetune = PRETRAINED_VOCAB_PATH.as_posix()\n shutil.copy2(file_vocab_finetune, voca_out_path)\n else:\n with open(voca_out_path.as_posix(), \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n\n dataset_name = out_dir.stem\n print(f\"\\nFor {dataset_name}, sample count: {len(result)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\")\n print(f\"For {dataset_name}, total {sum(duration_list) / 3600:.2f} hours\")\n\n\ndef prepare_and_save_set(inp_dir, out_dir, is_finetune: bool = True, num_workers: int = None):\n if is_finetune:\n assert PRETRAINED_VOCAB_PATH.exists(), f\"pretrained vocab.txt not found: {PRETRAINED_VOCAB_PATH}\"\n sub_result, durations, vocab_set = prepare_csv_wavs_dir(inp_dir, num_workers=num_workers)\n save_prepped_dataset(out_dir, sub_result, durations, vocab_set, is_finetune)\n\n\ndef cli():\n try:\n # Before processing, check if ffprobe is available.\n if shutil.which(\"ffprobe\") is None:\n print(\n \"Warning: ffprobe is not available. Duration extraction will rely on torchaudio (which may be slower).\"\n )\n\n # Usage examples in help text\n parser = argparse.ArgumentParser(\n description=\"Prepare and save dataset.\",\n epilog=\"\"\"\nExamples:\n # For fine-tuning (default):\n python prepare_csv_wavs.py /input/dataset/path /output/dataset/path\n \n # For pre-training:\n python prepare_csv_wavs.py /input/dataset/path /output/dataset/path --pretrain","source_hash":"a3a4a9fe211097c49fdbdd87c343f81bb992e127d473b27a46a970336eaff9e0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_csv_wavs.cli","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_csv_wavs.cli#L245-L279","kind":"function","name":"cli","path":"src/f5_tts/train/datasets/prepare_csv_wavs.py","language":"python","start_line":245,"end_line":279,"context_start_line":225,"context_end_line":283,"code":" file_vocab_finetune = PRETRAINED_VOCAB_PATH.as_posix()\n shutil.copy2(file_vocab_finetune, voca_out_path)\n else:\n with open(voca_out_path.as_posix(), \"w\") as f:\n for vocab in sorted(text_vocab_set):\n f.write(vocab + \"\\n\")\n\n dataset_name = out_dir.stem\n print(f\"\\nFor {dataset_name}, sample count: {len(result)}\")\n print(f\"For {dataset_name}, vocab size is: {len(text_vocab_set)}\")\n print(f\"For {dataset_name}, total {sum(duration_list) / 3600:.2f} hours\")\n\n\ndef prepare_and_save_set(inp_dir, out_dir, is_finetune: bool = True, num_workers: int = None):\n if is_finetune:\n assert PRETRAINED_VOCAB_PATH.exists(), f\"pretrained vocab.txt not found: {PRETRAINED_VOCAB_PATH}\"\n sub_result, durations, vocab_set = prepare_csv_wavs_dir(inp_dir, num_workers=num_workers)\n save_prepped_dataset(out_dir, sub_result, durations, vocab_set, is_finetune)\n\n\ndef cli():\n try:\n # Before processing, check if ffprobe is available.\n if shutil.which(\"ffprobe\") is None:\n print(\n \"Warning: ffprobe is not available. Duration extraction will rely on torchaudio (which may be slower).\"\n )\n\n # Usage examples in help text\n parser = argparse.ArgumentParser(\n description=\"Prepare and save dataset.\",\n epilog=\"\"\"\nExamples:\n # For fine-tuning (default):\n python prepare_csv_wavs.py /input/dataset/path /output/dataset/path\n \n # For pre-training:\n python prepare_csv_wavs.py /input/dataset/path /output/dataset/path --pretrain\n \n # With custom worker count:\n python prepare_csv_wavs.py /input/dataset/path /output/dataset/path --workers 4\n \"\"\",\n )\n parser.add_argument(\"inp_dir\", type=str, help=\"Input directory containing the data.\")\n parser.add_argument(\"out_dir\", type=str, help=\"Output directory to save the prepared data.\")\n parser.add_argument(\"--pretrain\", action=\"store_true\", help=\"Enable for new pretrain, otherwise is a fine-tune\")\n parser.add_argument(\"--workers\", type=int, help=f\"Number of worker threads (default: {MAX_WORKERS})\")\n args = parser.parse_args()\n\n prepare_and_save_set(args.inp_dir, args.out_dir, is_finetune=not args.pretrain, num_workers=args.workers)\n except KeyboardInterrupt:\n print(\"\\nOperation cancelled by user. Cleaning up...\")\n if executor is not None:\n executor.shutdown(wait=False, cancel_futures=True)\n sys.exit(1)\n\n\nif __name__ == \"__main__\":\n cli()","source_hash":"a3a4a9fe211097c49fdbdd87c343f81bb992e127d473b27a46a970336eaff9e0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.train.datasets.prepare_csv_wavs.signal_handler","uri":"program://DMOSpeech2/function/src.f5_tts.train.datasets.prepare_csv_wavs.signal_handler#L49-L54","kind":"function","name":"signal_handler","path":"src/f5_tts/train/datasets/prepare_csv_wavs.py","language":"python","start_line":49,"end_line":54,"context_start_line":29,"context_end_line":74,"code":"def is_csv_wavs_format(input_dataset_dir):\n fpath = Path(input_dataset_dir)\n metadata = fpath / \"metadata.csv\"\n wavs = fpath / \"wavs\"\n return metadata.exists() and metadata.is_file() and wavs.exists() and wavs.is_dir()\n\n\n# Configuration constants\nBATCH_SIZE = 100 # Batch size for text conversion\nMAX_WORKERS = max(1, multiprocessing.cpu_count() - 1) # Leave one CPU free\nTHREAD_NAME_PREFIX = \"AudioProcessor\"\nCHUNK_SIZE = 100 # Number of files to process per worker batch\n\nexecutor = None # Global executor for cleanup\n\n\n@contextmanager\ndef graceful_exit():\n \"\"\"Context manager for graceful shutdown on signals\"\"\"\n\n def signal_handler(signum, frame):\n print(\"\\nReceived signal to terminate. Cleaning up...\")\n if executor is not None:\n print(\"Shutting down executor...\")\n executor.shutdown(wait=False, cancel_futures=True)\n sys.exit(1)\n\n # Set up signal handlers\n signal.signal(signal.SIGINT, signal_handler)\n signal.signal(signal.SIGTERM, signal_handler)\n\n try:\n yield\n finally:\n if executor is not None:\n executor.shutdown(wait=False)\n\n\ndef process_audio_file(audio_path, text, polyphone):\n \"\"\"Process a single audio file by checking its existence and extracting duration.\"\"\"\n if not Path(audio_path).exists():\n print(f\"audio {audio_path} not found, skipping\")\n return None\n try:\n audio_duration = get_audio_duration(audio_path)\n if audio_duration <= 0:","source_hash":"a3a4a9fe211097c49fdbdd87c343f81bb992e127d473b27a46a970336eaff9e0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.trainer","uri":"program://DMOSpeech2/module/src.f5_tts.model.trainer#L1-L415","kind":"module","name":"src.f5_tts.model.trainer","path":"src/f5_tts/model/trainer.py","language":"python","start_line":1,"end_line":415,"context_start_line":1,"context_end_line":415,"code":"from __future__ import annotations\n\nimport gc\nimport os\n\nimport torch\nimport torchaudio\nimport wandb\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nfrom ema_pytorch import EMA\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model import CFM\nfrom f5_tts.model.dataset import DynamicBatchSampler, collate_fn\nfrom f5_tts.model.utils import default, exists\n\n# trainer\n\n\nclass Trainer:\n def __init__(\n self,\n model: CFM,\n epochs,\n learning_rate,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n noise_scheduler: str | None = None,\n duration_predictor: torch.nn.Module | None = None,\n logger: str | None = \"wandb\", # \"wandb\" | \"tensorboard\" | None\n wandb_project=\"test_e2-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n log_samples: bool = False,\n last_per_steps=None,\n accelerate_kwargs: dict = dict(),\n ema_kwargs: dict = dict(),\n bnb_optimizer: bool = False,\n mel_spec_type: str = \"vocos\", # \"vocos\" | \"bigvgan\"\n is_local_vocoder: bool = False, # use local path vocoder\n local_vocoder_path: str = \"\", # local vocoder path\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=True)\n\n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n print(f\"Using logger: {logger}\")\n self.log_samples = log_samples\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n \"noise_scheduler\": noise_scheduler,\n },\n )\n\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n self.model = model\n\n if self.is_main:\n self.ema_model = EMA(model, include_online_model=False, **ema_kwargs)\n self.ema_model.to(self.accelerator.device)\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n # mel vocoder config\n self.vocoder_name = mel_spec_type\n self.is_local_vocoder = is_local_vocoder\n self.local_vocoder_path = local_vocoder_path\n\n self.noise_scheduler = noise_scheduler\n\n self.duration_predictor = duration_predictor\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n self.scale = None\n self.count = 0\n\n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n ema_model_state_dict=self.ema_model.state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n scale=self.scale,\n count=self.count,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at step {step}\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n # patch for backward compatibility, 305e3ea\n for key in [\"ema_model.mel_spec.mel_stft.mel_scale.fb\", \"ema_model.mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"ema_model_state_dict\"]:\n del checkpoint[\"ema_model_state_dict\"][key]\n\n if self.is_main:\n self.ema_model.load_state_dict(checkpoint[\"ema_model_state_dict\"])\n\n if \"step\" in checkpoint:\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n\n if \"scale\" in checkpoint:\n self.scale = float(checkpoint[\"scale\"])\n self.model.scale = self.scale\n \n if \"count\" in checkpoint:\n self.count = int(checkpoint[\"count\"])\n \n del checkpoint\n gc.collect()\n return step\n\n def train(self, train_dataset: Dataset, num_workers=16, resumable_with_seed: int = None):\n if self.log_samples:\n from f5_tts.infer.utils_infer import cfg_strength, load_vocoder, nfe_step, sway_sampling_coef\n\n vocoder = load_vocoder(\n vocoder_name=self.vocoder_name, is_local=self.is_local_vocoder, local_path=self.local_vocoder_path\n )\n target_sample_rate = self.accelerator.unwrap_model(self.model).mel_spec.target_sample_rate\n log_samples_path = f\"{self.checkpoint_path}/samples\"\n os.makedirs(log_samples_path, exist_ok=True)\n\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n\n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_steps = (\n self.num_warmup_updates * self.accelerator.num_processes\n ) # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_steps = len(train_dataloader) * self.epochs / self.grad_accumulation_steps\n decay_steps = total_steps - warmup_steps\n warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)\n decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)\n self.scheduler = SequentialLR(\n self.optimizer, schedulers=[warmup_scheduler, decay_scheduler], milestones=[warmup_steps]\n )\n train_dataloader, self.scheduler = self.accelerator.prepare(\n train_dataloader, self.scheduler\n ) # actual steps = 1 gpu steps / gpus\n start_step = self.load_checkpoint()\n global_step = start_step\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar = tqdm(\n skipped_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n initial=skipped_batch,\n total=orig_epoch_step,\n )\n else:\n progress_bar = tqdm(\n train_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n )\n\n for batch in progress_bar:\n with self.accelerator.accumulate(self.model):\n text_inputs = batch[\"text\"]\n mel_spec = batch[\"mel\"].permute(0, 2, 1)\n mel_lengths = batch[\"mel_lengths\"]\n \n self.count += 1\n \n if self.scale is None:\n self.scale = mel_spec.std()\n else:\n self.scale += (mel_spec.std() - self.scale) / self.count\n \n mel_spec = mel_spec / self.scale # normalize mel spectrogram\n \n # TODO. add duration predictor training\n if self.duration_predictor is not None and self.accelerator.is_local_main_process:\n dur_loss = self.duration_predictor(mel_spec, lens=batch.get(\"durations\"))\n self.accelerator.log({\"duration loss\": dur_loss.item()}, step=global_step)\n\n loss, cond, pred, t = self.model(\n mel_spec, text=text_inputs, lens=mel_lengths, noise_scheduler=self.noise_scheduler\n )\n self.accelerator.backward(loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n self.optimizer.step()\n self.scheduler.step()\n self.optimizer.zero_grad()\n\n if self.is_main and self.accelerator.sync_gradients:\n self.ema_model.update()\n\n global_step += 1\n\n if self.accelerator.is_local_main_process:\n self.accelerator.log({\"loss\": loss.item(), \"lr\": self.scheduler.get_last_lr()[0]}, step=global_step)\n if self.logger == \"tensorboard\":\n self.writer.add_scalar(\"loss\", loss.item(), global_step)\n self.writer.add_scalar(\"lr\", self.scheduler.get_last_lr()[0], global_step)\n\n progress_bar.set_postfix(step=str(global_step), loss=loss.item())\n\n if global_step % (self.save_per_updates * self.grad_accumulation_steps) == 0:\n self.save_checkpoint(global_step)\n if self.log_samples and self.accelerator.is_local_main_process:\n gen_mel_spec = pred[0].unsqueeze(0).permute(0, 2, 1) * self.scale\n ref_mel_spec = cond[0].unsqueeze(0).permute(0, 2, 1) * self.scale\n with torch.inference_mode():\n if self.vocoder_name == \"vocos\":\n gen_audio = vocoder.decode(gen_mel_spec).cpu()\n ref_audio = vocoder.decode(ref_mel_spec).cpu()\n elif self.vocoder_name == \"bigvgan\":\n gen_audio = vocoder(gen_mel_spec).squeeze(0).cpu()\n ref_audio = vocoder(ref_mel_spec).squeeze(0).cpu()\n \n gen_audio = wandb.Audio(\n gen_audio.float().numpy().squeeze(),\n sample_rate=24000,\n caption=\"time: \" + str(t[0].squeeze().float().cpu().numpy())\n )\n ref_audio = wandb.Audio(\n ref_audio.float().numpy().squeeze(),\n sample_rate=24000,\n caption=\"time: \" + str(t[0].squeeze().float().cpu().numpy())\n )\n\n self.accelerator.log({\"gen_audio\": gen_audio, \n \"ref_audio\": ref_audio,\n }, step=global_step)\n\n\n# if self.log_samples and self.accelerator.is_local_main_process:\n# ref_audio_len = mel_lengths[0]\n# infer_text = [\n# text_inputs[0] + ([\" \"] if isinstance(text_inputs[0], list) else \" \") + text_inputs[0]\n# ]\n# with torch.inference_mode():\n# # generated, _ = self.accelerator.unwrap_model(self.model).sample(\n# # cond=mel_spec[0][:ref_audio_len].unsqueeze(0),\n# # text=infer_text,\n# # duration=ref_audio_len * 2,\n# # steps=nfe_step,\n# # cfg_strength=cfg_strength,\n# # sway_sampling_coef=sway_sampling_coef,\n# # )\n# # generated = generated.to(torch.float32)\n# # gen_mel_spec = generated[:, ref_audio_len:, :].permute(0, 2, 1).to(self.accelerator.device)\n# # ref_mel_spec = batch[\"mel\"][0].unsqueeze(0)\n# gen_mel_spec = pred[0].unsqueeze(0).permute(0, 2, 1)\n# ref_mel_spec = cond[0].unsqueeze(0).permute(0, 2, 1)\n# if self.vocoder_name == \"vocos\":\n# gen_audio = vocoder.decode(gen_mel_spec).cpu()\n# ref_audio = vocoder.decode(ref_mel_spec).cpu()\n# elif self.vocoder_name == \"bigvgan\":\n# gen_audio = vocoder(gen_mel_spec).squeeze(0).cpu()\n# ref_audio = vocoder(ref_mel_spec).squeeze(0).cpu()\n\n# torchaudio.save(f\"{log_samples_path}/step_{global_step}_gen.wav\", gen_audio, target_sample_rate)\n# torchaudio.save(f\"{log_samples_path}/step_{global_step}_ref.wav\", ref_audio, target_sample_rate)\n\n if global_step % self.last_per_steps == 0:\n self.save_checkpoint(global_step, last=True)\n\n self.save_checkpoint(global_step, last=True)\n\n self.accelerator.end_training()","source_hash":"2b9b27e8647e8f0a32d4b60a3a0bb41365c787bea1168dc1099632ac1d210ea7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.trainer.Trainer","uri":"program://DMOSpeech2/class/src.f5_tts.model.trainer.Trainer#L24-L415","kind":"class","name":"Trainer","path":"src/f5_tts/model/trainer.py","language":"python","start_line":24,"end_line":415,"context_start_line":4,"context_end_line":415,"code":"import os\n\nimport torch\nimport torchaudio\nimport wandb\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nfrom ema_pytorch import EMA\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model import CFM\nfrom f5_tts.model.dataset import DynamicBatchSampler, collate_fn\nfrom f5_tts.model.utils import default, exists\n\n# trainer\n\n\nclass Trainer:\n def __init__(\n self,\n model: CFM,\n epochs,\n learning_rate,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n noise_scheduler: str | None = None,\n duration_predictor: torch.nn.Module | None = None,\n logger: str | None = \"wandb\", # \"wandb\" | \"tensorboard\" | None\n wandb_project=\"test_e2-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n log_samples: bool = False,\n last_per_steps=None,\n accelerate_kwargs: dict = dict(),\n ema_kwargs: dict = dict(),\n bnb_optimizer: bool = False,\n mel_spec_type: str = \"vocos\", # \"vocos\" | \"bigvgan\"\n is_local_vocoder: bool = False, # use local path vocoder\n local_vocoder_path: str = \"\", # local vocoder path\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=True)\n\n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n print(f\"Using logger: {logger}\")\n self.log_samples = log_samples\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n \"noise_scheduler\": noise_scheduler,\n },\n )\n\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n self.model = model\n\n if self.is_main:\n self.ema_model = EMA(model, include_online_model=False, **ema_kwargs)\n self.ema_model.to(self.accelerator.device)\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n # mel vocoder config\n self.vocoder_name = mel_spec_type\n self.is_local_vocoder = is_local_vocoder\n self.local_vocoder_path = local_vocoder_path\n\n self.noise_scheduler = noise_scheduler\n\n self.duration_predictor = duration_predictor\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n self.scale = None\n self.count = 0\n\n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n ema_model_state_dict=self.ema_model.state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n scale=self.scale,\n count=self.count,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at step {step}\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n # patch for backward compatibility, 305e3ea\n for key in [\"ema_model.mel_spec.mel_stft.mel_scale.fb\", \"ema_model.mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"ema_model_state_dict\"]:\n del checkpoint[\"ema_model_state_dict\"][key]\n\n if self.is_main:\n self.ema_model.load_state_dict(checkpoint[\"ema_model_state_dict\"])\n\n if \"step\" in checkpoint:\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n\n if \"scale\" in checkpoint:\n self.scale = float(checkpoint[\"scale\"])\n self.model.scale = self.scale\n \n if \"count\" in checkpoint:\n self.count = int(checkpoint[\"count\"])\n \n del checkpoint\n gc.collect()\n return step\n\n def train(self, train_dataset: Dataset, num_workers=16, resumable_with_seed: int = None):\n if self.log_samples:\n from f5_tts.infer.utils_infer import cfg_strength, load_vocoder, nfe_step, sway_sampling_coef\n\n vocoder = load_vocoder(\n vocoder_name=self.vocoder_name, is_local=self.is_local_vocoder, local_path=self.local_vocoder_path\n )\n target_sample_rate = self.accelerator.unwrap_model(self.model).mel_spec.target_sample_rate\n log_samples_path = f\"{self.checkpoint_path}/samples\"\n os.makedirs(log_samples_path, exist_ok=True)\n\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n\n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_steps = (\n self.num_warmup_updates * self.accelerator.num_processes\n ) # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_steps = len(train_dataloader) * self.epochs / self.grad_accumulation_steps\n decay_steps = total_steps - warmup_steps\n warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)\n decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)\n self.scheduler = SequentialLR(\n self.optimizer, schedulers=[warmup_scheduler, decay_scheduler], milestones=[warmup_steps]\n )\n train_dataloader, self.scheduler = self.accelerator.prepare(\n train_dataloader, self.scheduler\n ) # actual steps = 1 gpu steps / gpus\n start_step = self.load_checkpoint()\n global_step = start_step\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar = tqdm(\n skipped_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n initial=skipped_batch,\n total=orig_epoch_step,\n )\n else:\n progress_bar = tqdm(\n train_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n )\n\n for batch in progress_bar:\n with self.accelerator.accumulate(self.model):\n text_inputs = batch[\"text\"]\n mel_spec = batch[\"mel\"].permute(0, 2, 1)\n mel_lengths = batch[\"mel_lengths\"]\n \n self.count += 1\n \n if self.scale is None:\n self.scale = mel_spec.std()\n else:\n self.scale += (mel_spec.std() - self.scale) / self.count\n \n mel_spec = mel_spec / self.scale # normalize mel spectrogram\n \n # TODO. add duration predictor training\n if self.duration_predictor is not None and self.accelerator.is_local_main_process:\n dur_loss = self.duration_predictor(mel_spec, lens=batch.get(\"durations\"))\n self.accelerator.log({\"duration loss\": dur_loss.item()}, step=global_step)\n\n loss, cond, pred, t = self.model(\n mel_spec, text=text_inputs, lens=mel_lengths, noise_scheduler=self.noise_scheduler\n )\n self.accelerator.backward(loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n self.optimizer.step()\n self.scheduler.step()\n self.optimizer.zero_grad()\n\n if self.is_main and self.accelerator.sync_gradients:\n self.ema_model.update()\n\n global_step += 1\n\n if self.accelerator.is_local_main_process:\n self.accelerator.log({\"loss\": loss.item(), \"lr\": self.scheduler.get_last_lr()[0]}, step=global_step)\n if self.logger == \"tensorboard\":\n self.writer.add_scalar(\"loss\", loss.item(), global_step)\n self.writer.add_scalar(\"lr\", self.scheduler.get_last_lr()[0], global_step)\n\n progress_bar.set_postfix(step=str(global_step), loss=loss.item())\n\n if global_step % (self.save_per_updates * self.grad_accumulation_steps) == 0:\n self.save_checkpoint(global_step)\n if self.log_samples and self.accelerator.is_local_main_process:\n gen_mel_spec = pred[0].unsqueeze(0).permute(0, 2, 1) * self.scale\n ref_mel_spec = cond[0].unsqueeze(0).permute(0, 2, 1) * self.scale\n with torch.inference_mode():\n if self.vocoder_name == \"vocos\":\n gen_audio = vocoder.decode(gen_mel_spec).cpu()\n ref_audio = vocoder.decode(ref_mel_spec).cpu()\n elif self.vocoder_name == \"bigvgan\":\n gen_audio = vocoder(gen_mel_spec).squeeze(0).cpu()\n ref_audio = vocoder(ref_mel_spec).squeeze(0).cpu()\n \n gen_audio = wandb.Audio(\n gen_audio.float().numpy().squeeze(),\n sample_rate=24000,\n caption=\"time: \" + str(t[0].squeeze().float().cpu().numpy())\n )\n ref_audio = wandb.Audio(\n ref_audio.float().numpy().squeeze(),\n sample_rate=24000,\n caption=\"time: \" + str(t[0].squeeze().float().cpu().numpy())\n )\n\n self.accelerator.log({\"gen_audio\": gen_audio, \n \"ref_audio\": ref_audio,\n }, step=global_step)\n\n\n# if self.log_samples and self.accelerator.is_local_main_process:\n# ref_audio_len = mel_lengths[0]\n# infer_text = [\n# text_inputs[0] + ([\" \"] if isinstance(text_inputs[0], list) else \" \") + text_inputs[0]\n# ]\n# with torch.inference_mode():\n# # generated, _ = self.accelerator.unwrap_model(self.model).sample(\n# # cond=mel_spec[0][:ref_audio_len].unsqueeze(0),\n# # text=infer_text,\n# # duration=ref_audio_len * 2,\n# # steps=nfe_step,\n# # cfg_strength=cfg_strength,\n# # sway_sampling_coef=sway_sampling_coef,\n# # )\n# # generated = generated.to(torch.float32)\n# # gen_mel_spec = generated[:, ref_audio_len:, :].permute(0, 2, 1).to(self.accelerator.device)\n# # ref_mel_spec = batch[\"mel\"][0].unsqueeze(0)\n# gen_mel_spec = pred[0].unsqueeze(0).permute(0, 2, 1)\n# ref_mel_spec = cond[0].unsqueeze(0).permute(0, 2, 1)\n# if self.vocoder_name == \"vocos\":\n# gen_audio = vocoder.decode(gen_mel_spec).cpu()\n# ref_audio = vocoder.decode(ref_mel_spec).cpu()\n# elif self.vocoder_name == \"bigvgan\":\n# gen_audio = vocoder(gen_mel_spec).squeeze(0).cpu()\n# ref_audio = vocoder(ref_mel_spec).squeeze(0).cpu()\n\n# torchaudio.save(f\"{log_samples_path}/step_{global_step}_gen.wav\", gen_audio, target_sample_rate)\n# torchaudio.save(f\"{log_samples_path}/step_{global_step}_ref.wav\", ref_audio, target_sample_rate)\n\n if global_step % self.last_per_steps == 0:\n self.save_checkpoint(global_step, last=True)\n\n self.save_checkpoint(global_step, last=True)\n\n self.accelerator.end_training()","source_hash":"2b9b27e8647e8f0a32d4b60a3a0bb41365c787bea1168dc1099632ac1d210ea7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.trainer.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model.trainer.__init__#L25-L132","kind":"function","name":"__init__","path":"src/f5_tts/model/trainer.py","language":"python","start_line":25,"end_line":132,"context_start_line":5,"context_end_line":152,"code":"\nimport torch\nimport torchaudio\nimport wandb\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nfrom ema_pytorch import EMA\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model import CFM\nfrom f5_tts.model.dataset import DynamicBatchSampler, collate_fn\nfrom f5_tts.model.utils import default, exists\n\n# trainer\n\n\nclass Trainer:\n def __init__(\n self,\n model: CFM,\n epochs,\n learning_rate,\n num_warmup_updates=20000,\n save_per_updates=1000,\n checkpoint_path=None,\n batch_size=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n noise_scheduler: str | None = None,\n duration_predictor: torch.nn.Module | None = None,\n logger: str | None = \"wandb\", # \"wandb\" | \"tensorboard\" | None\n wandb_project=\"test_e2-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n log_samples: bool = False,\n last_per_steps=None,\n accelerate_kwargs: dict = dict(),\n ema_kwargs: dict = dict(),\n bnb_optimizer: bool = False,\n mel_spec_type: str = \"vocos\", # \"vocos\" | \"bigvgan\"\n is_local_vocoder: bool = False, # use local path vocoder\n local_vocoder_path: str = \"\", # local vocoder path\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=True)\n\n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n print(f\"Using logger: {logger}\")\n self.log_samples = log_samples\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config={\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size\": batch_size,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"gpus\": self.accelerator.num_processes,\n \"noise_scheduler\": noise_scheduler,\n },\n )\n\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n self.model = model\n\n if self.is_main:\n self.ema_model = EMA(model, include_online_model=False, **ema_kwargs)\n self.ema_model.to(self.accelerator.device)\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_e2-tts\")\n\n self.batch_size = batch_size\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n # mel vocoder config\n self.vocoder_name = mel_spec_type\n self.is_local_vocoder = is_local_vocoder\n self.local_vocoder_path = local_vocoder_path\n\n self.noise_scheduler = noise_scheduler\n\n self.duration_predictor = duration_predictor\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n self.scale = None\n self.count = 0\n\n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n ema_model_state_dict=self.ema_model.state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n scale=self.scale,\n count=self.count,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:","source_hash":"2b9b27e8647e8f0a32d4b60a3a0bb41365c787bea1168dc1099632ac1d210ea7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.trainer.is_main","uri":"program://DMOSpeech2/function/src.f5_tts.model.trainer.is_main#L135-L136","kind":"function","name":"is_main","path":"src/f5_tts/model/trainer.py","language":"python","start_line":135,"end_line":136,"context_start_line":115,"context_end_line":156,"code":" self.vocoder_name = mel_spec_type\n self.is_local_vocoder = is_local_vocoder\n self.local_vocoder_path = local_vocoder_path\n\n self.noise_scheduler = noise_scheduler\n\n self.duration_predictor = duration_predictor\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n self.scale = None\n self.count = 0\n\n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n ema_model_state_dict=self.ema_model.state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n scale=self.scale,\n count=self.count,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at step {step}\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")","source_hash":"2b9b27e8647e8f0a32d4b60a3a0bb41365c787bea1168dc1099632ac1d210ea7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.trainer.save_checkpoint","uri":"program://DMOSpeech2/function/src.f5_tts.model.trainer.save_checkpoint#L138-L156","kind":"function","name":"save_checkpoint","path":"src/f5_tts/model/trainer.py","language":"python","start_line":138,"end_line":156,"context_start_line":118,"context_end_line":176,"code":"\n self.noise_scheduler = noise_scheduler\n\n self.duration_predictor = duration_predictor\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n \n self.scale = None\n self.count = 0\n\n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n ema_model_state_dict=self.ema_model.state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n scale=self.scale,\n count=self.count,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at step {step}\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n","source_hash":"2b9b27e8647e8f0a32d4b60a3a0bb41365c787bea1168dc1099632ac1d210ea7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.trainer.load_checkpoint","uri":"program://DMOSpeech2/function/src.f5_tts.model.trainer.load_checkpoint#L158-L214","kind":"function","name":"load_checkpoint","path":"src/f5_tts/model/trainer.py","language":"python","start_line":158,"end_line":214,"context_start_line":138,"context_end_line":234,"code":" def save_checkpoint(self, step, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main: \n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.accelerator.unwrap_model(self.optimizer).state_dict(),\n ema_model_state_dict=self.ema_model.state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n step=step,\n scale=self.scale,\n count=self.count,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at step {step}\")\n else:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{step}.pt\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith(\".pt\") for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n latest_checkpoint = sorted(\n [f for f in os.listdir(self.checkpoint_path) if f.endswith(\".pt\")],\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\")\n\n # patch for backward compatibility, 305e3ea\n for key in [\"ema_model.mel_spec.mel_stft.mel_scale.fb\", \"ema_model.mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"ema_model_state_dict\"]:\n del checkpoint[\"ema_model_state_dict\"][key]\n\n if self.is_main:\n self.ema_model.load_state_dict(checkpoint[\"ema_model_state_dict\"])\n\n if \"step\" in checkpoint:\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n step = checkpoint[\"step\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n\n if \"scale\" in checkpoint:\n self.scale = float(checkpoint[\"scale\"])\n self.model.scale = self.scale\n \n if \"count\" in checkpoint:\n self.count = int(checkpoint[\"count\"])\n \n del checkpoint\n gc.collect()\n return step\n\n def train(self, train_dataset: Dataset, num_workers=16, resumable_with_seed: int = None):\n if self.log_samples:\n from f5_tts.infer.utils_infer import cfg_strength, load_vocoder, nfe_step, sway_sampling_coef\n\n vocoder = load_vocoder(\n vocoder_name=self.vocoder_name, is_local=self.is_local_vocoder, local_path=self.local_vocoder_path\n )\n target_sample_rate = self.accelerator.unwrap_model(self.model).mel_spec.target_sample_rate\n log_samples_path = f\"{self.checkpoint_path}/samples\"\n os.makedirs(log_samples_path, exist_ok=True)\n\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(","source_hash":"2b9b27e8647e8f0a32d4b60a3a0bb41365c787bea1168dc1099632ac1d210ea7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.trainer.train","uri":"program://DMOSpeech2/function/src.f5_tts.model.trainer.train#L216-L415","kind":"function","name":"train","path":"src/f5_tts/model/trainer.py","language":"python","start_line":216,"end_line":415,"context_start_line":196,"context_end_line":415,"code":" else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n step = 0\n\n if \"scale\" in checkpoint:\n self.scale = float(checkpoint[\"scale\"])\n self.model.scale = self.scale\n \n if \"count\" in checkpoint:\n self.count = int(checkpoint[\"count\"])\n \n del checkpoint\n gc.collect()\n return step\n\n def train(self, train_dataset: Dataset, num_workers=16, resumable_with_seed: int = None):\n if self.log_samples:\n from f5_tts.infer.utils_infer import cfg_strength, load_vocoder, nfe_step, sway_sampling_coef\n\n vocoder = load_vocoder(\n vocoder_name=self.vocoder_name, is_local=self.is_local_vocoder, local_path=self.local_vocoder_path\n )\n target_sample_rate = self.accelerator.unwrap_model(self.model).mel_spec.target_sample_rate\n log_samples_path = f\"{self.checkpoint_path}/samples\"\n os.makedirs(log_samples_path, exist_ok=True)\n\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size,\n shuffle=True,\n generator=generator,\n )\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n\n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_steps = (\n self.num_warmup_updates * self.accelerator.num_processes\n ) # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_steps = len(train_dataloader) * self.epochs / self.grad_accumulation_steps\n decay_steps = total_steps - warmup_steps\n warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)\n decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)\n self.scheduler = SequentialLR(\n self.optimizer, schedulers=[warmup_scheduler, decay_scheduler], milestones=[warmup_steps]\n )\n train_dataloader, self.scheduler = self.accelerator.prepare(\n train_dataloader, self.scheduler\n ) # actual steps = 1 gpu steps / gpus\n start_step = self.load_checkpoint()\n global_step = start_step\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar = tqdm(\n skipped_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n initial=skipped_batch,\n total=orig_epoch_step,\n )\n else:\n progress_bar = tqdm(\n train_dataloader,\n desc=f\"Epoch {epoch+1}/{self.epochs}\",\n unit=\"step\",\n disable=not self.accelerator.is_local_main_process,\n )\n\n for batch in progress_bar:\n with self.accelerator.accumulate(self.model):\n text_inputs = batch[\"text\"]\n mel_spec = batch[\"mel\"].permute(0, 2, 1)\n mel_lengths = batch[\"mel_lengths\"]\n \n self.count += 1\n \n if self.scale is None:\n self.scale = mel_spec.std()\n else:\n self.scale += (mel_spec.std() - self.scale) / self.count\n \n mel_spec = mel_spec / self.scale # normalize mel spectrogram\n \n # TODO. add duration predictor training\n if self.duration_predictor is not None and self.accelerator.is_local_main_process:\n dur_loss = self.duration_predictor(mel_spec, lens=batch.get(\"durations\"))\n self.accelerator.log({\"duration loss\": dur_loss.item()}, step=global_step)\n\n loss, cond, pred, t = self.model(\n mel_spec, text=text_inputs, lens=mel_lengths, noise_scheduler=self.noise_scheduler\n )\n self.accelerator.backward(loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n self.optimizer.step()\n self.scheduler.step()\n self.optimizer.zero_grad()\n\n if self.is_main and self.accelerator.sync_gradients:\n self.ema_model.update()\n\n global_step += 1\n\n if self.accelerator.is_local_main_process:\n self.accelerator.log({\"loss\": loss.item(), \"lr\": self.scheduler.get_last_lr()[0]}, step=global_step)\n if self.logger == \"tensorboard\":\n self.writer.add_scalar(\"loss\", loss.item(), global_step)\n self.writer.add_scalar(\"lr\", self.scheduler.get_last_lr()[0], global_step)\n\n progress_bar.set_postfix(step=str(global_step), loss=loss.item())\n\n if global_step % (self.save_per_updates * self.grad_accumulation_steps) == 0:\n self.save_checkpoint(global_step)\n if self.log_samples and self.accelerator.is_local_main_process:\n gen_mel_spec = pred[0].unsqueeze(0).permute(0, 2, 1) * self.scale\n ref_mel_spec = cond[0].unsqueeze(0).permute(0, 2, 1) * self.scale\n with torch.inference_mode():\n if self.vocoder_name == \"vocos\":\n gen_audio = vocoder.decode(gen_mel_spec).cpu()\n ref_audio = vocoder.decode(ref_mel_spec).cpu()\n elif self.vocoder_name == \"bigvgan\":\n gen_audio = vocoder(gen_mel_spec).squeeze(0).cpu()\n ref_audio = vocoder(ref_mel_spec).squeeze(0).cpu()\n \n gen_audio = wandb.Audio(\n gen_audio.float().numpy().squeeze(),\n sample_rate=24000,\n caption=\"time: \" + str(t[0].squeeze().float().cpu().numpy())\n )\n ref_audio = wandb.Audio(\n ref_audio.float().numpy().squeeze(),\n sample_rate=24000,\n caption=\"time: \" + str(t[0].squeeze().float().cpu().numpy())\n )\n\n self.accelerator.log({\"gen_audio\": gen_audio, \n \"ref_audio\": ref_audio,\n }, step=global_step)\n\n\n# if self.log_samples and self.accelerator.is_local_main_process:\n# ref_audio_len = mel_lengths[0]\n# infer_text = [\n# text_inputs[0] + ([\" \"] if isinstance(text_inputs[0], list) else \" \") + text_inputs[0]\n# ]\n# with torch.inference_mode():\n# # generated, _ = self.accelerator.unwrap_model(self.model).sample(\n# # cond=mel_spec[0][:ref_audio_len].unsqueeze(0),\n# # text=infer_text,\n# # duration=ref_audio_len * 2,\n# # steps=nfe_step,\n# # cfg_strength=cfg_strength,\n# # sway_sampling_coef=sway_sampling_coef,\n# # )\n# # generated = generated.to(torch.float32)\n# # gen_mel_spec = generated[:, ref_audio_len:, :].permute(0, 2, 1).to(self.accelerator.device)\n# # ref_mel_spec = batch[\"mel\"][0].unsqueeze(0)\n# gen_mel_spec = pred[0].unsqueeze(0).permute(0, 2, 1)\n# ref_mel_spec = cond[0].unsqueeze(0).permute(0, 2, 1)\n# if self.vocoder_name == \"vocos\":\n# gen_audio = vocoder.decode(gen_mel_spec).cpu()\n# ref_audio = vocoder.decode(ref_mel_spec).cpu()\n# elif self.vocoder_name == \"bigvgan\":\n# gen_audio = vocoder(gen_mel_spec).squeeze(0).cpu()\n# ref_audio = vocoder(ref_mel_spec).squeeze(0).cpu()\n\n# torchaudio.save(f\"{log_samples_path}/step_{global_step}_gen.wav\", gen_audio, target_sample_rate)\n# torchaudio.save(f\"{log_samples_path}/step_{global_step}_ref.wav\", ref_audio, target_sample_rate)\n\n if global_step % self.last_per_steps == 0:\n self.save_checkpoint(global_step, last=True)\n\n self.save_checkpoint(global_step, last=True)\n\n self.accelerator.end_training()","source_hash":"2b9b27e8647e8f0a32d4b60a3a0bb41365c787bea1168dc1099632ac1d210ea7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.cfm","uri":"program://DMOSpeech2/module/src.f5_tts.model.cfm#L1-L293","kind":"module","name":"src.f5_tts.model.cfm","path":"src/f5_tts/model/cfm.py","language":"python","start_line":1,"end_line":293,"context_start_line":1,"context_end_line":293,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nfrom random import random\nfrom typing import Callable\n\nimport torch\nimport torch.nn.functional as F\nfrom torch import nn\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torchdiffeq import odeint\n\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import (\n default,\n exists,\n lens_to_mask,\n list_str_to_idx,\n list_str_to_tensor,\n mask_from_frac_lengths,\n)\n\n\nclass CFM(nn.Module):\n def __init__(\n self,\n transformer: nn.Module,\n sigma=0.0,\n odeint_kwargs: dict = dict(\n # atol = 1e-5,\n # rtol = 1e-5,\n method=\"euler\" # 'midpoint'\n ),\n audio_drop_prob=0.3,\n cond_drop_prob=0.2,\n num_channels=None,\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n frac_lengths_mask: tuple[float, float] = (0.7, 1.0),\n vocab_char_map: dict[str:int] | None = None,\n scale: float = 1.0,\n ):\n super().__init__()\n\n self.frac_lengths_mask = frac_lengths_mask\n\n # mel spec\n self.mel_spec = default(mel_spec_module, MelSpec(**mel_spec_kwargs))\n num_channels = default(num_channels, self.mel_spec.n_mel_channels)\n self.num_channels = num_channels\n\n # classifier-free guidance\n self.audio_drop_prob = audio_drop_prob\n self.cond_drop_prob = cond_drop_prob\n\n # transformer\n self.transformer = transformer\n dim = transformer.dim\n self.dim = dim\n\n # conditional flow related\n self.sigma = sigma\n\n # sampling related\n self.odeint_kwargs = odeint_kwargs\n\n # vocab map for tokenization\n self.vocab_char_map = vocab_char_map\n \n self.scale = scale\n\n @property\n def device(self):\n return next(self.parameters()).device\n\n @torch.no_grad()\n def sample(\n self,\n cond: float[\"b n d\"] | float[\"b nw\"], # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n duration: int | int[\"b\"], # noqa: F821\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n steps=32,\n cfg_strength=1.0,\n sway_sampling_coef=None,\n seed: int | None = None,\n max_duration=4096,\n vocoder: Callable[[float[\"b d n\"]], float[\"b nw\"]] | None = None, # noqa: F722\n no_ref_audio=False,\n duplicate_test=False,\n t_inter=0.1,\n edit_mask=None,\n ):\n self.eval()\n # raw wave\n\n if cond.ndim == 2:\n cond = self.mel_spec(cond)\n cond = cond.permute(0, 2, 1)\n assert cond.shape[-1] == self.num_channels\n\n cond = cond.to(next(self.parameters()).dtype)\n \n print(self.scale)\n\n cond = cond / self.scale\n \n batch, cond_seq_len, device = *cond.shape[:2], cond.device\n if not exists(lens):\n lens = torch.full((batch,), cond_seq_len, device=device, dtype=torch.long)\n\n # text\n\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n if exists(text):\n text_lens = (text != -1).sum(dim=-1)\n lens = torch.maximum(text_lens, lens) # make sure lengths are at least those of the text characters\n\n # duration\n\n cond_mask = lens_to_mask(lens)\n if edit_mask is not None:\n cond_mask = cond_mask & edit_mask\n\n if isinstance(duration, int):\n duration = torch.full((batch,), duration, device=device, dtype=torch.long)\n\n duration = torch.maximum(lens + 1, duration) # just add one token so something is generated\n duration = duration.clamp(max=max_duration)\n max_duration = duration.amax()\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n test_cond = F.pad(cond, (0, 0, cond_seq_len, max_duration - 2 * cond_seq_len), value=0.0)\n\n cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value=0.0)\n if no_ref_audio:\n cond = torch.zeros_like(cond)\n\n cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value=False)\n cond_mask = cond_mask.unsqueeze(-1)\n step_cond = torch.where(\n cond_mask, cond, torch.zeros_like(cond)\n ) # allow direct control (cut cond audio) with lens passed in\n\n if batch > 1:\n mask = lens_to_mask(duration)\n else: # save memory and speed up, as single inference need no mask currently\n mask = None\n\n # neural ode\n\n def fn(t, x):\n # at each step, conditioning is fixed\n # step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))\n\n # predict flow\n pred = self.transformer(\n x=x, cond=step_cond, text=text, time=t, mask=mask, drop_audio_cond=False, drop_text=False\n )\n if cfg_strength < 1e-5:\n return pred\n\n null_pred = self.transformer(\n x=x, cond=step_cond, text=text, time=t, mask=mask, drop_audio_cond=True, drop_text=True\n )\n return pred + (pred - null_pred) * cfg_strength\n\n # noise input\n # to make sure batch inference result is same with different batch size, and for sure single inference\n # still some difference maybe due to convolutional layers\n y0 = []\n for dur in duration:\n if exists(seed):\n torch.manual_seed(seed)\n y0.append(torch.randn(dur, self.num_channels, device=self.device, dtype=step_cond.dtype))\n y0 = pad_sequence(y0, padding_value=0, batch_first=True)\n\n t_start = 0\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n t_start = t_inter\n y0 = (1 - t_start) * y0 + t_start * test_cond\n steps = int(steps * (1 - t_start))\n\n t = torch.linspace(t_start, 1, steps + 1, device=self.device, dtype=step_cond.dtype)\n if sway_sampling_coef is not None:\n t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)\n\n trajectory = odeint(fn, y0, t, **self.odeint_kwargs)\n\n sampled = trajectory[-1]\n out = sampled\n out = torch.where(cond_mask, cond, out)\n\n out = out * self.scale\n \n if exists(vocoder):\n out = out.permute(0, 2, 1)\n out = vocoder(out)\n\n return out, trajectory\n\n def forward(\n self,\n inp: float[\"b n d\"] | float[\"b nw\"], # mel or raw wave # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n noise_scheduler: str | None = None,\n ):\n # handle raw wave\n if inp.ndim == 2:\n inp = self.mel_spec(inp)\n inp = inp.permute(0, 2, 1)\n assert inp.shape[-1] == self.num_channels\n\n batch, seq_len, dtype, device, _σ1 = *inp.shape[:2], inp.dtype, self.device, self.sigma\n\n # handle text as string\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # lens and mask\n if not exists(lens):\n lens = torch.full((batch,), seq_len, device=device)\n\n mask = lens_to_mask(lens, length=seq_len) # useless here, as collate_fn will pad to max length in batch\n\n # get a random span to mask out for training conditionally\n frac_lengths = torch.zeros((batch,), device=self.device).float().uniform_(*self.frac_lengths_mask)\n rand_span_mask = mask_from_frac_lengths(lens, frac_lengths)\n\n if exists(mask):\n rand_span_mask &= mask\n\n # mel is x1\n x1 = inp\n\n # x0 is gaussian noise\n x0 = torch.randn_like(x1)\n\n # time step\n time = torch.rand((batch,), dtype=dtype, device=self.device)\n # TODO. noise_scheduler\n\n # sample xt (φ_t(x) in the paper)\n t = time.unsqueeze(-1).unsqueeze(-1)\n φ = (1 - t) * x0 + t * x1\n flow = x1 - x0\n\n # only predict what is within the random mask span for infilling\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n\n # transformer and cfg training with a drop rate\n drop_audio_cond = random() < self.audio_drop_prob # p_drop in voicebox paper\n if random() < self.cond_drop_prob: # p_uncond in voicebox paper\n drop_audio_cond = True\n drop_text = True\n else:\n drop_text = False\n\n # if want rigourously mask out padding, record in collate_fn in dataset.py, and pass in here\n # adding mask will use more memory, thus also need to adjust batchsampler with scaled down threshold for long sequences\n pred = self.transformer(\n x=φ, cond=cond, text=text, time=time, drop_audio_cond=drop_audio_cond, drop_text=drop_text\n )\n\n # flow matching loss\n loss = F.mse_loss(pred, flow, reduction=\"none\")\n loss = loss[rand_span_mask]\n\n return loss.mean(), cond, pred, t","source_hash":"2e833f1d5f202c84edd5752b8412699b4cd2c92ca78790de2b19a8f31515d079","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.cfm.CFM","uri":"program://DMOSpeech2/class/src.f5_tts.model.cfm.CFM#L32-L293","kind":"class","name":"CFM","path":"src/f5_tts/model/cfm.py","language":"python","start_line":32,"end_line":293,"context_start_line":12,"context_end_line":293,"code":"from random import random\nfrom typing import Callable\n\nimport torch\nimport torch.nn.functional as F\nfrom torch import nn\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torchdiffeq import odeint\n\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import (\n default,\n exists,\n lens_to_mask,\n list_str_to_idx,\n list_str_to_tensor,\n mask_from_frac_lengths,\n)\n\n\nclass CFM(nn.Module):\n def __init__(\n self,\n transformer: nn.Module,\n sigma=0.0,\n odeint_kwargs: dict = dict(\n # atol = 1e-5,\n # rtol = 1e-5,\n method=\"euler\" # 'midpoint'\n ),\n audio_drop_prob=0.3,\n cond_drop_prob=0.2,\n num_channels=None,\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n frac_lengths_mask: tuple[float, float] = (0.7, 1.0),\n vocab_char_map: dict[str:int] | None = None,\n scale: float = 1.0,\n ):\n super().__init__()\n\n self.frac_lengths_mask = frac_lengths_mask\n\n # mel spec\n self.mel_spec = default(mel_spec_module, MelSpec(**mel_spec_kwargs))\n num_channels = default(num_channels, self.mel_spec.n_mel_channels)\n self.num_channels = num_channels\n\n # classifier-free guidance\n self.audio_drop_prob = audio_drop_prob\n self.cond_drop_prob = cond_drop_prob\n\n # transformer\n self.transformer = transformer\n dim = transformer.dim\n self.dim = dim\n\n # conditional flow related\n self.sigma = sigma\n\n # sampling related\n self.odeint_kwargs = odeint_kwargs\n\n # vocab map for tokenization\n self.vocab_char_map = vocab_char_map\n \n self.scale = scale\n\n @property\n def device(self):\n return next(self.parameters()).device\n\n @torch.no_grad()\n def sample(\n self,\n cond: float[\"b n d\"] | float[\"b nw\"], # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n duration: int | int[\"b\"], # noqa: F821\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n steps=32,\n cfg_strength=1.0,\n sway_sampling_coef=None,\n seed: int | None = None,\n max_duration=4096,\n vocoder: Callable[[float[\"b d n\"]], float[\"b nw\"]] | None = None, # noqa: F722\n no_ref_audio=False,\n duplicate_test=False,\n t_inter=0.1,\n edit_mask=None,\n ):\n self.eval()\n # raw wave\n\n if cond.ndim == 2:\n cond = self.mel_spec(cond)\n cond = cond.permute(0, 2, 1)\n assert cond.shape[-1] == self.num_channels\n\n cond = cond.to(next(self.parameters()).dtype)\n \n print(self.scale)\n\n cond = cond / self.scale\n \n batch, cond_seq_len, device = *cond.shape[:2], cond.device\n if not exists(lens):\n lens = torch.full((batch,), cond_seq_len, device=device, dtype=torch.long)\n\n # text\n\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n if exists(text):\n text_lens = (text != -1).sum(dim=-1)\n lens = torch.maximum(text_lens, lens) # make sure lengths are at least those of the text characters\n\n # duration\n\n cond_mask = lens_to_mask(lens)\n if edit_mask is not None:\n cond_mask = cond_mask & edit_mask\n\n if isinstance(duration, int):\n duration = torch.full((batch,), duration, device=device, dtype=torch.long)\n\n duration = torch.maximum(lens + 1, duration) # just add one token so something is generated\n duration = duration.clamp(max=max_duration)\n max_duration = duration.amax()\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n test_cond = F.pad(cond, (0, 0, cond_seq_len, max_duration - 2 * cond_seq_len), value=0.0)\n\n cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value=0.0)\n if no_ref_audio:\n cond = torch.zeros_like(cond)\n\n cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value=False)\n cond_mask = cond_mask.unsqueeze(-1)\n step_cond = torch.where(\n cond_mask, cond, torch.zeros_like(cond)\n ) # allow direct control (cut cond audio) with lens passed in\n\n if batch > 1:\n mask = lens_to_mask(duration)\n else: # save memory and speed up, as single inference need no mask currently\n mask = None\n\n # neural ode\n\n def fn(t, x):\n # at each step, conditioning is fixed\n # step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))\n\n # predict flow\n pred = self.transformer(\n x=x, cond=step_cond, text=text, time=t, mask=mask, drop_audio_cond=False, drop_text=False\n )\n if cfg_strength < 1e-5:\n return pred\n\n null_pred = self.transformer(\n x=x, cond=step_cond, text=text, time=t, mask=mask, drop_audio_cond=True, drop_text=True\n )\n return pred + (pred - null_pred) * cfg_strength\n\n # noise input\n # to make sure batch inference result is same with different batch size, and for sure single inference\n # still some difference maybe due to convolutional layers\n y0 = []\n for dur in duration:\n if exists(seed):\n torch.manual_seed(seed)\n y0.append(torch.randn(dur, self.num_channels, device=self.device, dtype=step_cond.dtype))\n y0 = pad_sequence(y0, padding_value=0, batch_first=True)\n\n t_start = 0\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n t_start = t_inter\n y0 = (1 - t_start) * y0 + t_start * test_cond\n steps = int(steps * (1 - t_start))\n\n t = torch.linspace(t_start, 1, steps + 1, device=self.device, dtype=step_cond.dtype)\n if sway_sampling_coef is not None:\n t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)\n\n trajectory = odeint(fn, y0, t, **self.odeint_kwargs)\n\n sampled = trajectory[-1]\n out = sampled\n out = torch.where(cond_mask, cond, out)\n\n out = out * self.scale\n \n if exists(vocoder):\n out = out.permute(0, 2, 1)\n out = vocoder(out)\n\n return out, trajectory\n\n def forward(\n self,\n inp: float[\"b n d\"] | float[\"b nw\"], # mel or raw wave # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n noise_scheduler: str | None = None,\n ):\n # handle raw wave\n if inp.ndim == 2:\n inp = self.mel_spec(inp)\n inp = inp.permute(0, 2, 1)\n assert inp.shape[-1] == self.num_channels\n\n batch, seq_len, dtype, device, _σ1 = *inp.shape[:2], inp.dtype, self.device, self.sigma\n\n # handle text as string\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # lens and mask\n if not exists(lens):\n lens = torch.full((batch,), seq_len, device=device)\n\n mask = lens_to_mask(lens, length=seq_len) # useless here, as collate_fn will pad to max length in batch\n\n # get a random span to mask out for training conditionally\n frac_lengths = torch.zeros((batch,), device=self.device).float().uniform_(*self.frac_lengths_mask)\n rand_span_mask = mask_from_frac_lengths(lens, frac_lengths)\n\n if exists(mask):\n rand_span_mask &= mask\n\n # mel is x1\n x1 = inp\n\n # x0 is gaussian noise\n x0 = torch.randn_like(x1)\n\n # time step\n time = torch.rand((batch,), dtype=dtype, device=self.device)\n # TODO. noise_scheduler\n\n # sample xt (φ_t(x) in the paper)\n t = time.unsqueeze(-1).unsqueeze(-1)\n φ = (1 - t) * x0 + t * x1\n flow = x1 - x0\n\n # only predict what is within the random mask span for infilling\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n\n # transformer and cfg training with a drop rate\n drop_audio_cond = random() < self.audio_drop_prob # p_drop in voicebox paper\n if random() < self.cond_drop_prob: # p_uncond in voicebox paper\n drop_audio_cond = True\n drop_text = True\n else:\n drop_text = False\n\n # if want rigourously mask out padding, record in collate_fn in dataset.py, and pass in here\n # adding mask will use more memory, thus also need to adjust batchsampler with scaled down threshold for long sequences\n pred = self.transformer(\n x=φ, cond=cond, text=text, time=time, drop_audio_cond=drop_audio_cond, drop_text=drop_text\n )\n\n # flow matching loss\n loss = F.mse_loss(pred, flow, reduction=\"none\")\n loss = loss[rand_span_mask]\n\n return loss.mean(), cond, pred, t","source_hash":"2e833f1d5f202c84edd5752b8412699b4cd2c92ca78790de2b19a8f31515d079","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.cfm.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model.cfm.__init__#L33-L78","kind":"function","name":"__init__","path":"src/f5_tts/model/cfm.py","language":"python","start_line":33,"end_line":78,"context_start_line":13,"context_end_line":98,"code":"from typing import Callable\n\nimport torch\nimport torch.nn.functional as F\nfrom torch import nn\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torchdiffeq import odeint\n\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import (\n default,\n exists,\n lens_to_mask,\n list_str_to_idx,\n list_str_to_tensor,\n mask_from_frac_lengths,\n)\n\n\nclass CFM(nn.Module):\n def __init__(\n self,\n transformer: nn.Module,\n sigma=0.0,\n odeint_kwargs: dict = dict(\n # atol = 1e-5,\n # rtol = 1e-5,\n method=\"euler\" # 'midpoint'\n ),\n audio_drop_prob=0.3,\n cond_drop_prob=0.2,\n num_channels=None,\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n frac_lengths_mask: tuple[float, float] = (0.7, 1.0),\n vocab_char_map: dict[str:int] | None = None,\n scale: float = 1.0,\n ):\n super().__init__()\n\n self.frac_lengths_mask = frac_lengths_mask\n\n # mel spec\n self.mel_spec = default(mel_spec_module, MelSpec(**mel_spec_kwargs))\n num_channels = default(num_channels, self.mel_spec.n_mel_channels)\n self.num_channels = num_channels\n\n # classifier-free guidance\n self.audio_drop_prob = audio_drop_prob\n self.cond_drop_prob = cond_drop_prob\n\n # transformer\n self.transformer = transformer\n dim = transformer.dim\n self.dim = dim\n\n # conditional flow related\n self.sigma = sigma\n\n # sampling related\n self.odeint_kwargs = odeint_kwargs\n\n # vocab map for tokenization\n self.vocab_char_map = vocab_char_map\n \n self.scale = scale\n\n @property\n def device(self):\n return next(self.parameters()).device\n\n @torch.no_grad()\n def sample(\n self,\n cond: float[\"b n d\"] | float[\"b nw\"], # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n duration: int | int[\"b\"], # noqa: F821\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n steps=32,\n cfg_strength=1.0,\n sway_sampling_coef=None,\n seed: int | None = None,\n max_duration=4096,\n vocoder: Callable[[float[\"b d n\"]], float[\"b nw\"]] | None = None, # noqa: F722\n no_ref_audio=False,","source_hash":"2e833f1d5f202c84edd5752b8412699b4cd2c92ca78790de2b19a8f31515d079","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.cfm.device","uri":"program://DMOSpeech2/function/src.f5_tts.model.cfm.device#L81-L82","kind":"function","name":"device","path":"src/f5_tts/model/cfm.py","language":"python","start_line":81,"end_line":82,"context_start_line":61,"context_end_line":102,"code":" self.audio_drop_prob = audio_drop_prob\n self.cond_drop_prob = cond_drop_prob\n\n # transformer\n self.transformer = transformer\n dim = transformer.dim\n self.dim = dim\n\n # conditional flow related\n self.sigma = sigma\n\n # sampling related\n self.odeint_kwargs = odeint_kwargs\n\n # vocab map for tokenization\n self.vocab_char_map = vocab_char_map\n \n self.scale = scale\n\n @property\n def device(self):\n return next(self.parameters()).device\n\n @torch.no_grad()\n def sample(\n self,\n cond: float[\"b n d\"] | float[\"b nw\"], # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n duration: int | int[\"b\"], # noqa: F821\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n steps=32,\n cfg_strength=1.0,\n sway_sampling_coef=None,\n seed: int | None = None,\n max_duration=4096,\n vocoder: Callable[[float[\"b d n\"]], float[\"b nw\"]] | None = None, # noqa: F722\n no_ref_audio=False,\n duplicate_test=False,\n t_inter=0.1,\n edit_mask=None,\n ):","source_hash":"2e833f1d5f202c84edd5752b8412699b4cd2c92ca78790de2b19a8f31515d079","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.cfm.sample","uri":"program://DMOSpeech2/function/src.f5_tts.model.cfm.sample#L85-L218","kind":"function","name":"sample","path":"src/f5_tts/model/cfm.py","language":"python","start_line":85,"end_line":218,"context_start_line":65,"context_end_line":238,"code":" self.transformer = transformer\n dim = transformer.dim\n self.dim = dim\n\n # conditional flow related\n self.sigma = sigma\n\n # sampling related\n self.odeint_kwargs = odeint_kwargs\n\n # vocab map for tokenization\n self.vocab_char_map = vocab_char_map\n \n self.scale = scale\n\n @property\n def device(self):\n return next(self.parameters()).device\n\n @torch.no_grad()\n def sample(\n self,\n cond: float[\"b n d\"] | float[\"b nw\"], # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n duration: int | int[\"b\"], # noqa: F821\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n steps=32,\n cfg_strength=1.0,\n sway_sampling_coef=None,\n seed: int | None = None,\n max_duration=4096,\n vocoder: Callable[[float[\"b d n\"]], float[\"b nw\"]] | None = None, # noqa: F722\n no_ref_audio=False,\n duplicate_test=False,\n t_inter=0.1,\n edit_mask=None,\n ):\n self.eval()\n # raw wave\n\n if cond.ndim == 2:\n cond = self.mel_spec(cond)\n cond = cond.permute(0, 2, 1)\n assert cond.shape[-1] == self.num_channels\n\n cond = cond.to(next(self.parameters()).dtype)\n \n print(self.scale)\n\n cond = cond / self.scale\n \n batch, cond_seq_len, device = *cond.shape[:2], cond.device\n if not exists(lens):\n lens = torch.full((batch,), cond_seq_len, device=device, dtype=torch.long)\n\n # text\n\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n if exists(text):\n text_lens = (text != -1).sum(dim=-1)\n lens = torch.maximum(text_lens, lens) # make sure lengths are at least those of the text characters\n\n # duration\n\n cond_mask = lens_to_mask(lens)\n if edit_mask is not None:\n cond_mask = cond_mask & edit_mask\n\n if isinstance(duration, int):\n duration = torch.full((batch,), duration, device=device, dtype=torch.long)\n\n duration = torch.maximum(lens + 1, duration) # just add one token so something is generated\n duration = duration.clamp(max=max_duration)\n max_duration = duration.amax()\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n test_cond = F.pad(cond, (0, 0, cond_seq_len, max_duration - 2 * cond_seq_len), value=0.0)\n\n cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value=0.0)\n if no_ref_audio:\n cond = torch.zeros_like(cond)\n\n cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value=False)\n cond_mask = cond_mask.unsqueeze(-1)\n step_cond = torch.where(\n cond_mask, cond, torch.zeros_like(cond)\n ) # allow direct control (cut cond audio) with lens passed in\n\n if batch > 1:\n mask = lens_to_mask(duration)\n else: # save memory and speed up, as single inference need no mask currently\n mask = None\n\n # neural ode\n\n def fn(t, x):\n # at each step, conditioning is fixed\n # step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))\n\n # predict flow\n pred = self.transformer(\n x=x, cond=step_cond, text=text, time=t, mask=mask, drop_audio_cond=False, drop_text=False\n )\n if cfg_strength < 1e-5:\n return pred\n\n null_pred = self.transformer(\n x=x, cond=step_cond, text=text, time=t, mask=mask, drop_audio_cond=True, drop_text=True\n )\n return pred + (pred - null_pred) * cfg_strength\n\n # noise input\n # to make sure batch inference result is same with different batch size, and for sure single inference\n # still some difference maybe due to convolutional layers\n y0 = []\n for dur in duration:\n if exists(seed):\n torch.manual_seed(seed)\n y0.append(torch.randn(dur, self.num_channels, device=self.device, dtype=step_cond.dtype))\n y0 = pad_sequence(y0, padding_value=0, batch_first=True)\n\n t_start = 0\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n t_start = t_inter\n y0 = (1 - t_start) * y0 + t_start * test_cond\n steps = int(steps * (1 - t_start))\n\n t = torch.linspace(t_start, 1, steps + 1, device=self.device, dtype=step_cond.dtype)\n if sway_sampling_coef is not None:\n t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)\n\n trajectory = odeint(fn, y0, t, **self.odeint_kwargs)\n\n sampled = trajectory[-1]\n out = sampled\n out = torch.where(cond_mask, cond, out)\n\n out = out * self.scale\n \n if exists(vocoder):\n out = out.permute(0, 2, 1)\n out = vocoder(out)\n\n return out, trajectory\n\n def forward(\n self,\n inp: float[\"b n d\"] | float[\"b nw\"], # mel or raw wave # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n noise_scheduler: str | None = None,\n ):\n # handle raw wave\n if inp.ndim == 2:\n inp = self.mel_spec(inp)\n inp = inp.permute(0, 2, 1)\n assert inp.shape[-1] == self.num_channels\n\n batch, seq_len, dtype, device, _σ1 = *inp.shape[:2], inp.dtype, self.device, self.sigma\n\n # handle text as string\n if isinstance(text, list):\n if exists(self.vocab_char_map):","source_hash":"2e833f1d5f202c84edd5752b8412699b4cd2c92ca78790de2b19a8f31515d079","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.cfm.forward","uri":"program://DMOSpeech2/function/src.f5_tts.model.cfm.forward#L220-L293","kind":"function","name":"forward","path":"src/f5_tts/model/cfm.py","language":"python","start_line":220,"end_line":293,"context_start_line":200,"context_end_line":293,"code":" steps = int(steps * (1 - t_start))\n\n t = torch.linspace(t_start, 1, steps + 1, device=self.device, dtype=step_cond.dtype)\n if sway_sampling_coef is not None:\n t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)\n\n trajectory = odeint(fn, y0, t, **self.odeint_kwargs)\n\n sampled = trajectory[-1]\n out = sampled\n out = torch.where(cond_mask, cond, out)\n\n out = out * self.scale\n \n if exists(vocoder):\n out = out.permute(0, 2, 1)\n out = vocoder(out)\n\n return out, trajectory\n\n def forward(\n self,\n inp: float[\"b n d\"] | float[\"b nw\"], # mel or raw wave # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n noise_scheduler: str | None = None,\n ):\n # handle raw wave\n if inp.ndim == 2:\n inp = self.mel_spec(inp)\n inp = inp.permute(0, 2, 1)\n assert inp.shape[-1] == self.num_channels\n\n batch, seq_len, dtype, device, _σ1 = *inp.shape[:2], inp.dtype, self.device, self.sigma\n\n # handle text as string\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # lens and mask\n if not exists(lens):\n lens = torch.full((batch,), seq_len, device=device)\n\n mask = lens_to_mask(lens, length=seq_len) # useless here, as collate_fn will pad to max length in batch\n\n # get a random span to mask out for training conditionally\n frac_lengths = torch.zeros((batch,), device=self.device).float().uniform_(*self.frac_lengths_mask)\n rand_span_mask = mask_from_frac_lengths(lens, frac_lengths)\n\n if exists(mask):\n rand_span_mask &= mask\n\n # mel is x1\n x1 = inp\n\n # x0 is gaussian noise\n x0 = torch.randn_like(x1)\n\n # time step\n time = torch.rand((batch,), dtype=dtype, device=self.device)\n # TODO. noise_scheduler\n\n # sample xt (φ_t(x) in the paper)\n t = time.unsqueeze(-1).unsqueeze(-1)\n φ = (1 - t) * x0 + t * x1\n flow = x1 - x0\n\n # only predict what is within the random mask span for infilling\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n\n # transformer and cfg training with a drop rate\n drop_audio_cond = random() < self.audio_drop_prob # p_drop in voicebox paper\n if random() < self.cond_drop_prob: # p_uncond in voicebox paper\n drop_audio_cond = True\n drop_text = True\n else:\n drop_text = False\n\n # if want rigourously mask out padding, record in collate_fn in dataset.py, and pass in here\n # adding mask will use more memory, thus also need to adjust batchsampler with scaled down threshold for long sequences\n pred = self.transformer(\n x=φ, cond=cond, text=text, time=time, drop_audio_cond=drop_audio_cond, drop_text=drop_text\n )\n\n # flow matching loss\n loss = F.mse_loss(pred, flow, reduction=\"none\")\n loss = loss[rand_span_mask]\n\n return loss.mean(), cond, pred, t","source_hash":"2e833f1d5f202c84edd5752b8412699b4cd2c92ca78790de2b19a8f31515d079","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.cfm.fn","uri":"program://DMOSpeech2/function/src.f5_tts.model.cfm.fn#L168-L182","kind":"function","name":"fn","path":"src/f5_tts/model/cfm.py","language":"python","start_line":168,"end_line":182,"context_start_line":148,"context_end_line":202,"code":" if duplicate_test:\n test_cond = F.pad(cond, (0, 0, cond_seq_len, max_duration - 2 * cond_seq_len), value=0.0)\n\n cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value=0.0)\n if no_ref_audio:\n cond = torch.zeros_like(cond)\n\n cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value=False)\n cond_mask = cond_mask.unsqueeze(-1)\n step_cond = torch.where(\n cond_mask, cond, torch.zeros_like(cond)\n ) # allow direct control (cut cond audio) with lens passed in\n\n if batch > 1:\n mask = lens_to_mask(duration)\n else: # save memory and speed up, as single inference need no mask currently\n mask = None\n\n # neural ode\n\n def fn(t, x):\n # at each step, conditioning is fixed\n # step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))\n\n # predict flow\n pred = self.transformer(\n x=x, cond=step_cond, text=text, time=t, mask=mask, drop_audio_cond=False, drop_text=False\n )\n if cfg_strength < 1e-5:\n return pred\n\n null_pred = self.transformer(\n x=x, cond=step_cond, text=text, time=t, mask=mask, drop_audio_cond=True, drop_text=True\n )\n return pred + (pred - null_pred) * cfg_strength\n\n # noise input\n # to make sure batch inference result is same with different batch size, and for sure single inference\n # still some difference maybe due to convolutional layers\n y0 = []\n for dur in duration:\n if exists(seed):\n torch.manual_seed(seed)\n y0.append(torch.randn(dur, self.num_channels, device=self.device, dtype=step_cond.dtype))\n y0 = pad_sequence(y0, padding_value=0, batch_first=True)\n\n t_start = 0\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n t_start = t_inter\n y0 = (1 - t_start) * y0 + t_start * test_cond\n steps = int(steps * (1 - t_start))\n\n t = torch.linspace(t_start, 1, steps + 1, device=self.device, dtype=step_cond.dtype)","source_hash":"2e833f1d5f202c84edd5752b8412699b4cd2c92ca78790de2b19a8f31515d079","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.dataset","uri":"program://DMOSpeech2/module/src.f5_tts.model.dataset#L1-L460","kind":"module","name":"src.f5_tts.model.dataset","path":"src/f5_tts/model/dataset.py","language":"python","start_line":1,"end_line":460,"context_start_line":1,"context_end_line":460,"code":"import re\nimport json\nimport random\nfrom importlib.resources import files\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom datasets import Dataset as Dataset_\nfrom datasets import load_from_disk\nfrom torch import nn\nfrom torch.utils.data import Dataset, Sampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import default\n\n\n\n\ndef get_speaker_id(path):\n parts = path.split('/')\n speaker_id = parts[-3]\n return speaker_id\n\n\nclass CustomDataset(Dataset):\n def __init__(\n self,\n custom_dataset: Dataset,\n durations=None,\n target_sample_rate=24_000,\n hop_length=256,\n n_mel_channels=100,\n n_fft=1024,\n win_length=1024,\n mel_spec_type=\"vocos\",\n preprocessed_mel=False,\n mel_spec_module: nn.Module | None = None,\n validation=False,\n validation_num=5000,\n data_augmentation=False,\n return_wavform=False,\n remove_starting_space=True,\n need_prompt_speech=False,\n prompt_repository: dict=None,\n ):\n self.data = custom_dataset\n self.durations = durations\n self.target_sample_rate = target_sample_rate\n self.hop_length = hop_length\n self.n_fft = n_fft\n self.win_length = win_length\n self.mel_spec_type = mel_spec_type\n self.preprocessed_mel = preprocessed_mel\n\n if not preprocessed_mel:\n self.mel_spectrogram = default(\n mel_spec_module,\n MelSpec(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n ),\n )\n \n self.validation = validation\n self.validation_num = validation_num\n\n if (not validation) and data_augmentation:\n print('Using data augmentation.')\n self.augment = Compose([\n AddBackgroundNoise(\n sounds_path=\"/data5/ESC-50-master\",\n min_snr_db=3.0,\n max_snr_db=30.0,\n noise_transform=PolarityInversion(),\n p=0.5\n ),\n AddGaussianNoise(\n min_amplitude=0.001,\n max_amplitude=0.015,\n p=0.5\n ), \n PitchShift(\n min_semitones=-12.0,\n max_semitones=12.0,\n p=0.8\n ),\n ApplyImpulseResponse(ir_path=\"/data5/Audio\", p=1.0),\n Aliasing(min_sample_rate=4000, max_sample_rate=30000, p=0.3),\n BandPassFilter(min_center_freq=100.0, max_center_freq=6000, p=0.2),\n SevenBandParametricEQ(p=0.2),\n TanhDistortion(\n min_distortion=0.01,\n max_distortion=0.7,\n p=0.2\n ),\n ])\n else:\n print('No data augmentation.')\n self.augment = None\n\n self.return_wavform = return_wavform\n self.remove_starting_space = remove_starting_space\n\n if need_prompt_speech:\n if prompt_repository == None:\n self.prompt_repository = {}\n for row in tqdm(self.data):\n audio_path = row[\"audio_path\"]\n text = row[\"text\"]\n duration = row[\"duration\"]\n spk_id = get_speaker_id(audio_path)\n assert spk_id != None and spk_id != 'mp3'\n if spk_id not in self.prompt_repository:\n self.prompt_repository[spk_id] = [row]\n else:\n self.prompt_repository[spk_id].append(row)\n else:\n self.prompt_repository = prompt_repository\n\n print(f'Grouped samples into {len(self.prompt_repository.keys())} speakers.') \n self.need_prompt_speech = True\n\n else:\n self.need_prompt_speech = False\n\n\n def get_frame_len(self, index):\n if self.validation:\n index += len(self.data) - self.validation_num\n\n if (\n self.durations is not None\n ): # Please make sure the separately provided durations are correct, otherwise 99.99% OOM\n return self.durations[index] * self.target_sample_rate / self.hop_length\n return self.data[index][\"duration\"] * self.target_sample_rate / self.hop_length\n\n def __len__(self):\n if not self.validation:\n return len(self.data) - self.validation_num\n return self.validation_num\n\n def __getitem__(self, index, return_row=True, return_path=False):\n if self.validation:\n index += len(self.data) - self.validation_num\n\n out = {}\n\n while True:\n row = self.data[index]\n audio_path = row[\"audio_path\"]\n text = row[\"text\"]\n duration = row[\"duration\"]\n\n if not isinstance(text, list):\n text = list(text)\n\n # filter by given length\n if (0.3 <= duration <= 30) and (0 < len(text) < 2048):\n break # valid\n\n index = (index + 1) % len(self.data)\n\n if self.remove_starting_space:\n while len(text) > 1 and text[0] == ' ':\n text = text[1:]\n \n if self.preprocessed_mel:\n mel_spec = torch.tensor(row[\"mel_spec\"])\n else:\n audio, source_sample_rate = torchaudio.load(audio_path)\n\n # make sure mono input\n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n\n # resample if necessary\n if source_sample_rate != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(source_sample_rate, self.target_sample_rate)\n audio = resampler(audio)\n\n if not self.validation:\n if self.augment != None:\n audio = self.augment(audio.squeeze().numpy(), sample_rate=self.target_sample_rate)\n audio = torch.from_numpy(audio).float().unsqueeze(0)\n\n # to mel spectrogram\n mel_spec = self.mel_spectrogram(audio)\n mel_spec = mel_spec.squeeze(0) # '1 d t -> d t'\n\n out['mel_spec'] = mel_spec\n out['text'] = text\n out['duration'] = duration\n out['target_text'] = self.data[(index + len(self.data) // 2) % len(self.data)][\"text\"]\n\n if self.return_wavform:\n out['wav'] = audio\n\n if return_path:\n out['path'] = audio_path\n\n if return_row:\n out['row'] = row\n\n # Sample a prompt speech of the same speaker\n # From prompt_repository\n if self.need_prompt_speech:\n spk = get_speaker_id(audio_path)\n spk_repository = self.prompt_repository[spk]\n _count = 100\n while True:\n pmt_row = random.choice(spk_repository)\n pmt_audio_path = pmt_row['audio_path']\n pmt_text = pmt_row['text']\n pmt_duration = pmt_row['duration']\n\n if not isinstance(pmt_text, list):\n pmt_text = list(pmt_text)\n\n # filter by given length\n if 0.3 <= pmt_duration <= 30 and (0 < len(pmt_text) < 2048):\n if pmt_text != text:\n break\n _count = _count - 1\n if _count <= 0:\n break\n\n if self.remove_starting_space:\n while len(pmt_text) > 1 and pmt_text[0] == ' ':\n pmt_text = pmt_text[1:]\n \n if self.preprocessed_mel:\n pmt_mel_spec = torch.tensor(pmt_row[\"mel_spec\"])\n else:\n pmt_audio, source_sample_rate = torchaudio.load(pmt_audio_path)\n\n # make sure mono input\n if pmt_audio.shape[0] > 1:\n pmt_audio = torch.mean(pmt_audio, dim=0, keepdim=True)\n\n # resample if necessary\n if source_sample_rate != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(source_sample_rate, self.target_sample_rate)\n pmt_audio = resampler(pmt_audio)\n\n if not self.validation:\n if self.augment != None:\n pmt_audio = self.augment(pmt_audio.squeeze().numpy(), sample_rate=self.target_sample_rate)\n pmt_audio = torch.from_numpy(pmt_audio).float().unsqueeze(0)\n\n # to mel spectrogram\n pmt_mel_spec = self.mel_spectrogram(pmt_audio)\n pmt_mel_spec = pmt_mel_spec.squeeze(0) # '1 d t -> d t'\n\n out['pmt_mel_spec'] = pmt_mel_spec\n out['pmt_text'] = pmt_text\n out['pmt_duration'] = pmt_duration\n\n if self.return_wavform:\n out['pmt_wav'] = pmt_audio\n\n if return_path:\n out['pmt_path'] = pmt_audio_path\n\n if return_row:\n out['pmt_row'] = pmt_row\n\n return out\n\n\n# Dynamic Batch Sampler\nclass DynamicBatchSampler(Sampler[list[int]]):\n \"\"\"Extension of Sampler that will do the following:\n 1. Change the batch size (essentially number of sequences)\n in a batch to ensure that the total number of frames are less\n than a certain threshold.\n 2. Make sure the padding efficiency in the batch is high.\n \"\"\"\n\n def __init__(\n self, sampler: Sampler[int], frames_threshold: int, max_samples=0, random_seed=None, drop_last: bool = False\n ):\n self.sampler = sampler\n self.frames_threshold = frames_threshold\n self.max_samples = max_samples\n\n indices, batches = [], []\n data_source = self.sampler.data_source\n\n # for idx in tqdm(\n # self.sampler, desc=\"Sorting with sampler... if slow, check whether dataset is provided with duration\"\n # ):\n for idx in self.sampler:\n indices.append((idx, data_source.get_frame_len(idx)))\n indices.sort(key=lambda elem: elem[1])\n\n batch = []\n batch_frames = 0\n # for idx, frame_len in tqdm(\n # indices, desc=f\"Creating dynamic batches with {frames_threshold} audio frames per gpu\"\n # ):\n for idx, frame_len in indices:\n if batch_frames + frame_len <= self.frames_threshold and (max_samples == 0 or len(batch) < max_samples):\n batch.append(idx)\n batch_frames += frame_len\n else:\n if len(batch) > 0:\n batches.append(batch)\n if frame_len <= self.frames_threshold:\n batch = [idx]\n batch_frames = frame_len\n else:\n batch = []\n batch_frames = 0\n\n if not drop_last and len(batch) > 0:\n batches.append(batch)\n\n del indices\n\n # if want to have different batches between epochs, may just set a seed and log it in ckpt\n # cuz during multi-gpu training, although the batch on per gpu not change between epochs, the formed general minibatch is different\n # e.g. for epoch n, use (random_seed + n)\n random.seed(random_seed)\n random.shuffle(batches)\n\n self.batches = batches\n\n def __iter__(self):\n return iter(self.batches)\n\n def __len__(self):\n return len(self.batches)\n\n\n# Load dataset\n\ndef load_dataset(\n dataset_name: str,\n tokenizer: str = \"pinyin\",\n dataset_type: str = \"CustomDataset\",\n audio_type: str = \"raw\",\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n split: str = \"train\",\n data_augmentation: bool = False,\n return_wavform: bool = False,\n remove_starting_space: bool = True,\n need_prompt_speech: bool = False,\n prompt_repository: dict = None\n) -> CustomDataset:\n \"\"\"\n dataset_type - \"CustomDataset\" if you want to use tokenizer name and default data path to load for train_dataset\n - \"CustomDatasetPath\" if you just want to pass the full path to a preprocessed dataset without relying on tokenizer\n \"\"\"\n\n print(\"Loading dataset ...\")\n\n if dataset_type == \"CustomDataset\":\n rel_data_path = str(f'/home/yl4579/F5-TTS-diff/F5-TTS-DMD-flow-ds/data/{dataset_name}_{tokenizer}')\n if 'LibriTTS_100_360_500_char_pinyin' in rel_data_path:\n rel_data_path = rel_data_path.replace('LibriTTS_100_360_500_char_pinyin', 'LibriTTS_100_360_500_char')\n if audio_type == \"raw\":\n try:\n train_dataset = load_from_disk(f\"{rel_data_path}/raw\")\n except: # noqa: E722\n train_dataset = Dataset_.from_file(f\"{rel_data_path}/raw.arrow\")\n preprocessed_mel = False\n elif audio_type == \"mel\":\n train_dataset = Dataset_.from_file(f\"{rel_data_path}/mel.arrow\")\n preprocessed_mel = True\n with open(f\"{rel_data_path}/duration.json\", \"r\", encoding=\"utf-8\") as f:\n data_dict = json.load(f)\n durations = data_dict[\"duration\"]\n train_dataset = CustomDataset(\n train_dataset,\n durations=durations,\n preprocessed_mel=preprocessed_mel,\n mel_spec_module=mel_spec_module,\n **mel_spec_kwargs,\n validation=split == \"val\",\n data_augmentation=data_augmentation,\n return_wavform=return_wavform,\n remove_starting_space=remove_starting_space,\n need_prompt_speech=need_prompt_speech,\n prompt_repository=prompt_repository\n )\n\n elif dataset_type == \"CustomDatasetPath\":\n try:\n train_dataset = load_from_disk(f\"{dataset_name}/raw\")\n except: # noqa: E722\n train_dataset = Dataset_.from_file(f\"{dataset_name}/raw.arrow\")\n\n with open(f\"{dataset_name}/duration.json\", \"r\", encoding=\"utf-8\") as f:\n data_dict = json.load(f)\n durations = data_dict[\"duration\"]\n train_dataset = CustomDataset(\n train_dataset, durations=durations, preprocessed_mel=preprocessed_mel, **mel_spec_kwargs\n )\n\n return train_dataset\n\n\n# collation\ndef collate_fn(batch):\n # Extract mel_specs and their lengths\n mel_specs = [item[\"mel_spec\"].squeeze(0) for item in batch]\n mel_lengths = torch.LongTensor([spec.shape[-1] for spec in mel_specs])\n max_mel_length = mel_lengths.amax()\n \n # Pad mel_specs\n padded_mel_specs = []\n for spec in mel_specs: # TODO. maybe records mask for attention here\n padding = (0, max_mel_length - spec.size(-1))\n padded_spec = F.pad(spec, padding, value=0)\n padded_mel_specs.append(padded_spec)\n mel_specs = torch.stack(padded_mel_specs)\n\n text = [item['text'] for item in batch]\n target_text = [item['target_text'] for item in batch]\n\n text_lengths = torch.LongTensor([len(item) for item in text])\n\n out = dict(\n mel=mel_specs,\n mel_lengths=mel_lengths,\n text=text,\n text_lengths=text_lengths,\n target_text=target_text,\n )\n\n if 'pmt_mel_spec' in batch[0]:\n pmt_mel_specs = [item[\"pmt_mel_spec\"].squeeze(0) for item in batch]\n pmt_mel_lengths = torch.LongTensor([spec.shape[-1] for spec in pmt_mel_specs])\n max_pmt_mel_length = pmt_mel_lengths.amax()\n \n # Pad mel_specs\n padded_pmt_mel_specs = []\n for spec in pmt_mel_specs: \n padding = (0, max_pmt_mel_length - spec.size(-1))\n padded_spec = F.pad(spec, padding, value=0)\n padded_pmt_mel_specs.append(padded_spec)\n pmt_mel_specs = torch.stack(padded_pmt_mel_specs)\n\n out['pmt_mel_specs'] = pmt_mel_specs\n\n if 'pmt_text' in batch[0]:\n pmt_text = [item['pmt_text'] for item in batch]\n pmt_text_lengths = torch.LongTensor([len(item) for item in pmt_text])\n\n out['pmt_text'] = pmt_text\n out['pmt_text_lengths'] = pmt_text_lengths\n\n return out","source_hash":"2ed7497956c0ce04b71926e3f330898dec956dea8a40d270c1682a03f8899787","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.dataset.get_speaker_id","uri":"program://DMOSpeech2/function/src.f5_tts.model.dataset.get_speaker_id#L21-L24","kind":"function","name":"get_speaker_id","path":"src/f5_tts/model/dataset.py","language":"python","start_line":21,"end_line":24,"context_start_line":1,"context_end_line":44,"code":"import re\nimport json\nimport random\nfrom importlib.resources import files\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom datasets import Dataset as Dataset_\nfrom datasets import load_from_disk\nfrom torch import nn\nfrom torch.utils.data import Dataset, Sampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import default\n\n\n\n\ndef get_speaker_id(path):\n parts = path.split('/')\n speaker_id = parts[-3]\n return speaker_id\n\n\nclass CustomDataset(Dataset):\n def __init__(\n self,\n custom_dataset: Dataset,\n durations=None,\n target_sample_rate=24_000,\n hop_length=256,\n n_mel_channels=100,\n n_fft=1024,\n win_length=1024,\n mel_spec_type=\"vocos\",\n preprocessed_mel=False,\n mel_spec_module: nn.Module | None = None,\n validation=False,\n validation_num=5000,\n data_augmentation=False,\n return_wavform=False,\n remove_starting_space=True,","source_hash":"2ed7497956c0ce04b71926e3f330898dec956dea8a40d270c1682a03f8899787","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.dataset.CustomDataset","uri":"program://DMOSpeech2/class/src.f5_tts.model.dataset.CustomDataset#L27-L273","kind":"class","name":"CustomDataset","path":"src/f5_tts/model/dataset.py","language":"python","start_line":27,"end_line":273,"context_start_line":7,"context_end_line":293,"code":"import torch.nn.functional as F\nimport torchaudio\nfrom datasets import Dataset as Dataset_\nfrom datasets import load_from_disk\nfrom torch import nn\nfrom torch.utils.data import Dataset, Sampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import default\n\n\n\n\ndef get_speaker_id(path):\n parts = path.split('/')\n speaker_id = parts[-3]\n return speaker_id\n\n\nclass CustomDataset(Dataset):\n def __init__(\n self,\n custom_dataset: Dataset,\n durations=None,\n target_sample_rate=24_000,\n hop_length=256,\n n_mel_channels=100,\n n_fft=1024,\n win_length=1024,\n mel_spec_type=\"vocos\",\n preprocessed_mel=False,\n mel_spec_module: nn.Module | None = None,\n validation=False,\n validation_num=5000,\n data_augmentation=False,\n return_wavform=False,\n remove_starting_space=True,\n need_prompt_speech=False,\n prompt_repository: dict=None,\n ):\n self.data = custom_dataset\n self.durations = durations\n self.target_sample_rate = target_sample_rate\n self.hop_length = hop_length\n self.n_fft = n_fft\n self.win_length = win_length\n self.mel_spec_type = mel_spec_type\n self.preprocessed_mel = preprocessed_mel\n\n if not preprocessed_mel:\n self.mel_spectrogram = default(\n mel_spec_module,\n MelSpec(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n ),\n )\n \n self.validation = validation\n self.validation_num = validation_num\n\n if (not validation) and data_augmentation:\n print('Using data augmentation.')\n self.augment = Compose([\n AddBackgroundNoise(\n sounds_path=\"/data5/ESC-50-master\",\n min_snr_db=3.0,\n max_snr_db=30.0,\n noise_transform=PolarityInversion(),\n p=0.5\n ),\n AddGaussianNoise(\n min_amplitude=0.001,\n max_amplitude=0.015,\n p=0.5\n ), \n PitchShift(\n min_semitones=-12.0,\n max_semitones=12.0,\n p=0.8\n ),\n ApplyImpulseResponse(ir_path=\"/data5/Audio\", p=1.0),\n Aliasing(min_sample_rate=4000, max_sample_rate=30000, p=0.3),\n BandPassFilter(min_center_freq=100.0, max_center_freq=6000, p=0.2),\n SevenBandParametricEQ(p=0.2),\n TanhDistortion(\n min_distortion=0.01,\n max_distortion=0.7,\n p=0.2\n ),\n ])\n else:\n print('No data augmentation.')\n self.augment = None\n\n self.return_wavform = return_wavform\n self.remove_starting_space = remove_starting_space\n\n if need_prompt_speech:\n if prompt_repository == None:\n self.prompt_repository = {}\n for row in tqdm(self.data):\n audio_path = row[\"audio_path\"]\n text = row[\"text\"]\n duration = row[\"duration\"]\n spk_id = get_speaker_id(audio_path)\n assert spk_id != None and spk_id != 'mp3'\n if spk_id not in self.prompt_repository:\n self.prompt_repository[spk_id] = [row]\n else:\n self.prompt_repository[spk_id].append(row)\n else:\n self.prompt_repository = prompt_repository\n\n print(f'Grouped samples into {len(self.prompt_repository.keys())} speakers.') \n self.need_prompt_speech = True\n\n else:\n self.need_prompt_speech = False\n\n\n def get_frame_len(self, index):\n if self.validation:\n index += len(self.data) - self.validation_num\n\n if (\n self.durations is not None\n ): # Please make sure the separately provided durations are correct, otherwise 99.99% OOM\n return self.durations[index] * self.target_sample_rate / self.hop_length\n return self.data[index][\"duration\"] * self.target_sample_rate / self.hop_length\n\n def __len__(self):\n if not self.validation:\n return len(self.data) - self.validation_num\n return self.validation_num\n\n def __getitem__(self, index, return_row=True, return_path=False):\n if self.validation:\n index += len(self.data) - self.validation_num\n\n out = {}\n\n while True:\n row = self.data[index]\n audio_path = row[\"audio_path\"]\n text = row[\"text\"]\n duration = row[\"duration\"]\n\n if not isinstance(text, list):\n text = list(text)\n\n # filter by given length\n if (0.3 <= duration <= 30) and (0 < len(text) < 2048):\n break # valid\n\n index = (index + 1) % len(self.data)\n\n if self.remove_starting_space:\n while len(text) > 1 and text[0] == ' ':\n text = text[1:]\n \n if self.preprocessed_mel:\n mel_spec = torch.tensor(row[\"mel_spec\"])\n else:\n audio, source_sample_rate = torchaudio.load(audio_path)\n\n # make sure mono input\n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n\n # resample if necessary\n if source_sample_rate != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(source_sample_rate, self.target_sample_rate)\n audio = resampler(audio)\n\n if not self.validation:\n if self.augment != None:\n audio = self.augment(audio.squeeze().numpy(), sample_rate=self.target_sample_rate)\n audio = torch.from_numpy(audio).float().unsqueeze(0)\n\n # to mel spectrogram\n mel_spec = self.mel_spectrogram(audio)\n mel_spec = mel_spec.squeeze(0) # '1 d t -> d t'\n\n out['mel_spec'] = mel_spec\n out['text'] = text\n out['duration'] = duration\n out['target_text'] = self.data[(index + len(self.data) // 2) % len(self.data)][\"text\"]\n\n if self.return_wavform:\n out['wav'] = audio\n\n if return_path:\n out['path'] = audio_path\n\n if return_row:\n out['row'] = row\n\n # Sample a prompt speech of the same speaker\n # From prompt_repository\n if self.need_prompt_speech:\n spk = get_speaker_id(audio_path)\n spk_repository = self.prompt_repository[spk]\n _count = 100\n while True:\n pmt_row = random.choice(spk_repository)\n pmt_audio_path = pmt_row['audio_path']\n pmt_text = pmt_row['text']\n pmt_duration = pmt_row['duration']\n\n if not isinstance(pmt_text, list):\n pmt_text = list(pmt_text)\n\n # filter by given length\n if 0.3 <= pmt_duration <= 30 and (0 < len(pmt_text) < 2048):\n if pmt_text != text:\n break\n _count = _count - 1\n if _count <= 0:\n break\n\n if self.remove_starting_space:\n while len(pmt_text) > 1 and pmt_text[0] == ' ':\n pmt_text = pmt_text[1:]\n \n if self.preprocessed_mel:\n pmt_mel_spec = torch.tensor(pmt_row[\"mel_spec\"])\n else:\n pmt_audio, source_sample_rate = torchaudio.load(pmt_audio_path)\n\n # make sure mono input\n if pmt_audio.shape[0] > 1:\n pmt_audio = torch.mean(pmt_audio, dim=0, keepdim=True)\n\n # resample if necessary\n if source_sample_rate != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(source_sample_rate, self.target_sample_rate)\n pmt_audio = resampler(pmt_audio)\n\n if not self.validation:\n if self.augment != None:\n pmt_audio = self.augment(pmt_audio.squeeze().numpy(), sample_rate=self.target_sample_rate)\n pmt_audio = torch.from_numpy(pmt_audio).float().unsqueeze(0)\n\n # to mel spectrogram\n pmt_mel_spec = self.mel_spectrogram(pmt_audio)\n pmt_mel_spec = pmt_mel_spec.squeeze(0) # '1 d t -> d t'\n\n out['pmt_mel_spec'] = pmt_mel_spec\n out['pmt_text'] = pmt_text\n out['pmt_duration'] = pmt_duration\n\n if self.return_wavform:\n out['pmt_wav'] = pmt_audio\n\n if return_path:\n out['pmt_path'] = pmt_audio_path\n\n if return_row:\n out['pmt_row'] = pmt_row\n\n return out\n\n\n# Dynamic Batch Sampler\nclass DynamicBatchSampler(Sampler[list[int]]):\n \"\"\"Extension of Sampler that will do the following:\n 1. Change the batch size (essentially number of sequences)\n in a batch to ensure that the total number of frames are less\n than a certain threshold.\n 2. Make sure the padding efficiency in the batch is high.\n \"\"\"\n\n def __init__(\n self, sampler: Sampler[int], frames_threshold: int, max_samples=0, random_seed=None, drop_last: bool = False\n ):\n self.sampler = sampler\n self.frames_threshold = frames_threshold\n self.max_samples = max_samples\n\n indices, batches = [], []\n data_source = self.sampler.data_source","source_hash":"2ed7497956c0ce04b71926e3f330898dec956dea8a40d270c1682a03f8899787","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.dataset.DynamicBatchSampler","uri":"program://DMOSpeech2/class/src.f5_tts.model.dataset.DynamicBatchSampler#L277-L338","kind":"class","name":"DynamicBatchSampler","path":"src/f5_tts/model/dataset.py","language":"python","start_line":277,"end_line":338,"context_start_line":257,"context_end_line":358,"code":" pmt_mel_spec = self.mel_spectrogram(pmt_audio)\n pmt_mel_spec = pmt_mel_spec.squeeze(0) # '1 d t -> d t'\n\n out['pmt_mel_spec'] = pmt_mel_spec\n out['pmt_text'] = pmt_text\n out['pmt_duration'] = pmt_duration\n\n if self.return_wavform:\n out['pmt_wav'] = pmt_audio\n\n if return_path:\n out['pmt_path'] = pmt_audio_path\n\n if return_row:\n out['pmt_row'] = pmt_row\n\n return out\n\n\n# Dynamic Batch Sampler\nclass DynamicBatchSampler(Sampler[list[int]]):\n \"\"\"Extension of Sampler that will do the following:\n 1. Change the batch size (essentially number of sequences)\n in a batch to ensure that the total number of frames are less\n than a certain threshold.\n 2. Make sure the padding efficiency in the batch is high.\n \"\"\"\n\n def __init__(\n self, sampler: Sampler[int], frames_threshold: int, max_samples=0, random_seed=None, drop_last: bool = False\n ):\n self.sampler = sampler\n self.frames_threshold = frames_threshold\n self.max_samples = max_samples\n\n indices, batches = [], []\n data_source = self.sampler.data_source\n\n # for idx in tqdm(\n # self.sampler, desc=\"Sorting with sampler... if slow, check whether dataset is provided with duration\"\n # ):\n for idx in self.sampler:\n indices.append((idx, data_source.get_frame_len(idx)))\n indices.sort(key=lambda elem: elem[1])\n\n batch = []\n batch_frames = 0\n # for idx, frame_len in tqdm(\n # indices, desc=f\"Creating dynamic batches with {frames_threshold} audio frames per gpu\"\n # ):\n for idx, frame_len in indices:\n if batch_frames + frame_len <= self.frames_threshold and (max_samples == 0 or len(batch) < max_samples):\n batch.append(idx)\n batch_frames += frame_len\n else:\n if len(batch) > 0:\n batches.append(batch)\n if frame_len <= self.frames_threshold:\n batch = [idx]\n batch_frames = frame_len\n else:\n batch = []\n batch_frames = 0\n\n if not drop_last and len(batch) > 0:\n batches.append(batch)\n\n del indices\n\n # if want to have different batches between epochs, may just set a seed and log it in ckpt\n # cuz during multi-gpu training, although the batch on per gpu not change between epochs, the formed general minibatch is different\n # e.g. for epoch n, use (random_seed + n)\n random.seed(random_seed)\n random.shuffle(batches)\n\n self.batches = batches\n\n def __iter__(self):\n return iter(self.batches)\n\n def __len__(self):\n return len(self.batches)\n\n\n# Load dataset\n\ndef load_dataset(\n dataset_name: str,\n tokenizer: str = \"pinyin\",\n dataset_type: str = \"CustomDataset\",\n audio_type: str = \"raw\",\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n split: str = \"train\",\n data_augmentation: bool = False,\n return_wavform: bool = False,\n remove_starting_space: bool = True,\n need_prompt_speech: bool = False,\n prompt_repository: dict = None\n) -> CustomDataset:\n \"\"\"\n dataset_type - \"CustomDataset\" if you want to use tokenizer name and default data path to load for train_dataset","source_hash":"2ed7497956c0ce04b71926e3f330898dec956dea8a40d270c1682a03f8899787","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.dataset.load_dataset","uri":"program://DMOSpeech2/function/src.f5_tts.model.dataset.load_dataset#L343-L407","kind":"function","name":"load_dataset","path":"src/f5_tts/model/dataset.py","language":"python","start_line":343,"end_line":407,"context_start_line":323,"context_end_line":427,"code":"\n del indices\n\n # if want to have different batches between epochs, may just set a seed and log it in ckpt\n # cuz during multi-gpu training, although the batch on per gpu not change between epochs, the formed general minibatch is different\n # e.g. for epoch n, use (random_seed + n)\n random.seed(random_seed)\n random.shuffle(batches)\n\n self.batches = batches\n\n def __iter__(self):\n return iter(self.batches)\n\n def __len__(self):\n return len(self.batches)\n\n\n# Load dataset\n\ndef load_dataset(\n dataset_name: str,\n tokenizer: str = \"pinyin\",\n dataset_type: str = \"CustomDataset\",\n audio_type: str = \"raw\",\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n split: str = \"train\",\n data_augmentation: bool = False,\n return_wavform: bool = False,\n remove_starting_space: bool = True,\n need_prompt_speech: bool = False,\n prompt_repository: dict = None\n) -> CustomDataset:\n \"\"\"\n dataset_type - \"CustomDataset\" if you want to use tokenizer name and default data path to load for train_dataset\n - \"CustomDatasetPath\" if you just want to pass the full path to a preprocessed dataset without relying on tokenizer\n \"\"\"\n\n print(\"Loading dataset ...\")\n\n if dataset_type == \"CustomDataset\":\n rel_data_path = str(f'/home/yl4579/F5-TTS-diff/F5-TTS-DMD-flow-ds/data/{dataset_name}_{tokenizer}')\n if 'LibriTTS_100_360_500_char_pinyin' in rel_data_path:\n rel_data_path = rel_data_path.replace('LibriTTS_100_360_500_char_pinyin', 'LibriTTS_100_360_500_char')\n if audio_type == \"raw\":\n try:\n train_dataset = load_from_disk(f\"{rel_data_path}/raw\")\n except: # noqa: E722\n train_dataset = Dataset_.from_file(f\"{rel_data_path}/raw.arrow\")\n preprocessed_mel = False\n elif audio_type == \"mel\":\n train_dataset = Dataset_.from_file(f\"{rel_data_path}/mel.arrow\")\n preprocessed_mel = True\n with open(f\"{rel_data_path}/duration.json\", \"r\", encoding=\"utf-8\") as f:\n data_dict = json.load(f)\n durations = data_dict[\"duration\"]\n train_dataset = CustomDataset(\n train_dataset,\n durations=durations,\n preprocessed_mel=preprocessed_mel,\n mel_spec_module=mel_spec_module,\n **mel_spec_kwargs,\n validation=split == \"val\",\n data_augmentation=data_augmentation,\n return_wavform=return_wavform,\n remove_starting_space=remove_starting_space,\n need_prompt_speech=need_prompt_speech,\n prompt_repository=prompt_repository\n )\n\n elif dataset_type == \"CustomDatasetPath\":\n try:\n train_dataset = load_from_disk(f\"{dataset_name}/raw\")\n except: # noqa: E722\n train_dataset = Dataset_.from_file(f\"{dataset_name}/raw.arrow\")\n\n with open(f\"{dataset_name}/duration.json\", \"r\", encoding=\"utf-8\") as f:\n data_dict = json.load(f)\n durations = data_dict[\"duration\"]\n train_dataset = CustomDataset(\n train_dataset, durations=durations, preprocessed_mel=preprocessed_mel, **mel_spec_kwargs\n )\n\n return train_dataset\n\n\n# collation\ndef collate_fn(batch):\n # Extract mel_specs and their lengths\n mel_specs = [item[\"mel_spec\"].squeeze(0) for item in batch]\n mel_lengths = torch.LongTensor([spec.shape[-1] for spec in mel_specs])\n max_mel_length = mel_lengths.amax()\n \n # Pad mel_specs\n padded_mel_specs = []\n for spec in mel_specs: # TODO. maybe records mask for attention here\n padding = (0, max_mel_length - spec.size(-1))\n padded_spec = F.pad(spec, padding, value=0)\n padded_mel_specs.append(padded_spec)\n mel_specs = torch.stack(padded_mel_specs)\n\n text = [item['text'] for item in batch]\n target_text = [item['target_text'] for item in batch]\n","source_hash":"2ed7497956c0ce04b71926e3f330898dec956dea8a40d270c1682a03f8899787","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.dataset.collate_fn","uri":"program://DMOSpeech2/function/src.f5_tts.model.dataset.collate_fn#L411-L460","kind":"function","name":"collate_fn","path":"src/f5_tts/model/dataset.py","language":"python","start_line":411,"end_line":460,"context_start_line":391,"context_end_line":460,"code":" prompt_repository=prompt_repository\n )\n\n elif dataset_type == \"CustomDatasetPath\":\n try:\n train_dataset = load_from_disk(f\"{dataset_name}/raw\")\n except: # noqa: E722\n train_dataset = Dataset_.from_file(f\"{dataset_name}/raw.arrow\")\n\n with open(f\"{dataset_name}/duration.json\", \"r\", encoding=\"utf-8\") as f:\n data_dict = json.load(f)\n durations = data_dict[\"duration\"]\n train_dataset = CustomDataset(\n train_dataset, durations=durations, preprocessed_mel=preprocessed_mel, **mel_spec_kwargs\n )\n\n return train_dataset\n\n\n# collation\ndef collate_fn(batch):\n # Extract mel_specs and their lengths\n mel_specs = [item[\"mel_spec\"].squeeze(0) for item in batch]\n mel_lengths = torch.LongTensor([spec.shape[-1] for spec in mel_specs])\n max_mel_length = mel_lengths.amax()\n \n # Pad mel_specs\n padded_mel_specs = []\n for spec in mel_specs: # TODO. maybe records mask for attention here\n padding = (0, max_mel_length - spec.size(-1))\n padded_spec = F.pad(spec, padding, value=0)\n padded_mel_specs.append(padded_spec)\n mel_specs = torch.stack(padded_mel_specs)\n\n text = [item['text'] for item in batch]\n target_text = [item['target_text'] for item in batch]\n\n text_lengths = torch.LongTensor([len(item) for item in text])\n\n out = dict(\n mel=mel_specs,\n mel_lengths=mel_lengths,\n text=text,\n text_lengths=text_lengths,\n target_text=target_text,\n )\n\n if 'pmt_mel_spec' in batch[0]:\n pmt_mel_specs = [item[\"pmt_mel_spec\"].squeeze(0) for item in batch]\n pmt_mel_lengths = torch.LongTensor([spec.shape[-1] for spec in pmt_mel_specs])\n max_pmt_mel_length = pmt_mel_lengths.amax()\n \n # Pad mel_specs\n padded_pmt_mel_specs = []\n for spec in pmt_mel_specs: \n padding = (0, max_pmt_mel_length - spec.size(-1))\n padded_spec = F.pad(spec, padding, value=0)\n padded_pmt_mel_specs.append(padded_spec)\n pmt_mel_specs = torch.stack(padded_pmt_mel_specs)\n\n out['pmt_mel_specs'] = pmt_mel_specs\n\n if 'pmt_text' in batch[0]:\n pmt_text = [item['pmt_text'] for item in batch]\n pmt_text_lengths = torch.LongTensor([len(item) for item in pmt_text])\n\n out['pmt_text'] = pmt_text\n out['pmt_text_lengths'] = pmt_text_lengths\n\n return out","source_hash":"2ed7497956c0ce04b71926e3f330898dec956dea8a40d270c1682a03f8899787","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.dataset.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model.dataset.__init__#L285-L332","kind":"function","name":"__init__","path":"src/f5_tts/model/dataset.py","language":"python","start_line":285,"end_line":332,"context_start_line":265,"context_end_line":352,"code":" out['pmt_wav'] = pmt_audio\n\n if return_path:\n out['pmt_path'] = pmt_audio_path\n\n if return_row:\n out['pmt_row'] = pmt_row\n\n return out\n\n\n# Dynamic Batch Sampler\nclass DynamicBatchSampler(Sampler[list[int]]):\n \"\"\"Extension of Sampler that will do the following:\n 1. Change the batch size (essentially number of sequences)\n in a batch to ensure that the total number of frames are less\n than a certain threshold.\n 2. Make sure the padding efficiency in the batch is high.\n \"\"\"\n\n def __init__(\n self, sampler: Sampler[int], frames_threshold: int, max_samples=0, random_seed=None, drop_last: bool = False\n ):\n self.sampler = sampler\n self.frames_threshold = frames_threshold\n self.max_samples = max_samples\n\n indices, batches = [], []\n data_source = self.sampler.data_source\n\n # for idx in tqdm(\n # self.sampler, desc=\"Sorting with sampler... if slow, check whether dataset is provided with duration\"\n # ):\n for idx in self.sampler:\n indices.append((idx, data_source.get_frame_len(idx)))\n indices.sort(key=lambda elem: elem[1])\n\n batch = []\n batch_frames = 0\n # for idx, frame_len in tqdm(\n # indices, desc=f\"Creating dynamic batches with {frames_threshold} audio frames per gpu\"\n # ):\n for idx, frame_len in indices:\n if batch_frames + frame_len <= self.frames_threshold and (max_samples == 0 or len(batch) < max_samples):\n batch.append(idx)\n batch_frames += frame_len\n else:\n if len(batch) > 0:\n batches.append(batch)\n if frame_len <= self.frames_threshold:\n batch = [idx]\n batch_frames = frame_len\n else:\n batch = []\n batch_frames = 0\n\n if not drop_last and len(batch) > 0:\n batches.append(batch)\n\n del indices\n\n # if want to have different batches between epochs, may just set a seed and log it in ckpt\n # cuz during multi-gpu training, although the batch on per gpu not change between epochs, the formed general minibatch is different\n # e.g. for epoch n, use (random_seed + n)\n random.seed(random_seed)\n random.shuffle(batches)\n\n self.batches = batches\n\n def __iter__(self):\n return iter(self.batches)\n\n def __len__(self):\n return len(self.batches)\n\n\n# Load dataset\n\ndef load_dataset(\n dataset_name: str,\n tokenizer: str = \"pinyin\",\n dataset_type: str = \"CustomDataset\",\n audio_type: str = \"raw\",\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n split: str = \"train\",\n data_augmentation: bool = False,\n return_wavform: bool = False,","source_hash":"2ed7497956c0ce04b71926e3f330898dec956dea8a40d270c1682a03f8899787","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.dataset.get_frame_len","uri":"program://DMOSpeech2/function/src.f5_tts.model.dataset.get_frame_len#L133-L141","kind":"function","name":"get_frame_len","path":"src/f5_tts/model/dataset.py","language":"python","start_line":133,"end_line":141,"context_start_line":113,"context_end_line":161,"code":" for row in tqdm(self.data):\n audio_path = row[\"audio_path\"]\n text = row[\"text\"]\n duration = row[\"duration\"]\n spk_id = get_speaker_id(audio_path)\n assert spk_id != None and spk_id != 'mp3'\n if spk_id not in self.prompt_repository:\n self.prompt_repository[spk_id] = [row]\n else:\n self.prompt_repository[spk_id].append(row)\n else:\n self.prompt_repository = prompt_repository\n\n print(f'Grouped samples into {len(self.prompt_repository.keys())} speakers.') \n self.need_prompt_speech = True\n\n else:\n self.need_prompt_speech = False\n\n\n def get_frame_len(self, index):\n if self.validation:\n index += len(self.data) - self.validation_num\n\n if (\n self.durations is not None\n ): # Please make sure the separately provided durations are correct, otherwise 99.99% OOM\n return self.durations[index] * self.target_sample_rate / self.hop_length\n return self.data[index][\"duration\"] * self.target_sample_rate / self.hop_length\n\n def __len__(self):\n if not self.validation:\n return len(self.data) - self.validation_num\n return self.validation_num\n\n def __getitem__(self, index, return_row=True, return_path=False):\n if self.validation:\n index += len(self.data) - self.validation_num\n\n out = {}\n\n while True:\n row = self.data[index]\n audio_path = row[\"audio_path\"]\n text = row[\"text\"]\n duration = row[\"duration\"]\n\n if not isinstance(text, list):\n text = list(text)","source_hash":"2ed7497956c0ce04b71926e3f330898dec956dea8a40d270c1682a03f8899787","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.dataset.__len__","uri":"program://DMOSpeech2/function/src.f5_tts.model.dataset.__len__#L337-L338","kind":"function","name":"__len__","path":"src/f5_tts/model/dataset.py","language":"python","start_line":337,"end_line":338,"context_start_line":317,"context_end_line":358,"code":" else:\n batch = []\n batch_frames = 0\n\n if not drop_last and len(batch) > 0:\n batches.append(batch)\n\n del indices\n\n # if want to have different batches between epochs, may just set a seed and log it in ckpt\n # cuz during multi-gpu training, although the batch on per gpu not change between epochs, the formed general minibatch is different\n # e.g. for epoch n, use (random_seed + n)\n random.seed(random_seed)\n random.shuffle(batches)\n\n self.batches = batches\n\n def __iter__(self):\n return iter(self.batches)\n\n def __len__(self):\n return len(self.batches)\n\n\n# Load dataset\n\ndef load_dataset(\n dataset_name: str,\n tokenizer: str = \"pinyin\",\n dataset_type: str = \"CustomDataset\",\n audio_type: str = \"raw\",\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n split: str = \"train\",\n data_augmentation: bool = False,\n return_wavform: bool = False,\n remove_starting_space: bool = True,\n need_prompt_speech: bool = False,\n prompt_repository: dict = None\n) -> CustomDataset:\n \"\"\"\n dataset_type - \"CustomDataset\" if you want to use tokenizer name and default data path to load for train_dataset","source_hash":"2ed7497956c0ce04b71926e3f330898dec956dea8a40d270c1682a03f8899787","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.dataset.__getitem__","uri":"program://DMOSpeech2/function/src.f5_tts.model.dataset.__getitem__#L148-L273","kind":"function","name":"__getitem__","path":"src/f5_tts/model/dataset.py","language":"python","start_line":148,"end_line":273,"context_start_line":128,"context_end_line":293,"code":"\n else:\n self.need_prompt_speech = False\n\n\n def get_frame_len(self, index):\n if self.validation:\n index += len(self.data) - self.validation_num\n\n if (\n self.durations is not None\n ): # Please make sure the separately provided durations are correct, otherwise 99.99% OOM\n return self.durations[index] * self.target_sample_rate / self.hop_length\n return self.data[index][\"duration\"] * self.target_sample_rate / self.hop_length\n\n def __len__(self):\n if not self.validation:\n return len(self.data) - self.validation_num\n return self.validation_num\n\n def __getitem__(self, index, return_row=True, return_path=False):\n if self.validation:\n index += len(self.data) - self.validation_num\n\n out = {}\n\n while True:\n row = self.data[index]\n audio_path = row[\"audio_path\"]\n text = row[\"text\"]\n duration = row[\"duration\"]\n\n if not isinstance(text, list):\n text = list(text)\n\n # filter by given length\n if (0.3 <= duration <= 30) and (0 < len(text) < 2048):\n break # valid\n\n index = (index + 1) % len(self.data)\n\n if self.remove_starting_space:\n while len(text) > 1 and text[0] == ' ':\n text = text[1:]\n \n if self.preprocessed_mel:\n mel_spec = torch.tensor(row[\"mel_spec\"])\n else:\n audio, source_sample_rate = torchaudio.load(audio_path)\n\n # make sure mono input\n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n\n # resample if necessary\n if source_sample_rate != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(source_sample_rate, self.target_sample_rate)\n audio = resampler(audio)\n\n if not self.validation:\n if self.augment != None:\n audio = self.augment(audio.squeeze().numpy(), sample_rate=self.target_sample_rate)\n audio = torch.from_numpy(audio).float().unsqueeze(0)\n\n # to mel spectrogram\n mel_spec = self.mel_spectrogram(audio)\n mel_spec = mel_spec.squeeze(0) # '1 d t -> d t'\n\n out['mel_spec'] = mel_spec\n out['text'] = text\n out['duration'] = duration\n out['target_text'] = self.data[(index + len(self.data) // 2) % len(self.data)][\"text\"]\n\n if self.return_wavform:\n out['wav'] = audio\n\n if return_path:\n out['path'] = audio_path\n\n if return_row:\n out['row'] = row\n\n # Sample a prompt speech of the same speaker\n # From prompt_repository\n if self.need_prompt_speech:\n spk = get_speaker_id(audio_path)\n spk_repository = self.prompt_repository[spk]\n _count = 100\n while True:\n pmt_row = random.choice(spk_repository)\n pmt_audio_path = pmt_row['audio_path']\n pmt_text = pmt_row['text']\n pmt_duration = pmt_row['duration']\n\n if not isinstance(pmt_text, list):\n pmt_text = list(pmt_text)\n\n # filter by given length\n if 0.3 <= pmt_duration <= 30 and (0 < len(pmt_text) < 2048):\n if pmt_text != text:\n break\n _count = _count - 1\n if _count <= 0:\n break\n\n if self.remove_starting_space:\n while len(pmt_text) > 1 and pmt_text[0] == ' ':\n pmt_text = pmt_text[1:]\n \n if self.preprocessed_mel:\n pmt_mel_spec = torch.tensor(pmt_row[\"mel_spec\"])\n else:\n pmt_audio, source_sample_rate = torchaudio.load(pmt_audio_path)\n\n # make sure mono input\n if pmt_audio.shape[0] > 1:\n pmt_audio = torch.mean(pmt_audio, dim=0, keepdim=True)\n\n # resample if necessary\n if source_sample_rate != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(source_sample_rate, self.target_sample_rate)\n pmt_audio = resampler(pmt_audio)\n\n if not self.validation:\n if self.augment != None:\n pmt_audio = self.augment(pmt_audio.squeeze().numpy(), sample_rate=self.target_sample_rate)\n pmt_audio = torch.from_numpy(pmt_audio).float().unsqueeze(0)\n\n # to mel spectrogram\n pmt_mel_spec = self.mel_spectrogram(pmt_audio)\n pmt_mel_spec = pmt_mel_spec.squeeze(0) # '1 d t -> d t'\n\n out['pmt_mel_spec'] = pmt_mel_spec\n out['pmt_text'] = pmt_text\n out['pmt_duration'] = pmt_duration\n\n if self.return_wavform:\n out['pmt_wav'] = pmt_audio\n\n if return_path:\n out['pmt_path'] = pmt_audio_path\n\n if return_row:\n out['pmt_row'] = pmt_row\n\n return out\n\n\n# Dynamic Batch Sampler\nclass DynamicBatchSampler(Sampler[list[int]]):\n \"\"\"Extension of Sampler that will do the following:\n 1. Change the batch size (essentially number of sequences)\n in a batch to ensure that the total number of frames are less\n than a certain threshold.\n 2. Make sure the padding efficiency in the batch is high.\n \"\"\"\n\n def __init__(\n self, sampler: Sampler[int], frames_threshold: int, max_samples=0, random_seed=None, drop_last: bool = False\n ):\n self.sampler = sampler\n self.frames_threshold = frames_threshold\n self.max_samples = max_samples\n\n indices, batches = [], []\n data_source = self.sampler.data_source","source_hash":"2ed7497956c0ce04b71926e3f330898dec956dea8a40d270c1682a03f8899787","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.dataset.__iter__","uri":"program://DMOSpeech2/function/src.f5_tts.model.dataset.__iter__#L334-L335","kind":"function","name":"__iter__","path":"src/f5_tts/model/dataset.py","language":"python","start_line":334,"end_line":335,"context_start_line":314,"context_end_line":355,"code":" if frame_len <= self.frames_threshold:\n batch = [idx]\n batch_frames = frame_len\n else:\n batch = []\n batch_frames = 0\n\n if not drop_last and len(batch) > 0:\n batches.append(batch)\n\n del indices\n\n # if want to have different batches between epochs, may just set a seed and log it in ckpt\n # cuz during multi-gpu training, although the batch on per gpu not change between epochs, the formed general minibatch is different\n # e.g. for epoch n, use (random_seed + n)\n random.seed(random_seed)\n random.shuffle(batches)\n\n self.batches = batches\n\n def __iter__(self):\n return iter(self.batches)\n\n def __len__(self):\n return len(self.batches)\n\n\n# Load dataset\n\ndef load_dataset(\n dataset_name: str,\n tokenizer: str = \"pinyin\",\n dataset_type: str = \"CustomDataset\",\n audio_type: str = \"raw\",\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n split: str = \"train\",\n data_augmentation: bool = False,\n return_wavform: bool = False,\n remove_starting_space: bool = True,\n need_prompt_speech: bool = False,\n prompt_repository: dict = None","source_hash":"2ed7497956c0ce04b71926e3f330898dec956dea8a40d270c1682a03f8899787","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils","uri":"program://DMOSpeech2/module/src.f5_tts.model.utils#L1-L250","kind":"module","name":"src.f5_tts.model.utils","path":"src/f5_tts/model/utils.py","language":"python","start_line":1,"end_line":250,"context_start_line":1,"context_end_line":250,"code":"from __future__ import annotations\n\nimport os\nimport random\nfrom collections import defaultdict\nfrom importlib.resources import files\n\nimport torch\nfrom torch.nn.utils.rnn import pad_sequence\n\nimport jieba\nfrom pypinyin import lazy_pinyin, Style\n\n\n# seed everything\n\n\ndef seed_everything(seed=0):\n random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.cuda.manual_seed_all(seed)\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n\n\n# helpers\n\n\ndef exists(v):\n return v is not None\n\n\ndef default(v, d):\n return v if exists(v) else d\n\n\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821\n if not exists(length):\n length = t.amax()\n\n seq = torch.arange(length, device=t.device)\n return seq[None, :] < t[:, None]\n\n\ndef mask_from_start_end_indices(seq_len: int[\"b\"], start: int[\"b\"], end: int[\"b\"]): # noqa: F722 F821\n max_seq_len = seq_len.max().item()\n seq = torch.arange(max_seq_len, device=start.device).long()\n start_mask = seq[None, :] >= start[:, None]\n end_mask = seq[None, :] < end[:, None]\n return start_mask & end_mask\n\n\ndef mask_from_frac_lengths(seq_len: int[\"b\"], frac_lengths: float[\"b\"]): # noqa: F722 F821\n lengths = (frac_lengths * seq_len).long()\n max_start = seq_len - lengths\n\n rand = torch.rand_like(frac_lengths)\n start = (max_start * rand).long().clamp(min=0)\n end = start + lengths\n\n return mask_from_start_end_indices(seq_len, start, end)\n\n\ndef maybe_masked_mean(t: float[\"b n d\"], mask: bool[\"b n\"] = None) -> float[\"b d\"]: # noqa: F722\n if not exists(mask):\n return t.mean(dim=1)\n\n t = torch.where(mask[:, :, None], t, torch.tensor(0.0, device=t.device))\n num = t.sum(dim=1)\n den = mask.float().sum(dim=1)\n\n return num / den.clamp(min=1.0)\n\n\n# simple utf-8 tokenizer, since paper went character based\ndef list_str_to_tensor(text: list[str], padding_value=-1) -> int[\"b nt\"]: # noqa: F722\n list_tensors = [torch.tensor([*bytes(t, \"UTF-8\")]) for t in text] # ByT5 style\n text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# char tokenizer, based on custom dataset's extracted .txt file\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n) -> int[\"b nt\"]: # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# Get tokenizer\n\n\ndef get_tokenizer(dataset_name, tokenizer: str = \"pinyin\"):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file\n - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n if tokenizer in [\"pinyin\", \"char\"]:\n tokenizer_path = os.path.join(files(\"f5_tts\").joinpath(\"../../data\"), f\"{dataset_name}_{tokenizer}/vocab.txt\")\n with open(tokenizer_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n assert vocab_char_map[\" \"] == 0, \"make sure space is of idx 0 in vocab.txt, cuz 0 is used for unknown char\"\n\n elif tokenizer == \"byte\":\n vocab_char_map = None\n vocab_size = 256\n\n elif tokenizer == \"custom\":\n with open(dataset_name, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n\n return vocab_char_map, vocab_size\n\n\n\n# convert char to pinyin\n\njieba.initialize()\nprint(\"Word segmentation module jieba initialized.\\n\")\n\n\ndef convert_char_to_pinyin(text_list, polyphone=True):\n final_text_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return (\n \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n )\n\n for text in text_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)\n elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)\n elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_text_list.append(char_list)\n\n return final_text_list\n\n\n# filter func for dirty data with many repetitions\n\n\ndef repetition_found(text, length=2, tolerance=10):\n pattern_count = defaultdict(int)\n for i in range(len(text) - length + 1):\n pattern = text[i : i + length]\n pattern_count[pattern] += 1\n for pattern, count in pattern_count.items():\n if count > tolerance:\n return True\n return False\n\n\ndef load_checkpoint(model, ckpt_path, device, use_ema=True):\n if device == \"cuda\":\n model = model.half()\n\n ckpt_type = ckpt_path.split(\".\")[-1]\n if ckpt_type == \"safetensors\":\n from safetensors.torch import load_file\n\n checkpoint = load_file(ckpt_path)\n else:\n checkpoint = torch.load(ckpt_path, weights_only=True)\n\n if use_ema:\n if ckpt_type == \"safetensors\":\n checkpoint = {\"ema_model_state_dict\": checkpoint}\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n model.load_state_dict(checkpoint[\"model_state_dict\"], strict=False)\n else:\n if ckpt_type == \"safetensors\":\n checkpoint = {\"model_state_dict\": checkpoint}\n model.load_state_dict(checkpoint[\"model_state_dict\"], strict=False)\n\n return model.to(device)\n\n\ndef sample_consecutive_steps(float_list):\n idx = torch.randint(0, len(float_list), size=(1,))\n next_idx = idx - 1\n \n if next_idx < 0:\n next_idx = 0\n else:\n next_idx = idx - 1\n return float(float_list[idx]), float(float_list[next_idx])\n\n\ndef sample_from_list(float_list, N):\n # Convert list to PyTorch tensor\n float_tensor = torch.tensor(float_list)\n list_length = len(float_list)\n\n if N <= list_length:\n # Generate a random permutation of indices for sampling without replacement\n random_indices = torch.randperm(list_length)[:N]\n random_samples = float_tensor[random_indices]\n else:\n # Generate random indices with replacement\n random_indices = torch.randint(list_length, (N,))\n random_samples = float_tensor[random_indices]\n\n return random_samples\n","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.seed_everything","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.seed_everything#L18-L25","kind":"function","name":"seed_everything","path":"src/f5_tts/model/utils.py","language":"python","start_line":18,"end_line":25,"context_start_line":1,"context_end_line":45,"code":"from __future__ import annotations\n\nimport os\nimport random\nfrom collections import defaultdict\nfrom importlib.resources import files\n\nimport torch\nfrom torch.nn.utils.rnn import pad_sequence\n\nimport jieba\nfrom pypinyin import lazy_pinyin, Style\n\n\n# seed everything\n\n\ndef seed_everything(seed=0):\n random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.cuda.manual_seed_all(seed)\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n\n\n# helpers\n\n\ndef exists(v):\n return v is not None\n\n\ndef default(v, d):\n return v if exists(v) else d\n\n\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821\n if not exists(length):\n length = t.amax()\n","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.exists","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.exists#L31-L32","kind":"function","name":"exists","path":"src/f5_tts/model/utils.py","language":"python","start_line":31,"end_line":32,"context_start_line":11,"context_end_line":52,"code":"import jieba\nfrom pypinyin import lazy_pinyin, Style\n\n\n# seed everything\n\n\ndef seed_everything(seed=0):\n random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.cuda.manual_seed_all(seed)\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n\n\n# helpers\n\n\ndef exists(v):\n return v is not None\n\n\ndef default(v, d):\n return v if exists(v) else d\n\n\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821\n if not exists(length):\n length = t.amax()\n\n seq = torch.arange(length, device=t.device)\n return seq[None, :] < t[:, None]\n\n\ndef mask_from_start_end_indices(seq_len: int[\"b\"], start: int[\"b\"], end: int[\"b\"]): # noqa: F722 F821\n max_seq_len = seq_len.max().item()\n seq = torch.arange(max_seq_len, device=start.device).long()","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.default","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.default#L35-L36","kind":"function","name":"default","path":"src/f5_tts/model/utils.py","language":"python","start_line":35,"end_line":36,"context_start_line":15,"context_end_line":56,"code":"# seed everything\n\n\ndef seed_everything(seed=0):\n random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.cuda.manual_seed_all(seed)\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n\n\n# helpers\n\n\ndef exists(v):\n return v is not None\n\n\ndef default(v, d):\n return v if exists(v) else d\n\n\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821\n if not exists(length):\n length = t.amax()\n\n seq = torch.arange(length, device=t.device)\n return seq[None, :] < t[:, None]\n\n\ndef mask_from_start_end_indices(seq_len: int[\"b\"], start: int[\"b\"], end: int[\"b\"]): # noqa: F722 F821\n max_seq_len = seq_len.max().item()\n seq = torch.arange(max_seq_len, device=start.device).long()\n start_mask = seq[None, :] >= start[:, None]\n end_mask = seq[None, :] < end[:, None]\n return start_mask & end_mask\n","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.lens_to_mask","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.lens_to_mask#L42-L47","kind":"function","name":"lens_to_mask","path":"src/f5_tts/model/utils.py","language":"python","start_line":42,"end_line":47,"context_start_line":22,"context_end_line":67,"code":" torch.cuda.manual_seed(seed)\n torch.cuda.manual_seed_all(seed)\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n\n\n# helpers\n\n\ndef exists(v):\n return v is not None\n\n\ndef default(v, d):\n return v if exists(v) else d\n\n\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821\n if not exists(length):\n length = t.amax()\n\n seq = torch.arange(length, device=t.device)\n return seq[None, :] < t[:, None]\n\n\ndef mask_from_start_end_indices(seq_len: int[\"b\"], start: int[\"b\"], end: int[\"b\"]): # noqa: F722 F821\n max_seq_len = seq_len.max().item()\n seq = torch.arange(max_seq_len, device=start.device).long()\n start_mask = seq[None, :] >= start[:, None]\n end_mask = seq[None, :] < end[:, None]\n return start_mask & end_mask\n\n\ndef mask_from_frac_lengths(seq_len: int[\"b\"], frac_lengths: float[\"b\"]): # noqa: F722 F821\n lengths = (frac_lengths * seq_len).long()\n max_start = seq_len - lengths\n\n rand = torch.rand_like(frac_lengths)\n start = (max_start * rand).long().clamp(min=0)\n end = start + lengths\n\n return mask_from_start_end_indices(seq_len, start, end)\n","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.mask_from_start_end_indices","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.mask_from_start_end_indices#L50-L55","kind":"function","name":"mask_from_start_end_indices","path":"src/f5_tts/model/utils.py","language":"python","start_line":50,"end_line":55,"context_start_line":30,"context_end_line":75,"code":"\ndef exists(v):\n return v is not None\n\n\ndef default(v, d):\n return v if exists(v) else d\n\n\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821\n if not exists(length):\n length = t.amax()\n\n seq = torch.arange(length, device=t.device)\n return seq[None, :] < t[:, None]\n\n\ndef mask_from_start_end_indices(seq_len: int[\"b\"], start: int[\"b\"], end: int[\"b\"]): # noqa: F722 F821\n max_seq_len = seq_len.max().item()\n seq = torch.arange(max_seq_len, device=start.device).long()\n start_mask = seq[None, :] >= start[:, None]\n end_mask = seq[None, :] < end[:, None]\n return start_mask & end_mask\n\n\ndef mask_from_frac_lengths(seq_len: int[\"b\"], frac_lengths: float[\"b\"]): # noqa: F722 F821\n lengths = (frac_lengths * seq_len).long()\n max_start = seq_len - lengths\n\n rand = torch.rand_like(frac_lengths)\n start = (max_start * rand).long().clamp(min=0)\n end = start + lengths\n\n return mask_from_start_end_indices(seq_len, start, end)\n\n\ndef maybe_masked_mean(t: float[\"b n d\"], mask: bool[\"b n\"] = None) -> float[\"b d\"]: # noqa: F722\n if not exists(mask):\n return t.mean(dim=1)\n\n t = torch.where(mask[:, :, None], t, torch.tensor(0.0, device=t.device))\n num = t.sum(dim=1)\n den = mask.float().sum(dim=1)","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.mask_from_frac_lengths","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.mask_from_frac_lengths#L58-L66","kind":"function","name":"mask_from_frac_lengths","path":"src/f5_tts/model/utils.py","language":"python","start_line":58,"end_line":66,"context_start_line":38,"context_end_line":86,"code":"\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821\n if not exists(length):\n length = t.amax()\n\n seq = torch.arange(length, device=t.device)\n return seq[None, :] < t[:, None]\n\n\ndef mask_from_start_end_indices(seq_len: int[\"b\"], start: int[\"b\"], end: int[\"b\"]): # noqa: F722 F821\n max_seq_len = seq_len.max().item()\n seq = torch.arange(max_seq_len, device=start.device).long()\n start_mask = seq[None, :] >= start[:, None]\n end_mask = seq[None, :] < end[:, None]\n return start_mask & end_mask\n\n\ndef mask_from_frac_lengths(seq_len: int[\"b\"], frac_lengths: float[\"b\"]): # noqa: F722 F821\n lengths = (frac_lengths * seq_len).long()\n max_start = seq_len - lengths\n\n rand = torch.rand_like(frac_lengths)\n start = (max_start * rand).long().clamp(min=0)\n end = start + lengths\n\n return mask_from_start_end_indices(seq_len, start, end)\n\n\ndef maybe_masked_mean(t: float[\"b n d\"], mask: bool[\"b n\"] = None) -> float[\"b d\"]: # noqa: F722\n if not exists(mask):\n return t.mean(dim=1)\n\n t = torch.where(mask[:, :, None], t, torch.tensor(0.0, device=t.device))\n num = t.sum(dim=1)\n den = mask.float().sum(dim=1)\n\n return num / den.clamp(min=1.0)\n\n\n# simple utf-8 tokenizer, since paper went character based\ndef list_str_to_tensor(text: list[str], padding_value=-1) -> int[\"b nt\"]: # noqa: F722\n list_tensors = [torch.tensor([*bytes(t, \"UTF-8\")]) for t in text] # ByT5 style\n text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.maybe_masked_mean","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.maybe_masked_mean#L69-L77","kind":"function","name":"maybe_masked_mean","path":"src/f5_tts/model/utils.py","language":"python","start_line":69,"end_line":77,"context_start_line":49,"context_end_line":97,"code":"\ndef mask_from_start_end_indices(seq_len: int[\"b\"], start: int[\"b\"], end: int[\"b\"]): # noqa: F722 F821\n max_seq_len = seq_len.max().item()\n seq = torch.arange(max_seq_len, device=start.device).long()\n start_mask = seq[None, :] >= start[:, None]\n end_mask = seq[None, :] < end[:, None]\n return start_mask & end_mask\n\n\ndef mask_from_frac_lengths(seq_len: int[\"b\"], frac_lengths: float[\"b\"]): # noqa: F722 F821\n lengths = (frac_lengths * seq_len).long()\n max_start = seq_len - lengths\n\n rand = torch.rand_like(frac_lengths)\n start = (max_start * rand).long().clamp(min=0)\n end = start + lengths\n\n return mask_from_start_end_indices(seq_len, start, end)\n\n\ndef maybe_masked_mean(t: float[\"b n d\"], mask: bool[\"b n\"] = None) -> float[\"b d\"]: # noqa: F722\n if not exists(mask):\n return t.mean(dim=1)\n\n t = torch.where(mask[:, :, None], t, torch.tensor(0.0, device=t.device))\n num = t.sum(dim=1)\n den = mask.float().sum(dim=1)\n\n return num / den.clamp(min=1.0)\n\n\n# simple utf-8 tokenizer, since paper went character based\ndef list_str_to_tensor(text: list[str], padding_value=-1) -> int[\"b nt\"]: # noqa: F722\n list_tensors = [torch.tensor([*bytes(t, \"UTF-8\")]) for t in text] # ByT5 style\n text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# char tokenizer, based on custom dataset's extracted .txt file\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n) -> int[\"b nt\"]: # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.list_str_to_tensor","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.list_str_to_tensor#L81-L84","kind":"function","name":"list_str_to_tensor","path":"src/f5_tts/model/utils.py","language":"python","start_line":81,"end_line":84,"context_start_line":61,"context_end_line":104,"code":"\n rand = torch.rand_like(frac_lengths)\n start = (max_start * rand).long().clamp(min=0)\n end = start + lengths\n\n return mask_from_start_end_indices(seq_len, start, end)\n\n\ndef maybe_masked_mean(t: float[\"b n d\"], mask: bool[\"b n\"] = None) -> float[\"b d\"]: # noqa: F722\n if not exists(mask):\n return t.mean(dim=1)\n\n t = torch.where(mask[:, :, None], t, torch.tensor(0.0, device=t.device))\n num = t.sum(dim=1)\n den = mask.float().sum(dim=1)\n\n return num / den.clamp(min=1.0)\n\n\n# simple utf-8 tokenizer, since paper went character based\ndef list_str_to_tensor(text: list[str], padding_value=-1) -> int[\"b nt\"]: # noqa: F722\n list_tensors = [torch.tensor([*bytes(t, \"UTF-8\")]) for t in text] # ByT5 style\n text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# char tokenizer, based on custom dataset's extracted .txt file\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n) -> int[\"b nt\"]: # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# Get tokenizer\n\n\ndef get_tokenizer(dataset_name, tokenizer: str = \"pinyin\"):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.list_str_to_idx","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.list_str_to_idx#L88-L95","kind":"function","name":"list_str_to_idx","path":"src/f5_tts/model/utils.py","language":"python","start_line":88,"end_line":95,"context_start_line":68,"context_end_line":115,"code":"\ndef maybe_masked_mean(t: float[\"b n d\"], mask: bool[\"b n\"] = None) -> float[\"b d\"]: # noqa: F722\n if not exists(mask):\n return t.mean(dim=1)\n\n t = torch.where(mask[:, :, None], t, torch.tensor(0.0, device=t.device))\n num = t.sum(dim=1)\n den = mask.float().sum(dim=1)\n\n return num / den.clamp(min=1.0)\n\n\n# simple utf-8 tokenizer, since paper went character based\ndef list_str_to_tensor(text: list[str], padding_value=-1) -> int[\"b nt\"]: # noqa: F722\n list_tensors = [torch.tensor([*bytes(t, \"UTF-8\")]) for t in text] # ByT5 style\n text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# char tokenizer, based on custom dataset's extracted .txt file\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n) -> int[\"b nt\"]: # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# Get tokenizer\n\n\ndef get_tokenizer(dataset_name, tokenizer: str = \"pinyin\"):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file\n - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n if tokenizer in [\"pinyin\", \"char\"]:\n tokenizer_path = os.path.join(files(\"f5_tts\").joinpath(\"../../data\"), f\"{dataset_name}_{tokenizer}/vocab.txt\")\n with open(tokenizer_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.get_tokenizer","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.get_tokenizer#L101-L131","kind":"function","name":"get_tokenizer","path":"src/f5_tts/model/utils.py","language":"python","start_line":101,"end_line":131,"context_start_line":81,"context_end_line":151,"code":"def list_str_to_tensor(text: list[str], padding_value=-1) -> int[\"b nt\"]: # noqa: F722\n list_tensors = [torch.tensor([*bytes(t, \"UTF-8\")]) for t in text] # ByT5 style\n text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# char tokenizer, based on custom dataset's extracted .txt file\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n) -> int[\"b nt\"]: # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# Get tokenizer\n\n\ndef get_tokenizer(dataset_name, tokenizer: str = \"pinyin\"):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file\n - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n if tokenizer in [\"pinyin\", \"char\"]:\n tokenizer_path = os.path.join(files(\"f5_tts\").joinpath(\"../../data\"), f\"{dataset_name}_{tokenizer}/vocab.txt\")\n with open(tokenizer_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n assert vocab_char_map[\" \"] == 0, \"make sure space is of idx 0 in vocab.txt, cuz 0 is used for unknown char\"\n\n elif tokenizer == \"byte\":\n vocab_char_map = None\n vocab_size = 256\n\n elif tokenizer == \"custom\":\n with open(dataset_name, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n\n return vocab_char_map, vocab_size\n\n\n\n# convert char to pinyin\n\njieba.initialize()\nprint(\"Word segmentation module jieba initialized.\\n\")\n\n\ndef convert_char_to_pinyin(text_list, polyphone=True):\n final_text_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return (\n \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n )\n","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.convert_char_to_pinyin","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.convert_char_to_pinyin#L141-L178","kind":"function","name":"convert_char_to_pinyin","path":"src/f5_tts/model/utils.py","language":"python","start_line":141,"end_line":178,"context_start_line":121,"context_end_line":198,"code":" vocab_char_map = None\n vocab_size = 256\n\n elif tokenizer == \"custom\":\n with open(dataset_name, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n\n return vocab_char_map, vocab_size\n\n\n\n# convert char to pinyin\n\njieba.initialize()\nprint(\"Word segmentation module jieba initialized.\\n\")\n\n\ndef convert_char_to_pinyin(text_list, polyphone=True):\n final_text_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return (\n \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n )\n\n for text in text_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)\n elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)\n elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_text_list.append(char_list)\n\n return final_text_list\n\n\n# filter func for dirty data with many repetitions\n\n\ndef repetition_found(text, length=2, tolerance=10):\n pattern_count = defaultdict(int)\n for i in range(len(text) - length + 1):\n pattern = text[i : i + length]\n pattern_count[pattern] += 1\n for pattern, count in pattern_count.items():\n if count > tolerance:\n return True\n return False\n\n\ndef load_checkpoint(model, ckpt_path, device, use_ema=True):\n if device == \"cuda\":\n model = model.half()\n","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.repetition_found","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.repetition_found#L184-L192","kind":"function","name":"repetition_found","path":"src/f5_tts/model/utils.py","language":"python","start_line":184,"end_line":192,"context_start_line":164,"context_end_line":212,"code":" if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)\n elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_text_list.append(char_list)\n\n return final_text_list\n\n\n# filter func for dirty data with many repetitions\n\n\ndef repetition_found(text, length=2, tolerance=10):\n pattern_count = defaultdict(int)\n for i in range(len(text) - length + 1):\n pattern = text[i : i + length]\n pattern_count[pattern] += 1\n for pattern, count in pattern_count.items():\n if count > tolerance:\n return True\n return False\n\n\ndef load_checkpoint(model, ckpt_path, device, use_ema=True):\n if device == \"cuda\":\n model = model.half()\n\n ckpt_type = ckpt_path.split(\".\")[-1]\n if ckpt_type == \"safetensors\":\n from safetensors.torch import load_file\n\n checkpoint = load_file(ckpt_path)\n else:\n checkpoint = torch.load(ckpt_path, weights_only=True)\n\n if use_ema:\n if ckpt_type == \"safetensors\":\n checkpoint = {\"ema_model_state_dict\": checkpoint}\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.load_checkpoint","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.load_checkpoint#L195-L221","kind":"function","name":"load_checkpoint","path":"src/f5_tts/model/utils.py","language":"python","start_line":195,"end_line":221,"context_start_line":175,"context_end_line":241,"code":" char_list.append(c)\n final_text_list.append(char_list)\n\n return final_text_list\n\n\n# filter func for dirty data with many repetitions\n\n\ndef repetition_found(text, length=2, tolerance=10):\n pattern_count = defaultdict(int)\n for i in range(len(text) - length + 1):\n pattern = text[i : i + length]\n pattern_count[pattern] += 1\n for pattern, count in pattern_count.items():\n if count > tolerance:\n return True\n return False\n\n\ndef load_checkpoint(model, ckpt_path, device, use_ema=True):\n if device == \"cuda\":\n model = model.half()\n\n ckpt_type = ckpt_path.split(\".\")[-1]\n if ckpt_type == \"safetensors\":\n from safetensors.torch import load_file\n\n checkpoint = load_file(ckpt_path)\n else:\n checkpoint = torch.load(ckpt_path, weights_only=True)\n\n if use_ema:\n if ckpt_type == \"safetensors\":\n checkpoint = {\"ema_model_state_dict\": checkpoint}\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n model.load_state_dict(checkpoint[\"model_state_dict\"], strict=False)\n else:\n if ckpt_type == \"safetensors\":\n checkpoint = {\"model_state_dict\": checkpoint}\n model.load_state_dict(checkpoint[\"model_state_dict\"], strict=False)\n\n return model.to(device)\n\n\ndef sample_consecutive_steps(float_list):\n idx = torch.randint(0, len(float_list), size=(1,))\n next_idx = idx - 1\n \n if next_idx < 0:\n next_idx = 0\n else:\n next_idx = idx - 1\n return float(float_list[idx]), float(float_list[next_idx])\n\n\ndef sample_from_list(float_list, N):\n # Convert list to PyTorch tensor\n float_tensor = torch.tensor(float_list)\n list_length = len(float_list)\n\n if N <= list_length:\n # Generate a random permutation of indices for sampling without replacement","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.sample_consecutive_steps","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.sample_consecutive_steps#L224-L232","kind":"function","name":"sample_consecutive_steps","path":"src/f5_tts/model/utils.py","language":"python","start_line":224,"end_line":232,"context_start_line":204,"context_end_line":250,"code":" else:\n checkpoint = torch.load(ckpt_path, weights_only=True)\n\n if use_ema:\n if ckpt_type == \"safetensors\":\n checkpoint = {\"ema_model_state_dict\": checkpoint}\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n model.load_state_dict(checkpoint[\"model_state_dict\"], strict=False)\n else:\n if ckpt_type == \"safetensors\":\n checkpoint = {\"model_state_dict\": checkpoint}\n model.load_state_dict(checkpoint[\"model_state_dict\"], strict=False)\n\n return model.to(device)\n\n\ndef sample_consecutive_steps(float_list):\n idx = torch.randint(0, len(float_list), size=(1,))\n next_idx = idx - 1\n \n if next_idx < 0:\n next_idx = 0\n else:\n next_idx = idx - 1\n return float(float_list[idx]), float(float_list[next_idx])\n\n\ndef sample_from_list(float_list, N):\n # Convert list to PyTorch tensor\n float_tensor = torch.tensor(float_list)\n list_length = len(float_list)\n\n if N <= list_length:\n # Generate a random permutation of indices for sampling without replacement\n random_indices = torch.randperm(list_length)[:N]\n random_samples = float_tensor[random_indices]\n else:\n # Generate random indices with replacement\n random_indices = torch.randint(list_length, (N,))\n random_samples = float_tensor[random_indices]\n\n return random_samples\n","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.sample_from_list","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.sample_from_list#L235-L249","kind":"function","name":"sample_from_list","path":"src/f5_tts/model/utils.py","language":"python","start_line":235,"end_line":249,"context_start_line":215,"context_end_line":250,"code":" model.load_state_dict(checkpoint[\"model_state_dict\"], strict=False)\n else:\n if ckpt_type == \"safetensors\":\n checkpoint = {\"model_state_dict\": checkpoint}\n model.load_state_dict(checkpoint[\"model_state_dict\"], strict=False)\n\n return model.to(device)\n\n\ndef sample_consecutive_steps(float_list):\n idx = torch.randint(0, len(float_list), size=(1,))\n next_idx = idx - 1\n \n if next_idx < 0:\n next_idx = 0\n else:\n next_idx = idx - 1\n return float(float_list[idx]), float(float_list[next_idx])\n\n\ndef sample_from_list(float_list, N):\n # Convert list to PyTorch tensor\n float_tensor = torch.tensor(float_list)\n list_length = len(float_list)\n\n if N <= list_length:\n # Generate a random permutation of indices for sampling without replacement\n random_indices = torch.randperm(list_length)[:N]\n random_samples = float_tensor[random_indices]\n else:\n # Generate random indices with replacement\n random_indices = torch.randint(list_length, (N,))\n random_samples = float_tensor[random_indices]\n\n return random_samples\n","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.utils.is_chinese","uri":"program://DMOSpeech2/function/src.f5_tts.model.utils.is_chinese#L147-L150","kind":"function","name":"is_chinese","path":"src/f5_tts/model/utils.py","language":"python","start_line":147,"end_line":150,"context_start_line":127,"context_end_line":170,"code":" for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n\n return vocab_char_map, vocab_size\n\n\n\n# convert char to pinyin\n\njieba.initialize()\nprint(\"Word segmentation module jieba initialized.\\n\")\n\n\ndef convert_char_to_pinyin(text_list, polyphone=True):\n final_text_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return (\n \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n )\n\n for text in text_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)\n elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules","uri":"program://DMOSpeech2/module/src.f5_tts.model.modules#L1-L669","kind":"module","name":"src.f5_tts.model.modules","path":"src/f5_tts/model/modules.py","language":"python","start_line":1,"end_line":669,"context_start_line":1,"context_end_line":669,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nimport math\nfrom typing import Optional\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom librosa.filters import mel as librosa_mel_fn\nfrom torch import nn\nfrom x_transformers.x_transformers import apply_rotary_pos_emb\n\n\n# raw wav to mel spec\n\n\nmel_basis_cache = {}\nhann_window_cache = {}\n\n\ndef get_bigvgan_mel_spectrogram(\n waveform,\n n_fft=1024,\n n_mel_channels=100,\n target_sample_rate=24000,\n hop_length=256,\n win_length=1024,\n fmin=0,\n fmax=None,\n center=False,\n): # Copy from https://github.com/NVIDIA/BigVGAN/tree/main\n device = waveform.device\n key = f\"{n_fft}_{n_mel_channels}_{target_sample_rate}_{hop_length}_{win_length}_{fmin}_{fmax}_{device}\"\n\n if key not in mel_basis_cache:\n mel = librosa_mel_fn(sr=target_sample_rate, n_fft=n_fft, n_mels=n_mel_channels, fmin=fmin, fmax=fmax)\n mel_basis_cache[key] = torch.from_numpy(mel).float().to(device) # TODO: why they need .float()?\n hann_window_cache[key] = torch.hann_window(win_length).to(device)\n\n mel_basis = mel_basis_cache[key]\n hann_window = hann_window_cache[key]\n\n padding = (n_fft - hop_length) // 2\n waveform = torch.nn.functional.pad(waveform.unsqueeze(1), (padding, padding), mode=\"reflect\").squeeze(1)\n\n spec = torch.stft(\n waveform,\n n_fft,\n hop_length=hop_length,\n win_length=win_length,\n window=hann_window,\n center=center,\n pad_mode=\"reflect\",\n normalized=False,\n onesided=True,\n return_complex=True,\n )\n spec = torch.sqrt(torch.view_as_real(spec).pow(2).sum(-1) + 1e-9)\n\n mel_spec = torch.matmul(mel_basis, spec)\n mel_spec = torch.log(torch.clamp(mel_spec, min=1e-5))\n\n return mel_spec\n\n\ndef get_vocos_mel_spectrogram(\n waveform,\n n_fft=1024,\n n_mel_channels=100,\n target_sample_rate=24000,\n hop_length=256,\n win_length=1024,\n):\n mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=target_sample_rate,\n n_fft=n_fft,\n win_length=win_length,\n hop_length=hop_length,\n n_mels=n_mel_channels,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(waveform.device)\n if len(waveform.shape) == 3:\n waveform = waveform.squeeze(1) # 'b 1 nw -> b nw'\n\n assert len(waveform.shape) == 2\n\n mel = mel_stft(waveform)\n mel = mel.clamp(min=1e-5).log()\n return mel\n\n\nclass MelSpec(nn.Module):\n def __init__(\n self,\n n_fft=1024,\n hop_length=256,\n win_length=1024,\n n_mel_channels=100,\n target_sample_rate=24_000,\n mel_spec_type=\"vocos\",\n ):\n super().__init__()\n assert mel_spec_type in [\"vocos\", \"bigvgan\"], print(\"We only support two extract mel backend: vocos or bigvgan\")\n\n self.n_fft = n_fft\n self.hop_length = hop_length\n self.win_length = win_length\n self.n_mel_channels = n_mel_channels\n self.target_sample_rate = target_sample_rate\n\n if mel_spec_type == \"vocos\":\n self.extractor = get_vocos_mel_spectrogram\n elif mel_spec_type == \"bigvgan\":\n self.extractor = get_bigvgan_mel_spectrogram\n\n self.register_buffer(\"dummy\", torch.tensor(0), persistent=False)\n\n def forward(self, wav):\n if self.dummy.device != wav.device:\n self.to(wav.device)\n\n mel = self.extractor(\n waveform=wav,\n n_fft=self.n_fft,\n n_mel_channels=self.n_mel_channels,\n target_sample_rate=self.target_sample_rate,\n hop_length=self.hop_length,\n win_length=self.win_length,\n )\n\n return mel\n\n\n# sinusoidal position embedding\n\n\nclass SinusPositionEmbedding(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.dim = dim\n\n def forward(self, x, scale=1000):\n device = x.device\n half_dim = self.dim // 2\n emb = math.log(10000) / (half_dim - 1)\n emb = torch.exp(torch.arange(half_dim, device=device).float() * -emb)\n emb = scale * x.unsqueeze(1) * emb.unsqueeze(0)\n emb = torch.cat((emb.sin(), emb.cos()), dim=-1)\n return emb\n\n\n# convolutional position embedding\n\n\nclass ConvPositionEmbedding(nn.Module):\n def __init__(self, dim, kernel_size=31, groups=16):\n super().__init__()\n assert kernel_size % 2 != 0\n self.conv1d = nn.Sequential(\n nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),\n nn.Mish(),\n nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),\n nn.Mish(),\n )\n\n def forward(self, x: float[\"b n d\"], mask: bool[\"b n\"] | None = None): # noqa: F722\n if mask is not None:\n mask = mask[..., None]\n x = x.masked_fill(~mask, 0.0)\n\n x = x.permute(0, 2, 1)\n x = self.conv1d(x)\n out = x.permute(0, 2, 1)\n\n if mask is not None:\n out = out.masked_fill(~mask, 0.0)\n\n return out\n\n\n# rotary positional embedding related\n\n\ndef precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0):\n # proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning\n # has some connection to NTK literature\n # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/\n # https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py\n theta *= theta_rescale_factor ** (dim / (dim - 2))\n freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))\n t = torch.arange(end, device=freqs.device) # type: ignore\n freqs = torch.outer(t, freqs).float() # type: ignore\n freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef get_pos_embed_indices(start, length, max_pos, scale=1.0):\n # length = length if isinstance(length, int) else length.max()\n scale = scale * torch.ones_like(start, dtype=torch.float32) # in case scale is a scalar\n pos = (\n start.unsqueeze(1)\n + (torch.arange(length, device=start.device, dtype=torch.float32).unsqueeze(0) * scale.unsqueeze(1)).long()\n )\n # avoid extra long error.\n pos = torch.where(pos < max_pos, pos, max_pos - 1)\n return pos\n\n\n# Global Response Normalization layer (Instance Normalization ?)\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.gamma = nn.Parameter(torch.zeros(1, 1, dim))\n self.beta = nn.Parameter(torch.zeros(1, 1, dim))\n\n def forward(self, x):\n Gx = torch.norm(x, p=2, dim=1, keepdim=True)\n Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)\n return self.gamma * (x * Nx) + self.beta + x\n\n\n# ConvNeXt-V2 Block https://github.com/facebookresearch/ConvNeXt-V2/blob/main/models/convnextv2.py\n# ref: https://github.com/bfs18/e2_tts/blob/main/rfwave/modules.py#L108\n\n\nclass ConvNeXtV2Block(nn.Module):\n def __init__(\n self,\n dim: int,\n intermediate_dim: int,\n dilation: int = 1,\n ):\n super().__init__()\n padding = (dilation * (7 - 1)) // 2\n self.dwconv = nn.Conv1d(\n dim, dim, kernel_size=7, padding=padding, groups=dim, dilation=dilation\n ) # depthwise conv\n self.norm = nn.LayerNorm(dim, eps=1e-6)\n self.pwconv1 = nn.Linear(dim, intermediate_dim) # pointwise/1x1 convs, implemented with linear layers\n self.act = nn.GELU()\n self.grn = GRN(intermediate_dim)\n self.pwconv2 = nn.Linear(intermediate_dim, dim)\n\n def forward(self, x: torch.Tensor) -> torch.Tensor:\n residual = x\n x = x.transpose(1, 2) # b n d -> b d n\n x = self.dwconv(x)\n x = x.transpose(1, 2) # b d n -> b n d\n x = self.norm(x)\n x = self.pwconv1(x)\n x = self.act(x)\n x = self.grn(x)\n x = self.pwconv2(x)\n return residual + x\n\n\n# AdaLayerNormZero\n# return with modulated x for attn input, and params for later mlp modulation\n\n\nclass AdaLayerNormZero(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 6)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb=None):\n emb = self.linear(self.silu(emb))\n shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = torch.chunk(emb, 6, dim=1)\n\n x = self.norm(x) * (1 + scale_msa[:, None]) + shift_msa[:, None]\n return x, gate_msa, shift_mlp, scale_mlp, gate_mlp\n\n\n# AdaLayerNormZero for final layer\n# return only with modulated x for attn input, cuz no more mlp modulation\n\n\nclass AdaLayerNormZero_Final(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 2)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb):\n emb = self.linear(self.silu(emb))\n scale, shift = torch.chunk(emb, 2, dim=1)\n\n x = self.norm(x) * (1 + scale)[:, None, :] + shift[:, None, :]\n return x\n\n\n# FeedForward\n\n\nclass FeedForward(nn.Module):\n def __init__(self, dim, dim_out=None, mult=4, dropout=0.0, approximate: str = \"none\"):\n super().__init__()\n inner_dim = int(dim * mult)\n dim_out = dim_out if dim_out is not None else dim\n\n activation = nn.GELU(approximate=approximate)\n project_in = nn.Sequential(nn.Linear(dim, inner_dim), activation)\n self.ff = nn.Sequential(project_in, nn.Dropout(dropout), nn.Linear(inner_dim, dim_out))\n\n def forward(self, x):\n return self.ff(x)\n\n\n# Attention with possible joint part\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass Attention(nn.Module):\n def __init__(\n self,\n processor: JointAttnProcessor | AttnProcessor,\n dim: int,\n heads: int = 8,\n dim_head: int = 64,\n dropout: float = 0.0,\n context_dim: Optional[int] = None, # if not None -> joint attention\n context_pre_only=None,\n ):\n super().__init__()\n\n if not hasattr(F, \"scaled_dot_product_attention\"):\n raise ImportError(\"Attention equires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.\")\n\n self.processor = processor\n\n self.dim = dim\n self.heads = heads\n self.inner_dim = dim_head * heads\n self.dropout = dropout\n\n self.context_dim = context_dim\n self.context_pre_only = context_pre_only\n\n self.to_q = nn.Linear(dim, self.inner_dim)\n self.to_k = nn.Linear(dim, self.inner_dim)\n self.to_v = nn.Linear(dim, self.inner_dim)\n\n if self.context_dim is not None:\n self.to_k_c = nn.Linear(context_dim, self.inner_dim)\n self.to_v_c = nn.Linear(context_dim, self.inner_dim)\n if self.context_pre_only is not None:\n self.to_q_c = nn.Linear(context_dim, self.inner_dim)\n\n self.to_out = nn.ModuleList([])\n self.to_out.append(nn.Linear(self.inner_dim, dim))\n self.to_out.append(nn.Dropout(dropout))\n\n if self.context_pre_only is not None and not self.context_pre_only:\n self.to_out_c = nn.Linear(self.inner_dim, dim)\n\n def forward(\n self,\n x: float[\"b n d\"], # noised input x # noqa: F722\n c: float[\"b n d\"] = None, # context c # noqa: F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n src_mask: bool[\"b nt\"] | None = None, # noqa: F722\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.Tensor:\n if c is not None:\n return self.processor(self, x, c=c, mask=mask, src_mask=src_mask, rope=rope, c_rope=c_rope)\n else:\n return self.processor(self, x, mask=mask, rope=rope)\n\n\n# Attention processor\n\n\nclass AttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn: Attention,\n x: float[\"b n d\"], # noised input x # noqa: F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n rope=None, # rotary position embedding\n ) -> torch.FloatTensor:\n batch_size = x.shape[0]\n\n # `sample` projections.\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n\n # apply rotary position embedding\n if rope is not None:\n freqs, xpos_scale = rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n\n query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)\n key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)\n\n # attention\n inner_dim = key.shape[-1]\n head_dim = inner_dim // attn.heads\n query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n\n # mask. e.g. inference got a batch with different target durations, mask out the padding\n if mask is not None:\n attn_mask = mask\n attn_mask = attn_mask.unsqueeze(1).unsqueeze(1) # 'b n -> b 1 1 n'\n attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])\n else:\n attn_mask = None\n\n x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)\n x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)\n x = x.to(query.dtype)\n\n # linear proj\n x = attn.to_out[0](x)\n # dropout\n x = attn.to_out[1](x)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n\n return x\n\n\n# Joint Attention processor for MM-DiT\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass JointAttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn: Attention,\n x: float[\"b n d\"], # noised input x\n c: float[\"b nt d\"] = None, # context c, here text\n mask: bool[\"b n\"] | None = None,\n src_mask: bool[\"b nt\"] | None = None,\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.FloatTensor:\n residual = x\n batch_size = c.shape[0]\n\n # `sample` projections\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n\n # `context` projections\n c_query = attn.to_q_c(c)\n c_key = attn.to_k_c(c)\n c_value = attn.to_v_c(c)\n\n # apply rope for x\n if rope is not None:\n freqs, xpos_scale = rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)\n key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)\n\n # apply rope for c\n if c_rope is not None:\n freqs, xpos_scale = c_rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n c_query = apply_rotary_pos_emb(c_query, freqs, q_xpos_scale)\n c_key = apply_rotary_pos_emb(c_key, freqs, k_xpos_scale)\n\n # attention\n query = torch.cat([query, c_query], dim=1)\n key = torch.cat([key, c_key], dim=1)\n value = torch.cat([value, c_value], dim=1)\n\n inner_dim = key.shape[-1]\n head_dim = inner_dim // attn.heads\n query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n\n # mask: combine 'mask' (for x) + 'src_mask' (for c) minimally\n if mask is not None:\n attn_mask = F.pad(mask, (0, c.shape[1]), value=True) # no mask for c (text)\n if src_mask is not None:\n # pad src_mask in front so it aligns with c positions\n # and combine with x-mask by logical AND\n attn_mask_c = F.pad(src_mask, (x.shape[1], 0), value=True)\n attn_mask = attn_mask & attn_mask_c\n attn_mask = attn_mask.unsqueeze(1).unsqueeze(1)\n attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])\n else:\n if src_mask is not None:\n # if there's no mask for x but there's src_mask\n attn_mask = F.pad(src_mask, (x.shape[1], 0), value=True)\n attn_mask = attn_mask.unsqueeze(1).unsqueeze(1)\n attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])\n else:\n attn_mask = None\n\n x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)\n x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)\n x = x.to(query.dtype)\n\n # Split the attention outputs\n x, c = x[:, : residual.shape[1]], x[:, residual.shape[1] :]\n\n # linear proj\n x = attn.to_out[0](x)\n x = attn.to_out[1](x)\n if not attn.context_pre_only:\n c = attn.to_out_c(c)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n # c is left as-is (no mask for c by default)\n\n return x, c\n\n\n\n# DiT Block\n\n\nclass DiTBlock(nn.Module):\n def __init__(self, dim, heads, dim_head, ff_mult=4, dropout=0.1):\n super().__init__()\n\n self.attn_norm = AdaLayerNormZero(dim)\n self.attn = Attention(\n processor=AttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n )\n\n self.ff_norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n self.ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n\n def forward(self, x, t, mask=None, rope=None): # x: noised input, t: time embedding\n # pre-norm & modulation for attention input\n norm, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.attn_norm(x, emb=t)\n\n # attention\n attn_output = self.attn(x=norm, mask=mask, rope=rope)\n\n # process attention output for input x\n x = x + gate_msa.unsqueeze(1) * attn_output\n\n norm = self.ff_norm(x) * (1 + scale_mlp[:, None]) + shift_mlp[:, None]\n ff_output = self.ff(norm)\n x = x + gate_mlp.unsqueeze(1) * ff_output\n\n return x\n\n\n# MMDiT Block https://arxiv.org/abs/2403.03206\n\n\nclass MMDiTBlock(nn.Module):\n r\"\"\"\n modified from diffusers/src/diffusers/models/attention.py\n\n notes.\n _c: context related. text, cond, etc. (left part in sd3 fig2.b)\n _x: noised input related. (right part)\n context_pre_only: last layer only do prenorm + modulation cuz no more ffn\n \"\"\"\n\n def __init__(self, dim, heads, dim_head, ff_mult=4, dropout=0.1, context_pre_only=False):\n super().__init__()\n\n self.context_pre_only = context_pre_only\n\n self.attn_norm_c = AdaLayerNormZero_Final(dim) if context_pre_only else AdaLayerNormZero(dim)\n self.attn_norm_x = AdaLayerNormZero(dim)\n self.attn = Attention(\n processor=JointAttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n cont\n# ... truncated ...","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.get_bigvgan_mel_spectrogram","uri":"program://DMOSpeech2/function/src.f5_tts.model.modules.get_bigvgan_mel_spectrogram#L30-L72","kind":"function","name":"get_bigvgan_mel_spectrogram","path":"src/f5_tts/model/modules.py","language":"python","start_line":30,"end_line":72,"context_start_line":10,"context_end_line":92,"code":"from __future__ import annotations\n\nimport math\nfrom typing import Optional\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom librosa.filters import mel as librosa_mel_fn\nfrom torch import nn\nfrom x_transformers.x_transformers import apply_rotary_pos_emb\n\n\n# raw wav to mel spec\n\n\nmel_basis_cache = {}\nhann_window_cache = {}\n\n\ndef get_bigvgan_mel_spectrogram(\n waveform,\n n_fft=1024,\n n_mel_channels=100,\n target_sample_rate=24000,\n hop_length=256,\n win_length=1024,\n fmin=0,\n fmax=None,\n center=False,\n): # Copy from https://github.com/NVIDIA/BigVGAN/tree/main\n device = waveform.device\n key = f\"{n_fft}_{n_mel_channels}_{target_sample_rate}_{hop_length}_{win_length}_{fmin}_{fmax}_{device}\"\n\n if key not in mel_basis_cache:\n mel = librosa_mel_fn(sr=target_sample_rate, n_fft=n_fft, n_mels=n_mel_channels, fmin=fmin, fmax=fmax)\n mel_basis_cache[key] = torch.from_numpy(mel).float().to(device) # TODO: why they need .float()?\n hann_window_cache[key] = torch.hann_window(win_length).to(device)\n\n mel_basis = mel_basis_cache[key]\n hann_window = hann_window_cache[key]\n\n padding = (n_fft - hop_length) // 2\n waveform = torch.nn.functional.pad(waveform.unsqueeze(1), (padding, padding), mode=\"reflect\").squeeze(1)\n\n spec = torch.stft(\n waveform,\n n_fft,\n hop_length=hop_length,\n win_length=win_length,\n window=hann_window,\n center=center,\n pad_mode=\"reflect\",\n normalized=False,\n onesided=True,\n return_complex=True,\n )\n spec = torch.sqrt(torch.view_as_real(spec).pow(2).sum(-1) + 1e-9)\n\n mel_spec = torch.matmul(mel_basis, spec)\n mel_spec = torch.log(torch.clamp(mel_spec, min=1e-5))\n\n return mel_spec\n\n\ndef get_vocos_mel_spectrogram(\n waveform,\n n_fft=1024,\n n_mel_channels=100,\n target_sample_rate=24000,\n hop_length=256,\n win_length=1024,\n):\n mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=target_sample_rate,\n n_fft=n_fft,\n win_length=win_length,\n hop_length=hop_length,\n n_mels=n_mel_channels,\n power=1,\n center=True,\n normalized=False,\n norm=None,","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.get_vocos_mel_spectrogram","uri":"program://DMOSpeech2/function/src.f5_tts.model.modules.get_vocos_mel_spectrogram#L75-L101","kind":"function","name":"get_vocos_mel_spectrogram","path":"src/f5_tts/model/modules.py","language":"python","start_line":75,"end_line":101,"context_start_line":55,"context_end_line":121,"code":" spec = torch.stft(\n waveform,\n n_fft,\n hop_length=hop_length,\n win_length=win_length,\n window=hann_window,\n center=center,\n pad_mode=\"reflect\",\n normalized=False,\n onesided=True,\n return_complex=True,\n )\n spec = torch.sqrt(torch.view_as_real(spec).pow(2).sum(-1) + 1e-9)\n\n mel_spec = torch.matmul(mel_basis, spec)\n mel_spec = torch.log(torch.clamp(mel_spec, min=1e-5))\n\n return mel_spec\n\n\ndef get_vocos_mel_spectrogram(\n waveform,\n n_fft=1024,\n n_mel_channels=100,\n target_sample_rate=24000,\n hop_length=256,\n win_length=1024,\n):\n mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=target_sample_rate,\n n_fft=n_fft,\n win_length=win_length,\n hop_length=hop_length,\n n_mels=n_mel_channels,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(waveform.device)\n if len(waveform.shape) == 3:\n waveform = waveform.squeeze(1) # 'b 1 nw -> b nw'\n\n assert len(waveform.shape) == 2\n\n mel = mel_stft(waveform)\n mel = mel.clamp(min=1e-5).log()\n return mel\n\n\nclass MelSpec(nn.Module):\n def __init__(\n self,\n n_fft=1024,\n hop_length=256,\n win_length=1024,\n n_mel_channels=100,\n target_sample_rate=24_000,\n mel_spec_type=\"vocos\",\n ):\n super().__init__()\n assert mel_spec_type in [\"vocos\", \"bigvgan\"], print(\"We only support two extract mel backend: vocos or bigvgan\")\n\n self.n_fft = n_fft\n self.hop_length = hop_length\n self.win_length = win_length\n self.n_mel_channels = n_mel_channels\n self.target_sample_rate = target_sample_rate","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.MelSpec","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.MelSpec#L104-L143","kind":"class","name":"MelSpec","path":"src/f5_tts/model/modules.py","language":"python","start_line":104,"end_line":143,"context_start_line":84,"context_end_line":163,"code":" sample_rate=target_sample_rate,\n n_fft=n_fft,\n win_length=win_length,\n hop_length=hop_length,\n n_mels=n_mel_channels,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(waveform.device)\n if len(waveform.shape) == 3:\n waveform = waveform.squeeze(1) # 'b 1 nw -> b nw'\n\n assert len(waveform.shape) == 2\n\n mel = mel_stft(waveform)\n mel = mel.clamp(min=1e-5).log()\n return mel\n\n\nclass MelSpec(nn.Module):\n def __init__(\n self,\n n_fft=1024,\n hop_length=256,\n win_length=1024,\n n_mel_channels=100,\n target_sample_rate=24_000,\n mel_spec_type=\"vocos\",\n ):\n super().__init__()\n assert mel_spec_type in [\"vocos\", \"bigvgan\"], print(\"We only support two extract mel backend: vocos or bigvgan\")\n\n self.n_fft = n_fft\n self.hop_length = hop_length\n self.win_length = win_length\n self.n_mel_channels = n_mel_channels\n self.target_sample_rate = target_sample_rate\n\n if mel_spec_type == \"vocos\":\n self.extractor = get_vocos_mel_spectrogram\n elif mel_spec_type == \"bigvgan\":\n self.extractor = get_bigvgan_mel_spectrogram\n\n self.register_buffer(\"dummy\", torch.tensor(0), persistent=False)\n\n def forward(self, wav):\n if self.dummy.device != wav.device:\n self.to(wav.device)\n\n mel = self.extractor(\n waveform=wav,\n n_fft=self.n_fft,\n n_mel_channels=self.n_mel_channels,\n target_sample_rate=self.target_sample_rate,\n hop_length=self.hop_length,\n win_length=self.win_length,\n )\n\n return mel\n\n\n# sinusoidal position embedding\n\n\nclass SinusPositionEmbedding(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.dim = dim\n\n def forward(self, x, scale=1000):\n device = x.device\n half_dim = self.dim // 2\n emb = math.log(10000) / (half_dim - 1)\n emb = torch.exp(torch.arange(half_dim, device=device).float() * -emb)\n emb = scale * x.unsqueeze(1) * emb.unsqueeze(0)\n emb = torch.cat((emb.sin(), emb.cos()), dim=-1)\n return emb\n\n","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.SinusPositionEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.SinusPositionEmbedding#L149-L161","kind":"class","name":"SinusPositionEmbedding","path":"src/f5_tts/model/modules.py","language":"python","start_line":149,"end_line":161,"context_start_line":129,"context_end_line":181,"code":"\n def forward(self, wav):\n if self.dummy.device != wav.device:\n self.to(wav.device)\n\n mel = self.extractor(\n waveform=wav,\n n_fft=self.n_fft,\n n_mel_channels=self.n_mel_channels,\n target_sample_rate=self.target_sample_rate,\n hop_length=self.hop_length,\n win_length=self.win_length,\n )\n\n return mel\n\n\n# sinusoidal position embedding\n\n\nclass SinusPositionEmbedding(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.dim = dim\n\n def forward(self, x, scale=1000):\n device = x.device\n half_dim = self.dim // 2\n emb = math.log(10000) / (half_dim - 1)\n emb = torch.exp(torch.arange(half_dim, device=device).float() * -emb)\n emb = scale * x.unsqueeze(1) * emb.unsqueeze(0)\n emb = torch.cat((emb.sin(), emb.cos()), dim=-1)\n return emb\n\n\n# convolutional position embedding\n\n\nclass ConvPositionEmbedding(nn.Module):\n def __init__(self, dim, kernel_size=31, groups=16):\n super().__init__()\n assert kernel_size % 2 != 0\n self.conv1d = nn.Sequential(\n nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),\n nn.Mish(),\n nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),\n nn.Mish(),\n )\n\n def forward(self, x: float[\"b n d\"], mask: bool[\"b n\"] | None = None): # noqa: F722\n if mask is not None:\n mask = mask[..., None]\n x = x.masked_fill(~mask, 0.0)","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.ConvPositionEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.ConvPositionEmbedding#L167-L190","kind":"class","name":"ConvPositionEmbedding","path":"src/f5_tts/model/modules.py","language":"python","start_line":167,"end_line":190,"context_start_line":147,"context_end_line":210,"code":"\n\nclass SinusPositionEmbedding(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.dim = dim\n\n def forward(self, x, scale=1000):\n device = x.device\n half_dim = self.dim // 2\n emb = math.log(10000) / (half_dim - 1)\n emb = torch.exp(torch.arange(half_dim, device=device).float() * -emb)\n emb = scale * x.unsqueeze(1) * emb.unsqueeze(0)\n emb = torch.cat((emb.sin(), emb.cos()), dim=-1)\n return emb\n\n\n# convolutional position embedding\n\n\nclass ConvPositionEmbedding(nn.Module):\n def __init__(self, dim, kernel_size=31, groups=16):\n super().__init__()\n assert kernel_size % 2 != 0\n self.conv1d = nn.Sequential(\n nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),\n nn.Mish(),\n nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),\n nn.Mish(),\n )\n\n def forward(self, x: float[\"b n d\"], mask: bool[\"b n\"] | None = None): # noqa: F722\n if mask is not None:\n mask = mask[..., None]\n x = x.masked_fill(~mask, 0.0)\n\n x = x.permute(0, 2, 1)\n x = self.conv1d(x)\n out = x.permute(0, 2, 1)\n\n if mask is not None:\n out = out.masked_fill(~mask, 0.0)\n\n return out\n\n\n# rotary positional embedding related\n\n\ndef precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0):\n # proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning\n # has some connection to NTK literature\n # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/\n # https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py\n theta *= theta_rescale_factor ** (dim / (dim - 2))\n freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))\n t = torch.arange(end, device=freqs.device) # type: ignore\n freqs = torch.outer(t, freqs).float() # type: ignore\n freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef get_pos_embed_indices(start, length, max_pos, scale=1.0):","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.precompute_freqs_cis","uri":"program://DMOSpeech2/function/src.f5_tts.model.modules.precompute_freqs_cis#L196-L207","kind":"function","name":"precompute_freqs_cis","path":"src/f5_tts/model/modules.py","language":"python","start_line":196,"end_line":207,"context_start_line":176,"context_end_line":227,"code":" )\n\n def forward(self, x: float[\"b n d\"], mask: bool[\"b n\"] | None = None): # noqa: F722\n if mask is not None:\n mask = mask[..., None]\n x = x.masked_fill(~mask, 0.0)\n\n x = x.permute(0, 2, 1)\n x = self.conv1d(x)\n out = x.permute(0, 2, 1)\n\n if mask is not None:\n out = out.masked_fill(~mask, 0.0)\n\n return out\n\n\n# rotary positional embedding related\n\n\ndef precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0):\n # proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning\n # has some connection to NTK literature\n # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/\n # https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py\n theta *= theta_rescale_factor ** (dim / (dim - 2))\n freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))\n t = torch.arange(end, device=freqs.device) # type: ignore\n freqs = torch.outer(t, freqs).float() # type: ignore\n freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef get_pos_embed_indices(start, length, max_pos, scale=1.0):\n # length = length if isinstance(length, int) else length.max()\n scale = scale * torch.ones_like(start, dtype=torch.float32) # in case scale is a scalar\n pos = (\n start.unsqueeze(1)\n + (torch.arange(length, device=start.device, dtype=torch.float32).unsqueeze(0) * scale.unsqueeze(1)).long()\n )\n # avoid extra long error.\n pos = torch.where(pos < max_pos, pos, max_pos - 1)\n return pos\n\n\n# Global Response Normalization layer (Instance Normalization ?)\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.get_pos_embed_indices","uri":"program://DMOSpeech2/function/src.f5_tts.model.modules.get_pos_embed_indices#L210-L219","kind":"function","name":"get_pos_embed_indices","path":"src/f5_tts/model/modules.py","language":"python","start_line":210,"end_line":219,"context_start_line":190,"context_end_line":239,"code":" return out\n\n\n# rotary positional embedding related\n\n\ndef precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0):\n # proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning\n # has some connection to NTK literature\n # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/\n # https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py\n theta *= theta_rescale_factor ** (dim / (dim - 2))\n freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))\n t = torch.arange(end, device=freqs.device) # type: ignore\n freqs = torch.outer(t, freqs).float() # type: ignore\n freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef get_pos_embed_indices(start, length, max_pos, scale=1.0):\n # length = length if isinstance(length, int) else length.max()\n scale = scale * torch.ones_like(start, dtype=torch.float32) # in case scale is a scalar\n pos = (\n start.unsqueeze(1)\n + (torch.arange(length, device=start.device, dtype=torch.float32).unsqueeze(0) * scale.unsqueeze(1)).long()\n )\n # avoid extra long error.\n pos = torch.where(pos < max_pos, pos, max_pos - 1)\n return pos\n\n\n# Global Response Normalization layer (Instance Normalization ?)\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.gamma = nn.Parameter(torch.zeros(1, 1, dim))\n self.beta = nn.Parameter(torch.zeros(1, 1, dim))\n\n def forward(self, x):\n Gx = torch.norm(x, p=2, dim=1, keepdim=True)\n Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)\n return self.gamma * (x * Nx) + self.beta + x\n\n\n# ConvNeXt-V2 Block https://github.com/facebookresearch/ConvNeXt-V2/blob/main/models/convnextv2.py\n# ref: https://github.com/bfs18/e2_tts/blob/main/rfwave/modules.py#L108\n","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.GRN","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.GRN#L225-L234","kind":"class","name":"GRN","path":"src/f5_tts/model/modules.py","language":"python","start_line":225,"end_line":234,"context_start_line":205,"context_end_line":254,"code":" freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef get_pos_embed_indices(start, length, max_pos, scale=1.0):\n # length = length if isinstance(length, int) else length.max()\n scale = scale * torch.ones_like(start, dtype=torch.float32) # in case scale is a scalar\n pos = (\n start.unsqueeze(1)\n + (torch.arange(length, device=start.device, dtype=torch.float32).unsqueeze(0) * scale.unsqueeze(1)).long()\n )\n # avoid extra long error.\n pos = torch.where(pos < max_pos, pos, max_pos - 1)\n return pos\n\n\n# Global Response Normalization layer (Instance Normalization ?)\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.gamma = nn.Parameter(torch.zeros(1, 1, dim))\n self.beta = nn.Parameter(torch.zeros(1, 1, dim))\n\n def forward(self, x):\n Gx = torch.norm(x, p=2, dim=1, keepdim=True)\n Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)\n return self.gamma * (x * Nx) + self.beta + x\n\n\n# ConvNeXt-V2 Block https://github.com/facebookresearch/ConvNeXt-V2/blob/main/models/convnextv2.py\n# ref: https://github.com/bfs18/e2_tts/blob/main/rfwave/modules.py#L108\n\n\nclass ConvNeXtV2Block(nn.Module):\n def __init__(\n self,\n dim: int,\n intermediate_dim: int,\n dilation: int = 1,\n ):\n super().__init__()\n padding = (dilation * (7 - 1)) // 2\n self.dwconv = nn.Conv1d(\n dim, dim, kernel_size=7, padding=padding, groups=dim, dilation=dilation\n ) # depthwise conv\n self.norm = nn.LayerNorm(dim, eps=1e-6)\n self.pwconv1 = nn.Linear(dim, intermediate_dim) # pointwise/1x1 convs, implemented with linear layers","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.ConvNeXtV2Block","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.ConvNeXtV2Block#L241-L269","kind":"class","name":"ConvNeXtV2Block","path":"src/f5_tts/model/modules.py","language":"python","start_line":241,"end_line":269,"context_start_line":221,"context_end_line":289,"code":"\n# Global Response Normalization layer (Instance Normalization ?)\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.gamma = nn.Parameter(torch.zeros(1, 1, dim))\n self.beta = nn.Parameter(torch.zeros(1, 1, dim))\n\n def forward(self, x):\n Gx = torch.norm(x, p=2, dim=1, keepdim=True)\n Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)\n return self.gamma * (x * Nx) + self.beta + x\n\n\n# ConvNeXt-V2 Block https://github.com/facebookresearch/ConvNeXt-V2/blob/main/models/convnextv2.py\n# ref: https://github.com/bfs18/e2_tts/blob/main/rfwave/modules.py#L108\n\n\nclass ConvNeXtV2Block(nn.Module):\n def __init__(\n self,\n dim: int,\n intermediate_dim: int,\n dilation: int = 1,\n ):\n super().__init__()\n padding = (dilation * (7 - 1)) // 2\n self.dwconv = nn.Conv1d(\n dim, dim, kernel_size=7, padding=padding, groups=dim, dilation=dilation\n ) # depthwise conv\n self.norm = nn.LayerNorm(dim, eps=1e-6)\n self.pwconv1 = nn.Linear(dim, intermediate_dim) # pointwise/1x1 convs, implemented with linear layers\n self.act = nn.GELU()\n self.grn = GRN(intermediate_dim)\n self.pwconv2 = nn.Linear(intermediate_dim, dim)\n\n def forward(self, x: torch.Tensor) -> torch.Tensor:\n residual = x\n x = x.transpose(1, 2) # b n d -> b d n\n x = self.dwconv(x)\n x = x.transpose(1, 2) # b d n -> b n d\n x = self.norm(x)\n x = self.pwconv1(x)\n x = self.act(x)\n x = self.grn(x)\n x = self.pwconv2(x)\n return residual + x\n\n\n# AdaLayerNormZero\n# return with modulated x for attn input, and params for later mlp modulation\n\n\nclass AdaLayerNormZero(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 6)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb=None):\n emb = self.linear(self.silu(emb))\n shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = torch.chunk(emb, 6, dim=1)\n\n x = self.norm(x) * (1 + scale_msa[:, None]) + shift_msa[:, None]","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.AdaLayerNormZero","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.AdaLayerNormZero#L276-L290","kind":"class","name":"AdaLayerNormZero","path":"src/f5_tts/model/modules.py","language":"python","start_line":276,"end_line":290,"context_start_line":256,"context_end_line":310,"code":" self.grn = GRN(intermediate_dim)\n self.pwconv2 = nn.Linear(intermediate_dim, dim)\n\n def forward(self, x: torch.Tensor) -> torch.Tensor:\n residual = x\n x = x.transpose(1, 2) # b n d -> b d n\n x = self.dwconv(x)\n x = x.transpose(1, 2) # b d n -> b n d\n x = self.norm(x)\n x = self.pwconv1(x)\n x = self.act(x)\n x = self.grn(x)\n x = self.pwconv2(x)\n return residual + x\n\n\n# AdaLayerNormZero\n# return with modulated x for attn input, and params for later mlp modulation\n\n\nclass AdaLayerNormZero(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 6)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb=None):\n emb = self.linear(self.silu(emb))\n shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = torch.chunk(emb, 6, dim=1)\n\n x = self.norm(x) * (1 + scale_msa[:, None]) + shift_msa[:, None]\n return x, gate_msa, shift_mlp, scale_mlp, gate_mlp\n\n\n# AdaLayerNormZero for final layer\n# return only with modulated x for attn input, cuz no more mlp modulation\n\n\nclass AdaLayerNormZero_Final(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 2)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb):\n emb = self.linear(self.silu(emb))\n scale, shift = torch.chunk(emb, 2, dim=1)\n\n x = self.norm(x) * (1 + scale)[:, None, :] + shift[:, None, :]","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.AdaLayerNormZero_Final","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.AdaLayerNormZero_Final#L297-L311","kind":"class","name":"AdaLayerNormZero_Final","path":"src/f5_tts/model/modules.py","language":"python","start_line":297,"end_line":311,"context_start_line":277,"context_end_line":331,"code":" def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 6)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb=None):\n emb = self.linear(self.silu(emb))\n shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = torch.chunk(emb, 6, dim=1)\n\n x = self.norm(x) * (1 + scale_msa[:, None]) + shift_msa[:, None]\n return x, gate_msa, shift_mlp, scale_mlp, gate_mlp\n\n\n# AdaLayerNormZero for final layer\n# return only with modulated x for attn input, cuz no more mlp modulation\n\n\nclass AdaLayerNormZero_Final(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 2)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb):\n emb = self.linear(self.silu(emb))\n scale, shift = torch.chunk(emb, 2, dim=1)\n\n x = self.norm(x) * (1 + scale)[:, None, :] + shift[:, None, :]\n return x\n\n\n# FeedForward\n\n\nclass FeedForward(nn.Module):\n def __init__(self, dim, dim_out=None, mult=4, dropout=0.0, approximate: str = \"none\"):\n super().__init__()\n inner_dim = int(dim * mult)\n dim_out = dim_out if dim_out is not None else dim\n\n activation = nn.GELU(approximate=approximate)\n project_in = nn.Sequential(nn.Linear(dim, inner_dim), activation)\n self.ff = nn.Sequential(project_in, nn.Dropout(dropout), nn.Linear(inner_dim, dim_out))\n\n def forward(self, x):\n return self.ff(x)\n\n\n# Attention with possible joint part","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.FeedForward","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.FeedForward#L317-L328","kind":"class","name":"FeedForward","path":"src/f5_tts/model/modules.py","language":"python","start_line":317,"end_line":328,"context_start_line":297,"context_end_line":348,"code":"class AdaLayerNormZero_Final(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 2)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb):\n emb = self.linear(self.silu(emb))\n scale, shift = torch.chunk(emb, 2, dim=1)\n\n x = self.norm(x) * (1 + scale)[:, None, :] + shift[:, None, :]\n return x\n\n\n# FeedForward\n\n\nclass FeedForward(nn.Module):\n def __init__(self, dim, dim_out=None, mult=4, dropout=0.0, approximate: str = \"none\"):\n super().__init__()\n inner_dim = int(dim * mult)\n dim_out = dim_out if dim_out is not None else dim\n\n activation = nn.GELU(approximate=approximate)\n project_in = nn.Sequential(nn.Linear(dim, inner_dim), activation)\n self.ff = nn.Sequential(project_in, nn.Dropout(dropout), nn.Linear(inner_dim, dim_out))\n\n def forward(self, x):\n return self.ff(x)\n\n\n# Attention with possible joint part\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass Attention(nn.Module):\n def __init__(\n self,\n processor: JointAttnProcessor | AttnProcessor,\n dim: int,\n heads: int = 8,\n dim_head: int = 64,\n dropout: float = 0.0,\n context_dim: Optional[int] = None, # if not None -> joint attention\n context_pre_only=None,\n ):\n super().__init__()\n\n if not hasattr(F, \"scaled_dot_product_attention\"):","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.Attention","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.Attention#L335-L390","kind":"class","name":"Attention","path":"src/f5_tts/model/modules.py","language":"python","start_line":335,"end_line":390,"context_start_line":315,"context_end_line":410,"code":"\n\nclass FeedForward(nn.Module):\n def __init__(self, dim, dim_out=None, mult=4, dropout=0.0, approximate: str = \"none\"):\n super().__init__()\n inner_dim = int(dim * mult)\n dim_out = dim_out if dim_out is not None else dim\n\n activation = nn.GELU(approximate=approximate)\n project_in = nn.Sequential(nn.Linear(dim, inner_dim), activation)\n self.ff = nn.Sequential(project_in, nn.Dropout(dropout), nn.Linear(inner_dim, dim_out))\n\n def forward(self, x):\n return self.ff(x)\n\n\n# Attention with possible joint part\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass Attention(nn.Module):\n def __init__(\n self,\n processor: JointAttnProcessor | AttnProcessor,\n dim: int,\n heads: int = 8,\n dim_head: int = 64,\n dropout: float = 0.0,\n context_dim: Optional[int] = None, # if not None -> joint attention\n context_pre_only=None,\n ):\n super().__init__()\n\n if not hasattr(F, \"scaled_dot_product_attention\"):\n raise ImportError(\"Attention equires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.\")\n\n self.processor = processor\n\n self.dim = dim\n self.heads = heads\n self.inner_dim = dim_head * heads\n self.dropout = dropout\n\n self.context_dim = context_dim\n self.context_pre_only = context_pre_only\n\n self.to_q = nn.Linear(dim, self.inner_dim)\n self.to_k = nn.Linear(dim, self.inner_dim)\n self.to_v = nn.Linear(dim, self.inner_dim)\n\n if self.context_dim is not None:\n self.to_k_c = nn.Linear(context_dim, self.inner_dim)\n self.to_v_c = nn.Linear(context_dim, self.inner_dim)\n if self.context_pre_only is not None:\n self.to_q_c = nn.Linear(context_dim, self.inner_dim)\n\n self.to_out = nn.ModuleList([])\n self.to_out.append(nn.Linear(self.inner_dim, dim))\n self.to_out.append(nn.Dropout(dropout))\n\n if self.context_pre_only is not None and not self.context_pre_only:\n self.to_out_c = nn.Linear(self.inner_dim, dim)\n\n def forward(\n self,\n x: float[\"b n d\"], # noised input x # noqa: F722\n c: float[\"b n d\"] = None, # context c # noqa: F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n src_mask: bool[\"b nt\"] | None = None, # noqa: F722\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.Tensor:\n if c is not None:\n return self.processor(self, x, c=c, mask=mask, src_mask=src_mask, rope=rope, c_rope=c_rope)\n else:\n return self.processor(self, x, mask=mask, rope=rope)\n\n\n# Attention processor\n\n\nclass AttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn: Attention,\n x: float[\"b n d\"], # noised input x # noqa: F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n rope=None, # rotary position embedding\n ) -> torch.FloatTensor:\n batch_size = x.shape[0]\n\n # `sample` projections.\n query = attn.to_q(x)","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.AttnProcessor","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.AttnProcessor#L396-L450","kind":"class","name":"AttnProcessor","path":"src/f5_tts/model/modules.py","language":"python","start_line":396,"end_line":450,"context_start_line":376,"context_end_line":470,"code":" self.to_out_c = nn.Linear(self.inner_dim, dim)\n\n def forward(\n self,\n x: float[\"b n d\"], # noised input x # noqa: F722\n c: float[\"b n d\"] = None, # context c # noqa: F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n src_mask: bool[\"b nt\"] | None = None, # noqa: F722\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.Tensor:\n if c is not None:\n return self.processor(self, x, c=c, mask=mask, src_mask=src_mask, rope=rope, c_rope=c_rope)\n else:\n return self.processor(self, x, mask=mask, rope=rope)\n\n\n# Attention processor\n\n\nclass AttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn: Attention,\n x: float[\"b n d\"], # noised input x # noqa: F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n rope=None, # rotary position embedding\n ) -> torch.FloatTensor:\n batch_size = x.shape[0]\n\n # `sample` projections.\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n\n # apply rotary position embedding\n if rope is not None:\n freqs, xpos_scale = rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n\n query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)\n key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)\n\n # attention\n inner_dim = key.shape[-1]\n head_dim = inner_dim // attn.heads\n query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n\n # mask. e.g. inference got a batch with different target durations, mask out the padding\n if mask is not None:\n attn_mask = mask\n attn_mask = attn_mask.unsqueeze(1).unsqueeze(1) # 'b n -> b 1 1 n'\n attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])\n else:\n attn_mask = None\n\n x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)\n x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)\n x = x.to(query.dtype)\n\n # linear proj\n x = attn.to_out[0](x)\n # dropout\n x = attn.to_out[1](x)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n\n return x\n\n\n# Joint Attention processor for MM-DiT\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass JointAttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn: Attention,\n x: float[\"b n d\"], # noised input x\n c: float[\"b nt d\"] = None, # context c, here text\n mask: bool[\"b n\"] | None = None,\n src_mask: bool[\"b nt\"] | None = None,\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.FloatTensor:","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.JointAttnProcessor","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.JointAttnProcessor#L457-L546","kind":"class","name":"JointAttnProcessor","path":"src/f5_tts/model/modules.py","language":"python","start_line":457,"end_line":546,"context_start_line":437,"context_end_line":566,"code":" x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)\n x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)\n x = x.to(query.dtype)\n\n # linear proj\n x = attn.to_out[0](x)\n # dropout\n x = attn.to_out[1](x)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n\n return x\n\n\n# Joint Attention processor for MM-DiT\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass JointAttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn: Attention,\n x: float[\"b n d\"], # noised input x\n c: float[\"b nt d\"] = None, # context c, here text\n mask: bool[\"b n\"] | None = None,\n src_mask: bool[\"b nt\"] | None = None,\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.FloatTensor:\n residual = x\n batch_size = c.shape[0]\n\n # `sample` projections\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n\n # `context` projections\n c_query = attn.to_q_c(c)\n c_key = attn.to_k_c(c)\n c_value = attn.to_v_c(c)\n\n # apply rope for x\n if rope is not None:\n freqs, xpos_scale = rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)\n key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)\n\n # apply rope for c\n if c_rope is not None:\n freqs, xpos_scale = c_rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n c_query = apply_rotary_pos_emb(c_query, freqs, q_xpos_scale)\n c_key = apply_rotary_pos_emb(c_key, freqs, k_xpos_scale)\n\n # attention\n query = torch.cat([query, c_query], dim=1)\n key = torch.cat([key, c_key], dim=1)\n value = torch.cat([value, c_value], dim=1)\n\n inner_dim = key.shape[-1]\n head_dim = inner_dim // attn.heads\n query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n\n # mask: combine 'mask' (for x) + 'src_mask' (for c) minimally\n if mask is not None:\n attn_mask = F.pad(mask, (0, c.shape[1]), value=True) # no mask for c (text)\n if src_mask is not None:\n # pad src_mask in front so it aligns with c positions\n # and combine with x-mask by logical AND\n attn_mask_c = F.pad(src_mask, (x.shape[1], 0), value=True)\n attn_mask = attn_mask & attn_mask_c\n attn_mask = attn_mask.unsqueeze(1).unsqueeze(1)\n attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])\n else:\n if src_mask is not None:\n # if there's no mask for x but there's src_mask\n attn_mask = F.pad(src_mask, (x.shape[1], 0), value=True)\n attn_mask = attn_mask.unsqueeze(1).unsqueeze(1)\n attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])\n else:\n attn_mask = None\n\n x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)\n x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)\n x = x.to(query.dtype)\n\n # Split the attention outputs\n x, c = x[:, : residual.shape[1]], x[:, residual.shape[1] :]\n\n # linear proj\n x = attn.to_out[0](x)\n x = attn.to_out[1](x)\n if not attn.context_pre_only:\n c = attn.to_out_c(c)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n # c is left as-is (no mask for c by default)\n\n return x, c\n\n\n\n# DiT Block\n\n\nclass DiTBlock(nn.Module):\n def __init__(self, dim, heads, dim_head, ff_mult=4, dropout=0.1):\n super().__init__()\n\n self.attn_norm = AdaLayerNormZero(dim)\n self.attn = Attention(\n processor=AttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n )\n\n self.ff_norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.DiTBlock","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.DiTBlock#L553-L583","kind":"class","name":"DiTBlock","path":"src/f5_tts/model/modules.py","language":"python","start_line":553,"end_line":583,"context_start_line":533,"context_end_line":603,"code":" x, c = x[:, : residual.shape[1]], x[:, residual.shape[1] :]\n\n # linear proj\n x = attn.to_out[0](x)\n x = attn.to_out[1](x)\n if not attn.context_pre_only:\n c = attn.to_out_c(c)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n # c is left as-is (no mask for c by default)\n\n return x, c\n\n\n\n# DiT Block\n\n\nclass DiTBlock(nn.Module):\n def __init__(self, dim, heads, dim_head, ff_mult=4, dropout=0.1):\n super().__init__()\n\n self.attn_norm = AdaLayerNormZero(dim)\n self.attn = Attention(\n processor=AttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n )\n\n self.ff_norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n self.ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n\n def forward(self, x, t, mask=None, rope=None): # x: noised input, t: time embedding\n # pre-norm & modulation for attention input\n norm, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.attn_norm(x, emb=t)\n\n # attention\n attn_output = self.attn(x=norm, mask=mask, rope=rope)\n\n # process attention output for input x\n x = x + gate_msa.unsqueeze(1) * attn_output\n\n norm = self.ff_norm(x) * (1 + scale_mlp[:, None]) + shift_mlp[:, None]\n ff_output = self.ff(norm)\n x = x + gate_mlp.unsqueeze(1) * ff_output\n\n return x\n\n\n# MMDiT Block https://arxiv.org/abs/2403.03206\n\n\nclass MMDiTBlock(nn.Module):\n r\"\"\"\n modified from diffusers/src/diffusers/models/attention.py\n\n notes.\n _c: context related. text, cond, etc. (left part in sd3 fig2.b)\n _x: noised input related. (right part)\n context_pre_only: last layer only do prenorm + modulation cuz no more ffn\n \"\"\"\n\n def __init__(self, dim, heads, dim_head, ff_mult=4, dropout=0.1, context_pre_only=False):\n super().__init__()\n\n self.context_pre_only = context_pre_only\n","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.MMDiTBlock","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.MMDiTBlock#L589-L653","kind":"class","name":"MMDiTBlock","path":"src/f5_tts/model/modules.py","language":"python","start_line":589,"end_line":653,"context_start_line":569,"context_end_line":669,"code":" def forward(self, x, t, mask=None, rope=None): # x: noised input, t: time embedding\n # pre-norm & modulation for attention input\n norm, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.attn_norm(x, emb=t)\n\n # attention\n attn_output = self.attn(x=norm, mask=mask, rope=rope)\n\n # process attention output for input x\n x = x + gate_msa.unsqueeze(1) * attn_output\n\n norm = self.ff_norm(x) * (1 + scale_mlp[:, None]) + shift_mlp[:, None]\n ff_output = self.ff(norm)\n x = x + gate_mlp.unsqueeze(1) * ff_output\n\n return x\n\n\n# MMDiT Block https://arxiv.org/abs/2403.03206\n\n\nclass MMDiTBlock(nn.Module):\n r\"\"\"\n modified from diffusers/src/diffusers/models/attention.py\n\n notes.\n _c: context related. text, cond, etc. (left part in sd3 fig2.b)\n _x: noised input related. (right part)\n context_pre_only: last layer only do prenorm + modulation cuz no more ffn\n \"\"\"\n\n def __init__(self, dim, heads, dim_head, ff_mult=4, dropout=0.1, context_pre_only=False):\n super().__init__()\n\n self.context_pre_only = context_pre_only\n\n self.attn_norm_c = AdaLayerNormZero_Final(dim) if context_pre_only else AdaLayerNormZero(dim)\n self.attn_norm_x = AdaLayerNormZero(dim)\n self.attn = Attention(\n processor=JointAttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n context_dim=dim,\n context_pre_only=context_pre_only,\n )\n\n if not context_pre_only:\n self.ff_norm_c = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n self.ff_c = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n else:\n self.ff_norm_c = None\n self.ff_c = None\n self.ff_norm_x = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n self.ff_x = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n\n def forward(self, x, c, t, mask=None, src_mask=None, rope=None, c_rope=None): # x: noised input, c: context, t: time embedding\n # pre-norm & modulation for attention input\n if self.context_pre_only:\n norm_c = self.attn_norm_c(c, t)\n else:\n norm_c, c_gate_msa, c_shift_mlp, c_scale_mlp, c_gate_mlp = self.attn_norm_c(c, emb=t)\n norm_x, x_gate_msa, x_shift_mlp, x_scale_mlp, x_gate_mlp = self.attn_norm_x(x, emb=t)\n\n # attention\n x_attn_output, c_attn_output = self.attn(x=norm_x, c=norm_c, mask=mask, src_mask=src_mask, rope=rope, c_rope=c_rope)\n\n # process attention output for context c\n if self.context_pre_only:\n c = None\n else: # if not last layer\n c = c + c_gate_msa.unsqueeze(1) * c_attn_output\n\n norm_c = self.ff_norm_c(c) * (1 + c_scale_mlp[:, None]) + c_shift_mlp[:, None]\n c_ff_output = self.ff_c(norm_c)\n c = c + c_gate_mlp.unsqueeze(1) * c_ff_output\n\n # process attention output for input x\n x = x + x_gate_msa.unsqueeze(1) * x_attn_output\n\n norm_x = self.ff_norm_x(x) * (1 + x_scale_mlp[:, None]) + x_shift_mlp[:, None]\n x_ff_output = self.ff_x(norm_x)\n x = x + x_gate_mlp.unsqueeze(1) * x_ff_output\n\n return c, x\n\n\n# time step conditioning embedding\n\n\nclass TimestepEmbedding(nn.Module):\n def __init__(self, dim, freq_embed_dim=256):\n super().__init__()\n self.time_embed = SinusPositionEmbedding(freq_embed_dim)\n self.time_mlp = nn.Sequential(nn.Linear(freq_embed_dim, dim), nn.SiLU(), nn.Linear(dim, dim))\n\n def forward(self, timestep: float[\"b\"]): # noqa: F821\n time_hidden = self.time_embed(timestep)\n time_hidden = time_hidden.to(timestep.dtype)\n time = self.time_mlp(time_hidden) # b d\n return time","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.TimestepEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model.modules.TimestepEmbedding#L659-L669","kind":"class","name":"TimestepEmbedding","path":"src/f5_tts/model/modules.py","language":"python","start_line":659,"end_line":669,"context_start_line":639,"context_end_line":669,"code":" else: # if not last layer\n c = c + c_gate_msa.unsqueeze(1) * c_attn_output\n\n norm_c = self.ff_norm_c(c) * (1 + c_scale_mlp[:, None]) + c_shift_mlp[:, None]\n c_ff_output = self.ff_c(norm_c)\n c = c + c_gate_mlp.unsqueeze(1) * c_ff_output\n\n # process attention output for input x\n x = x + x_gate_msa.unsqueeze(1) * x_attn_output\n\n norm_x = self.ff_norm_x(x) * (1 + x_scale_mlp[:, None]) + x_shift_mlp[:, None]\n x_ff_output = self.ff_x(norm_x)\n x = x + x_gate_mlp.unsqueeze(1) * x_ff_output\n\n return c, x\n\n\n# time step conditioning embedding\n\n\nclass TimestepEmbedding(nn.Module):\n def __init__(self, dim, freq_embed_dim=256):\n super().__init__()\n self.time_embed = SinusPositionEmbedding(freq_embed_dim)\n self.time_mlp = nn.Sequential(nn.Linear(freq_embed_dim, dim), nn.SiLU(), nn.Linear(dim, dim))\n\n def forward(self, timestep: float[\"b\"]): # noqa: F821\n time_hidden = self.time_embed(timestep)\n time_hidden = time_hidden.to(timestep.dtype)\n time = self.time_mlp(time_hidden) # b d\n return time","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model.modules.__init__#L660-L663","kind":"function","name":"__init__","path":"src/f5_tts/model/modules.py","language":"python","start_line":660,"end_line":663,"context_start_line":640,"context_end_line":669,"code":" c = c + c_gate_msa.unsqueeze(1) * c_attn_output\n\n norm_c = self.ff_norm_c(c) * (1 + c_scale_mlp[:, None]) + c_shift_mlp[:, None]\n c_ff_output = self.ff_c(norm_c)\n c = c + c_gate_mlp.unsqueeze(1) * c_ff_output\n\n # process attention output for input x\n x = x + x_gate_msa.unsqueeze(1) * x_attn_output\n\n norm_x = self.ff_norm_x(x) * (1 + x_scale_mlp[:, None]) + x_shift_mlp[:, None]\n x_ff_output = self.ff_x(norm_x)\n x = x + x_gate_mlp.unsqueeze(1) * x_ff_output\n\n return c, x\n\n\n# time step conditioning embedding\n\n\nclass TimestepEmbedding(nn.Module):\n def __init__(self, dim, freq_embed_dim=256):\n super().__init__()\n self.time_embed = SinusPositionEmbedding(freq_embed_dim)\n self.time_mlp = nn.Sequential(nn.Linear(freq_embed_dim, dim), nn.SiLU(), nn.Linear(dim, dim))\n\n def forward(self, timestep: float[\"b\"]): # noqa: F821\n time_hidden = self.time_embed(timestep)\n time_hidden = time_hidden.to(timestep.dtype)\n time = self.time_mlp(time_hidden) # b d\n return time","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.forward","uri":"program://DMOSpeech2/function/src.f5_tts.model.modules.forward#L665-L669","kind":"function","name":"forward","path":"src/f5_tts/model/modules.py","language":"python","start_line":665,"end_line":669,"context_start_line":645,"context_end_line":669,"code":"\n # process attention output for input x\n x = x + x_gate_msa.unsqueeze(1) * x_attn_output\n\n norm_x = self.ff_norm_x(x) * (1 + x_scale_mlp[:, None]) + x_shift_mlp[:, None]\n x_ff_output = self.ff_x(norm_x)\n x = x + x_gate_mlp.unsqueeze(1) * x_ff_output\n\n return c, x\n\n\n# time step conditioning embedding\n\n\nclass TimestepEmbedding(nn.Module):\n def __init__(self, dim, freq_embed_dim=256):\n super().__init__()\n self.time_embed = SinusPositionEmbedding(freq_embed_dim)\n self.time_mlp = nn.Sequential(nn.Linear(freq_embed_dim, dim), nn.SiLU(), nn.Linear(dim, dim))\n\n def forward(self, timestep: float[\"b\"]): # noqa: F821\n time_hidden = self.time_embed(timestep)\n time_hidden = time_hidden.to(timestep.dtype)\n time = self.time_mlp(time_hidden) # b d\n return time","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.modules.__call__","uri":"program://DMOSpeech2/function/src.f5_tts.model.modules.__call__#L461-L546","kind":"function","name":"__call__","path":"src/f5_tts/model/modules.py","language":"python","start_line":461,"end_line":546,"context_start_line":441,"context_end_line":566,"code":" # linear proj\n x = attn.to_out[0](x)\n # dropout\n x = attn.to_out[1](x)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n\n return x\n\n\n# Joint Attention processor for MM-DiT\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass JointAttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn: Attention,\n x: float[\"b n d\"], # noised input x\n c: float[\"b nt d\"] = None, # context c, here text\n mask: bool[\"b n\"] | None = None,\n src_mask: bool[\"b nt\"] | None = None,\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.FloatTensor:\n residual = x\n batch_size = c.shape[0]\n\n # `sample` projections\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n\n # `context` projections\n c_query = attn.to_q_c(c)\n c_key = attn.to_k_c(c)\n c_value = attn.to_v_c(c)\n\n # apply rope for x\n if rope is not None:\n freqs, xpos_scale = rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)\n key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)\n\n # apply rope for c\n if c_rope is not None:\n freqs, xpos_scale = c_rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n c_query = apply_rotary_pos_emb(c_query, freqs, q_xpos_scale)\n c_key = apply_rotary_pos_emb(c_key, freqs, k_xpos_scale)\n\n # attention\n query = torch.cat([query, c_query], dim=1)\n key = torch.cat([key, c_key], dim=1)\n value = torch.cat([value, c_value], dim=1)\n\n inner_dim = key.shape[-1]\n head_dim = inner_dim // attn.heads\n query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n\n # mask: combine 'mask' (for x) + 'src_mask' (for c) minimally\n if mask is not None:\n attn_mask = F.pad(mask, (0, c.shape[1]), value=True) # no mask for c (text)\n if src_mask is not None:\n # pad src_mask in front so it aligns with c positions\n # and combine with x-mask by logical AND\n attn_mask_c = F.pad(src_mask, (x.shape[1], 0), value=True)\n attn_mask = attn_mask & attn_mask_c\n attn_mask = attn_mask.unsqueeze(1).unsqueeze(1)\n attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])\n else:\n if src_mask is not None:\n # if there's no mask for x but there's src_mask\n attn_mask = F.pad(src_mask, (x.shape[1], 0), value=True)\n attn_mask = attn_mask.unsqueeze(1).unsqueeze(1)\n attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])\n else:\n attn_mask = None\n\n x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)\n x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)\n x = x.to(query.dtype)\n\n # Split the attention outputs\n x, c = x[:, : residual.shape[1]], x[:, residual.shape[1] :]\n\n # linear proj\n x = attn.to_out[0](x)\n x = attn.to_out[1](x)\n if not attn.context_pre_only:\n c = attn.to_out_c(c)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n # c is left as-is (no mask for c by default)\n\n return x, c\n\n\n\n# DiT Block\n\n\nclass DiTBlock(nn.Module):\n def __init__(self, dim, heads, dim_head, ff_mult=4, dropout=0.1):\n super().__init__()\n\n self.attn_norm = AdaLayerNormZero(dim)\n self.attn = Attention(\n processor=AttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n )\n\n self.ff_norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.dit","uri":"program://DMOSpeech2/module/src.f5_tts.model.backbones.dit#L1-L204","kind":"module","name":"src.f5_tts.model.backbones.dit","path":"src/f5_tts/model/backbones/dit.py","language":"python","start_line":1,"end_line":204,"context_start_line":1,"context_end_line":204,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nimport torch\nfrom torch import nn\nimport torch.nn.functional as F\n\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model.modules import (\n TimestepEmbedding,\n ConvNeXtV2Block,\n ConvPositionEmbedding,\n DiTBlock,\n AdaLayerNormZero_Final,\n precompute_freqs_cis,\n get_pos_embed_indices,\n)\n\n\n# Text embedding\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, text_num_embeds, text_dim, conv_layers=0, conv_mult=2):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim) # use 0 as filler token\n\n if conv_layers > 0:\n self.extra_modeling = True\n self.precompute_max_pos = 4096 # ~44s of 24khz audio\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(text_dim, self.precompute_max_pos), persistent=False)\n self.text_blocks = nn.Sequential(\n *[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)]\n )\n else:\n self.extra_modeling = False\n\n def forward(self, text: int[\"b nt\"], seq_len, drop_text=False): # noqa: F722\n text = text + 1 # use 0 as filler token. preprocess of batch pad -1, see list_str_to_idx()\n text = text[:, :seq_len] # curtail if character tokens are more than the mel spec tokens\n batch, text_len = text.shape[0], text.shape[1]\n text = F.pad(text, (0, seq_len - text_len), value=0)\n\n if drop_text: # cfg for text\n text = torch.zeros_like(text)\n \n text = self.text_embed(text) # b n -> b n d\n\n # possible extra modeling\n if self.extra_modeling:\n # sinus pos emb\n batch_start = torch.zeros((batch,), dtype=torch.long)\n pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n text = text + text_pos_embed\n\n # convnextv2 blocks\n text = self.text_blocks(text)\n\n return text\n\n\n# noised input audio and context mixing embedding\n\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using DiT blocks\n\n\nclass DiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n conv_layers=0,\n long_skip_connection=False,\n checkpoint_activations=False,\n second_time=False,\n ):\n super().__init__()\n\n self.time_embed = TimestepEmbedding(dim)\n if second_time:\n self.time_embed2 = TimestepEmbedding(dim)\n # Zero-init the weights and biases of the first and last Linear layers in time_mlp\n nn.init.zeros_(self.time_embed2.time_mlp[0].weight) # First Linear layer weights\n nn.init.zeros_(self.time_embed2.time_mlp[0].bias) # First Linear layer bias\n nn.init.zeros_(self.time_embed2.time_mlp[-1].weight) # Last Linear layer weights\n nn.init.zeros_(self.time_embed2.time_mlp[-1].bias) # Last Linear layer bias\n else:\n self.time_embed2 = None\n \n if text_dim is None:\n text_dim = mel_dim\n self.vocab_size = text_num_embeds\n self.text_embed = TextEmbedding(text_num_embeds, text_dim, conv_layers=conv_layers)\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n self.dim = dim\n self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [DiTBlock(dim=dim, heads=heads, dim_head=dim_head, ff_mult=ff_mult, dropout=dropout) for _ in range(depth)]\n )\n self.long_skip_connection = nn.Linear(dim * 2, dim, bias=False) if long_skip_connection else None\n\n self.norm_out = AdaLayerNormZero_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.checkpoint_activations = checkpoint_activations\n\n def ckpt_wrapper(self, module):\n # https://github.com/chuanyangjin/fast-DiT/blob/main/models.py\n def ckpt_forward(*inputs):\n outputs = module(*inputs)\n return outputs\n\n return ckpt_forward\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n drop_audio_cond, # cfg for cond audio\n drop_text, # cfg for text\n mask: bool[\"b n\"] | None = None, # noqa: F722\n second_time: float[\"b\"] | float[\"\"] = None, # noqa: F821 F722\n classify_mode: bool = False, # noqa: F821\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if second_time is not None and self.time_embed2 is not None:\n t2 = self.time_embed2(second_time)\n t = t + t2\n \n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n rope = self.rotary_embed.forward_from_seq_len(seq_len)\n\n if self.long_skip_connection is not None:\n residual = x\n\n if classify_mode:\n layers = [x]\n\n for block in self.transformer_blocks:\n if self.checkpoint_activations:\n x = torch.utils.checkpoint.checkpoint(self.ckpt_wrapper(block), x, t, mask, rope)\n else:\n x = block(x, t, mask=mask, rope=rope)\n\n if classify_mode:\n layers.append(x)\n\n if self.long_skip_connection is not None:\n x = self.long_skip_connection(torch.cat((x, residual), dim=-1))\n\n x = self.norm_out(x, t)\n output = self.proj_out(x)\n\n if classify_mode:\n return layers\n\n return output","source_hash":"4647e46533416eff5d28067e6dbaf5a86f7fec6b3a8466774fbaeb77ed0f432d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.dit.TextEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model.backbones.dit.TextEmbedding#L32-L69","kind":"class","name":"TextEmbedding","path":"src/f5_tts/model/backbones/dit.py","language":"python","start_line":32,"end_line":69,"context_start_line":12,"context_end_line":89,"code":"import torch\nfrom torch import nn\nimport torch.nn.functional as F\n\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model.modules import (\n TimestepEmbedding,\n ConvNeXtV2Block,\n ConvPositionEmbedding,\n DiTBlock,\n AdaLayerNormZero_Final,\n precompute_freqs_cis,\n get_pos_embed_indices,\n)\n\n\n# Text embedding\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, text_num_embeds, text_dim, conv_layers=0, conv_mult=2):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim) # use 0 as filler token\n\n if conv_layers > 0:\n self.extra_modeling = True\n self.precompute_max_pos = 4096 # ~44s of 24khz audio\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(text_dim, self.precompute_max_pos), persistent=False)\n self.text_blocks = nn.Sequential(\n *[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)]\n )\n else:\n self.extra_modeling = False\n\n def forward(self, text: int[\"b nt\"], seq_len, drop_text=False): # noqa: F722\n text = text + 1 # use 0 as filler token. preprocess of batch pad -1, see list_str_to_idx()\n text = text[:, :seq_len] # curtail if character tokens are more than the mel spec tokens\n batch, text_len = text.shape[0], text.shape[1]\n text = F.pad(text, (0, seq_len - text_len), value=0)\n\n if drop_text: # cfg for text\n text = torch.zeros_like(text)\n \n text = self.text_embed(text) # b n -> b n d\n\n # possible extra modeling\n if self.extra_modeling:\n # sinus pos emb\n batch_start = torch.zeros((batch,), dtype=torch.long)\n pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n text = text + text_pos_embed\n\n # convnextv2 blocks\n text = self.text_blocks(text)\n\n return text\n\n\n# noised input audio and context mixing embedding\n\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n","source_hash":"4647e46533416eff5d28067e6dbaf5a86f7fec6b3a8466774fbaeb77ed0f432d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.dit.InputEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model.backbones.dit.InputEmbedding#L75-L87","kind":"class","name":"InputEmbedding","path":"src/f5_tts/model/backbones/dit.py","language":"python","start_line":75,"end_line":87,"context_start_line":55,"context_end_line":107,"code":" \n text = self.text_embed(text) # b n -> b n d\n\n # possible extra modeling\n if self.extra_modeling:\n # sinus pos emb\n batch_start = torch.zeros((batch,), dtype=torch.long)\n pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n text = text + text_pos_embed\n\n # convnextv2 blocks\n text = self.text_blocks(text)\n\n return text\n\n\n# noised input audio and context mixing embedding\n\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using DiT blocks\n\n\nclass DiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n conv_layers=0,\n long_skip_connection=False,","source_hash":"4647e46533416eff5d28067e6dbaf5a86f7fec6b3a8466774fbaeb77ed0f432d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.dit.DiT","uri":"program://DMOSpeech2/class/src.f5_tts.model.backbones.dit.DiT#L93-L204","kind":"class","name":"DiT","path":"src/f5_tts/model/backbones/dit.py","language":"python","start_line":93,"end_line":204,"context_start_line":73,"context_end_line":204,"code":"\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using DiT blocks\n\n\nclass DiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n conv_layers=0,\n long_skip_connection=False,\n checkpoint_activations=False,\n second_time=False,\n ):\n super().__init__()\n\n self.time_embed = TimestepEmbedding(dim)\n if second_time:\n self.time_embed2 = TimestepEmbedding(dim)\n # Zero-init the weights and biases of the first and last Linear layers in time_mlp\n nn.init.zeros_(self.time_embed2.time_mlp[0].weight) # First Linear layer weights\n nn.init.zeros_(self.time_embed2.time_mlp[0].bias) # First Linear layer bias\n nn.init.zeros_(self.time_embed2.time_mlp[-1].weight) # Last Linear layer weights\n nn.init.zeros_(self.time_embed2.time_mlp[-1].bias) # Last Linear layer bias\n else:\n self.time_embed2 = None\n \n if text_dim is None:\n text_dim = mel_dim\n self.vocab_size = text_num_embeds\n self.text_embed = TextEmbedding(text_num_embeds, text_dim, conv_layers=conv_layers)\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n self.dim = dim\n self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [DiTBlock(dim=dim, heads=heads, dim_head=dim_head, ff_mult=ff_mult, dropout=dropout) for _ in range(depth)]\n )\n self.long_skip_connection = nn.Linear(dim * 2, dim, bias=False) if long_skip_connection else None\n\n self.norm_out = AdaLayerNormZero_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.checkpoint_activations = checkpoint_activations\n\n def ckpt_wrapper(self, module):\n # https://github.com/chuanyangjin/fast-DiT/blob/main/models.py\n def ckpt_forward(*inputs):\n outputs = module(*inputs)\n return outputs\n\n return ckpt_forward\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n drop_audio_cond, # cfg for cond audio\n drop_text, # cfg for text\n mask: bool[\"b n\"] | None = None, # noqa: F722\n second_time: float[\"b\"] | float[\"\"] = None, # noqa: F821 F722\n classify_mode: bool = False, # noqa: F821\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if second_time is not None and self.time_embed2 is not None:\n t2 = self.time_embed2(second_time)\n t = t + t2\n \n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n rope = self.rotary_embed.forward_from_seq_len(seq_len)\n\n if self.long_skip_connection is not None:\n residual = x\n\n if classify_mode:\n layers = [x]\n\n for block in self.transformer_blocks:\n if self.checkpoint_activations:\n x = torch.utils.checkpoint.checkpoint(self.ckpt_wrapper(block), x, t, mask, rope)\n else:\n x = block(x, t, mask=mask, rope=rope)\n\n if classify_mode:\n layers.append(x)\n\n if self.long_skip_connection is not None:\n x = self.long_skip_connection(torch.cat((x, residual), dim=-1))\n\n x = self.norm_out(x, t)\n output = self.proj_out(x)\n\n if classify_mode:\n return layers\n\n return output","source_hash":"4647e46533416eff5d28067e6dbaf5a86f7fec6b3a8466774fbaeb77ed0f432d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.dit.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model.backbones.dit.__init__#L94-L143","kind":"function","name":"__init__","path":"src/f5_tts/model/backbones/dit.py","language":"python","start_line":94,"end_line":143,"context_start_line":74,"context_end_line":163,"code":"\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using DiT blocks\n\n\nclass DiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n conv_layers=0,\n long_skip_connection=False,\n checkpoint_activations=False,\n second_time=False,\n ):\n super().__init__()\n\n self.time_embed = TimestepEmbedding(dim)\n if second_time:\n self.time_embed2 = TimestepEmbedding(dim)\n # Zero-init the weights and biases of the first and last Linear layers in time_mlp\n nn.init.zeros_(self.time_embed2.time_mlp[0].weight) # First Linear layer weights\n nn.init.zeros_(self.time_embed2.time_mlp[0].bias) # First Linear layer bias\n nn.init.zeros_(self.time_embed2.time_mlp[-1].weight) # Last Linear layer weights\n nn.init.zeros_(self.time_embed2.time_mlp[-1].bias) # Last Linear layer bias\n else:\n self.time_embed2 = None\n \n if text_dim is None:\n text_dim = mel_dim\n self.vocab_size = text_num_embeds\n self.text_embed = TextEmbedding(text_num_embeds, text_dim, conv_layers=conv_layers)\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n self.dim = dim\n self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [DiTBlock(dim=dim, heads=heads, dim_head=dim_head, ff_mult=ff_mult, dropout=dropout) for _ in range(depth)]\n )\n self.long_skip_connection = nn.Linear(dim * 2, dim, bias=False) if long_skip_connection else None\n\n self.norm_out = AdaLayerNormZero_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.checkpoint_activations = checkpoint_activations\n\n def ckpt_wrapper(self, module):\n # https://github.com/chuanyangjin/fast-DiT/blob/main/models.py\n def ckpt_forward(*inputs):\n outputs = module(*inputs)\n return outputs\n\n return ckpt_forward\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n drop_audio_cond, # cfg for cond audio\n drop_text, # cfg for text\n mask: bool[\"b n\"] | None = None, # noqa: F722\n second_time: float[\"b\"] | float[\"\"] = None, # noqa: F821 F722\n classify_mode: bool = False, # noqa: F821","source_hash":"4647e46533416eff5d28067e6dbaf5a86f7fec6b3a8466774fbaeb77ed0f432d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.dit.forward","uri":"program://DMOSpeech2/function/src.f5_tts.model.backbones.dit.forward#L153-L204","kind":"function","name":"forward","path":"src/f5_tts/model/backbones/dit.py","language":"python","start_line":153,"end_line":204,"context_start_line":133,"context_end_line":204,"code":" self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [DiTBlock(dim=dim, heads=heads, dim_head=dim_head, ff_mult=ff_mult, dropout=dropout) for _ in range(depth)]\n )\n self.long_skip_connection = nn.Linear(dim * 2, dim, bias=False) if long_skip_connection else None\n\n self.norm_out = AdaLayerNormZero_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.checkpoint_activations = checkpoint_activations\n\n def ckpt_wrapper(self, module):\n # https://github.com/chuanyangjin/fast-DiT/blob/main/models.py\n def ckpt_forward(*inputs):\n outputs = module(*inputs)\n return outputs\n\n return ckpt_forward\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n drop_audio_cond, # cfg for cond audio\n drop_text, # cfg for text\n mask: bool[\"b n\"] | None = None, # noqa: F722\n second_time: float[\"b\"] | float[\"\"] = None, # noqa: F821 F722\n classify_mode: bool = False, # noqa: F821\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if second_time is not None and self.time_embed2 is not None:\n t2 = self.time_embed2(second_time)\n t = t + t2\n \n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n rope = self.rotary_embed.forward_from_seq_len(seq_len)\n\n if self.long_skip_connection is not None:\n residual = x\n\n if classify_mode:\n layers = [x]\n\n for block in self.transformer_blocks:\n if self.checkpoint_activations:\n x = torch.utils.checkpoint.checkpoint(self.ckpt_wrapper(block), x, t, mask, rope)\n else:\n x = block(x, t, mask=mask, rope=rope)\n\n if classify_mode:\n layers.append(x)\n\n if self.long_skip_connection is not None:\n x = self.long_skip_connection(torch.cat((x, residual), dim=-1))\n\n x = self.norm_out(x, t)\n output = self.proj_out(x)\n\n if classify_mode:\n return layers\n\n return output","source_hash":"4647e46533416eff5d28067e6dbaf5a86f7fec6b3a8466774fbaeb77ed0f432d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.dit.ckpt_wrapper","uri":"program://DMOSpeech2/function/src.f5_tts.model.backbones.dit.ckpt_wrapper#L145-L151","kind":"function","name":"ckpt_wrapper","path":"src/f5_tts/model/backbones/dit.py","language":"python","start_line":145,"end_line":151,"context_start_line":125,"context_end_line":171,"code":" text_dim = mel_dim\n self.vocab_size = text_num_embeds\n self.text_embed = TextEmbedding(text_num_embeds, text_dim, conv_layers=conv_layers)\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n self.dim = dim\n self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [DiTBlock(dim=dim, heads=heads, dim_head=dim_head, ff_mult=ff_mult, dropout=dropout) for _ in range(depth)]\n )\n self.long_skip_connection = nn.Linear(dim * 2, dim, bias=False) if long_skip_connection else None\n\n self.norm_out = AdaLayerNormZero_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.checkpoint_activations = checkpoint_activations\n\n def ckpt_wrapper(self, module):\n # https://github.com/chuanyangjin/fast-DiT/blob/main/models.py\n def ckpt_forward(*inputs):\n outputs = module(*inputs)\n return outputs\n\n return ckpt_forward\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n drop_audio_cond, # cfg for cond audio\n drop_text, # cfg for text\n mask: bool[\"b n\"] | None = None, # noqa: F722\n second_time: float[\"b\"] | float[\"\"] = None, # noqa: F821 F722\n classify_mode: bool = False, # noqa: F821\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if second_time is not None and self.time_embed2 is not None:","source_hash":"4647e46533416eff5d28067e6dbaf5a86f7fec6b3a8466774fbaeb77ed0f432d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.dit.ckpt_forward","uri":"program://DMOSpeech2/function/src.f5_tts.model.backbones.dit.ckpt_forward#L147-L149","kind":"function","name":"ckpt_forward","path":"src/f5_tts/model/backbones/dit.py","language":"python","start_line":147,"end_line":149,"context_start_line":127,"context_end_line":169,"code":" self.text_embed = TextEmbedding(text_num_embeds, text_dim, conv_layers=conv_layers)\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n self.dim = dim\n self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [DiTBlock(dim=dim, heads=heads, dim_head=dim_head, ff_mult=ff_mult, dropout=dropout) for _ in range(depth)]\n )\n self.long_skip_connection = nn.Linear(dim * 2, dim, bias=False) if long_skip_connection else None\n\n self.norm_out = AdaLayerNormZero_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.checkpoint_activations = checkpoint_activations\n\n def ckpt_wrapper(self, module):\n # https://github.com/chuanyangjin/fast-DiT/blob/main/models.py\n def ckpt_forward(*inputs):\n outputs = module(*inputs)\n return outputs\n\n return ckpt_forward\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n drop_audio_cond, # cfg for cond audio\n drop_text, # cfg for text\n mask: bool[\"b n\"] | None = None, # noqa: F722\n second_time: float[\"b\"] | float[\"\"] = None, # noqa: F821 F722\n classify_mode: bool = False, # noqa: F821\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, c: context (text + masked cond audio), x: noised input audio","source_hash":"4647e46533416eff5d28067e6dbaf5a86f7fec6b3a8466774fbaeb77ed0f432d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.mmdit","uri":"program://DMOSpeech2/module/src.f5_tts.model.backbones.mmdit#L1-L336","kind":"module","name":"src.f5_tts.model.backbones.mmdit","path":"src/f5_tts/model/backbones/mmdit.py","language":"python","start_line":1,"end_line":336,"context_start_line":1,"context_end_line":336,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nimport torch\nfrom torch import nn\nimport torch.nn.functional as F\n\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model.modules import (\n TimestepEmbedding,\n ConvPositionEmbedding,\n MMDiTBlock,\n DiTBlock,\n AdaLayerNormZero_Final,\n precompute_freqs_cis,\n get_pos_embed_indices,\n)\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n lens_to_mask,\n list_str_to_idx,\n list_str_to_tensor,\n mask_from_frac_lengths,\n)\n# text embedding\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, out_dim, text_num_embeds):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, out_dim) # will use 0 as filler token\n\n self.precompute_max_pos = 1024\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(out_dim, self.precompute_max_pos), persistent=False)\n\n def forward(self, text: int[\"b nt\"], drop_text=False) -> int[\"b nt d\"]: # noqa: F722\n text = text + 1\n if drop_text:\n text = torch.zeros_like(text)\n text = self.text_embed(text)\n\n # sinus pos emb\n batch_start = torch.zeros((text.shape[0],), dtype=torch.long)\n batch_text_len = text.shape[1]\n pos_idx = get_pos_embed_indices(batch_start, batch_text_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n\n text = text + text_pos_embed\n\n return text\n\n\n# noised input & masked cond audio embedding\n\n\nclass AudioEmbedding(nn.Module):\n def __init__(self, in_dim, out_dim):\n super().__init__()\n self.linear = nn.Linear(2 * in_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond:\n cond = torch.zeros_like(cond)\n x = torch.cat((x, cond), dim=-1)\n x = self.linear(x)\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using MM-DiT blocks\n\n\nclass MMDiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n text_depth=4,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n text_num_embeds=256,\n mel_dim=100,\n checkpoint_activations=False,\n text_encoder=True,\n\n ):\n super().__init__()\n\n self.time_embed = TimestepEmbedding(dim)\n if text_encoder:\n self.text_encoder = TextEncoder(text_num_embeds=text_num_embeds, \n text_dim=dim,\n depth=text_depth,\n heads=heads,\n dim_head=dim_head,\n ff_mult=ff_mult,\n dropout=dropout)\n else:\n self.text_encoder = None\n self.text_embed = TextEmbedding(dim, text_num_embeds)\n \n self.audio_embed = AudioEmbedding(mel_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n self.dim = dim\n self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [\n MMDiTBlock(\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n ff_mult=ff_mult,\n context_pre_only=i == depth - 1,\n )\n for i in range(depth)\n ]\n )\n self.norm_out = AdaLayerNormZero_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n \n self.checkpoint_activations = checkpoint_activations\n\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n drop_audio_cond, # cfg for cond audio\n drop_text, # cfg for text\n mask: bool[\"b n\"] | None = None, # noqa: F722\n text_mask: bool[\"b nt\"] | None = None, # noqa: F722\n ):\n batch = x.shape[0]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning (time), c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if self.text_encoder is not None:\n c = self.text_encoder(text, t, mask=text_mask, drop_text=drop_text)\n else:\n c = self.text_embed(text, drop_text=drop_text)\n \n x = self.audio_embed(x, cond, drop_audio_cond=drop_audio_cond)\n\n seq_len = x.shape[1]\n text_len = text.shape[1]\n rope_audio = self.rotary_embed.forward_from_seq_len(seq_len)\n rope_text = self.rotary_embed.forward_from_seq_len(text_len)\n \n # if mask is not None:\n # rope_audio = self.rotary_embed.forward_from_seq_len(seq_len + 1)\n \n # dummy_token = torch.zeros((x.shape[0], 1, x.shape[-1]), device=x.device, dtype=x.dtype)\n # x = torch.cat([x, dummy_token], dim=1) # shape is now [b, nw+1, d]\n \n # # pad the mask so that new dummy token is always masked out\n # # mask: [b, nw] -> [b, nw+1]\n # false_col = torch.zeros((x.shape[0], 1), dtype=torch.bool, device=x.device)\n # mask = torch.cat([mask, false_col], dim=1)\n \n # if text_mask is not None:\n # rope_text = self.rotary_embed.forward_from_seq_len(text_len + 1)\n\n # dummy_token = torch.zeros((c.shape[0], 1, c.shape[-1]), device=c.device, dtype=c.dtype)\n # c = torch.cat([c, dummy_token], dim=1) # shape is now [b, nt+1, d]\n \n # # pad the text mask so that new dummy token is always masked out\n # # text_mask: [b, nt] -> [b, nt+1]\n # false_col = torch.zeros((c.shape[0], 1), dtype=torch.bool, device=c.device)\n # text_mask = torch.cat([text_mask, false_col], dim=1)\n \n for block in self.transformer_blocks:\n c, x = block(x, c, t, mask=mask, src_mask=text_mask, rope=rope_audio, c_rope=rope_text)\n\n x = self.norm_out(x, t)\n output = self.proj_out(x)\n \n\n return output\n\nclass TextEncoder(nn.Module):\n def __init__(\n self,\n text_num_embeds: int,\n text_dim: int = 512,\n depth: int = 4,\n heads: int = 8,\n dim_head: int = 64,\n ff_mult: int = 4,\n dropout: float = 0.1,\n ):\n \"\"\"\n A simple text encoder: an embedding layer + multiple DiTBlocks or any other\n transformer blocks for text-only self-attention.\n \"\"\"\n super().__init__()\n # Embeddings\n self.text_embed = TextEmbedding(text_dim, text_num_embeds)\n self.rotary_embed = RotaryEmbedding(dim_head)\n \n # Example stack of DiTBlocks or any custom blocks\n self.transformer_blocks = nn.ModuleList(\n [\n DiTBlock(\n dim=text_dim,\n heads=heads,\n dim_head=dim_head,\n ff_mult=ff_mult,\n dropout=dropout,\n )\n for _ in range(depth)\n ]\n )\n\n def forward(\n self,\n text: int[\"b nt\"], # noqa: F821\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b nt\"] | None = None, # noqa: F821 F722\n drop_text: bool = False\n ):\n \"\"\"\n Encode text into hidden states of shape [b, nt, d].\n \"\"\"\n batch, seq_len, device = text.shape[0], text.shape[1], text.device\n\n if drop_text:\n text = torch.zeros_like(text)\n\n # Basic embedding\n hidden_states = self.text_embed(text, seq_len) # [b, nt, d]\n \n # lens and mask\n rope = self.rotary_embed.forward_from_seq_len(seq_len)\n\n # Pass through self-attention blocks\n for block in self.transformer_blocks:\n # Here, you likely want standard self-attn, so no cross-attn\n hidden_states = block(\n x=hidden_states,\n t=time, # no time embedding for the text encoder by default\n mask=mask, # or pass a text mask if needed\n rope=rope # pass a rope if you want rotary embeddings for text\n )\n return hidden_states\n\nif __name__ == \"__main__\":\n from f5_tts.model.utils import get_tokenizer\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n \n text = [\"hello world\"] * bsz\n text_lens = torch.ones((bsz, ), dtype=torch.long) * len(\"hello world\")\n text_lens[-1] = 5\n device = \"cuda\"\n batch = bsz\n time_embed = TimestepEmbedding(512).to(device)\n \n \n # handle text as string\n if isinstance(text, list):\n if exists(vocab_char_map):\n text = list_str_to_idx(text, vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch \n \n time = torch.rand((batch,), device=device)\n text_mask = lens_to_mask(text_lens).to(device)\n\n # # test text encoder\n # text_encoder = TextEncoder(\n # text_num_embeds=vocab_size,\n # text_dim=512,\n # depth=4,\n # heads=8,\n # dim_head=64,\n # ff_mult=4,\n # dropout=0.1\n # ).to('cuda')\n # hidden_states = text_encoder(text, time_embed(time), mask)\n # print(hidden_states.shape) # [bsz, seq_len, text_dim]\n \n # test MMDiT\n mel_dim = 80\n model = MMDiT(\n dim=512,\n text_depth=4,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n text_num_embeds=vocab_size,\n mel_dim=mel_dim\n ).to(device)\n \n x = torch.rand((batch, 100, mel_dim), device=device)\n cond = torch.rand((batch, 100, mel_dim), device=device)\n lens = torch.ones((batch,), dtype=torch.long) * 100\n mask = lens_to_mask(lens).to(device)\n \n output = model(x, cond, text, time, drop_audio_cond=False, drop_text=False, mask=mask, text_mask=text_mask)\n \n print(output.shape) # [bsz, seq_len, mel_dim]","source_hash":"127234f75c330a1063a9e8032ed491029c301ed85c79db764881b250600ebdb5","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.mmdit.TextEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model.backbones.mmdit.TextEmbedding#L39-L61","kind":"class","name":"TextEmbedding","path":"src/f5_tts/model/backbones/mmdit.py","language":"python","start_line":39,"end_line":61,"context_start_line":19,"context_end_line":81,"code":" TimestepEmbedding,\n ConvPositionEmbedding,\n MMDiTBlock,\n DiTBlock,\n AdaLayerNormZero_Final,\n precompute_freqs_cis,\n get_pos_embed_indices,\n)\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n lens_to_mask,\n list_str_to_idx,\n list_str_to_tensor,\n mask_from_frac_lengths,\n)\n# text embedding\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, out_dim, text_num_embeds):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, out_dim) # will use 0 as filler token\n\n self.precompute_max_pos = 1024\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(out_dim, self.precompute_max_pos), persistent=False)\n\n def forward(self, text: int[\"b nt\"], drop_text=False) -> int[\"b nt d\"]: # noqa: F722\n text = text + 1\n if drop_text:\n text = torch.zeros_like(text)\n text = self.text_embed(text)\n\n # sinus pos emb\n batch_start = torch.zeros((text.shape[0],), dtype=torch.long)\n batch_text_len = text.shape[1]\n pos_idx = get_pos_embed_indices(batch_start, batch_text_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n\n text = text + text_pos_embed\n\n return text\n\n\n# noised input & masked cond audio embedding\n\n\nclass AudioEmbedding(nn.Module):\n def __init__(self, in_dim, out_dim):\n super().__init__()\n self.linear = nn.Linear(2 * in_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond:\n cond = torch.zeros_like(cond)\n x = torch.cat((x, cond), dim=-1)\n x = self.linear(x)\n x = self.conv_pos_embed(x) + x\n return x\n\n","source_hash":"127234f75c330a1063a9e8032ed491029c301ed85c79db764881b250600ebdb5","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.mmdit.AudioEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model.backbones.mmdit.AudioEmbedding#L67-L79","kind":"class","name":"AudioEmbedding","path":"src/f5_tts/model/backbones/mmdit.py","language":"python","start_line":67,"end_line":79,"context_start_line":47,"context_end_line":99,"code":" def forward(self, text: int[\"b nt\"], drop_text=False) -> int[\"b nt d\"]: # noqa: F722\n text = text + 1\n if drop_text:\n text = torch.zeros_like(text)\n text = self.text_embed(text)\n\n # sinus pos emb\n batch_start = torch.zeros((text.shape[0],), dtype=torch.long)\n batch_text_len = text.shape[1]\n pos_idx = get_pos_embed_indices(batch_start, batch_text_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n\n text = text + text_pos_embed\n\n return text\n\n\n# noised input & masked cond audio embedding\n\n\nclass AudioEmbedding(nn.Module):\n def __init__(self, in_dim, out_dim):\n super().__init__()\n self.linear = nn.Linear(2 * in_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond:\n cond = torch.zeros_like(cond)\n x = torch.cat((x, cond), dim=-1)\n x = self.linear(x)\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using MM-DiT blocks\n\n\nclass MMDiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n text_depth=4,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n text_num_embeds=256,\n mel_dim=100,\n checkpoint_activations=False,\n text_encoder=True,","source_hash":"127234f75c330a1063a9e8032ed491029c301ed85c79db764881b250600ebdb5","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.mmdit.MMDiT","uri":"program://DMOSpeech2/class/src.f5_tts.model.backbones.mmdit.MMDiT#L85-L201","kind":"class","name":"MMDiT","path":"src/f5_tts/model/backbones/mmdit.py","language":"python","start_line":85,"end_line":201,"context_start_line":65,"context_end_line":221,"code":"\n\nclass AudioEmbedding(nn.Module):\n def __init__(self, in_dim, out_dim):\n super().__init__()\n self.linear = nn.Linear(2 * in_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond:\n cond = torch.zeros_like(cond)\n x = torch.cat((x, cond), dim=-1)\n x = self.linear(x)\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using MM-DiT blocks\n\n\nclass MMDiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n text_depth=4,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n text_num_embeds=256,\n mel_dim=100,\n checkpoint_activations=False,\n text_encoder=True,\n\n ):\n super().__init__()\n\n self.time_embed = TimestepEmbedding(dim)\n if text_encoder:\n self.text_encoder = TextEncoder(text_num_embeds=text_num_embeds, \n text_dim=dim,\n depth=text_depth,\n heads=heads,\n dim_head=dim_head,\n ff_mult=ff_mult,\n dropout=dropout)\n else:\n self.text_encoder = None\n self.text_embed = TextEmbedding(dim, text_num_embeds)\n \n self.audio_embed = AudioEmbedding(mel_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n self.dim = dim\n self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [\n MMDiTBlock(\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n ff_mult=ff_mult,\n context_pre_only=i == depth - 1,\n )\n for i in range(depth)\n ]\n )\n self.norm_out = AdaLayerNormZero_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n \n self.checkpoint_activations = checkpoint_activations\n\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n drop_audio_cond, # cfg for cond audio\n drop_text, # cfg for text\n mask: bool[\"b n\"] | None = None, # noqa: F722\n text_mask: bool[\"b nt\"] | None = None, # noqa: F722\n ):\n batch = x.shape[0]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning (time), c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if self.text_encoder is not None:\n c = self.text_encoder(text, t, mask=text_mask, drop_text=drop_text)\n else:\n c = self.text_embed(text, drop_text=drop_text)\n \n x = self.audio_embed(x, cond, drop_audio_cond=drop_audio_cond)\n\n seq_len = x.shape[1]\n text_len = text.shape[1]\n rope_audio = self.rotary_embed.forward_from_seq_len(seq_len)\n rope_text = self.rotary_embed.forward_from_seq_len(text_len)\n \n # if mask is not None:\n # rope_audio = self.rotary_embed.forward_from_seq_len(seq_len + 1)\n \n # dummy_token = torch.zeros((x.shape[0], 1, x.shape[-1]), device=x.device, dtype=x.dtype)\n # x = torch.cat([x, dummy_token], dim=1) # shape is now [b, nw+1, d]\n \n # # pad the mask so that new dummy token is always masked out\n # # mask: [b, nw] -> [b, nw+1]\n # false_col = torch.zeros((x.shape[0], 1), dtype=torch.bool, device=x.device)\n # mask = torch.cat([mask, false_col], dim=1)\n \n # if text_mask is not None:\n # rope_text = self.rotary_embed.forward_from_seq_len(text_len + 1)\n\n # dummy_token = torch.zeros((c.shape[0], 1, c.shape[-1]), device=c.device, dtype=c.dtype)\n # c = torch.cat([c, dummy_token], dim=1) # shape is now [b, nt+1, d]\n \n # # pad the text mask so that new dummy token is always masked out\n # # text_mask: [b, nt] -> [b, nt+1]\n # false_col = torch.zeros((c.shape[0], 1), dtype=torch.bool, device=c.device)\n # text_mask = torch.cat([text_mask, false_col], dim=1)\n \n for block in self.transformer_blocks:\n c, x = block(x, c, t, mask=mask, src_mask=text_mask, rope=rope_audio, c_rope=rope_text)\n\n x = self.norm_out(x, t)\n output = self.proj_out(x)\n \n\n return output\n\nclass TextEncoder(nn.Module):\n def __init__(\n self,\n text_num_embeds: int,\n text_dim: int = 512,\n depth: int = 4,\n heads: int = 8,\n dim_head: int = 64,\n ff_mult: int = 4,\n dropout: float = 0.1,\n ):\n \"\"\"\n A simple text encoder: an embedding layer + multiple DiTBlocks or any other\n transformer blocks for text-only self-attention.\n \"\"\"\n super().__init__()\n # Embeddings\n self.text_embed = TextEmbedding(text_dim, text_num_embeds)\n self.rotary_embed = RotaryEmbedding(dim_head)","source_hash":"127234f75c330a1063a9e8032ed491029c301ed85c79db764881b250600ebdb5","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.mmdit.TextEncoder","uri":"program://DMOSpeech2/class/src.f5_tts.model.backbones.mmdit.TextEncoder#L203-L267","kind":"class","name":"TextEncoder","path":"src/f5_tts/model/backbones/mmdit.py","language":"python","start_line":203,"end_line":267,"context_start_line":183,"context_end_line":287,"code":" # if text_mask is not None:\n # rope_text = self.rotary_embed.forward_from_seq_len(text_len + 1)\n\n # dummy_token = torch.zeros((c.shape[0], 1, c.shape[-1]), device=c.device, dtype=c.dtype)\n # c = torch.cat([c, dummy_token], dim=1) # shape is now [b, nt+1, d]\n \n # # pad the text mask so that new dummy token is always masked out\n # # text_mask: [b, nt] -> [b, nt+1]\n # false_col = torch.zeros((c.shape[0], 1), dtype=torch.bool, device=c.device)\n # text_mask = torch.cat([text_mask, false_col], dim=1)\n \n for block in self.transformer_blocks:\n c, x = block(x, c, t, mask=mask, src_mask=text_mask, rope=rope_audio, c_rope=rope_text)\n\n x = self.norm_out(x, t)\n output = self.proj_out(x)\n \n\n return output\n\nclass TextEncoder(nn.Module):\n def __init__(\n self,\n text_num_embeds: int,\n text_dim: int = 512,\n depth: int = 4,\n heads: int = 8,\n dim_head: int = 64,\n ff_mult: int = 4,\n dropout: float = 0.1,\n ):\n \"\"\"\n A simple text encoder: an embedding layer + multiple DiTBlocks or any other\n transformer blocks for text-only self-attention.\n \"\"\"\n super().__init__()\n # Embeddings\n self.text_embed = TextEmbedding(text_dim, text_num_embeds)\n self.rotary_embed = RotaryEmbedding(dim_head)\n \n # Example stack of DiTBlocks or any custom blocks\n self.transformer_blocks = nn.ModuleList(\n [\n DiTBlock(\n dim=text_dim,\n heads=heads,\n dim_head=dim_head,\n ff_mult=ff_mult,\n dropout=dropout,\n )\n for _ in range(depth)\n ]\n )\n\n def forward(\n self,\n text: int[\"b nt\"], # noqa: F821\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b nt\"] | None = None, # noqa: F821 F722\n drop_text: bool = False\n ):\n \"\"\"\n Encode text into hidden states of shape [b, nt, d].\n \"\"\"\n batch, seq_len, device = text.shape[0], text.shape[1], text.device\n\n if drop_text:\n text = torch.zeros_like(text)\n\n # Basic embedding\n hidden_states = self.text_embed(text, seq_len) # [b, nt, d]\n \n # lens and mask\n rope = self.rotary_embed.forward_from_seq_len(seq_len)\n\n # Pass through self-attention blocks\n for block in self.transformer_blocks:\n # Here, you likely want standard self-attn, so no cross-attn\n hidden_states = block(\n x=hidden_states,\n t=time, # no time embedding for the text encoder by default\n mask=mask, # or pass a text mask if needed\n rope=rope # pass a rope if you want rotary embeddings for text\n )\n return hidden_states\n\nif __name__ == \"__main__\":\n from f5_tts.model.utils import get_tokenizer\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n \n text = [\"hello world\"] * bsz\n text_lens = torch.ones((bsz, ), dtype=torch.long) * len(\"hello world\")\n text_lens[-1] = 5\n device = \"cuda\"\n batch = bsz","source_hash":"127234f75c330a1063a9e8032ed491029c301ed85c79db764881b250600ebdb5","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.mmdit.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model.backbones.mmdit.__init__#L204-L235","kind":"function","name":"__init__","path":"src/f5_tts/model/backbones/mmdit.py","language":"python","start_line":204,"end_line":235,"context_start_line":184,"context_end_line":255,"code":" # rope_text = self.rotary_embed.forward_from_seq_len(text_len + 1)\n\n # dummy_token = torch.zeros((c.shape[0], 1, c.shape[-1]), device=c.device, dtype=c.dtype)\n # c = torch.cat([c, dummy_token], dim=1) # shape is now [b, nt+1, d]\n \n # # pad the text mask so that new dummy token is always masked out\n # # text_mask: [b, nt] -> [b, nt+1]\n # false_col = torch.zeros((c.shape[0], 1), dtype=torch.bool, device=c.device)\n # text_mask = torch.cat([text_mask, false_col], dim=1)\n \n for block in self.transformer_blocks:\n c, x = block(x, c, t, mask=mask, src_mask=text_mask, rope=rope_audio, c_rope=rope_text)\n\n x = self.norm_out(x, t)\n output = self.proj_out(x)\n \n\n return output\n\nclass TextEncoder(nn.Module):\n def __init__(\n self,\n text_num_embeds: int,\n text_dim: int = 512,\n depth: int = 4,\n heads: int = 8,\n dim_head: int = 64,\n ff_mult: int = 4,\n dropout: float = 0.1,\n ):\n \"\"\"\n A simple text encoder: an embedding layer + multiple DiTBlocks or any other\n transformer blocks for text-only self-attention.\n \"\"\"\n super().__init__()\n # Embeddings\n self.text_embed = TextEmbedding(text_dim, text_num_embeds)\n self.rotary_embed = RotaryEmbedding(dim_head)\n \n # Example stack of DiTBlocks or any custom blocks\n self.transformer_blocks = nn.ModuleList(\n [\n DiTBlock(\n dim=text_dim,\n heads=heads,\n dim_head=dim_head,\n ff_mult=ff_mult,\n dropout=dropout,\n )\n for _ in range(depth)\n ]\n )\n\n def forward(\n self,\n text: int[\"b nt\"], # noqa: F821\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b nt\"] | None = None, # noqa: F821 F722\n drop_text: bool = False\n ):\n \"\"\"\n Encode text into hidden states of shape [b, nt, d].\n \"\"\"\n batch, seq_len, device = text.shape[0], text.shape[1], text.device\n\n if drop_text:\n text = torch.zeros_like(text)\n\n # Basic embedding\n hidden_states = self.text_embed(text, seq_len) # [b, nt, d]\n \n # lens and mask","source_hash":"127234f75c330a1063a9e8032ed491029c301ed85c79db764881b250600ebdb5","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.mmdit.forward","uri":"program://DMOSpeech2/function/src.f5_tts.model.backbones.mmdit.forward#L237-L267","kind":"function","name":"forward","path":"src/f5_tts/model/backbones/mmdit.py","language":"python","start_line":237,"end_line":267,"context_start_line":217,"context_end_line":287,"code":" \"\"\"\n super().__init__()\n # Embeddings\n self.text_embed = TextEmbedding(text_dim, text_num_embeds)\n self.rotary_embed = RotaryEmbedding(dim_head)\n \n # Example stack of DiTBlocks or any custom blocks\n self.transformer_blocks = nn.ModuleList(\n [\n DiTBlock(\n dim=text_dim,\n heads=heads,\n dim_head=dim_head,\n ff_mult=ff_mult,\n dropout=dropout,\n )\n for _ in range(depth)\n ]\n )\n\n def forward(\n self,\n text: int[\"b nt\"], # noqa: F821\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b nt\"] | None = None, # noqa: F821 F722\n drop_text: bool = False\n ):\n \"\"\"\n Encode text into hidden states of shape [b, nt, d].\n \"\"\"\n batch, seq_len, device = text.shape[0], text.shape[1], text.device\n\n if drop_text:\n text = torch.zeros_like(text)\n\n # Basic embedding\n hidden_states = self.text_embed(text, seq_len) # [b, nt, d]\n \n # lens and mask\n rope = self.rotary_embed.forward_from_seq_len(seq_len)\n\n # Pass through self-attention blocks\n for block in self.transformer_blocks:\n # Here, you likely want standard self-attn, so no cross-attn\n hidden_states = block(\n x=hidden_states,\n t=time, # no time embedding for the text encoder by default\n mask=mask, # or pass a text mask if needed\n rope=rope # pass a rope if you want rotary embeddings for text\n )\n return hidden_states\n\nif __name__ == \"__main__\":\n from f5_tts.model.utils import get_tokenizer\n\n bsz = 16\n \n tokenizer = \"pinyin\" # 'pinyin', 'char', or 'custom'\n tokenizer_path = None # if tokenizer = 'custom', define the path to the tokenizer you want to use (should be vocab.txt)\n dataset_name = \"Emilia_ZH_EN\"\n if tokenizer == \"custom\":\n tokenizer_path = tokenizer_path\n else:\n tokenizer_path = dataset_name\n vocab_char_map, vocab_size = get_tokenizer(tokenizer_path, tokenizer)\n \n text = [\"hello world\"] * bsz\n text_lens = torch.ones((bsz, ), dtype=torch.long) * len(\"hello world\")\n text_lens[-1] = 5\n device = \"cuda\"\n batch = bsz","source_hash":"127234f75c330a1063a9e8032ed491029c301ed85c79db764881b250600ebdb5","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.unett","uri":"program://DMOSpeech2/module/src.f5_tts.model.backbones.unett#L1-L219","kind":"module","name":"src.f5_tts.model.backbones.unett","path":"src/f5_tts/model/backbones/unett.py","language":"python","start_line":1,"end_line":219,"context_start_line":1,"context_end_line":219,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\nfrom typing import Literal\n\nimport torch\nfrom torch import nn\nimport torch.nn.functional as F\n\nfrom x_transformers import RMSNorm\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model.modules import (\n TimestepEmbedding,\n ConvNeXtV2Block,\n ConvPositionEmbedding,\n Attention,\n AttnProcessor,\n FeedForward,\n precompute_freqs_cis,\n get_pos_embed_indices,\n)\n\n\n# Text embedding\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, text_num_embeds, text_dim, conv_layers=0, conv_mult=2):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim) # use 0 as filler token\n\n if conv_layers > 0:\n self.extra_modeling = True\n self.precompute_max_pos = 4096 # ~44s of 24khz audio\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(text_dim, self.precompute_max_pos), persistent=False)\n self.text_blocks = nn.Sequential(\n *[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)]\n )\n else:\n self.extra_modeling = False\n\n def forward(self, text: int[\"b nt\"], seq_len, drop_text=False): # noqa: F722\n text = text + 1 # use 0 as filler token. preprocess of batch pad -1, see list_str_to_idx()\n text = text[:, :seq_len] # curtail if character tokens are more than the mel spec tokens\n batch, text_len = text.shape[0], text.shape[1]\n text = F.pad(text, (0, seq_len - text_len), value=0)\n\n if drop_text: # cfg for text\n text = torch.zeros_like(text)\n\n text = self.text_embed(text) # b n -> b n d\n\n # possible extra modeling\n if self.extra_modeling:\n # sinus pos emb\n batch_start = torch.zeros((batch,), dtype=torch.long)\n pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n text = text + text_pos_embed\n\n # convnextv2 blocks\n text = self.text_blocks(text)\n\n return text\n\n\n# noised input audio and context mixing embedding\n\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Flat UNet Transformer backbone\n\n\nclass UNetT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n conv_layers=0,\n skip_connect_type: Literal[\"add\", \"concat\", \"none\"] = \"concat\",\n ):\n super().__init__()\n assert depth % 2 == 0, \"UNet-Transformer's depth should be even.\"\n\n self.time_embed = TimestepEmbedding(dim)\n if text_dim is None:\n text_dim = mel_dim\n self.text_embed = TextEmbedding(text_num_embeds, text_dim, conv_layers=conv_layers)\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n # transformer layers & skip connections\n\n self.dim = dim\n self.skip_connect_type = skip_connect_type\n needs_skip_proj = skip_connect_type == \"concat\"\n\n self.depth = depth\n self.layers = nn.ModuleList([])\n\n for idx in range(depth):\n is_later_half = idx >= (depth // 2)\n\n attn_norm = RMSNorm(dim)\n attn = Attention(\n processor=AttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n )\n\n ff_norm = RMSNorm(dim)\n ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n\n skip_proj = nn.Linear(dim * 2, dim, bias=False) if needs_skip_proj and is_later_half else None\n\n self.layers.append(\n nn.ModuleList(\n [\n skip_proj,\n attn_norm,\n attn,\n ff_norm,\n ff,\n ]\n )\n )\n\n self.norm_out = RMSNorm(dim)\n self.proj_out = nn.Linear(dim, mel_dim)\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n drop_audio_cond, # cfg for cond audio\n drop_text, # cfg for text\n mask: bool[\"b n\"] | None = None, # noqa: F722\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n # postfix time t to input x, [b n d] -> [b n+1 d]\n x = torch.cat([t.unsqueeze(1), x], dim=1) # pack t to x\n if mask is not None:\n mask = F.pad(mask, (1, 0), value=1)\n\n rope = self.rotary_embed.forward_from_seq_len(seq_len + 1)\n\n # flat unet transformer\n skip_connect_type = self.skip_connect_type\n skips = []\n for idx, (maybe_skip_proj, attn_norm, attn, ff_norm, ff) in enumerate(self.layers):\n layer = idx + 1\n\n # skip connection logic\n is_first_half = layer <= (self.depth // 2)\n is_later_half = not is_first_half\n\n if is_first_half:\n skips.append(x)\n\n if is_later_half:\n skip = skips.pop()\n if skip_connect_type == \"concat\":\n x = torch.cat((x, skip), dim=-1)\n x = maybe_skip_proj(x)\n elif skip_connect_type == \"add\":\n x = x + skip\n\n # attention and feedforward blocks\n x = attn(attn_norm(x), rope=rope, mask=mask) + x\n x = ff(ff_norm(x)) + x\n\n assert len(skips) == 0\n\n x = self.norm_out(x)[:, 1:, :] # unpack t from x\n\n return self.proj_out(x)","source_hash":"852e63fcd735376abb5572df733fb3d7ce64af8d3377c4930a0a4d02090be47a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.unett.TextEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model.backbones.unett.TextEmbedding#L35-L72","kind":"class","name":"TextEmbedding","path":"src/f5_tts/model/backbones/unett.py","language":"python","start_line":35,"end_line":72,"context_start_line":15,"context_end_line":92,"code":"import torch.nn.functional as F\n\nfrom x_transformers import RMSNorm\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model.modules import (\n TimestepEmbedding,\n ConvNeXtV2Block,\n ConvPositionEmbedding,\n Attention,\n AttnProcessor,\n FeedForward,\n precompute_freqs_cis,\n get_pos_embed_indices,\n)\n\n\n# Text embedding\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, text_num_embeds, text_dim, conv_layers=0, conv_mult=2):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim) # use 0 as filler token\n\n if conv_layers > 0:\n self.extra_modeling = True\n self.precompute_max_pos = 4096 # ~44s of 24khz audio\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(text_dim, self.precompute_max_pos), persistent=False)\n self.text_blocks = nn.Sequential(\n *[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)]\n )\n else:\n self.extra_modeling = False\n\n def forward(self, text: int[\"b nt\"], seq_len, drop_text=False): # noqa: F722\n text = text + 1 # use 0 as filler token. preprocess of batch pad -1, see list_str_to_idx()\n text = text[:, :seq_len] # curtail if character tokens are more than the mel spec tokens\n batch, text_len = text.shape[0], text.shape[1]\n text = F.pad(text, (0, seq_len - text_len), value=0)\n\n if drop_text: # cfg for text\n text = torch.zeros_like(text)\n\n text = self.text_embed(text) # b n -> b n d\n\n # possible extra modeling\n if self.extra_modeling:\n # sinus pos emb\n batch_start = torch.zeros((batch,), dtype=torch.long)\n pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n text = text + text_pos_embed\n\n # convnextv2 blocks\n text = self.text_blocks(text)\n\n return text\n\n\n# noised input audio and context mixing embedding\n\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n","source_hash":"852e63fcd735376abb5572df733fb3d7ce64af8d3377c4930a0a4d02090be47a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.unett.InputEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model.backbones.unett.InputEmbedding#L78-L90","kind":"class","name":"InputEmbedding","path":"src/f5_tts/model/backbones/unett.py","language":"python","start_line":78,"end_line":90,"context_start_line":58,"context_end_line":110,"code":"\n text = self.text_embed(text) # b n -> b n d\n\n # possible extra modeling\n if self.extra_modeling:\n # sinus pos emb\n batch_start = torch.zeros((batch,), dtype=torch.long)\n pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n text = text + text_pos_embed\n\n # convnextv2 blocks\n text = self.text_blocks(text)\n\n return text\n\n\n# noised input audio and context mixing embedding\n\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Flat UNet Transformer backbone\n\n\nclass UNetT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n conv_layers=0,\n skip_connect_type: Literal[\"add\", \"concat\", \"none\"] = \"concat\",","source_hash":"852e63fcd735376abb5572df733fb3d7ce64af8d3377c4930a0a4d02090be47a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.unett.UNetT","uri":"program://DMOSpeech2/class/src.f5_tts.model.backbones.unett.UNetT#L96-L219","kind":"class","name":"UNetT","path":"src/f5_tts/model/backbones/unett.py","language":"python","start_line":96,"end_line":219,"context_start_line":76,"context_end_line":219,"code":"\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Flat UNet Transformer backbone\n\n\nclass UNetT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n conv_layers=0,\n skip_connect_type: Literal[\"add\", \"concat\", \"none\"] = \"concat\",\n ):\n super().__init__()\n assert depth % 2 == 0, \"UNet-Transformer's depth should be even.\"\n\n self.time_embed = TimestepEmbedding(dim)\n if text_dim is None:\n text_dim = mel_dim\n self.text_embed = TextEmbedding(text_num_embeds, text_dim, conv_layers=conv_layers)\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n # transformer layers & skip connections\n\n self.dim = dim\n self.skip_connect_type = skip_connect_type\n needs_skip_proj = skip_connect_type == \"concat\"\n\n self.depth = depth\n self.layers = nn.ModuleList([])\n\n for idx in range(depth):\n is_later_half = idx >= (depth // 2)\n\n attn_norm = RMSNorm(dim)\n attn = Attention(\n processor=AttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n )\n\n ff_norm = RMSNorm(dim)\n ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n\n skip_proj = nn.Linear(dim * 2, dim, bias=False) if needs_skip_proj and is_later_half else None\n\n self.layers.append(\n nn.ModuleList(\n [\n skip_proj,\n attn_norm,\n attn,\n ff_norm,\n ff,\n ]\n )\n )\n\n self.norm_out = RMSNorm(dim)\n self.proj_out = nn.Linear(dim, mel_dim)\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n drop_audio_cond, # cfg for cond audio\n drop_text, # cfg for text\n mask: bool[\"b n\"] | None = None, # noqa: F722\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n # postfix time t to input x, [b n d] -> [b n+1 d]\n x = torch.cat([t.unsqueeze(1), x], dim=1) # pack t to x\n if mask is not None:\n mask = F.pad(mask, (1, 0), value=1)\n\n rope = self.rotary_embed.forward_from_seq_len(seq_len + 1)\n\n # flat unet transformer\n skip_connect_type = self.skip_connect_type\n skips = []\n for idx, (maybe_skip_proj, attn_norm, attn, ff_norm, ff) in enumerate(self.layers):\n layer = idx + 1\n\n # skip connection logic\n is_first_half = layer <= (self.depth // 2)\n is_later_half = not is_first_half\n\n if is_first_half:\n skips.append(x)\n\n if is_later_half:\n skip = skips.pop()\n if skip_connect_type == \"concat\":\n x = torch.cat((x, skip), dim=-1)\n x = maybe_skip_proj(x)\n elif skip_connect_type == \"add\":\n x = x + skip\n\n # attention and feedforward blocks\n x = attn(attn_norm(x), rope=rope, mask=mask) + x\n x = ff(ff_norm(x)) + x\n\n assert len(skips) == 0\n\n x = self.norm_out(x)[:, 1:, :] # unpack t from x\n\n return self.proj_out(x)","source_hash":"852e63fcd735376abb5572df733fb3d7ce64af8d3377c4930a0a4d02090be47a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.unett.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model.backbones.unett.__init__#L97-L162","kind":"function","name":"__init__","path":"src/f5_tts/model/backbones/unett.py","language":"python","start_line":97,"end_line":162,"context_start_line":77,"context_end_line":182,"code":"\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Flat UNet Transformer backbone\n\n\nclass UNetT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n conv_layers=0,\n skip_connect_type: Literal[\"add\", \"concat\", \"none\"] = \"concat\",\n ):\n super().__init__()\n assert depth % 2 == 0, \"UNet-Transformer's depth should be even.\"\n\n self.time_embed = TimestepEmbedding(dim)\n if text_dim is None:\n text_dim = mel_dim\n self.text_embed = TextEmbedding(text_num_embeds, text_dim, conv_layers=conv_layers)\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n # transformer layers & skip connections\n\n self.dim = dim\n self.skip_connect_type = skip_connect_type\n needs_skip_proj = skip_connect_type == \"concat\"\n\n self.depth = depth\n self.layers = nn.ModuleList([])\n\n for idx in range(depth):\n is_later_half = idx >= (depth // 2)\n\n attn_norm = RMSNorm(dim)\n attn = Attention(\n processor=AttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n )\n\n ff_norm = RMSNorm(dim)\n ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n\n skip_proj = nn.Linear(dim * 2, dim, bias=False) if needs_skip_proj and is_later_half else None\n\n self.layers.append(\n nn.ModuleList(\n [\n skip_proj,\n attn_norm,\n attn,\n ff_norm,\n ff,\n ]\n )\n )\n\n self.norm_out = RMSNorm(dim)\n self.proj_out = nn.Linear(dim, mel_dim)\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n drop_audio_cond, # cfg for cond audio\n drop_text, # cfg for text\n mask: bool[\"b n\"] | None = None, # noqa: F722\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n","source_hash":"852e63fcd735376abb5572df733fb3d7ce64af8d3377c4930a0a4d02090be47a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model.backbones.unett.forward","uri":"program://DMOSpeech2/function/src.f5_tts.model.backbones.unett.forward#L164-L219","kind":"function","name":"forward","path":"src/f5_tts/model/backbones/unett.py","language":"python","start_line":164,"end_line":219,"context_start_line":144,"context_end_line":219,"code":" ff_norm = RMSNorm(dim)\n ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n\n skip_proj = nn.Linear(dim * 2, dim, bias=False) if needs_skip_proj and is_later_half else None\n\n self.layers.append(\n nn.ModuleList(\n [\n skip_proj,\n attn_norm,\n attn,\n ff_norm,\n ff,\n ]\n )\n )\n\n self.norm_out = RMSNorm(dim)\n self.proj_out = nn.Linear(dim, mel_dim)\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n drop_audio_cond, # cfg for cond audio\n drop_text, # cfg for text\n mask: bool[\"b n\"] | None = None, # noqa: F722\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n # postfix time t to input x, [b n d] -> [b n+1 d]\n x = torch.cat([t.unsqueeze(1), x], dim=1) # pack t to x\n if mask is not None:\n mask = F.pad(mask, (1, 0), value=1)\n\n rope = self.rotary_embed.forward_from_seq_len(seq_len + 1)\n\n # flat unet transformer\n skip_connect_type = self.skip_connect_type\n skips = []\n for idx, (maybe_skip_proj, attn_norm, attn, ff_norm, ff) in enumerate(self.layers):\n layer = idx + 1\n\n # skip connection logic\n is_first_half = layer <= (self.depth // 2)\n is_later_half = not is_first_half\n\n if is_first_half:\n skips.append(x)\n\n if is_later_half:\n skip = skips.pop()\n if skip_connect_type == \"concat\":\n x = torch.cat((x, skip), dim=-1)\n x = maybe_skip_proj(x)\n elif skip_connect_type == \"add\":\n x = x + skip\n\n # attention and feedforward blocks\n x = attn(attn_norm(x), rope=rope, mask=mask) + x\n x = ff(ff_norm(x)) + x\n\n assert len(skips) == 0\n\n x = self.norm_out(x)[:, 1:, :] # unpack t from x\n\n return self.proj_out(x)","source_hash":"852e63fcd735376abb5572df733fb3d7ce64af8d3377c4930a0a4d02090be47a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.trainer","uri":"program://DMOSpeech2/module/src.f5_tts.model_new.trainer#L1-L439","kind":"module","name":"src.f5_tts.model_new.trainer","path":"src/f5_tts/model_new/trainer.py","language":"python","start_line":1,"end_line":439,"context_start_line":1,"context_end_line":439,"code":"from __future__ import annotations\n\nimport gc\nimport math\nimport os\n\nimport torch\nimport torchaudio\nimport wandb\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nfrom ema_pytorch import EMA\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model import CFM\nfrom f5_tts.model.dataset import DynamicBatchSampler, collate_fn\nfrom f5_tts.model.utils import default, exists\n\n\n# trainer\n\n\nclass Trainer:\n def __init__(\n self,\n model: CFM,\n epochs,\n learning_rate,\n num_warmup_updates=20000,\n save_per_updates=1000,\n keep_last_n_checkpoints: int = -1, # -1 to keep all, 0 to not save intermediate, > 0 to keep last N checkpoints\n checkpoint_path=None,\n batch_size_per_gpu=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n noise_scheduler: str | None = None,\n duration_predictor: torch.nn.Module | None = None,\n logger: str | None = \"wandb\", # \"wandb\" | \"tensorboard\" | None\n wandb_project=\"test_f5-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n log_samples: bool = False,\n last_per_updates=None,\n accelerate_kwargs: dict = dict(),\n ema_kwargs: dict = dict(),\n bnb_optimizer: bool = False,\n mel_spec_type: str = \"vocos\", # \"vocos\" | \"bigvgan\"\n is_local_vocoder: bool = False, # use local path vocoder\n local_vocoder_path: str = \"\", # local vocoder path\n model_cfg_dict: dict = dict(), # training config\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=True)\n\n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n self.log_samples = log_samples\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n if not model_cfg_dict:\n model_cfg_dict = {\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size_per_gpu\": batch_size_per_gpu,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"noise_scheduler\": noise_scheduler,\n }\n model_cfg_dict[\"gpus\"] = self.accelerator.num_processes\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config=model_cfg_dict,\n )\n\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n self.model = model\n\n if self.is_main:\n self.ema_model = EMA(model, include_online_model=False, **ema_kwargs)\n self.ema_model.to(self.accelerator.device)\n\n print(f\"Using logger: {logger}\")\n if grad_accumulation_steps > 1:\n print(\n \"Gradient accumulation checkpointing with per_updates now, old logic per_steps used with before f992c4e\"\n )\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.keep_last_n_checkpoints = keep_last_n_checkpoints\n self.last_per_updates = default(last_per_updates, save_per_updates)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_f5-tts\")\n\n self.batch_size_per_gpu = batch_size_per_gpu\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n # mel vocoder config\n self.vocoder_name = mel_spec_type\n self.is_local_vocoder = is_local_vocoder\n self.local_vocoder_path = local_vocoder_path\n\n self.noise_scheduler = noise_scheduler\n\n self.duration_predictor = duration_predictor\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n\n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, update, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.optimizer.state_dict(),\n ema_model_state_dict=self.ema_model.state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n update=update,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at update {update}\")\n else:\n if self.keep_last_n_checkpoints == 0:\n return\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{update}.pt\")\n if self.keep_last_n_checkpoints > 0:\n # Updated logic to exclude pretrained model from rotation\n checkpoints = [\n f\n for f in os.listdir(self.checkpoint_path)\n if f.startswith(\"model_\")\n and not f.startswith(\"pretrained_\") # Exclude pretrained models\n and f.endswith(\".pt\")\n and f != \"model_last.pt\"\n ]\n checkpoints.sort(key=lambda x: int(x.split(\"_\")[1].split(\".\")[0]))\n while len(checkpoints) > self.keep_last_n_checkpoints:\n oldest_checkpoint = checkpoints.pop(0)\n os.remove(os.path.join(self.checkpoint_path, oldest_checkpoint))\n print(f\"Removed old checkpoint: {oldest_checkpoint}\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith((\".pt\", \".safetensors\")) for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n # Updated to consider pretrained models for loading but prioritize training checkpoints\n all_checkpoints = [\n f\n for f in os.listdir(self.checkpoint_path)\n if (f.startswith(\"model_\") or f.startswith(\"pretrained_\")) and f.endswith((\".pt\", \".safetensors\"))\n ]\n\n # First try to find regular training checkpoints\n training_checkpoints = [f for f in all_checkpoints if f.startswith(\"model_\") and f != \"model_last.pt\"]\n if training_checkpoints:\n latest_checkpoint = sorted(\n training_checkpoints,\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n else:\n # If no training checkpoints, use pretrained model\n latest_checkpoint = next(f for f in all_checkpoints if f.startswith(\"pretrained_\"))\n\n if latest_checkpoint.endswith(\".safetensors\"): # always a pretrained checkpoint\n from safetensors.torch import load_file\n\n checkpoint = load_file(f\"{self.checkpoint_path}/{latest_checkpoint}\", device=\"cpu\")\n checkpoint = {\"ema_model_state_dict\": checkpoint}\n elif latest_checkpoint.endswith(\".pt\"):\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(\n f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\"\n )\n\n # patch for backward compatibility, 305e3ea\n for key in [\"ema_model.mel_spec.mel_stft.mel_scale.fb\", \"ema_model.mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"ema_model_state_dict\"]:\n del checkpoint[\"ema_model_state_dict\"][key]\n\n if self.is_main:\n self.ema_model.load_state_dict(checkpoint[\"ema_model_state_dict\"])\n\n if \"update\" in checkpoint or \"step\" in checkpoint:\n # patch for backward compatibility, with before f992c4e\n if \"step\" in checkpoint:\n checkpoint[\"update\"] = checkpoint[\"step\"] // self.grad_accumulation_steps\n if self.grad_accumulation_steps > 1 and self.is_main:\n print(\n \"F5-TTS WARNING: Loading checkpoint saved with per_steps logic (before f992c4e), will convert to per_updates according to grad_accumulation_steps setting, may have unexpected behaviour.\"\n )\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.optimizer.load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n update = checkpoint[\"update\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"update\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n update = 0\n\n del checkpoint\n gc.collect()\n return update\n\n def train(self, train_dataset: Dataset, num_workers=16, resumable_with_seed: int = None):\n if self.log_samples:\n from f5_tts.infer.utils_infer import cfg_strength, load_vocoder, nfe_step, sway_sampling_coef\n\n vocoder = load_vocoder(\n vocoder_name=self.vocoder_name, is_local=self.is_local_vocoder, local_path=self.local_vocoder_path\n )\n target_sample_rate = self.accelerator.unwrap_model(self.model).mel_spec.target_sample_rate\n log_samples_path = f\"{self.checkpoint_path}/samples\"\n os.makedirs(log_samples_path, exist_ok=True)\n\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size_per_gpu,\n shuffle=True,\n generator=generator,\n )\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler,\n self.batch_size_per_gpu,\n max_samples=self.max_samples,\n random_seed=resumable_with_seed, # This enables reproducible shuffling\n drop_residual=False,\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n\n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_updates = (\n self.num_warmup_updates * self.accelerator.num_processes\n ) # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_updates = math.ceil(len(train_dataloader) / self.grad_accumulation_steps) * self.epochs\n decay_updates = total_updates - warmup_updates\n warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_updates)\n decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_updates)\n self.scheduler = SequentialLR(\n self.optimizer, schedulers=[warmup_scheduler, decay_scheduler], milestones=[warmup_updates]\n )\n train_dataloader, self.scheduler = self.accelerator.prepare(\n train_dataloader, self.scheduler\n ) # actual multi_gpu updates = single_gpu updates / gpu nums\n start_update = self.load_checkpoint()\n global_update = start_update\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n start_step = start_update * self.grad_accumulation_steps\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar_initial = math.ceil(skipped_batch / self.grad_accumulation_steps)\n current_dataloader = skipped_dataloader\n else:\n progress_bar_initial = 0\n current_dataloader = train_dataloader\n\n # Set epoch for the batch sampler if it exists\n if hasattr(train_dataloader, \"batch_sampler\") and hasattr(train_dataloader.batch_sampler, \"set_epoch\"):\n train_dataloader.batch_sampler.set_epoch(epoch)\n\n progress_bar = tqdm(\n range(math.ceil(len(train_dataloader) / self.grad_accumulation_steps)),\n desc=f\"Epoch {epoch + 1}/{self.epochs}\",\n unit=\"update\",\n disable=not self.accelerator.is_local_main_process,\n initial=progress_bar_initial,\n )\n\n for batch in current_dataloader:\n with self.accelerator.accumulate(self.model):\n text_inputs = batch[\"text\"]\n mel_spec = batch[\"mel\"].permute(0, 2, 1)\n mel_lengths = batch[\"mel_lengths\"]\n\n # TODO. add duration predictor training\n if self.duration_predictor is not None and self.accelerator.is_local_main_process:\n dur_loss = self.duration_predictor(mel_spec, lens=batch.get(\"durations\"))\n self.accelerator.log({\"duration loss\": dur_loss.item()}, step=global_update)\n\n loss, cond, pred = self.model(\n mel_spec, text=text_inputs, lens=mel_lengths, noise_scheduler=self.noise_scheduler\n )\n self.accelerator.backward(loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n self.optimizer.step()\n self.scheduler.step()\n self.optimizer.zero_grad()\n\n if self.accelerator.sync_gradients:\n if self.is_main:\n self.ema_model.update()\n\n global_update += 1\n progress_bar.update(1)\n progress_bar.set_postfix(update=str(global_update), loss=loss.item())\n\n if self.accelerator.is_local_main_process:\n self.accelerator.log(\n {\"loss\": loss.item(), \"lr\": self.scheduler.get_last_lr()[0]}, step=global_update\n )\n if self.logger == \"tensorboard\":\n self.writer.add_scalar(\"loss\", loss.item(), global_update)\n self.writer.add_scalar(\"lr\", self.scheduler.get_last_lr()[0], global_update)\n\n if global_update % self.last_per_updates == 0 and self.accelerator.sync_gradients:\n self.save_checkpoint(global_update, last=True)\n\n if global_update % self.save_per_updates == 0 and self.accelerator.sync_gradients:\n self.save_checkpoint(global_update)\n\n if self.log_samples and self.accelerator.is_local_main_process:\n ref_audio_len = mel_lengths[0]\n infer_text = [\n text_inputs[0] + ([\" \"] if isinstance(text_inputs[0], list) else \" \") + text_inputs[0]\n ]\n with torch.inference_mode():\n generated, _ = self.accelerator.unwrap_model(self.model).sample(\n cond=mel_spec[0][:ref_audio_len].unsqueeze(0),\n text=infer_text,\n duration=ref_audio_len * 2,\n steps=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n )\n generated = generated.to(torch.float32)\n gen_mel_spec = generated[:, ref_audio_len:, :].permute(0, 2, 1).to(self.accelerator.device)\n ref_mel_spec = batch[\"mel\"][0].unsqueeze(0)\n if self.vocoder_name == \"vocos\":\n gen_audio = vocoder.decode(gen_mel_spec).cpu()\n ref_audio = vocoder.decode(ref_mel_spec).cpu()\n elif self.vocoder_name == \"bigvgan\":\n gen_audio = vocoder(gen_mel_spec).squeeze(0).cpu()\n ref_audio = vocoder(ref_mel_spec).squeeze(0).cpu()\n\n torchaudio.save(\n f\"{log_samples_path}/update_{global_update}_gen.wav\", gen_audio,\n# ... truncated ...","source_hash":"f920112a2433c8a5c92100766e845ef7990d488705ac80de568096e6900fb11c","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.trainer.Trainer","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.trainer.Trainer#L26-L439","kind":"class","name":"Trainer","path":"src/f5_tts/model_new/trainer.py","language":"python","start_line":26,"end_line":439,"context_start_line":6,"context_end_line":439,"code":"\nimport torch\nimport torchaudio\nimport wandb\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nfrom ema_pytorch import EMA\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model import CFM\nfrom f5_tts.model.dataset import DynamicBatchSampler, collate_fn\nfrom f5_tts.model.utils import default, exists\n\n\n# trainer\n\n\nclass Trainer:\n def __init__(\n self,\n model: CFM,\n epochs,\n learning_rate,\n num_warmup_updates=20000,\n save_per_updates=1000,\n keep_last_n_checkpoints: int = -1, # -1 to keep all, 0 to not save intermediate, > 0 to keep last N checkpoints\n checkpoint_path=None,\n batch_size_per_gpu=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n noise_scheduler: str | None = None,\n duration_predictor: torch.nn.Module | None = None,\n logger: str | None = \"wandb\", # \"wandb\" | \"tensorboard\" | None\n wandb_project=\"test_f5-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n log_samples: bool = False,\n last_per_updates=None,\n accelerate_kwargs: dict = dict(),\n ema_kwargs: dict = dict(),\n bnb_optimizer: bool = False,\n mel_spec_type: str = \"vocos\", # \"vocos\" | \"bigvgan\"\n is_local_vocoder: bool = False, # use local path vocoder\n local_vocoder_path: str = \"\", # local vocoder path\n model_cfg_dict: dict = dict(), # training config\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=True)\n\n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n self.log_samples = log_samples\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n if not model_cfg_dict:\n model_cfg_dict = {\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size_per_gpu\": batch_size_per_gpu,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"noise_scheduler\": noise_scheduler,\n }\n model_cfg_dict[\"gpus\"] = self.accelerator.num_processes\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config=model_cfg_dict,\n )\n\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n self.model = model\n\n if self.is_main:\n self.ema_model = EMA(model, include_online_model=False, **ema_kwargs)\n self.ema_model.to(self.accelerator.device)\n\n print(f\"Using logger: {logger}\")\n if grad_accumulation_steps > 1:\n print(\n \"Gradient accumulation checkpointing with per_updates now, old logic per_steps used with before f992c4e\"\n )\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.keep_last_n_checkpoints = keep_last_n_checkpoints\n self.last_per_updates = default(last_per_updates, save_per_updates)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_f5-tts\")\n\n self.batch_size_per_gpu = batch_size_per_gpu\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n # mel vocoder config\n self.vocoder_name = mel_spec_type\n self.is_local_vocoder = is_local_vocoder\n self.local_vocoder_path = local_vocoder_path\n\n self.noise_scheduler = noise_scheduler\n\n self.duration_predictor = duration_predictor\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n\n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, update, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.optimizer.state_dict(),\n ema_model_state_dict=self.ema_model.state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n update=update,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at update {update}\")\n else:\n if self.keep_last_n_checkpoints == 0:\n return\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{update}.pt\")\n if self.keep_last_n_checkpoints > 0:\n # Updated logic to exclude pretrained model from rotation\n checkpoints = [\n f\n for f in os.listdir(self.checkpoint_path)\n if f.startswith(\"model_\")\n and not f.startswith(\"pretrained_\") # Exclude pretrained models\n and f.endswith(\".pt\")\n and f != \"model_last.pt\"\n ]\n checkpoints.sort(key=lambda x: int(x.split(\"_\")[1].split(\".\")[0]))\n while len(checkpoints) > self.keep_last_n_checkpoints:\n oldest_checkpoint = checkpoints.pop(0)\n os.remove(os.path.join(self.checkpoint_path, oldest_checkpoint))\n print(f\"Removed old checkpoint: {oldest_checkpoint}\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith((\".pt\", \".safetensors\")) for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n # Updated to consider pretrained models for loading but prioritize training checkpoints\n all_checkpoints = [\n f\n for f in os.listdir(self.checkpoint_path)\n if (f.startswith(\"model_\") or f.startswith(\"pretrained_\")) and f.endswith((\".pt\", \".safetensors\"))\n ]\n\n # First try to find regular training checkpoints\n training_checkpoints = [f for f in all_checkpoints if f.startswith(\"model_\") and f != \"model_last.pt\"]\n if training_checkpoints:\n latest_checkpoint = sorted(\n training_checkpoints,\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n else:\n # If no training checkpoints, use pretrained model\n latest_checkpoint = next(f for f in all_checkpoints if f.startswith(\"pretrained_\"))\n\n if latest_checkpoint.endswith(\".safetensors\"): # always a pretrained checkpoint\n from safetensors.torch import load_file\n\n checkpoint = load_file(f\"{self.checkpoint_path}/{latest_checkpoint}\", device=\"cpu\")\n checkpoint = {\"ema_model_state_dict\": checkpoint}\n elif latest_checkpoint.endswith(\".pt\"):\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(\n f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\"\n )\n\n # patch for backward compatibility, 305e3ea\n for key in [\"ema_model.mel_spec.mel_stft.mel_scale.fb\", \"ema_model.mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"ema_model_state_dict\"]:\n del checkpoint[\"ema_model_state_dict\"][key]\n\n if self.is_main:\n self.ema_model.load_state_dict(checkpoint[\"ema_model_state_dict\"])\n\n if \"update\" in checkpoint or \"step\" in checkpoint:\n # patch for backward compatibility, with before f992c4e\n if \"step\" in checkpoint:\n checkpoint[\"update\"] = checkpoint[\"step\"] // self.grad_accumulation_steps\n if self.grad_accumulation_steps > 1 and self.is_main:\n print(\n \"F5-TTS WARNING: Loading checkpoint saved with per_steps logic (before f992c4e), will convert to per_updates according to grad_accumulation_steps setting, may have unexpected behaviour.\"\n )\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.optimizer.load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n update = checkpoint[\"update\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"update\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n update = 0\n\n del checkpoint\n gc.collect()\n return update\n\n def train(self, train_dataset: Dataset, num_workers=16, resumable_with_seed: int = None):\n if self.log_samples:\n from f5_tts.infer.utils_infer import cfg_strength, load_vocoder, nfe_step, sway_sampling_coef\n\n vocoder = load_vocoder(\n vocoder_name=self.vocoder_name, is_local=self.is_local_vocoder, local_path=self.local_vocoder_path\n )\n target_sample_rate = self.accelerator.unwrap_model(self.model).mel_spec.target_sample_rate\n log_samples_path = f\"{self.checkpoint_path}/samples\"\n os.makedirs(log_samples_path, exist_ok=True)\n\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size_per_gpu,\n shuffle=True,\n generator=generator,\n )\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler,\n self.batch_size_per_gpu,\n max_samples=self.max_samples,\n random_seed=resumable_with_seed, # This enables reproducible shuffling\n drop_residual=False,\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n\n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_updates = (\n self.num_warmup_updates * self.accelerator.num_processes\n ) # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_updates = math.ceil(len(train_dataloader) / self.grad_accumulation_steps) * self.epochs\n decay_updates = total_updates - warmup_updates\n warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_updates)\n decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_updates)\n self.scheduler = SequentialLR(\n self.optimizer, schedulers=[warmup_scheduler, decay_scheduler], milestones=[warmup_updates]\n )\n train_dataloader, self.scheduler = self.accelerator.prepare(\n train_dataloader, self.scheduler\n ) # actual multi_gpu updates = single_gpu updates / gpu nums\n start_update = self.load_checkpoint()\n global_update = start_update\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n start_step = start_update * self.grad_accumulation_steps\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar_initial = math.ceil(skipped_batch / self.grad_accumulation_steps)\n current_dataloader = skipped_dataloader\n else:\n progress_bar_initial = 0\n current_dataloader = train_dataloader\n\n # Set epoch for the batch sampler if it exists\n if hasattr(train_dataloader, \"batch_sampler\") and hasattr(train_dataloader.batch_sampler, \"set_epoch\"):\n train_dataloader.batch_sampler.set_epoch(epoch)\n\n progress_bar = tqdm(\n range(math.ceil(len(train_dataloader) / self.grad_accumulation_steps)),\n desc=f\"Epoch {epoch + 1}/{self.epochs}\",\n unit=\"update\",\n disable=not self.accelerator.is_local_main_process,\n initial=progress_bar_initial,\n )\n\n for batch in current_dataloader:\n with self.accelerator.accumulate(self.model):\n text_inputs = batch[\"text\"]\n mel_spec = batch[\"mel\"].permute(0, 2, 1)\n mel_lengths = batch[\"mel_lengths\"]\n\n # TODO. add duration predictor training\n if self.duration_predictor is not None and self.accelerator.is_local_main_process:\n dur_loss = self.duration_predictor(mel_spec, lens=batch.get(\"durations\"))\n self.accelerator.log({\"duration loss\": dur_loss.item()}, step=global_update)\n\n loss, cond, pred = self.model(\n mel_spec, text=text_inputs, lens=mel_lengths, noise_scheduler=self.noise_scheduler\n )\n self.accelerator.backward(loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n self.optimizer.step()\n self.scheduler.step()\n self.optimizer.zero_grad()\n\n if self.accelerator.sync_gradients:\n if self.is_main:\n self.ema_model.update()\n\n global_update += 1\n progress_bar.update(1)\n progress_bar.set_postfix(update=str(global_update), loss=loss.item())\n\n if self.accelerator.is_local_main_process:\n self.accelerator.log(\n {\"loss\": loss.item(), \"lr\": self.scheduler.get_last_lr()[0]}, step=global_update\n )\n if self.logger == \"tensorboard\":\n self.writer.add_scalar(\"loss\", loss.item(), global_update)\n self.writer.add_scalar(\"lr\", self.scheduler.get_last_lr()[0], global_update)\n\n if global_update % self.last_per_updates == 0 and self.accelerator.sync_gradients:\n self.save_checkpoint(global_update, last=True)\n\n if global_update % self.save_per_updates == 0 and self.accelerator.sync_gradients:\n self.save_checkpoint(global_update)\n\n if self.log_samples and self.accelerator.is_local_main_process:\n ref_audio_len = mel_lengths[0]\n infer_text = [\n text_inputs[0] + ([\" \"] if isinstance(text_inputs[0], list) else \" \") + text_inputs[0]\n ]\n with torch.inference_mode():\n generated, _ = self.accelerator.unwrap_model(self.model).sample(\n cond=mel_spec[0][:ref_audio_len].unsqueeze(0),\n text=infer_text,\n duration=ref_audio_len * 2,\n steps=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n )\n generated = generated.to(torch.float32)\n gen_mel_spec = generated[:, ref_audio_len:, :].permute(0, 2, 1).to(self.accelerator.device)\n ref_mel_spec = batch[\"mel\"][0].unsqueeze(0)\n if self.vocoder_name == \"vocos\":\n gen_audio = vocoder.decode(gen_mel_spec).cpu()\n ref_audio = vocoder.decode(ref_mel_spec).cpu()\n elif self.vocoder_name == \"bigvgan\":\n gen_audio = vocoder(gen_mel_spec).squeeze(0).cpu()\n ref_audio = vocoder(ref_mel_spec).squeeze(0).cpu()\n\n torchaudio.save(\n f\"{log_samples_path}/update_{global_update}_gen.wav\", gen_audio, target_sample_rate\n )\n# ... truncated ...","source_hash":"f920112a2433c8a5c92100766e845ef7990d488705ac80de568096e6900fb11c","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.trainer.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.trainer.__init__#L27-L141","kind":"function","name":"__init__","path":"src/f5_tts/model_new/trainer.py","language":"python","start_line":27,"end_line":141,"context_start_line":7,"context_end_line":161,"code":"import torch\nimport torchaudio\nimport wandb\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nfrom ema_pytorch import EMA\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model import CFM\nfrom f5_tts.model.dataset import DynamicBatchSampler, collate_fn\nfrom f5_tts.model.utils import default, exists\n\n\n# trainer\n\n\nclass Trainer:\n def __init__(\n self,\n model: CFM,\n epochs,\n learning_rate,\n num_warmup_updates=20000,\n save_per_updates=1000,\n keep_last_n_checkpoints: int = -1, # -1 to keep all, 0 to not save intermediate, > 0 to keep last N checkpoints\n checkpoint_path=None,\n batch_size_per_gpu=32,\n batch_size_type: str = \"sample\",\n max_samples=32,\n grad_accumulation_steps=1,\n max_grad_norm=1.0,\n noise_scheduler: str | None = None,\n duration_predictor: torch.nn.Module | None = None,\n logger: str | None = \"wandb\", # \"wandb\" | \"tensorboard\" | None\n wandb_project=\"test_f5-tts\",\n wandb_run_name=\"test_run\",\n wandb_resume_id: str = None,\n log_samples: bool = False,\n last_per_updates=None,\n accelerate_kwargs: dict = dict(),\n ema_kwargs: dict = dict(),\n bnb_optimizer: bool = False,\n mel_spec_type: str = \"vocos\", # \"vocos\" | \"bigvgan\"\n is_local_vocoder: bool = False, # use local path vocoder\n local_vocoder_path: str = \"\", # local vocoder path\n model_cfg_dict: dict = dict(), # training config\n ):\n ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=True)\n\n if logger == \"wandb\" and not wandb.api.api_key:\n logger = None\n self.log_samples = log_samples\n\n self.accelerator = Accelerator(\n log_with=logger if logger == \"wandb\" else None,\n kwargs_handlers=[ddp_kwargs],\n gradient_accumulation_steps=grad_accumulation_steps,\n **accelerate_kwargs,\n )\n\n self.logger = logger\n if self.logger == \"wandb\":\n if exists(wandb_resume_id):\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name, \"id\": wandb_resume_id}}\n else:\n init_kwargs = {\"wandb\": {\"resume\": \"allow\", \"name\": wandb_run_name}}\n\n if not model_cfg_dict:\n model_cfg_dict = {\n \"epochs\": epochs,\n \"learning_rate\": learning_rate,\n \"num_warmup_updates\": num_warmup_updates,\n \"batch_size_per_gpu\": batch_size_per_gpu,\n \"batch_size_type\": batch_size_type,\n \"max_samples\": max_samples,\n \"grad_accumulation_steps\": grad_accumulation_steps,\n \"max_grad_norm\": max_grad_norm,\n \"noise_scheduler\": noise_scheduler,\n }\n model_cfg_dict[\"gpus\"] = self.accelerator.num_processes\n self.accelerator.init_trackers(\n project_name=wandb_project,\n init_kwargs=init_kwargs,\n config=model_cfg_dict,\n )\n\n elif self.logger == \"tensorboard\":\n from torch.utils.tensorboard import SummaryWriter\n\n self.writer = SummaryWriter(log_dir=f\"runs/{wandb_run_name}\")\n\n self.model = model\n\n if self.is_main:\n self.ema_model = EMA(model, include_online_model=False, **ema_kwargs)\n self.ema_model.to(self.accelerator.device)\n\n print(f\"Using logger: {logger}\")\n if grad_accumulation_steps > 1:\n print(\n \"Gradient accumulation checkpointing with per_updates now, old logic per_steps used with before f992c4e\"\n )\n\n self.epochs = epochs\n self.num_warmup_updates = num_warmup_updates\n self.save_per_updates = save_per_updates\n self.keep_last_n_checkpoints = keep_last_n_checkpoints\n self.last_per_updates = default(last_per_updates, save_per_updates)\n self.checkpoint_path = default(checkpoint_path, \"ckpts/test_f5-tts\")\n\n self.batch_size_per_gpu = batch_size_per_gpu\n self.batch_size_type = batch_size_type\n self.max_samples = max_samples\n self.grad_accumulation_steps = grad_accumulation_steps\n self.max_grad_norm = max_grad_norm\n\n # mel vocoder config\n self.vocoder_name = mel_spec_type\n self.is_local_vocoder = is_local_vocoder\n self.local_vocoder_path = local_vocoder_path\n\n self.noise_scheduler = noise_scheduler\n\n self.duration_predictor = duration_predictor\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n\n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, update, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.optimizer.state_dict(),\n ema_model_state_dict=self.ema_model.state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n update=update,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at update {update}\")","source_hash":"f920112a2433c8a5c92100766e845ef7990d488705ac80de568096e6900fb11c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.trainer.is_main","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.trainer.is_main#L144-L145","kind":"function","name":"is_main","path":"src/f5_tts/model_new/trainer.py","language":"python","start_line":144,"end_line":145,"context_start_line":124,"context_end_line":165,"code":" self.max_grad_norm = max_grad_norm\n\n # mel vocoder config\n self.vocoder_name = mel_spec_type\n self.is_local_vocoder = is_local_vocoder\n self.local_vocoder_path = local_vocoder_path\n\n self.noise_scheduler = noise_scheduler\n\n self.duration_predictor = duration_predictor\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n\n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, update, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.optimizer.state_dict(),\n ema_model_state_dict=self.ema_model.state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n update=update,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at update {update}\")\n else:\n if self.keep_last_n_checkpoints == 0:\n return\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{update}.pt\")","source_hash":"f920112a2433c8a5c92100766e845ef7990d488705ac80de568096e6900fb11c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.trainer.save_checkpoint","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.trainer.save_checkpoint#L147-L180","kind":"function","name":"save_checkpoint","path":"src/f5_tts/model_new/trainer.py","language":"python","start_line":147,"end_line":180,"context_start_line":127,"context_end_line":200,"code":" self.vocoder_name = mel_spec_type\n self.is_local_vocoder = is_local_vocoder\n self.local_vocoder_path = local_vocoder_path\n\n self.noise_scheduler = noise_scheduler\n\n self.duration_predictor = duration_predictor\n\n if bnb_optimizer:\n import bitsandbytes as bnb\n\n self.optimizer = bnb.optim.AdamW8bit(model.parameters(), lr=learning_rate)\n else:\n self.optimizer = AdamW(model.parameters(), lr=learning_rate)\n self.model, self.optimizer = self.accelerator.prepare(self.model, self.optimizer)\n\n @property\n def is_main(self):\n return self.accelerator.is_main_process\n\n def save_checkpoint(self, update, last=False):\n self.accelerator.wait_for_everyone()\n if self.is_main:\n checkpoint = dict(\n model_state_dict=self.accelerator.unwrap_model(self.model).state_dict(),\n optimizer_state_dict=self.optimizer.state_dict(),\n ema_model_state_dict=self.ema_model.state_dict(),\n scheduler_state_dict=self.scheduler.state_dict(),\n update=update,\n )\n if not os.path.exists(self.checkpoint_path):\n os.makedirs(self.checkpoint_path)\n if last:\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_last.pt\")\n print(f\"Saved last checkpoint at update {update}\")\n else:\n if self.keep_last_n_checkpoints == 0:\n return\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{update}.pt\")\n if self.keep_last_n_checkpoints > 0:\n # Updated logic to exclude pretrained model from rotation\n checkpoints = [\n f\n for f in os.listdir(self.checkpoint_path)\n if f.startswith(\"model_\")\n and not f.startswith(\"pretrained_\") # Exclude pretrained models\n and f.endswith(\".pt\")\n and f != \"model_last.pt\"\n ]\n checkpoints.sort(key=lambda x: int(x.split(\"_\")[1].split(\".\")[0]))\n while len(checkpoints) > self.keep_last_n_checkpoints:\n oldest_checkpoint = checkpoints.pop(0)\n os.remove(os.path.join(self.checkpoint_path, oldest_checkpoint))\n print(f\"Removed old checkpoint: {oldest_checkpoint}\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith((\".pt\", \".safetensors\")) for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n # Updated to consider pretrained models for loading but prioritize training checkpoints\n all_checkpoints = [\n f\n for f in os.listdir(self.checkpoint_path)\n if (f.startswith(\"model_\") or f.startswith(\"pretrained_\")) and f.endswith((\".pt\", \".safetensors\"))\n ]\n","source_hash":"f920112a2433c8a5c92100766e845ef7990d488705ac80de568096e6900fb11c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.trainer.load_checkpoint","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.trainer.load_checkpoint#L182-L260","kind":"function","name":"load_checkpoint","path":"src/f5_tts/model_new/trainer.py","language":"python","start_line":182,"end_line":260,"context_start_line":162,"context_end_line":280,"code":" else:\n if self.keep_last_n_checkpoints == 0:\n return\n self.accelerator.save(checkpoint, f\"{self.checkpoint_path}/model_{update}.pt\")\n if self.keep_last_n_checkpoints > 0:\n # Updated logic to exclude pretrained model from rotation\n checkpoints = [\n f\n for f in os.listdir(self.checkpoint_path)\n if f.startswith(\"model_\")\n and not f.startswith(\"pretrained_\") # Exclude pretrained models\n and f.endswith(\".pt\")\n and f != \"model_last.pt\"\n ]\n checkpoints.sort(key=lambda x: int(x.split(\"_\")[1].split(\".\")[0]))\n while len(checkpoints) > self.keep_last_n_checkpoints:\n oldest_checkpoint = checkpoints.pop(0)\n os.remove(os.path.join(self.checkpoint_path, oldest_checkpoint))\n print(f\"Removed old checkpoint: {oldest_checkpoint}\")\n\n def load_checkpoint(self):\n if (\n not exists(self.checkpoint_path)\n or not os.path.exists(self.checkpoint_path)\n or not any(filename.endswith((\".pt\", \".safetensors\")) for filename in os.listdir(self.checkpoint_path))\n ):\n return 0\n\n self.accelerator.wait_for_everyone()\n if \"model_last.pt\" in os.listdir(self.checkpoint_path):\n latest_checkpoint = \"model_last.pt\"\n else:\n # Updated to consider pretrained models for loading but prioritize training checkpoints\n all_checkpoints = [\n f\n for f in os.listdir(self.checkpoint_path)\n if (f.startswith(\"model_\") or f.startswith(\"pretrained_\")) and f.endswith((\".pt\", \".safetensors\"))\n ]\n\n # First try to find regular training checkpoints\n training_checkpoints = [f for f in all_checkpoints if f.startswith(\"model_\") and f != \"model_last.pt\"]\n if training_checkpoints:\n latest_checkpoint = sorted(\n training_checkpoints,\n key=lambda x: int(\"\".join(filter(str.isdigit, x))),\n )[-1]\n else:\n # If no training checkpoints, use pretrained model\n latest_checkpoint = next(f for f in all_checkpoints if f.startswith(\"pretrained_\"))\n\n if latest_checkpoint.endswith(\".safetensors\"): # always a pretrained checkpoint\n from safetensors.torch import load_file\n\n checkpoint = load_file(f\"{self.checkpoint_path}/{latest_checkpoint}\", device=\"cpu\")\n checkpoint = {\"ema_model_state_dict\": checkpoint}\n elif latest_checkpoint.endswith(\".pt\"):\n # checkpoint = torch.load(f\"{self.checkpoint_path}/{latest_checkpoint}\", map_location=self.accelerator.device) # rather use accelerator.load_state ಥ_ಥ\n checkpoint = torch.load(\n f\"{self.checkpoint_path}/{latest_checkpoint}\", weights_only=True, map_location=\"cpu\"\n )\n\n # patch for backward compatibility, 305e3ea\n for key in [\"ema_model.mel_spec.mel_stft.mel_scale.fb\", \"ema_model.mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"ema_model_state_dict\"]:\n del checkpoint[\"ema_model_state_dict\"][key]\n\n if self.is_main:\n self.ema_model.load_state_dict(checkpoint[\"ema_model_state_dict\"])\n\n if \"update\" in checkpoint or \"step\" in checkpoint:\n # patch for backward compatibility, with before f992c4e\n if \"step\" in checkpoint:\n checkpoint[\"update\"] = checkpoint[\"step\"] // self.grad_accumulation_steps\n if self.grad_accumulation_steps > 1 and self.is_main:\n print(\n \"F5-TTS WARNING: Loading checkpoint saved with per_steps logic (before f992c4e), will convert to per_updates according to grad_accumulation_steps setting, may have unexpected behaviour.\"\n )\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.optimizer.load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n update = checkpoint[\"update\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"update\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n update = 0\n\n del checkpoint\n gc.collect()\n return update\n\n def train(self, train_dataset: Dataset, num_workers=16, resumable_with_seed: int = None):\n if self.log_samples:\n from f5_tts.infer.utils_infer import cfg_strength, load_vocoder, nfe_step, sway_sampling_coef\n\n vocoder = load_vocoder(\n vocoder_name=self.vocoder_name, is_local=self.is_local_vocoder, local_path=self.local_vocoder_path\n )\n target_sample_rate = self.accelerator.unwrap_model(self.model).mel_spec.target_sample_rate\n log_samples_path = f\"{self.checkpoint_path}/samples\"\n os.makedirs(log_samples_path, exist_ok=True)\n\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(","source_hash":"f920112a2433c8a5c92100766e845ef7990d488705ac80de568096e6900fb11c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.trainer.train","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.trainer.train#L262-L439","kind":"function","name":"train","path":"src/f5_tts/model_new/trainer.py","language":"python","start_line":262,"end_line":439,"context_start_line":242,"context_end_line":439,"code":" del checkpoint[\"model_state_dict\"][key]\n\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n self.optimizer.load_state_dict(checkpoint[\"optimizer_state_dict\"])\n if self.scheduler:\n self.scheduler.load_state_dict(checkpoint[\"scheduler_state_dict\"])\n update = checkpoint[\"update\"]\n else:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"update\", \"step\"]\n }\n self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint[\"model_state_dict\"])\n update = 0\n\n del checkpoint\n gc.collect()\n return update\n\n def train(self, train_dataset: Dataset, num_workers=16, resumable_with_seed: int = None):\n if self.log_samples:\n from f5_tts.infer.utils_infer import cfg_strength, load_vocoder, nfe_step, sway_sampling_coef\n\n vocoder = load_vocoder(\n vocoder_name=self.vocoder_name, is_local=self.is_local_vocoder, local_path=self.local_vocoder_path\n )\n target_sample_rate = self.accelerator.unwrap_model(self.model).mel_spec.target_sample_rate\n log_samples_path = f\"{self.checkpoint_path}/samples\"\n os.makedirs(log_samples_path, exist_ok=True)\n\n if exists(resumable_with_seed):\n generator = torch.Generator()\n generator.manual_seed(resumable_with_seed)\n else:\n generator = None\n\n if self.batch_size_type == \"sample\":\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_size=self.batch_size_per_gpu,\n shuffle=True,\n generator=generator,\n )\n elif self.batch_size_type == \"frame\":\n self.accelerator.even_batches = False\n sampler = SequentialSampler(train_dataset)\n batch_sampler = DynamicBatchSampler(\n sampler,\n self.batch_size_per_gpu,\n max_samples=self.max_samples,\n random_seed=resumable_with_seed, # This enables reproducible shuffling\n drop_residual=False,\n )\n train_dataloader = DataLoader(\n train_dataset,\n collate_fn=collate_fn,\n num_workers=num_workers,\n pin_memory=True,\n persistent_workers=True,\n batch_sampler=batch_sampler,\n )\n else:\n raise ValueError(f\"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}\")\n\n # accelerator.prepare() dispatches batches to devices;\n # which means the length of dataloader calculated before, should consider the number of devices\n warmup_updates = (\n self.num_warmup_updates * self.accelerator.num_processes\n ) # consider a fixed warmup steps while using accelerate multi-gpu ddp\n # otherwise by default with split_batches=False, warmup steps change with num_processes\n total_updates = math.ceil(len(train_dataloader) / self.grad_accumulation_steps) * self.epochs\n decay_updates = total_updates - warmup_updates\n warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_updates)\n decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_updates)\n self.scheduler = SequentialLR(\n self.optimizer, schedulers=[warmup_scheduler, decay_scheduler], milestones=[warmup_updates]\n )\n train_dataloader, self.scheduler = self.accelerator.prepare(\n train_dataloader, self.scheduler\n ) # actual multi_gpu updates = single_gpu updates / gpu nums\n start_update = self.load_checkpoint()\n global_update = start_update\n\n if exists(resumable_with_seed):\n orig_epoch_step = len(train_dataloader)\n start_step = start_update * self.grad_accumulation_steps\n skipped_epoch = int(start_step // orig_epoch_step)\n skipped_batch = start_step % orig_epoch_step\n skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)\n else:\n skipped_epoch = 0\n\n for epoch in range(skipped_epoch, self.epochs):\n self.model.train()\n if exists(resumable_with_seed) and epoch == skipped_epoch:\n progress_bar_initial = math.ceil(skipped_batch / self.grad_accumulation_steps)\n current_dataloader = skipped_dataloader\n else:\n progress_bar_initial = 0\n current_dataloader = train_dataloader\n\n # Set epoch for the batch sampler if it exists\n if hasattr(train_dataloader, \"batch_sampler\") and hasattr(train_dataloader.batch_sampler, \"set_epoch\"):\n train_dataloader.batch_sampler.set_epoch(epoch)\n\n progress_bar = tqdm(\n range(math.ceil(len(train_dataloader) / self.grad_accumulation_steps)),\n desc=f\"Epoch {epoch + 1}/{self.epochs}\",\n unit=\"update\",\n disable=not self.accelerator.is_local_main_process,\n initial=progress_bar_initial,\n )\n\n for batch in current_dataloader:\n with self.accelerator.accumulate(self.model):\n text_inputs = batch[\"text\"]\n mel_spec = batch[\"mel\"].permute(0, 2, 1)\n mel_lengths = batch[\"mel_lengths\"]\n\n # TODO. add duration predictor training\n if self.duration_predictor is not None and self.accelerator.is_local_main_process:\n dur_loss = self.duration_predictor(mel_spec, lens=batch.get(\"durations\"))\n self.accelerator.log({\"duration loss\": dur_loss.item()}, step=global_update)\n\n loss, cond, pred = self.model(\n mel_spec, text=text_inputs, lens=mel_lengths, noise_scheduler=self.noise_scheduler\n )\n self.accelerator.backward(loss)\n\n if self.max_grad_norm > 0 and self.accelerator.sync_gradients:\n self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)\n\n self.optimizer.step()\n self.scheduler.step()\n self.optimizer.zero_grad()\n\n if self.accelerator.sync_gradients:\n if self.is_main:\n self.ema_model.update()\n\n global_update += 1\n progress_bar.update(1)\n progress_bar.set_postfix(update=str(global_update), loss=loss.item())\n\n if self.accelerator.is_local_main_process:\n self.accelerator.log(\n {\"loss\": loss.item(), \"lr\": self.scheduler.get_last_lr()[0]}, step=global_update\n )\n if self.logger == \"tensorboard\":\n self.writer.add_scalar(\"loss\", loss.item(), global_update)\n self.writer.add_scalar(\"lr\", self.scheduler.get_last_lr()[0], global_update)\n\n if global_update % self.last_per_updates == 0 and self.accelerator.sync_gradients:\n self.save_checkpoint(global_update, last=True)\n\n if global_update % self.save_per_updates == 0 and self.accelerator.sync_gradients:\n self.save_checkpoint(global_update)\n\n if self.log_samples and self.accelerator.is_local_main_process:\n ref_audio_len = mel_lengths[0]\n infer_text = [\n text_inputs[0] + ([\" \"] if isinstance(text_inputs[0], list) else \" \") + text_inputs[0]\n ]\n with torch.inference_mode():\n generated, _ = self.accelerator.unwrap_model(self.model).sample(\n cond=mel_spec[0][:ref_audio_len].unsqueeze(0),\n text=infer_text,\n duration=ref_audio_len * 2,\n steps=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n )\n generated = generated.to(torch.float32)\n gen_mel_spec = generated[:, ref_audio_len:, :].permute(0, 2, 1).to(self.accelerator.device)\n ref_mel_spec = batch[\"mel\"][0].unsqueeze(0)\n if self.vocoder_name == \"vocos\":\n gen_audio = vocoder.decode(gen_mel_spec).cpu()\n ref_audio = vocoder.decode(ref_mel_spec).cpu()\n elif self.vocoder_name == \"bigvgan\":\n gen_audio = vocoder(gen_mel_spec).squeeze(0).cpu()\n ref_audio = vocoder(ref_mel_spec).squeeze(0).cpu()\n\n torchaudio.save(\n f\"{log_samples_path}/update_{global_update}_gen.wav\", gen_audio, target_sample_rate\n )\n torchaudio.save(\n f\"{log_samples_path}/update_{global_update}_ref.wav\", ref_audio, target_sample_rate\n )\n self.model.train()\n\n self.save_checkpoint(global_update, last=True)\n\n self.accelerator.end_training()","source_hash":"f920112a2433c8a5c92100766e845ef7990d488705ac80de568096e6900fb11c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.cfm","uri":"program://DMOSpeech2/module/src.f5_tts.model_new.cfm#L1-L302","kind":"module","name":"src.f5_tts.model_new.cfm","path":"src/f5_tts/model_new/cfm.py","language":"python","start_line":1,"end_line":302,"context_start_line":1,"context_end_line":302,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nfrom random import random\nfrom typing import Callable\n\nimport torch\nimport torch.nn.functional as F\nfrom torch import nn\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torchdiffeq import odeint\n\nfrom f5_tts.model_new.modules import MelSpec\nfrom f5_tts.model_new.utils import (\n default,\n exists,\n get_epss_timesteps,\n lens_to_mask,\n list_str_to_idx,\n list_str_to_tensor,\n mask_from_frac_lengths,\n)\n\n\nclass CFM(nn.Module):\n def __init__(\n self,\n transformer: nn.Module,\n sigma=0.0,\n odeint_kwargs: dict = dict(\n # atol = 1e-5,\n # rtol = 1e-5,\n method=\"euler\" # 'midpoint'\n ),\n audio_drop_prob=0.3,\n cond_drop_prob=0.2,\n num_channels=None,\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n frac_lengths_mask: tuple[float, float] = (0.7, 1.0),\n vocab_char_map: dict[str:int] | None = None,\n ):\n super().__init__()\n\n self.frac_lengths_mask = frac_lengths_mask\n\n # mel spec\n self.mel_spec = default(mel_spec_module, MelSpec(**mel_spec_kwargs))\n num_channels = default(num_channels, self.mel_spec.n_mel_channels)\n self.num_channels = num_channels\n\n # classifier-free guidance\n self.audio_drop_prob = audio_drop_prob\n self.cond_drop_prob = cond_drop_prob\n\n # transformer\n self.transformer = transformer\n dim = transformer.dim\n self.dim = dim\n\n # conditional flow related\n self.sigma = sigma\n\n # sampling related\n self.odeint_kwargs = odeint_kwargs\n\n # vocab map for tokenization\n self.vocab_char_map = vocab_char_map\n\n @property\n def device(self):\n return next(self.parameters()).device\n\n @torch.no_grad()\n def sample(\n self,\n cond: float[\"b n d\"] | float[\"b nw\"], # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n duration: int | int[\"b\"], # noqa: F821\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n steps=32,\n cfg_strength=1.0,\n sway_sampling_coef=None,\n seed: int | None = None,\n max_duration=4096,\n vocoder: Callable[[float[\"b d n\"]], float[\"b nw\"]] | None = None, # noqa: F722\n use_epss=True,\n no_ref_audio=False,\n duplicate_test=False,\n t_inter=0.1,\n edit_mask=None,\n ):\n self.eval()\n # raw wave\n\n if cond.ndim == 2:\n cond = self.mel_spec(cond)\n cond = cond.permute(0, 2, 1)\n assert cond.shape[-1] == self.num_channels\n\n cond = cond.to(next(self.parameters()).dtype)\n\n batch, cond_seq_len, device = *cond.shape[:2], cond.device\n if not exists(lens):\n lens = torch.full((batch,), cond_seq_len, device=device, dtype=torch.long)\n\n # text\n\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # duration\n\n cond_mask = lens_to_mask(lens)\n if edit_mask is not None:\n cond_mask = cond_mask & edit_mask\n\n if isinstance(duration, int):\n duration = torch.full((batch,), duration, device=device, dtype=torch.long)\n\n duration = torch.maximum(\n torch.maximum((text != -1).sum(dim=-1), lens) + 1, duration\n ) # duration at least text/audio prompt length plus one token, so something is generated\n duration = duration.clamp(max=max_duration)\n max_duration = duration.amax()\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n test_cond = F.pad(cond, (0, 0, cond_seq_len, max_duration - 2 * cond_seq_len), value=0.0)\n\n cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value=0.0)\n if no_ref_audio:\n cond = torch.zeros_like(cond)\n\n cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value=False)\n cond_mask = cond_mask.unsqueeze(-1)\n step_cond = torch.where(\n cond_mask, cond, torch.zeros_like(cond)\n ) # allow direct control (cut cond audio) with lens passed in\n\n if batch > 1:\n mask = lens_to_mask(duration)\n else: # save memory and speed up, as single inference need no mask currently\n mask = None\n\n # neural ode\n\n def fn(t, x):\n # at each step, conditioning is fixed\n # step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))\n\n # predict flow (cond)\n if cfg_strength < 1e-5:\n pred = self.transformer(\n x=x,\n cond=step_cond,\n text=text,\n time=t,\n mask=mask,\n drop_audio_cond=False,\n drop_text=False,\n cache=True,\n )\n return pred\n\n # predict flow (cond and uncond), for classifier-free guidance\n pred_cfg = self.transformer(\n x=x,\n cond=step_cond,\n text=text,\n time=t,\n mask=mask,\n cfg_infer=True,\n cache=True,\n )\n pred, null_pred = torch.chunk(pred_cfg, 2, dim=0)\n return pred + (pred - null_pred) * cfg_strength\n\n # noise input\n # to make sure batch inference result is same with different batch size, and for sure single inference\n # still some difference maybe due to convolutional layers\n y0 = []\n for dur in duration:\n if exists(seed):\n torch.manual_seed(seed)\n y0.append(torch.randn(dur, self.num_channels, device=self.device, dtype=step_cond.dtype))\n y0 = pad_sequence(y0, padding_value=0, batch_first=True)\n\n t_start = 0\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n t_start = t_inter\n y0 = (1 - t_start) * y0 + t_start * test_cond\n steps = int(steps * (1 - t_start))\n\n if t_start == 0 and use_epss: # use Empirically Pruned Step Sampling for low NFE\n t = get_epss_timesteps(steps, device=self.device, dtype=step_cond.dtype)\n else:\n t = torch.linspace(t_start, 1, steps + 1, device=self.device, dtype=step_cond.dtype)\n if sway_sampling_coef is not None:\n t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)\n\n trajectory = odeint(fn, y0, t, **self.odeint_kwargs)\n self.transformer.clear_cache()\n\n sampled = trajectory[-1]\n out = sampled\n out = torch.where(cond_mask, cond, out)\n\n if exists(vocoder):\n out = out.permute(0, 2, 1)\n out = vocoder(out)\n\n return out, trajectory\n\n def forward(\n self,\n inp: float[\"b n d\"] | float[\"b nw\"], # mel or raw wave # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n noise_scheduler: str | None = None,\n ):\n # handle raw wave\n if inp.ndim == 2:\n inp = self.mel_spec(inp)\n inp = inp.permute(0, 2, 1)\n assert inp.shape[-1] == self.num_channels\n\n batch, seq_len, dtype, device, _σ1 = *inp.shape[:2], inp.dtype, self.device, self.sigma\n\n # handle text as string\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # lens and mask\n if not exists(lens):\n lens = torch.full((batch,), seq_len, device=device)\n\n mask = lens_to_mask(lens, length=seq_len) # useless here, as collate_fn will pad to max length in batch\n\n # get a random span to mask out for training conditionally\n frac_lengths = torch.zeros((batch,), device=self.device).float().uniform_(*self.frac_lengths_mask)\n rand_span_mask = mask_from_frac_lengths(lens, frac_lengths)\n\n if exists(mask):\n rand_span_mask &= mask\n\n # mel is x1\n x1 = inp\n\n # x0 is gaussian noise\n x0 = torch.randn_like(x1)\n\n # time step\n time = torch.rand((batch,), dtype=dtype, device=self.device)\n # TODO. noise_scheduler\n\n # sample xt (φ_t(x) in the paper)\n t = time.unsqueeze(-1).unsqueeze(-1)\n φ = (1 - t) * x0 + t * x1\n flow = x1 - x0\n\n # only predict what is within the random mask span for infilling\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n\n # transformer and cfg training with a drop rate\n drop_audio_cond = random() < self.audio_drop_prob # p_drop in voicebox paper\n if random() < self.cond_drop_prob: # p_uncond in voicebox paper\n drop_audio_cond = True\n drop_text = True\n else:\n drop_text = False\n\n # apply mask will use more memory; might adjust batchsize or batchsampler long sequence threshold\n pred = self.transformer(\n x=φ, cond=cond, text=text, time=time, drop_audio_cond=drop_audio_cond, drop_text=drop_text, mask=mask\n )\n\n # flow matching loss\n loss = F.mse_loss(pred, flow, reduction=\"none\")\n loss = loss[rand_span_mask]\n\n return loss.mean(), cond, pred","source_hash":"99fd339581ab0096169d1183c45c099bd69d572e9343efc3d9f19614c3f8629c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.cfm.CFM","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.cfm.CFM#L33-L302","kind":"class","name":"CFM","path":"src/f5_tts/model_new/cfm.py","language":"python","start_line":33,"end_line":302,"context_start_line":13,"context_end_line":302,"code":"from typing import Callable\n\nimport torch\nimport torch.nn.functional as F\nfrom torch import nn\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torchdiffeq import odeint\n\nfrom f5_tts.model_new.modules import MelSpec\nfrom f5_tts.model_new.utils import (\n default,\n exists,\n get_epss_timesteps,\n lens_to_mask,\n list_str_to_idx,\n list_str_to_tensor,\n mask_from_frac_lengths,\n)\n\n\nclass CFM(nn.Module):\n def __init__(\n self,\n transformer: nn.Module,\n sigma=0.0,\n odeint_kwargs: dict = dict(\n # atol = 1e-5,\n # rtol = 1e-5,\n method=\"euler\" # 'midpoint'\n ),\n audio_drop_prob=0.3,\n cond_drop_prob=0.2,\n num_channels=None,\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n frac_lengths_mask: tuple[float, float] = (0.7, 1.0),\n vocab_char_map: dict[str:int] | None = None,\n ):\n super().__init__()\n\n self.frac_lengths_mask = frac_lengths_mask\n\n # mel spec\n self.mel_spec = default(mel_spec_module, MelSpec(**mel_spec_kwargs))\n num_channels = default(num_channels, self.mel_spec.n_mel_channels)\n self.num_channels = num_channels\n\n # classifier-free guidance\n self.audio_drop_prob = audio_drop_prob\n self.cond_drop_prob = cond_drop_prob\n\n # transformer\n self.transformer = transformer\n dim = transformer.dim\n self.dim = dim\n\n # conditional flow related\n self.sigma = sigma\n\n # sampling related\n self.odeint_kwargs = odeint_kwargs\n\n # vocab map for tokenization\n self.vocab_char_map = vocab_char_map\n\n @property\n def device(self):\n return next(self.parameters()).device\n\n @torch.no_grad()\n def sample(\n self,\n cond: float[\"b n d\"] | float[\"b nw\"], # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n duration: int | int[\"b\"], # noqa: F821\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n steps=32,\n cfg_strength=1.0,\n sway_sampling_coef=None,\n seed: int | None = None,\n max_duration=4096,\n vocoder: Callable[[float[\"b d n\"]], float[\"b nw\"]] | None = None, # noqa: F722\n use_epss=True,\n no_ref_audio=False,\n duplicate_test=False,\n t_inter=0.1,\n edit_mask=None,\n ):\n self.eval()\n # raw wave\n\n if cond.ndim == 2:\n cond = self.mel_spec(cond)\n cond = cond.permute(0, 2, 1)\n assert cond.shape[-1] == self.num_channels\n\n cond = cond.to(next(self.parameters()).dtype)\n\n batch, cond_seq_len, device = *cond.shape[:2], cond.device\n if not exists(lens):\n lens = torch.full((batch,), cond_seq_len, device=device, dtype=torch.long)\n\n # text\n\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # duration\n\n cond_mask = lens_to_mask(lens)\n if edit_mask is not None:\n cond_mask = cond_mask & edit_mask\n\n if isinstance(duration, int):\n duration = torch.full((batch,), duration, device=device, dtype=torch.long)\n\n duration = torch.maximum(\n torch.maximum((text != -1).sum(dim=-1), lens) + 1, duration\n ) # duration at least text/audio prompt length plus one token, so something is generated\n duration = duration.clamp(max=max_duration)\n max_duration = duration.amax()\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n test_cond = F.pad(cond, (0, 0, cond_seq_len, max_duration - 2 * cond_seq_len), value=0.0)\n\n cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value=0.0)\n if no_ref_audio:\n cond = torch.zeros_like(cond)\n\n cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value=False)\n cond_mask = cond_mask.unsqueeze(-1)\n step_cond = torch.where(\n cond_mask, cond, torch.zeros_like(cond)\n ) # allow direct control (cut cond audio) with lens passed in\n\n if batch > 1:\n mask = lens_to_mask(duration)\n else: # save memory and speed up, as single inference need no mask currently\n mask = None\n\n # neural ode\n\n def fn(t, x):\n # at each step, conditioning is fixed\n # step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))\n\n # predict flow (cond)\n if cfg_strength < 1e-5:\n pred = self.transformer(\n x=x,\n cond=step_cond,\n text=text,\n time=t,\n mask=mask,\n drop_audio_cond=False,\n drop_text=False,\n cache=True,\n )\n return pred\n\n # predict flow (cond and uncond), for classifier-free guidance\n pred_cfg = self.transformer(\n x=x,\n cond=step_cond,\n text=text,\n time=t,\n mask=mask,\n cfg_infer=True,\n cache=True,\n )\n pred, null_pred = torch.chunk(pred_cfg, 2, dim=0)\n return pred + (pred - null_pred) * cfg_strength\n\n # noise input\n # to make sure batch inference result is same with different batch size, and for sure single inference\n # still some difference maybe due to convolutional layers\n y0 = []\n for dur in duration:\n if exists(seed):\n torch.manual_seed(seed)\n y0.append(torch.randn(dur, self.num_channels, device=self.device, dtype=step_cond.dtype))\n y0 = pad_sequence(y0, padding_value=0, batch_first=True)\n\n t_start = 0\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n t_start = t_inter\n y0 = (1 - t_start) * y0 + t_start * test_cond\n steps = int(steps * (1 - t_start))\n\n if t_start == 0 and use_epss: # use Empirically Pruned Step Sampling for low NFE\n t = get_epss_timesteps(steps, device=self.device, dtype=step_cond.dtype)\n else:\n t = torch.linspace(t_start, 1, steps + 1, device=self.device, dtype=step_cond.dtype)\n if sway_sampling_coef is not None:\n t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)\n\n trajectory = odeint(fn, y0, t, **self.odeint_kwargs)\n self.transformer.clear_cache()\n\n sampled = trajectory[-1]\n out = sampled\n out = torch.where(cond_mask, cond, out)\n\n if exists(vocoder):\n out = out.permute(0, 2, 1)\n out = vocoder(out)\n\n return out, trajectory\n\n def forward(\n self,\n inp: float[\"b n d\"] | float[\"b nw\"], # mel or raw wave # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n noise_scheduler: str | None = None,\n ):\n # handle raw wave\n if inp.ndim == 2:\n inp = self.mel_spec(inp)\n inp = inp.permute(0, 2, 1)\n assert inp.shape[-1] == self.num_channels\n\n batch, seq_len, dtype, device, _σ1 = *inp.shape[:2], inp.dtype, self.device, self.sigma\n\n # handle text as string\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # lens and mask\n if not exists(lens):\n lens = torch.full((batch,), seq_len, device=device)\n\n mask = lens_to_mask(lens, length=seq_len) # useless here, as collate_fn will pad to max length in batch\n\n # get a random span to mask out for training conditionally\n frac_lengths = torch.zeros((batch,), device=self.device).float().uniform_(*self.frac_lengths_mask)\n rand_span_mask = mask_from_frac_lengths(lens, frac_lengths)\n\n if exists(mask):\n rand_span_mask &= mask\n\n # mel is x1\n x1 = inp\n\n # x0 is gaussian noise\n x0 = torch.randn_like(x1)\n\n # time step\n time = torch.rand((batch,), dtype=dtype, device=self.device)\n # TODO. noise_scheduler\n\n # sample xt (φ_t(x) in the paper)\n t = time.unsqueeze(-1).unsqueeze(-1)\n φ = (1 - t) * x0 + t * x1\n flow = x1 - x0\n\n # only predict what is within the random mask span for infilling\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n\n # transformer and cfg training with a drop rate\n drop_audio_cond = random() < self.audio_drop_prob # p_drop in voicebox paper\n if random() < self.cond_drop_prob: # p_uncond in voicebox paper\n drop_audio_cond = True\n drop_text = True\n else:\n drop_text = False\n\n # apply mask will use more memory; might adjust batchsize or batchsampler long sequence threshold\n pred = self.transformer(\n x=φ, cond=cond, text=text, time=time, drop_audio_cond=drop_audio_cond, drop_text=drop_text, mask=mask\n )\n\n # flow matching loss\n loss = F.mse_loss(pred, flow, reduction=\"none\")\n loss = loss[rand_span_mask]\n\n return loss.mean(), cond, pred","source_hash":"99fd339581ab0096169d1183c45c099bd69d572e9343efc3d9f19614c3f8629c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.cfm.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.cfm.__init__#L34-L76","kind":"function","name":"__init__","path":"src/f5_tts/model_new/cfm.py","language":"python","start_line":34,"end_line":76,"context_start_line":14,"context_end_line":96,"code":"\nimport torch\nimport torch.nn.functional as F\nfrom torch import nn\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torchdiffeq import odeint\n\nfrom f5_tts.model_new.modules import MelSpec\nfrom f5_tts.model_new.utils import (\n default,\n exists,\n get_epss_timesteps,\n lens_to_mask,\n list_str_to_idx,\n list_str_to_tensor,\n mask_from_frac_lengths,\n)\n\n\nclass CFM(nn.Module):\n def __init__(\n self,\n transformer: nn.Module,\n sigma=0.0,\n odeint_kwargs: dict = dict(\n # atol = 1e-5,\n # rtol = 1e-5,\n method=\"euler\" # 'midpoint'\n ),\n audio_drop_prob=0.3,\n cond_drop_prob=0.2,\n num_channels=None,\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n frac_lengths_mask: tuple[float, float] = (0.7, 1.0),\n vocab_char_map: dict[str:int] | None = None,\n ):\n super().__init__()\n\n self.frac_lengths_mask = frac_lengths_mask\n\n # mel spec\n self.mel_spec = default(mel_spec_module, MelSpec(**mel_spec_kwargs))\n num_channels = default(num_channels, self.mel_spec.n_mel_channels)\n self.num_channels = num_channels\n\n # classifier-free guidance\n self.audio_drop_prob = audio_drop_prob\n self.cond_drop_prob = cond_drop_prob\n\n # transformer\n self.transformer = transformer\n dim = transformer.dim\n self.dim = dim\n\n # conditional flow related\n self.sigma = sigma\n\n # sampling related\n self.odeint_kwargs = odeint_kwargs\n\n # vocab map for tokenization\n self.vocab_char_map = vocab_char_map\n\n @property\n def device(self):\n return next(self.parameters()).device\n\n @torch.no_grad()\n def sample(\n self,\n cond: float[\"b n d\"] | float[\"b nw\"], # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n duration: int | int[\"b\"], # noqa: F821\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n steps=32,\n cfg_strength=1.0,\n sway_sampling_coef=None,\n seed: int | None = None,\n max_duration=4096,\n vocoder: Callable[[float[\"b d n\"]], float[\"b nw\"]] | None = None, # noqa: F722\n use_epss=True,","source_hash":"99fd339581ab0096169d1183c45c099bd69d572e9343efc3d9f19614c3f8629c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.cfm.device","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.cfm.device#L79-L80","kind":"function","name":"device","path":"src/f5_tts/model_new/cfm.py","language":"python","start_line":79,"end_line":80,"context_start_line":59,"context_end_line":100,"code":"\n # classifier-free guidance\n self.audio_drop_prob = audio_drop_prob\n self.cond_drop_prob = cond_drop_prob\n\n # transformer\n self.transformer = transformer\n dim = transformer.dim\n self.dim = dim\n\n # conditional flow related\n self.sigma = sigma\n\n # sampling related\n self.odeint_kwargs = odeint_kwargs\n\n # vocab map for tokenization\n self.vocab_char_map = vocab_char_map\n\n @property\n def device(self):\n return next(self.parameters()).device\n\n @torch.no_grad()\n def sample(\n self,\n cond: float[\"b n d\"] | float[\"b nw\"], # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n duration: int | int[\"b\"], # noqa: F821\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n steps=32,\n cfg_strength=1.0,\n sway_sampling_coef=None,\n seed: int | None = None,\n max_duration=4096,\n vocoder: Callable[[float[\"b d n\"]], float[\"b nw\"]] | None = None, # noqa: F722\n use_epss=True,\n no_ref_audio=False,\n duplicate_test=False,\n t_inter=0.1,\n edit_mask=None,","source_hash":"99fd339581ab0096169d1183c45c099bd69d572e9343efc3d9f19614c3f8629c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.cfm.sample","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.cfm.sample#L83-L228","kind":"function","name":"sample","path":"src/f5_tts/model_new/cfm.py","language":"python","start_line":83,"end_line":228,"context_start_line":63,"context_end_line":248,"code":"\n # transformer\n self.transformer = transformer\n dim = transformer.dim\n self.dim = dim\n\n # conditional flow related\n self.sigma = sigma\n\n # sampling related\n self.odeint_kwargs = odeint_kwargs\n\n # vocab map for tokenization\n self.vocab_char_map = vocab_char_map\n\n @property\n def device(self):\n return next(self.parameters()).device\n\n @torch.no_grad()\n def sample(\n self,\n cond: float[\"b n d\"] | float[\"b nw\"], # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n duration: int | int[\"b\"], # noqa: F821\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n steps=32,\n cfg_strength=1.0,\n sway_sampling_coef=None,\n seed: int | None = None,\n max_duration=4096,\n vocoder: Callable[[float[\"b d n\"]], float[\"b nw\"]] | None = None, # noqa: F722\n use_epss=True,\n no_ref_audio=False,\n duplicate_test=False,\n t_inter=0.1,\n edit_mask=None,\n ):\n self.eval()\n # raw wave\n\n if cond.ndim == 2:\n cond = self.mel_spec(cond)\n cond = cond.permute(0, 2, 1)\n assert cond.shape[-1] == self.num_channels\n\n cond = cond.to(next(self.parameters()).dtype)\n\n batch, cond_seq_len, device = *cond.shape[:2], cond.device\n if not exists(lens):\n lens = torch.full((batch,), cond_seq_len, device=device, dtype=torch.long)\n\n # text\n\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # duration\n\n cond_mask = lens_to_mask(lens)\n if edit_mask is not None:\n cond_mask = cond_mask & edit_mask\n\n if isinstance(duration, int):\n duration = torch.full((batch,), duration, device=device, dtype=torch.long)\n\n duration = torch.maximum(\n torch.maximum((text != -1).sum(dim=-1), lens) + 1, duration\n ) # duration at least text/audio prompt length plus one token, so something is generated\n duration = duration.clamp(max=max_duration)\n max_duration = duration.amax()\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n test_cond = F.pad(cond, (0, 0, cond_seq_len, max_duration - 2 * cond_seq_len), value=0.0)\n\n cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value=0.0)\n if no_ref_audio:\n cond = torch.zeros_like(cond)\n\n cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value=False)\n cond_mask = cond_mask.unsqueeze(-1)\n step_cond = torch.where(\n cond_mask, cond, torch.zeros_like(cond)\n ) # allow direct control (cut cond audio) with lens passed in\n\n if batch > 1:\n mask = lens_to_mask(duration)\n else: # save memory and speed up, as single inference need no mask currently\n mask = None\n\n # neural ode\n\n def fn(t, x):\n # at each step, conditioning is fixed\n # step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))\n\n # predict flow (cond)\n if cfg_strength < 1e-5:\n pred = self.transformer(\n x=x,\n cond=step_cond,\n text=text,\n time=t,\n mask=mask,\n drop_audio_cond=False,\n drop_text=False,\n cache=True,\n )\n return pred\n\n # predict flow (cond and uncond), for classifier-free guidance\n pred_cfg = self.transformer(\n x=x,\n cond=step_cond,\n text=text,\n time=t,\n mask=mask,\n cfg_infer=True,\n cache=True,\n )\n pred, null_pred = torch.chunk(pred_cfg, 2, dim=0)\n return pred + (pred - null_pred) * cfg_strength\n\n # noise input\n # to make sure batch inference result is same with different batch size, and for sure single inference\n # still some difference maybe due to convolutional layers\n y0 = []\n for dur in duration:\n if exists(seed):\n torch.manual_seed(seed)\n y0.append(torch.randn(dur, self.num_channels, device=self.device, dtype=step_cond.dtype))\n y0 = pad_sequence(y0, padding_value=0, batch_first=True)\n\n t_start = 0\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n t_start = t_inter\n y0 = (1 - t_start) * y0 + t_start * test_cond\n steps = int(steps * (1 - t_start))\n\n if t_start == 0 and use_epss: # use Empirically Pruned Step Sampling for low NFE\n t = get_epss_timesteps(steps, device=self.device, dtype=step_cond.dtype)\n else:\n t = torch.linspace(t_start, 1, steps + 1, device=self.device, dtype=step_cond.dtype)\n if sway_sampling_coef is not None:\n t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)\n\n trajectory = odeint(fn, y0, t, **self.odeint_kwargs)\n self.transformer.clear_cache()\n\n sampled = trajectory[-1]\n out = sampled\n out = torch.where(cond_mask, cond, out)\n\n if exists(vocoder):\n out = out.permute(0, 2, 1)\n out = vocoder(out)\n\n return out, trajectory\n\n def forward(\n self,\n inp: float[\"b n d\"] | float[\"b nw\"], # mel or raw wave # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n noise_scheduler: str | None = None,\n ):\n # handle raw wave\n if inp.ndim == 2:\n inp = self.mel_spec(inp)\n inp = inp.permute(0, 2, 1)\n assert inp.shape[-1] == self.num_channels\n\n batch, seq_len, dtype, device, _σ1 = *inp.shape[:2], inp.dtype, self.device, self.sigma\n\n # handle text as string\n if isinstance(text, list):\n if exists(self.vocab_char_map):","source_hash":"99fd339581ab0096169d1183c45c099bd69d572e9343efc3d9f19614c3f8629c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.cfm.forward","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.cfm.forward#L230-L302","kind":"function","name":"forward","path":"src/f5_tts/model_new/cfm.py","language":"python","start_line":230,"end_line":302,"context_start_line":210,"context_end_line":302,"code":" if t_start == 0 and use_epss: # use Empirically Pruned Step Sampling for low NFE\n t = get_epss_timesteps(steps, device=self.device, dtype=step_cond.dtype)\n else:\n t = torch.linspace(t_start, 1, steps + 1, device=self.device, dtype=step_cond.dtype)\n if sway_sampling_coef is not None:\n t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)\n\n trajectory = odeint(fn, y0, t, **self.odeint_kwargs)\n self.transformer.clear_cache()\n\n sampled = trajectory[-1]\n out = sampled\n out = torch.where(cond_mask, cond, out)\n\n if exists(vocoder):\n out = out.permute(0, 2, 1)\n out = vocoder(out)\n\n return out, trajectory\n\n def forward(\n self,\n inp: float[\"b n d\"] | float[\"b nw\"], # mel or raw wave # noqa: F722\n text: int[\"b nt\"] | list[str], # noqa: F722\n *,\n lens: int[\"b\"] | None = None, # noqa: F821\n noise_scheduler: str | None = None,\n ):\n # handle raw wave\n if inp.ndim == 2:\n inp = self.mel_spec(inp)\n inp = inp.permute(0, 2, 1)\n assert inp.shape[-1] == self.num_channels\n\n batch, seq_len, dtype, device, _σ1 = *inp.shape[:2], inp.dtype, self.device, self.sigma\n\n # handle text as string\n if isinstance(text, list):\n if exists(self.vocab_char_map):\n text = list_str_to_idx(text, self.vocab_char_map).to(device)\n else:\n text = list_str_to_tensor(text).to(device)\n assert text.shape[0] == batch\n\n # lens and mask\n if not exists(lens):\n lens = torch.full((batch,), seq_len, device=device)\n\n mask = lens_to_mask(lens, length=seq_len) # useless here, as collate_fn will pad to max length in batch\n\n # get a random span to mask out for training conditionally\n frac_lengths = torch.zeros((batch,), device=self.device).float().uniform_(*self.frac_lengths_mask)\n rand_span_mask = mask_from_frac_lengths(lens, frac_lengths)\n\n if exists(mask):\n rand_span_mask &= mask\n\n # mel is x1\n x1 = inp\n\n # x0 is gaussian noise\n x0 = torch.randn_like(x1)\n\n # time step\n time = torch.rand((batch,), dtype=dtype, device=self.device)\n # TODO. noise_scheduler\n\n # sample xt (φ_t(x) in the paper)\n t = time.unsqueeze(-1).unsqueeze(-1)\n φ = (1 - t) * x0 + t * x1\n flow = x1 - x0\n\n # only predict what is within the random mask span for infilling\n cond = torch.where(rand_span_mask[..., None], torch.zeros_like(x1), x1)\n\n # transformer and cfg training with a drop rate\n drop_audio_cond = random() < self.audio_drop_prob # p_drop in voicebox paper\n if random() < self.cond_drop_prob: # p_uncond in voicebox paper\n drop_audio_cond = True\n drop_text = True\n else:\n drop_text = False\n\n # apply mask will use more memory; might adjust batchsize or batchsampler long sequence threshold\n pred = self.transformer(\n x=φ, cond=cond, text=text, time=time, drop_audio_cond=drop_audio_cond, drop_text=drop_text, mask=mask\n )\n\n # flow matching loss\n loss = F.mse_loss(pred, flow, reduction=\"none\")\n loss = loss[rand_span_mask]\n\n return loss.mean(), cond, pred","source_hash":"99fd339581ab0096169d1183c45c099bd69d572e9343efc3d9f19614c3f8629c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.cfm.fn","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.cfm.fn#L161-L190","kind":"function","name":"fn","path":"src/f5_tts/model_new/cfm.py","language":"python","start_line":161,"end_line":190,"context_start_line":141,"context_end_line":210,"code":" if duplicate_test:\n test_cond = F.pad(cond, (0, 0, cond_seq_len, max_duration - 2 * cond_seq_len), value=0.0)\n\n cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value=0.0)\n if no_ref_audio:\n cond = torch.zeros_like(cond)\n\n cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value=False)\n cond_mask = cond_mask.unsqueeze(-1)\n step_cond = torch.where(\n cond_mask, cond, torch.zeros_like(cond)\n ) # allow direct control (cut cond audio) with lens passed in\n\n if batch > 1:\n mask = lens_to_mask(duration)\n else: # save memory and speed up, as single inference need no mask currently\n mask = None\n\n # neural ode\n\n def fn(t, x):\n # at each step, conditioning is fixed\n # step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))\n\n # predict flow (cond)\n if cfg_strength < 1e-5:\n pred = self.transformer(\n x=x,\n cond=step_cond,\n text=text,\n time=t,\n mask=mask,\n drop_audio_cond=False,\n drop_text=False,\n cache=True,\n )\n return pred\n\n # predict flow (cond and uncond), for classifier-free guidance\n pred_cfg = self.transformer(\n x=x,\n cond=step_cond,\n text=text,\n time=t,\n mask=mask,\n cfg_infer=True,\n cache=True,\n )\n pred, null_pred = torch.chunk(pred_cfg, 2, dim=0)\n return pred + (pred - null_pred) * cfg_strength\n\n # noise input\n # to make sure batch inference result is same with different batch size, and for sure single inference\n # still some difference maybe due to convolutional layers\n y0 = []\n for dur in duration:\n if exists(seed):\n torch.manual_seed(seed)\n y0.append(torch.randn(dur, self.num_channels, device=self.device, dtype=step_cond.dtype))\n y0 = pad_sequence(y0, padding_value=0, batch_first=True)\n\n t_start = 0\n\n # duplicate test corner for inner time step oberservation\n if duplicate_test:\n t_start = t_inter\n y0 = (1 - t_start) * y0 + t_start * test_cond\n steps = int(steps * (1 - t_start))\n\n if t_start == 0 and use_epss: # use Empirically Pruned Step Sampling for low NFE","source_hash":"99fd339581ab0096169d1183c45c099bd69d572e9343efc3d9f19614c3f8629c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.dataset","uri":"program://DMOSpeech2/module/src.f5_tts.model_new.dataset#L1-L330","kind":"module","name":"src.f5_tts.model_new.dataset","path":"src/f5_tts/model_new/dataset.py","language":"python","start_line":1,"end_line":330,"context_start_line":1,"context_end_line":330,"code":"import json\nfrom importlib.resources import files\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom datasets import Dataset as Dataset_\nfrom datasets import load_from_disk\nfrom torch import nn\nfrom torch.utils.data import Dataset, Sampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import default\n\n\nclass HFDataset(Dataset):\n def __init__(\n self,\n hf_dataset: Dataset,\n target_sample_rate=24_000,\n n_mel_channels=100,\n hop_length=256,\n n_fft=1024,\n win_length=1024,\n mel_spec_type=\"vocos\",\n ):\n self.data = hf_dataset\n self.target_sample_rate = target_sample_rate\n self.hop_length = hop_length\n\n self.mel_spectrogram = MelSpec(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n )\n\n def get_frame_len(self, index):\n row = self.data[index]\n audio = row[\"audio\"][\"array\"]\n sample_rate = row[\"audio\"][\"sampling_rate\"]\n return audio.shape[-1] / sample_rate * self.target_sample_rate / self.hop_length\n\n def __len__(self):\n return len(self.data)\n\n def __getitem__(self, index):\n row = self.data[index]\n audio = row[\"audio\"][\"array\"]\n\n # logger.info(f\"Audio shape: {audio.shape}\")\n\n sample_rate = row[\"audio\"][\"sampling_rate\"]\n duration = audio.shape[-1] / sample_rate\n\n if duration > 30 or duration < 0.3:\n return self.__getitem__((index + 1) % len(self.data))\n\n audio_tensor = torch.from_numpy(audio).float()\n\n if sample_rate != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sample_rate, self.target_sample_rate)\n audio_tensor = resampler(audio_tensor)\n\n audio_tensor = audio_tensor.unsqueeze(0) # 't -> 1 t')\n\n mel_spec = self.mel_spectrogram(audio_tensor)\n\n mel_spec = mel_spec.squeeze(0) # '1 d t -> d t'\n\n text = row[\"text\"]\n\n return dict(\n mel_spec=mel_spec,\n text=text,\n )\n\n\nclass CustomDataset(Dataset):\n def __init__(\n self,\n custom_dataset: Dataset,\n durations=None,\n target_sample_rate=24_000,\n hop_length=256,\n n_mel_channels=100,\n n_fft=1024,\n win_length=1024,\n mel_spec_type=\"vocos\",\n preprocessed_mel=False,\n mel_spec_module: nn.Module | None = None,\n ):\n self.data = custom_dataset\n self.durations = durations\n self.target_sample_rate = target_sample_rate\n self.hop_length = hop_length\n self.n_fft = n_fft\n self.win_length = win_length\n self.mel_spec_type = mel_spec_type\n self.preprocessed_mel = preprocessed_mel\n\n if not preprocessed_mel:\n self.mel_spectrogram = default(\n mel_spec_module,\n MelSpec(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n ),\n )\n\n def get_frame_len(self, index):\n if (\n self.durations is not None\n ): # Please make sure the separately provided durations are correct, otherwise 99.99% OOM\n return self.durations[index] * self.target_sample_rate / self.hop_length\n return self.data[index][\"duration\"] * self.target_sample_rate / self.hop_length\n\n def __len__(self):\n return len(self.data)\n\n def __getitem__(self, index):\n while True:\n row = self.data[index]\n audio_path = row[\"audio_path\"]\n text = row[\"text\"]\n duration = row[\"duration\"]\n\n # filter by given length\n if 0.3 <= duration <= 30:\n break # valid\n\n index = (index + 1) % len(self.data)\n\n if self.preprocessed_mel:\n mel_spec = torch.tensor(row[\"mel_spec\"])\n else:\n audio, source_sample_rate = torchaudio.load(audio_path)\n\n # make sure mono input\n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n\n # resample if necessary\n if source_sample_rate != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(source_sample_rate, self.target_sample_rate)\n audio = resampler(audio)\n\n # to mel spectrogram\n mel_spec = self.mel_spectrogram(audio)\n mel_spec = mel_spec.squeeze(0) # '1 d t -> d t'\n\n return {\n \"mel_spec\": mel_spec,\n \"text\": text,\n }\n\n\n# Dynamic Batch Sampler\nclass DynamicBatchSampler(Sampler[list[int]]):\n \"\"\"Extension of Sampler that will do the following:\n 1. Change the batch size (essentially number of sequences)\n in a batch to ensure that the total number of frames are less\n than a certain threshold.\n 2. Make sure the padding efficiency in the batch is high.\n 3. Shuffle batches each epoch while maintaining reproducibility.\n \"\"\"\n\n def __init__(\n self, sampler: Sampler[int], frames_threshold: int, max_samples=0, random_seed=None, drop_residual: bool = False\n ):\n self.sampler = sampler\n self.frames_threshold = frames_threshold\n self.max_samples = max_samples\n self.random_seed = random_seed\n self.epoch = 0\n\n indices, batches = [], []\n data_source = self.sampler.data_source\n\n for idx in tqdm(\n self.sampler, desc=\"Sorting with sampler... if slow, check whether dataset is provided with duration\"\n ):\n indices.append((idx, data_source.get_frame_len(idx)))\n indices.sort(key=lambda elem: elem[1])\n\n batch = []\n batch_frames = 0\n for idx, frame_len in tqdm(\n indices, desc=f\"Creating dynamic batches with {frames_threshold} audio frames per gpu\"\n ):\n if batch_frames + frame_len <= self.frames_threshold and (max_samples == 0 or len(batch) < max_samples):\n batch.append(idx)\n batch_frames += frame_len\n else:\n if len(batch) > 0:\n batches.append(batch)\n if frame_len <= self.frames_threshold:\n batch = [idx]\n batch_frames = frame_len\n else:\n batch = []\n batch_frames = 0\n\n if not drop_residual and len(batch) > 0:\n batches.append(batch)\n\n del indices\n self.batches = batches\n\n # Ensure even batches with accelerate BatchSamplerShard cls under frame_per_batch setting\n self.drop_last = True\n\n def set_epoch(self, epoch: int) -> None:\n \"\"\"Sets the epoch for this sampler.\"\"\"\n self.epoch = epoch\n\n def __iter__(self):\n # Use both random_seed and epoch for deterministic but different shuffling per epoch\n if self.random_seed is not None:\n g = torch.Generator()\n g.manual_seed(self.random_seed + self.epoch)\n # Use PyTorch's random permutation for better reproducibility across PyTorch versions\n indices = torch.randperm(len(self.batches), generator=g).tolist()\n batches = [self.batches[i] for i in indices]\n else:\n batches = self.batches\n return iter(batches)\n\n def __len__(self):\n return len(self.batches)\n\n\n# Load dataset\n\n\ndef load_dataset(\n dataset_name: str,\n tokenizer: str = \"pinyin\",\n dataset_type: str = \"CustomDataset\",\n audio_type: str = \"raw\",\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n) -> CustomDataset | HFDataset:\n \"\"\"\n dataset_type - \"CustomDataset\" if you want to use tokenizer name and default data path to load for train_dataset\n - \"CustomDatasetPath\" if you just want to pass the full path to a preprocessed dataset without relying on tokenizer\n \"\"\"\n\n print(\"Loading dataset ...\")\n\n if dataset_type == \"CustomDataset\":\n rel_data_path = str(files(\"f5_tts\").joinpath(f\"../../data/{dataset_name}_{tokenizer}\"))\n if audio_type == \"raw\":\n try:\n train_dataset = load_from_disk(f\"{rel_data_path}/raw\")\n except: # noqa: E722\n train_dataset = Dataset_.from_file(f\"{rel_data_path}/raw.arrow\")\n preprocessed_mel = False\n elif audio_type == \"mel\":\n train_dataset = Dataset_.from_file(f\"{rel_data_path}/mel.arrow\")\n preprocessed_mel = True\n with open(f\"{rel_data_path}/duration.json\", \"r\", encoding=\"utf-8\") as f:\n data_dict = json.load(f)\n durations = data_dict[\"duration\"]\n train_dataset = CustomDataset(\n train_dataset,\n durations=durations,\n preprocessed_mel=preprocessed_mel,\n mel_spec_module=mel_spec_module,\n **mel_spec_kwargs,\n )\n\n elif dataset_type == \"CustomDatasetPath\":\n try:\n train_dataset = load_from_disk(f\"{dataset_name}/raw\")\n except: # noqa: E722\n train_dataset = Dataset_.from_file(f\"{dataset_name}/raw.arrow\")\n\n with open(f\"{dataset_name}/duration.json\", \"r\", encoding=\"utf-8\") as f:\n data_dict = json.load(f)\n durations = data_dict[\"duration\"]\n train_dataset = CustomDataset(\n train_dataset, durations=durations, preprocessed_mel=preprocessed_mel, **mel_spec_kwargs\n )\n\n elif dataset_type == \"HFDataset\":\n print(\n \"Should manually modify the path of huggingface dataset to your need.\\n\"\n + \"May also the corresponding script cuz different dataset may have different format.\"\n )\n pre, post = dataset_name.split(\"_\")\n train_dataset = HFDataset(\n load_dataset(f\"{pre}/{pre}\", split=f\"train.{post}\", cache_dir=str(files(\"f5_tts\").joinpath(\"../../data\"))),\n )\n\n return train_dataset\n\n\n# collation\n\n\ndef collate_fn(batch):\n mel_specs = [item[\"mel_spec\"].squeeze(0) for item in batch]\n mel_lengths = torch.LongTensor([spec.shape[-1] for spec in mel_specs])\n max_mel_length = mel_lengths.amax()\n\n padded_mel_specs = []\n for spec in mel_specs:\n padding = (0, max_mel_length - spec.size(-1))\n padded_spec = F.pad(spec, padding, value=0)\n padded_mel_specs.append(padded_spec)\n\n mel_specs = torch.stack(padded_mel_specs)\n\n text = [item[\"text\"] for item in batch]\n text_lengths = torch.LongTensor([len(item) for item in text])\n\n return dict(\n mel=mel_specs,\n mel_lengths=mel_lengths, # records for padding mask\n text=text,\n text_lengths=text_lengths,\n )","source_hash":"5ddfadf6e712d32b7c55a455c51078896ebcd239a8e60bf2d70ac7033f14f697","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.dataset.HFDataset","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.dataset.HFDataset#L17-L79","kind":"class","name":"HFDataset","path":"src/f5_tts/model_new/dataset.py","language":"python","start_line":17,"end_line":79,"context_start_line":1,"context_end_line":99,"code":"import json\nfrom importlib.resources import files\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom datasets import Dataset as Dataset_\nfrom datasets import load_from_disk\nfrom torch import nn\nfrom torch.utils.data import Dataset, Sampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import default\n\n\nclass HFDataset(Dataset):\n def __init__(\n self,\n hf_dataset: Dataset,\n target_sample_rate=24_000,\n n_mel_channels=100,\n hop_length=256,\n n_fft=1024,\n win_length=1024,\n mel_spec_type=\"vocos\",\n ):\n self.data = hf_dataset\n self.target_sample_rate = target_sample_rate\n self.hop_length = hop_length\n\n self.mel_spectrogram = MelSpec(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n )\n\n def get_frame_len(self, index):\n row = self.data[index]\n audio = row[\"audio\"][\"array\"]\n sample_rate = row[\"audio\"][\"sampling_rate\"]\n return audio.shape[-1] / sample_rate * self.target_sample_rate / self.hop_length\n\n def __len__(self):\n return len(self.data)\n\n def __getitem__(self, index):\n row = self.data[index]\n audio = row[\"audio\"][\"array\"]\n\n # logger.info(f\"Audio shape: {audio.shape}\")\n\n sample_rate = row[\"audio\"][\"sampling_rate\"]\n duration = audio.shape[-1] / sample_rate\n\n if duration > 30 or duration < 0.3:\n return self.__getitem__((index + 1) % len(self.data))\n\n audio_tensor = torch.from_numpy(audio).float()\n\n if sample_rate != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sample_rate, self.target_sample_rate)\n audio_tensor = resampler(audio_tensor)\n\n audio_tensor = audio_tensor.unsqueeze(0) # 't -> 1 t')\n\n mel_spec = self.mel_spectrogram(audio_tensor)\n\n mel_spec = mel_spec.squeeze(0) # '1 d t -> d t'\n\n text = row[\"text\"]\n\n return dict(\n mel_spec=mel_spec,\n text=text,\n )\n\n\nclass CustomDataset(Dataset):\n def __init__(\n self,\n custom_dataset: Dataset,\n durations=None,\n target_sample_rate=24_000,\n hop_length=256,\n n_mel_channels=100,\n n_fft=1024,\n win_length=1024,\n mel_spec_type=\"vocos\",\n preprocessed_mel=False,\n mel_spec_module: nn.Module | None = None,\n ):\n self.data = custom_dataset\n self.durations = durations\n self.target_sample_rate = target_sample_rate\n self.hop_length = hop_length","source_hash":"5ddfadf6e712d32b7c55a455c51078896ebcd239a8e60bf2d70ac7033f14f697","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.dataset.CustomDataset","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.dataset.CustomDataset#L82-L162","kind":"class","name":"CustomDataset","path":"src/f5_tts/model_new/dataset.py","language":"python","start_line":82,"end_line":162,"context_start_line":62,"context_end_line":182,"code":" audio_tensor = torch.from_numpy(audio).float()\n\n if sample_rate != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(sample_rate, self.target_sample_rate)\n audio_tensor = resampler(audio_tensor)\n\n audio_tensor = audio_tensor.unsqueeze(0) # 't -> 1 t')\n\n mel_spec = self.mel_spectrogram(audio_tensor)\n\n mel_spec = mel_spec.squeeze(0) # '1 d t -> d t'\n\n text = row[\"text\"]\n\n return dict(\n mel_spec=mel_spec,\n text=text,\n )\n\n\nclass CustomDataset(Dataset):\n def __init__(\n self,\n custom_dataset: Dataset,\n durations=None,\n target_sample_rate=24_000,\n hop_length=256,\n n_mel_channels=100,\n n_fft=1024,\n win_length=1024,\n mel_spec_type=\"vocos\",\n preprocessed_mel=False,\n mel_spec_module: nn.Module | None = None,\n ):\n self.data = custom_dataset\n self.durations = durations\n self.target_sample_rate = target_sample_rate\n self.hop_length = hop_length\n self.n_fft = n_fft\n self.win_length = win_length\n self.mel_spec_type = mel_spec_type\n self.preprocessed_mel = preprocessed_mel\n\n if not preprocessed_mel:\n self.mel_spectrogram = default(\n mel_spec_module,\n MelSpec(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n ),\n )\n\n def get_frame_len(self, index):\n if (\n self.durations is not None\n ): # Please make sure the separately provided durations are correct, otherwise 99.99% OOM\n return self.durations[index] * self.target_sample_rate / self.hop_length\n return self.data[index][\"duration\"] * self.target_sample_rate / self.hop_length\n\n def __len__(self):\n return len(self.data)\n\n def __getitem__(self, index):\n while True:\n row = self.data[index]\n audio_path = row[\"audio_path\"]\n text = row[\"text\"]\n duration = row[\"duration\"]\n\n # filter by given length\n if 0.3 <= duration <= 30:\n break # valid\n\n index = (index + 1) % len(self.data)\n\n if self.preprocessed_mel:\n mel_spec = torch.tensor(row[\"mel_spec\"])\n else:\n audio, source_sample_rate = torchaudio.load(audio_path)\n\n # make sure mono input\n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n\n # resample if necessary\n if source_sample_rate != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(source_sample_rate, self.target_sample_rate)\n audio = resampler(audio)\n\n # to mel spectrogram\n mel_spec = self.mel_spectrogram(audio)\n mel_spec = mel_spec.squeeze(0) # '1 d t -> d t'\n\n return {\n \"mel_spec\": mel_spec,\n \"text\": text,\n }\n\n\n# Dynamic Batch Sampler\nclass DynamicBatchSampler(Sampler[list[int]]):\n \"\"\"Extension of Sampler that will do the following:\n 1. Change the batch size (essentially number of sequences)\n in a batch to ensure that the total number of frames are less\n than a certain threshold.\n 2. Make sure the padding efficiency in the batch is high.\n 3. Shuffle batches each epoch while maintaining reproducibility.\n \"\"\"\n\n def __init__(\n self, sampler: Sampler[int], frames_threshold: int, max_samples=0, random_seed=None, drop_residual: bool = False\n ):\n self.sampler = sampler\n self.frames_threshold = frames_threshold\n self.max_samples = max_samples\n self.random_seed = random_seed\n self.epoch = 0","source_hash":"5ddfadf6e712d32b7c55a455c51078896ebcd239a8e60bf2d70ac7033f14f697","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.dataset.DynamicBatchSampler","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.dataset.DynamicBatchSampler#L166-L237","kind":"class","name":"DynamicBatchSampler","path":"src/f5_tts/model_new/dataset.py","language":"python","start_line":166,"end_line":237,"context_start_line":146,"context_end_line":257,"code":" # make sure mono input\n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n\n # resample if necessary\n if source_sample_rate != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(source_sample_rate, self.target_sample_rate)\n audio = resampler(audio)\n\n # to mel spectrogram\n mel_spec = self.mel_spectrogram(audio)\n mel_spec = mel_spec.squeeze(0) # '1 d t -> d t'\n\n return {\n \"mel_spec\": mel_spec,\n \"text\": text,\n }\n\n\n# Dynamic Batch Sampler\nclass DynamicBatchSampler(Sampler[list[int]]):\n \"\"\"Extension of Sampler that will do the following:\n 1. Change the batch size (essentially number of sequences)\n in a batch to ensure that the total number of frames are less\n than a certain threshold.\n 2. Make sure the padding efficiency in the batch is high.\n 3. Shuffle batches each epoch while maintaining reproducibility.\n \"\"\"\n\n def __init__(\n self, sampler: Sampler[int], frames_threshold: int, max_samples=0, random_seed=None, drop_residual: bool = False\n ):\n self.sampler = sampler\n self.frames_threshold = frames_threshold\n self.max_samples = max_samples\n self.random_seed = random_seed\n self.epoch = 0\n\n indices, batches = [], []\n data_source = self.sampler.data_source\n\n for idx in tqdm(\n self.sampler, desc=\"Sorting with sampler... if slow, check whether dataset is provided with duration\"\n ):\n indices.append((idx, data_source.get_frame_len(idx)))\n indices.sort(key=lambda elem: elem[1])\n\n batch = []\n batch_frames = 0\n for idx, frame_len in tqdm(\n indices, desc=f\"Creating dynamic batches with {frames_threshold} audio frames per gpu\"\n ):\n if batch_frames + frame_len <= self.frames_threshold and (max_samples == 0 or len(batch) < max_samples):\n batch.append(idx)\n batch_frames += frame_len\n else:\n if len(batch) > 0:\n batches.append(batch)\n if frame_len <= self.frames_threshold:\n batch = [idx]\n batch_frames = frame_len\n else:\n batch = []\n batch_frames = 0\n\n if not drop_residual and len(batch) > 0:\n batches.append(batch)\n\n del indices\n self.batches = batches\n\n # Ensure even batches with accelerate BatchSamplerShard cls under frame_per_batch setting\n self.drop_last = True\n\n def set_epoch(self, epoch: int) -> None:\n \"\"\"Sets the epoch for this sampler.\"\"\"\n self.epoch = epoch\n\n def __iter__(self):\n # Use both random_seed and epoch for deterministic but different shuffling per epoch\n if self.random_seed is not None:\n g = torch.Generator()\n g.manual_seed(self.random_seed + self.epoch)\n # Use PyTorch's random permutation for better reproducibility across PyTorch versions\n indices = torch.randperm(len(self.batches), generator=g).tolist()\n batches = [self.batches[i] for i in indices]\n else:\n batches = self.batches\n return iter(batches)\n\n def __len__(self):\n return len(self.batches)\n\n\n# Load dataset\n\n\ndef load_dataset(\n dataset_name: str,\n tokenizer: str = \"pinyin\",\n dataset_type: str = \"CustomDataset\",\n audio_type: str = \"raw\",\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n) -> CustomDataset | HFDataset:\n \"\"\"\n dataset_type - \"CustomDataset\" if you want to use tokenizer name and default data path to load for train_dataset\n - \"CustomDatasetPath\" if you just want to pass the full path to a preprocessed dataset without relying on tokenizer\n \"\"\"\n\n print(\"Loading dataset ...\")\n","source_hash":"5ddfadf6e712d32b7c55a455c51078896ebcd239a8e60bf2d70ac7033f14f697","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.dataset.load_dataset","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.dataset.load_dataset#L243-L303","kind":"function","name":"load_dataset","path":"src/f5_tts/model_new/dataset.py","language":"python","start_line":243,"end_line":303,"context_start_line":223,"context_end_line":323,"code":"\n def __iter__(self):\n # Use both random_seed and epoch for deterministic but different shuffling per epoch\n if self.random_seed is not None:\n g = torch.Generator()\n g.manual_seed(self.random_seed + self.epoch)\n # Use PyTorch's random permutation for better reproducibility across PyTorch versions\n indices = torch.randperm(len(self.batches), generator=g).tolist()\n batches = [self.batches[i] for i in indices]\n else:\n batches = self.batches\n return iter(batches)\n\n def __len__(self):\n return len(self.batches)\n\n\n# Load dataset\n\n\ndef load_dataset(\n dataset_name: str,\n tokenizer: str = \"pinyin\",\n dataset_type: str = \"CustomDataset\",\n audio_type: str = \"raw\",\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n) -> CustomDataset | HFDataset:\n \"\"\"\n dataset_type - \"CustomDataset\" if you want to use tokenizer name and default data path to load for train_dataset\n - \"CustomDatasetPath\" if you just want to pass the full path to a preprocessed dataset without relying on tokenizer\n \"\"\"\n\n print(\"Loading dataset ...\")\n\n if dataset_type == \"CustomDataset\":\n rel_data_path = str(files(\"f5_tts\").joinpath(f\"../../data/{dataset_name}_{tokenizer}\"))\n if audio_type == \"raw\":\n try:\n train_dataset = load_from_disk(f\"{rel_data_path}/raw\")\n except: # noqa: E722\n train_dataset = Dataset_.from_file(f\"{rel_data_path}/raw.arrow\")\n preprocessed_mel = False\n elif audio_type == \"mel\":\n train_dataset = Dataset_.from_file(f\"{rel_data_path}/mel.arrow\")\n preprocessed_mel = True\n with open(f\"{rel_data_path}/duration.json\", \"r\", encoding=\"utf-8\") as f:\n data_dict = json.load(f)\n durations = data_dict[\"duration\"]\n train_dataset = CustomDataset(\n train_dataset,\n durations=durations,\n preprocessed_mel=preprocessed_mel,\n mel_spec_module=mel_spec_module,\n **mel_spec_kwargs,\n )\n\n elif dataset_type == \"CustomDatasetPath\":\n try:\n train_dataset = load_from_disk(f\"{dataset_name}/raw\")\n except: # noqa: E722\n train_dataset = Dataset_.from_file(f\"{dataset_name}/raw.arrow\")\n\n with open(f\"{dataset_name}/duration.json\", \"r\", encoding=\"utf-8\") as f:\n data_dict = json.load(f)\n durations = data_dict[\"duration\"]\n train_dataset = CustomDataset(\n train_dataset, durations=durations, preprocessed_mel=preprocessed_mel, **mel_spec_kwargs\n )\n\n elif dataset_type == \"HFDataset\":\n print(\n \"Should manually modify the path of huggingface dataset to your need.\\n\"\n + \"May also the corresponding script cuz different dataset may have different format.\"\n )\n pre, post = dataset_name.split(\"_\")\n train_dataset = HFDataset(\n load_dataset(f\"{pre}/{pre}\", split=f\"train.{post}\", cache_dir=str(files(\"f5_tts\").joinpath(\"../../data\"))),\n )\n\n return train_dataset\n\n\n# collation\n\n\ndef collate_fn(batch):\n mel_specs = [item[\"mel_spec\"].squeeze(0) for item in batch]\n mel_lengths = torch.LongTensor([spec.shape[-1] for spec in mel_specs])\n max_mel_length = mel_lengths.amax()\n\n padded_mel_specs = []\n for spec in mel_specs:\n padding = (0, max_mel_length - spec.size(-1))\n padded_spec = F.pad(spec, padding, value=0)\n padded_mel_specs.append(padded_spec)\n\n mel_specs = torch.stack(padded_mel_specs)\n\n text = [item[\"text\"] for item in batch]\n text_lengths = torch.LongTensor([len(item) for item in text])","source_hash":"5ddfadf6e712d32b7c55a455c51078896ebcd239a8e60bf2d70ac7033f14f697","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.dataset.collate_fn","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.dataset.collate_fn#L309-L330","kind":"function","name":"collate_fn","path":"src/f5_tts/model_new/dataset.py","language":"python","start_line":309,"end_line":330,"context_start_line":289,"context_end_line":330,"code":" train_dataset = CustomDataset(\n train_dataset, durations=durations, preprocessed_mel=preprocessed_mel, **mel_spec_kwargs\n )\n\n elif dataset_type == \"HFDataset\":\n print(\n \"Should manually modify the path of huggingface dataset to your need.\\n\"\n + \"May also the corresponding script cuz different dataset may have different format.\"\n )\n pre, post = dataset_name.split(\"_\")\n train_dataset = HFDataset(\n load_dataset(f\"{pre}/{pre}\", split=f\"train.{post}\", cache_dir=str(files(\"f5_tts\").joinpath(\"../../data\"))),\n )\n\n return train_dataset\n\n\n# collation\n\n\ndef collate_fn(batch):\n mel_specs = [item[\"mel_spec\"].squeeze(0) for item in batch]\n mel_lengths = torch.LongTensor([spec.shape[-1] for spec in mel_specs])\n max_mel_length = mel_lengths.amax()\n\n padded_mel_specs = []\n for spec in mel_specs:\n padding = (0, max_mel_length - spec.size(-1))\n padded_spec = F.pad(spec, padding, value=0)\n padded_mel_specs.append(padded_spec)\n\n mel_specs = torch.stack(padded_mel_specs)\n\n text = [item[\"text\"] for item in batch]\n text_lengths = torch.LongTensor([len(item) for item in text])\n\n return dict(\n mel=mel_specs,\n mel_lengths=mel_lengths, # records for padding mask\n text=text,\n text_lengths=text_lengths,\n )","source_hash":"5ddfadf6e712d32b7c55a455c51078896ebcd239a8e60bf2d70ac7033f14f697","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.dataset.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.dataset.__init__#L175-L218","kind":"function","name":"__init__","path":"src/f5_tts/model_new/dataset.py","language":"python","start_line":175,"end_line":218,"context_start_line":155,"context_end_line":238,"code":" # to mel spectrogram\n mel_spec = self.mel_spectrogram(audio)\n mel_spec = mel_spec.squeeze(0) # '1 d t -> d t'\n\n return {\n \"mel_spec\": mel_spec,\n \"text\": text,\n }\n\n\n# Dynamic Batch Sampler\nclass DynamicBatchSampler(Sampler[list[int]]):\n \"\"\"Extension of Sampler that will do the following:\n 1. Change the batch size (essentially number of sequences)\n in a batch to ensure that the total number of frames are less\n than a certain threshold.\n 2. Make sure the padding efficiency in the batch is high.\n 3. Shuffle batches each epoch while maintaining reproducibility.\n \"\"\"\n\n def __init__(\n self, sampler: Sampler[int], frames_threshold: int, max_samples=0, random_seed=None, drop_residual: bool = False\n ):\n self.sampler = sampler\n self.frames_threshold = frames_threshold\n self.max_samples = max_samples\n self.random_seed = random_seed\n self.epoch = 0\n\n indices, batches = [], []\n data_source = self.sampler.data_source\n\n for idx in tqdm(\n self.sampler, desc=\"Sorting with sampler... if slow, check whether dataset is provided with duration\"\n ):\n indices.append((idx, data_source.get_frame_len(idx)))\n indices.sort(key=lambda elem: elem[1])\n\n batch = []\n batch_frames = 0\n for idx, frame_len in tqdm(\n indices, desc=f\"Creating dynamic batches with {frames_threshold} audio frames per gpu\"\n ):\n if batch_frames + frame_len <= self.frames_threshold and (max_samples == 0 or len(batch) < max_samples):\n batch.append(idx)\n batch_frames += frame_len\n else:\n if len(batch) > 0:\n batches.append(batch)\n if frame_len <= self.frames_threshold:\n batch = [idx]\n batch_frames = frame_len\n else:\n batch = []\n batch_frames = 0\n\n if not drop_residual and len(batch) > 0:\n batches.append(batch)\n\n del indices\n self.batches = batches\n\n # Ensure even batches with accelerate BatchSamplerShard cls under frame_per_batch setting\n self.drop_last = True\n\n def set_epoch(self, epoch: int) -> None:\n \"\"\"Sets the epoch for this sampler.\"\"\"\n self.epoch = epoch\n\n def __iter__(self):\n # Use both random_seed and epoch for deterministic but different shuffling per epoch\n if self.random_seed is not None:\n g = torch.Generator()\n g.manual_seed(self.random_seed + self.epoch)\n # Use PyTorch's random permutation for better reproducibility across PyTorch versions\n indices = torch.randperm(len(self.batches), generator=g).tolist()\n batches = [self.batches[i] for i in indices]\n else:\n batches = self.batches\n return iter(batches)\n\n def __len__(self):\n return len(self.batches)\n","source_hash":"5ddfadf6e712d32b7c55a455c51078896ebcd239a8e60bf2d70ac7033f14f697","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.dataset.get_frame_len","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.dataset.get_frame_len#L118-L123","kind":"function","name":"get_frame_len","path":"src/f5_tts/model_new/dataset.py","language":"python","start_line":118,"end_line":123,"context_start_line":98,"context_end_line":143,"code":" self.target_sample_rate = target_sample_rate\n self.hop_length = hop_length\n self.n_fft = n_fft\n self.win_length = win_length\n self.mel_spec_type = mel_spec_type\n self.preprocessed_mel = preprocessed_mel\n\n if not preprocessed_mel:\n self.mel_spectrogram = default(\n mel_spec_module,\n MelSpec(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n ),\n )\n\n def get_frame_len(self, index):\n if (\n self.durations is not None\n ): # Please make sure the separately provided durations are correct, otherwise 99.99% OOM\n return self.durations[index] * self.target_sample_rate / self.hop_length\n return self.data[index][\"duration\"] * self.target_sample_rate / self.hop_length\n\n def __len__(self):\n return len(self.data)\n\n def __getitem__(self, index):\n while True:\n row = self.data[index]\n audio_path = row[\"audio_path\"]\n text = row[\"text\"]\n duration = row[\"duration\"]\n\n # filter by given length\n if 0.3 <= duration <= 30:\n break # valid\n\n index = (index + 1) % len(self.data)\n\n if self.preprocessed_mel:\n mel_spec = torch.tensor(row[\"mel_spec\"])\n else:","source_hash":"5ddfadf6e712d32b7c55a455c51078896ebcd239a8e60bf2d70ac7033f14f697","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.dataset.__len__","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.dataset.__len__#L236-L237","kind":"function","name":"__len__","path":"src/f5_tts/model_new/dataset.py","language":"python","start_line":236,"end_line":237,"context_start_line":216,"context_end_line":257,"code":"\n # Ensure even batches with accelerate BatchSamplerShard cls under frame_per_batch setting\n self.drop_last = True\n\n def set_epoch(self, epoch: int) -> None:\n \"\"\"Sets the epoch for this sampler.\"\"\"\n self.epoch = epoch\n\n def __iter__(self):\n # Use both random_seed and epoch for deterministic but different shuffling per epoch\n if self.random_seed is not None:\n g = torch.Generator()\n g.manual_seed(self.random_seed + self.epoch)\n # Use PyTorch's random permutation for better reproducibility across PyTorch versions\n indices = torch.randperm(len(self.batches), generator=g).tolist()\n batches = [self.batches[i] for i in indices]\n else:\n batches = self.batches\n return iter(batches)\n\n def __len__(self):\n return len(self.batches)\n\n\n# Load dataset\n\n\ndef load_dataset(\n dataset_name: str,\n tokenizer: str = \"pinyin\",\n dataset_type: str = \"CustomDataset\",\n audio_type: str = \"raw\",\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n) -> CustomDataset | HFDataset:\n \"\"\"\n dataset_type - \"CustomDataset\" if you want to use tokenizer name and default data path to load for train_dataset\n - \"CustomDatasetPath\" if you just want to pass the full path to a preprocessed dataset without relying on tokenizer\n \"\"\"\n\n print(\"Loading dataset ...\")\n","source_hash":"5ddfadf6e712d32b7c55a455c51078896ebcd239a8e60bf2d70ac7033f14f697","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.dataset.__getitem__","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.dataset.__getitem__#L128-L162","kind":"function","name":"__getitem__","path":"src/f5_tts/model_new/dataset.py","language":"python","start_line":128,"end_line":162,"context_start_line":108,"context_end_line":182,"code":" MelSpec(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n ),\n )\n\n def get_frame_len(self, index):\n if (\n self.durations is not None\n ): # Please make sure the separately provided durations are correct, otherwise 99.99% OOM\n return self.durations[index] * self.target_sample_rate / self.hop_length\n return self.data[index][\"duration\"] * self.target_sample_rate / self.hop_length\n\n def __len__(self):\n return len(self.data)\n\n def __getitem__(self, index):\n while True:\n row = self.data[index]\n audio_path = row[\"audio_path\"]\n text = row[\"text\"]\n duration = row[\"duration\"]\n\n # filter by given length\n if 0.3 <= duration <= 30:\n break # valid\n\n index = (index + 1) % len(self.data)\n\n if self.preprocessed_mel:\n mel_spec = torch.tensor(row[\"mel_spec\"])\n else:\n audio, source_sample_rate = torchaudio.load(audio_path)\n\n # make sure mono input\n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n\n # resample if necessary\n if source_sample_rate != self.target_sample_rate:\n resampler = torchaudio.transforms.Resample(source_sample_rate, self.target_sample_rate)\n audio = resampler(audio)\n\n # to mel spectrogram\n mel_spec = self.mel_spectrogram(audio)\n mel_spec = mel_spec.squeeze(0) # '1 d t -> d t'\n\n return {\n \"mel_spec\": mel_spec,\n \"text\": text,\n }\n\n\n# Dynamic Batch Sampler\nclass DynamicBatchSampler(Sampler[list[int]]):\n \"\"\"Extension of Sampler that will do the following:\n 1. Change the batch size (essentially number of sequences)\n in a batch to ensure that the total number of frames are less\n than a certain threshold.\n 2. Make sure the padding efficiency in the batch is high.\n 3. Shuffle batches each epoch while maintaining reproducibility.\n \"\"\"\n\n def __init__(\n self, sampler: Sampler[int], frames_threshold: int, max_samples=0, random_seed=None, drop_residual: bool = False\n ):\n self.sampler = sampler\n self.frames_threshold = frames_threshold\n self.max_samples = max_samples\n self.random_seed = random_seed\n self.epoch = 0","source_hash":"5ddfadf6e712d32b7c55a455c51078896ebcd239a8e60bf2d70ac7033f14f697","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.dataset.set_epoch","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.dataset.set_epoch#L220-L222","kind":"function","name":"set_epoch","path":"src/f5_tts/model_new/dataset.py","language":"python","start_line":220,"end_line":222,"context_start_line":200,"context_end_line":242,"code":" batch_frames += frame_len\n else:\n if len(batch) > 0:\n batches.append(batch)\n if frame_len <= self.frames_threshold:\n batch = [idx]\n batch_frames = frame_len\n else:\n batch = []\n batch_frames = 0\n\n if not drop_residual and len(batch) > 0:\n batches.append(batch)\n\n del indices\n self.batches = batches\n\n # Ensure even batches with accelerate BatchSamplerShard cls under frame_per_batch setting\n self.drop_last = True\n\n def set_epoch(self, epoch: int) -> None:\n \"\"\"Sets the epoch for this sampler.\"\"\"\n self.epoch = epoch\n\n def __iter__(self):\n # Use both random_seed and epoch for deterministic but different shuffling per epoch\n if self.random_seed is not None:\n g = torch.Generator()\n g.manual_seed(self.random_seed + self.epoch)\n # Use PyTorch's random permutation for better reproducibility across PyTorch versions\n indices = torch.randperm(len(self.batches), generator=g).tolist()\n batches = [self.batches[i] for i in indices]\n else:\n batches = self.batches\n return iter(batches)\n\n def __len__(self):\n return len(self.batches)\n\n\n# Load dataset\n\n","source_hash":"5ddfadf6e712d32b7c55a455c51078896ebcd239a8e60bf2d70ac7033f14f697","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.dataset.__iter__","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.dataset.__iter__#L224-L234","kind":"function","name":"__iter__","path":"src/f5_tts/model_new/dataset.py","language":"python","start_line":224,"end_line":234,"context_start_line":204,"context_end_line":254,"code":" if frame_len <= self.frames_threshold:\n batch = [idx]\n batch_frames = frame_len\n else:\n batch = []\n batch_frames = 0\n\n if not drop_residual and len(batch) > 0:\n batches.append(batch)\n\n del indices\n self.batches = batches\n\n # Ensure even batches with accelerate BatchSamplerShard cls under frame_per_batch setting\n self.drop_last = True\n\n def set_epoch(self, epoch: int) -> None:\n \"\"\"Sets the epoch for this sampler.\"\"\"\n self.epoch = epoch\n\n def __iter__(self):\n # Use both random_seed and epoch for deterministic but different shuffling per epoch\n if self.random_seed is not None:\n g = torch.Generator()\n g.manual_seed(self.random_seed + self.epoch)\n # Use PyTorch's random permutation for better reproducibility across PyTorch versions\n indices = torch.randperm(len(self.batches), generator=g).tolist()\n batches = [self.batches[i] for i in indices]\n else:\n batches = self.batches\n return iter(batches)\n\n def __len__(self):\n return len(self.batches)\n\n\n# Load dataset\n\n\ndef load_dataset(\n dataset_name: str,\n tokenizer: str = \"pinyin\",\n dataset_type: str = \"CustomDataset\",\n audio_type: str = \"raw\",\n mel_spec_module: nn.Module | None = None,\n mel_spec_kwargs: dict = dict(),\n) -> CustomDataset | HFDataset:\n \"\"\"\n dataset_type - \"CustomDataset\" if you want to use tokenizer name and default data path to load for train_dataset\n - \"CustomDatasetPath\" if you just want to pass the full path to a preprocessed dataset without relying on tokenizer\n \"\"\"","source_hash":"5ddfadf6e712d32b7c55a455c51078896ebcd239a8e60bf2d70ac7033f14f697","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils","uri":"program://DMOSpeech2/module/src.f5_tts.model_new.utils#L1-L220","kind":"module","name":"src.f5_tts.model_new.utils","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":1,"end_line":220,"context_start_line":1,"context_end_line":220,"code":"from __future__ import annotations\n\nimport os\nimport random\nfrom collections import defaultdict\nfrom importlib.resources import files\n\nimport jieba\nimport torch\nfrom pypinyin import Style, lazy_pinyin\nfrom torch.nn.utils.rnn import pad_sequence\n\n\n# seed everything\n\n\ndef seed_everything(seed=0):\n random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.cuda.manual_seed_all(seed)\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n\n\n# helpers\n\n\ndef exists(v):\n return v is not None\n\n\ndef default(v, d):\n return v if exists(v) else d\n\n\ndef is_package_available(package_name: str) -> bool:\n try:\n import importlib\n\n package_exists = importlib.util.find_spec(package_name) is not None\n return package_exists\n except Exception:\n return False\n\n\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821\n if not exists(length):\n length = t.amax()\n\n seq = torch.arange(length, device=t.device)\n return seq[None, :] < t[:, None]\n\n\ndef mask_from_start_end_indices(seq_len: int[\"b\"], start: int[\"b\"], end: int[\"b\"]): # noqa: F722 F821\n max_seq_len = seq_len.max().item()\n seq = torch.arange(max_seq_len, device=start.device).long()\n start_mask = seq[None, :] >= start[:, None]\n end_mask = seq[None, :] < end[:, None]\n return start_mask & end_mask\n\n\ndef mask_from_frac_lengths(seq_len: int[\"b\"], frac_lengths: float[\"b\"]): # noqa: F722 F821\n lengths = (frac_lengths * seq_len).long()\n max_start = seq_len - lengths\n\n rand = torch.rand_like(frac_lengths)\n start = (max_start * rand).long().clamp(min=0)\n end = start + lengths\n\n return mask_from_start_end_indices(seq_len, start, end)\n\n\ndef maybe_masked_mean(t: float[\"b n d\"], mask: bool[\"b n\"] = None) -> float[\"b d\"]: # noqa: F722\n if not exists(mask):\n return t.mean(dim=1)\n\n t = torch.where(mask[:, :, None], t, torch.tensor(0.0, device=t.device))\n num = t.sum(dim=1)\n den = mask.float().sum(dim=1)\n\n return num / den.clamp(min=1.0)\n\n\n# simple utf-8 tokenizer, since paper went character based\ndef list_str_to_tensor(text: list[str], padding_value=-1) -> int[\"b nt\"]: # noqa: F722\n list_tensors = [torch.tensor([*bytes(t, \"UTF-8\")]) for t in text] # ByT5 style\n text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# char tokenizer, based on custom dataset's extracted .txt file\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n) -> int[\"b nt\"]: # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# Get tokenizer\n\n\ndef get_tokenizer(dataset_name, tokenizer: str = \"pinyin\"):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file\n - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n if tokenizer in [\"pinyin\", \"char\"]:\n tokenizer_path = os.path.join(files(\"f5_tts\").joinpath(\"../../data\"), f\"{dataset_name}_{tokenizer}/vocab.txt\")\n with open(tokenizer_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n assert vocab_char_map[\" \"] == 0, \"make sure space is of idx 0 in vocab.txt, cuz 0 is used for unknown char\"\n\n elif tokenizer == \"byte\":\n vocab_char_map = None\n vocab_size = 256\n\n elif tokenizer == \"custom\":\n with open(dataset_name, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n\n return vocab_char_map, vocab_size\n\n\n# convert char to pinyin\n\n\ndef convert_char_to_pinyin(text_list, polyphone=True):\n if jieba.dt.initialized is False:\n jieba.default_logger.setLevel(50) # CRITICAL\n jieba.initialize()\n\n final_text_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return (\n \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n )\n\n for text in text_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)\n elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)\n elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_text_list.append(char_list)\n\n return final_text_list\n\n\n# filter func for dirty data with many repetitions\n\n\ndef repetition_found(text, length=2, tolerance=10):\n pattern_count = defaultdict(int)\n for i in range(len(text) - length + 1):\n pattern = text[i : i + length]\n pattern_count[pattern] += 1\n for pattern, count in pattern_count.items():\n if count > tolerance:\n return True\n return False\n\n\n# get the empirically pruned step for sampling\n\n\ndef get_epss_timesteps(n, device, dtype):\n dt = 1 / 32\n predefined_timesteps = {\n 5: [0, 2, 4, 8, 16, 32],\n 6: [0, 2, 4, 6, 8, 16, 32],\n 7: [0, 2, 4, 6, 8, 16, 24, 32],\n 10: [0, 2, 4, 6, 8, 12, 16, 20, 24, 28, 32],\n 12: [0, 2, 4, 6, 8, 10, 12, 14, 16, 20, 24, 28, 32],\n 16: [0, 1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 20, 24, 28, 32],\n }\n t = predefined_timesteps.get(n, [])\n if not t:\n return torch.linspace(0, 1, n + 1, device=device, dtype=dtype)\n return dt * torch.tensor(t, device=device, dtype=dtype)","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.seed_everything","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.seed_everything#L17-L24","kind":"function","name":"seed_everything","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":17,"end_line":24,"context_start_line":1,"context_end_line":44,"code":"from __future__ import annotations\n\nimport os\nimport random\nfrom collections import defaultdict\nfrom importlib.resources import files\n\nimport jieba\nimport torch\nfrom pypinyin import Style, lazy_pinyin\nfrom torch.nn.utils.rnn import pad_sequence\n\n\n# seed everything\n\n\ndef seed_everything(seed=0):\n random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.cuda.manual_seed_all(seed)\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n\n\n# helpers\n\n\ndef exists(v):\n return v is not None\n\n\ndef default(v, d):\n return v if exists(v) else d\n\n\ndef is_package_available(package_name: str) -> bool:\n try:\n import importlib\n\n package_exists = importlib.util.find_spec(package_name) is not None\n return package_exists\n except Exception:","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.exists","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.exists#L30-L31","kind":"function","name":"exists","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":30,"end_line":31,"context_start_line":10,"context_end_line":51,"code":"from pypinyin import Style, lazy_pinyin\nfrom torch.nn.utils.rnn import pad_sequence\n\n\n# seed everything\n\n\ndef seed_everything(seed=0):\n random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.cuda.manual_seed_all(seed)\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n\n\n# helpers\n\n\ndef exists(v):\n return v is not None\n\n\ndef default(v, d):\n return v if exists(v) else d\n\n\ndef is_package_available(package_name: str) -> bool:\n try:\n import importlib\n\n package_exists = importlib.util.find_spec(package_name) is not None\n return package_exists\n except Exception:\n return False\n\n\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.default","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.default#L34-L35","kind":"function","name":"default","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":34,"end_line":35,"context_start_line":14,"context_end_line":55,"code":"# seed everything\n\n\ndef seed_everything(seed=0):\n random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.cuda.manual_seed_all(seed)\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n\n\n# helpers\n\n\ndef exists(v):\n return v is not None\n\n\ndef default(v, d):\n return v if exists(v) else d\n\n\ndef is_package_available(package_name: str) -> bool:\n try:\n import importlib\n\n package_exists = importlib.util.find_spec(package_name) is not None\n return package_exists\n except Exception:\n return False\n\n\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821\n if not exists(length):\n length = t.amax()\n\n seq = torch.arange(length, device=t.device)","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.is_package_available","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.is_package_available#L38-L45","kind":"function","name":"is_package_available","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":38,"end_line":45,"context_start_line":18,"context_end_line":65,"code":" random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)\n torch.cuda.manual_seed_all(seed)\n torch.backends.cudnn.deterministic = True\n torch.backends.cudnn.benchmark = False\n\n\n# helpers\n\n\ndef exists(v):\n return v is not None\n\n\ndef default(v, d):\n return v if exists(v) else d\n\n\ndef is_package_available(package_name: str) -> bool:\n try:\n import importlib\n\n package_exists = importlib.util.find_spec(package_name) is not None\n return package_exists\n except Exception:\n return False\n\n\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821\n if not exists(length):\n length = t.amax()\n\n seq = torch.arange(length, device=t.device)\n return seq[None, :] < t[:, None]\n\n\ndef mask_from_start_end_indices(seq_len: int[\"b\"], start: int[\"b\"], end: int[\"b\"]): # noqa: F722 F821\n max_seq_len = seq_len.max().item()\n seq = torch.arange(max_seq_len, device=start.device).long()\n start_mask = seq[None, :] >= start[:, None]\n end_mask = seq[None, :] < end[:, None]\n return start_mask & end_mask\n","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.lens_to_mask","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.lens_to_mask#L51-L56","kind":"function","name":"lens_to_mask","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":51,"end_line":56,"context_start_line":31,"context_end_line":76,"code":" return v is not None\n\n\ndef default(v, d):\n return v if exists(v) else d\n\n\ndef is_package_available(package_name: str) -> bool:\n try:\n import importlib\n\n package_exists = importlib.util.find_spec(package_name) is not None\n return package_exists\n except Exception:\n return False\n\n\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821\n if not exists(length):\n length = t.amax()\n\n seq = torch.arange(length, device=t.device)\n return seq[None, :] < t[:, None]\n\n\ndef mask_from_start_end_indices(seq_len: int[\"b\"], start: int[\"b\"], end: int[\"b\"]): # noqa: F722 F821\n max_seq_len = seq_len.max().item()\n seq = torch.arange(max_seq_len, device=start.device).long()\n start_mask = seq[None, :] >= start[:, None]\n end_mask = seq[None, :] < end[:, None]\n return start_mask & end_mask\n\n\ndef mask_from_frac_lengths(seq_len: int[\"b\"], frac_lengths: float[\"b\"]): # noqa: F722 F821\n lengths = (frac_lengths * seq_len).long()\n max_start = seq_len - lengths\n\n rand = torch.rand_like(frac_lengths)\n start = (max_start * rand).long().clamp(min=0)\n end = start + lengths\n\n return mask_from_start_end_indices(seq_len, start, end)\n","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.mask_from_start_end_indices","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.mask_from_start_end_indices#L59-L64","kind":"function","name":"mask_from_start_end_indices","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":59,"end_line":64,"context_start_line":39,"context_end_line":84,"code":" try:\n import importlib\n\n package_exists = importlib.util.find_spec(package_name) is not None\n return package_exists\n except Exception:\n return False\n\n\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821\n if not exists(length):\n length = t.amax()\n\n seq = torch.arange(length, device=t.device)\n return seq[None, :] < t[:, None]\n\n\ndef mask_from_start_end_indices(seq_len: int[\"b\"], start: int[\"b\"], end: int[\"b\"]): # noqa: F722 F821\n max_seq_len = seq_len.max().item()\n seq = torch.arange(max_seq_len, device=start.device).long()\n start_mask = seq[None, :] >= start[:, None]\n end_mask = seq[None, :] < end[:, None]\n return start_mask & end_mask\n\n\ndef mask_from_frac_lengths(seq_len: int[\"b\"], frac_lengths: float[\"b\"]): # noqa: F722 F821\n lengths = (frac_lengths * seq_len).long()\n max_start = seq_len - lengths\n\n rand = torch.rand_like(frac_lengths)\n start = (max_start * rand).long().clamp(min=0)\n end = start + lengths\n\n return mask_from_start_end_indices(seq_len, start, end)\n\n\ndef maybe_masked_mean(t: float[\"b n d\"], mask: bool[\"b n\"] = None) -> float[\"b d\"]: # noqa: F722\n if not exists(mask):\n return t.mean(dim=1)\n\n t = torch.where(mask[:, :, None], t, torch.tensor(0.0, device=t.device))\n num = t.sum(dim=1)\n den = mask.float().sum(dim=1)","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.mask_from_frac_lengths","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.mask_from_frac_lengths#L67-L75","kind":"function","name":"mask_from_frac_lengths","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":67,"end_line":75,"context_start_line":47,"context_end_line":95,"code":"\n# tensor helpers\n\n\ndef lens_to_mask(t: int[\"b\"], length: int | None = None) -> bool[\"b n\"]: # noqa: F722 F821\n if not exists(length):\n length = t.amax()\n\n seq = torch.arange(length, device=t.device)\n return seq[None, :] < t[:, None]\n\n\ndef mask_from_start_end_indices(seq_len: int[\"b\"], start: int[\"b\"], end: int[\"b\"]): # noqa: F722 F821\n max_seq_len = seq_len.max().item()\n seq = torch.arange(max_seq_len, device=start.device).long()\n start_mask = seq[None, :] >= start[:, None]\n end_mask = seq[None, :] < end[:, None]\n return start_mask & end_mask\n\n\ndef mask_from_frac_lengths(seq_len: int[\"b\"], frac_lengths: float[\"b\"]): # noqa: F722 F821\n lengths = (frac_lengths * seq_len).long()\n max_start = seq_len - lengths\n\n rand = torch.rand_like(frac_lengths)\n start = (max_start * rand).long().clamp(min=0)\n end = start + lengths\n\n return mask_from_start_end_indices(seq_len, start, end)\n\n\ndef maybe_masked_mean(t: float[\"b n d\"], mask: bool[\"b n\"] = None) -> float[\"b d\"]: # noqa: F722\n if not exists(mask):\n return t.mean(dim=1)\n\n t = torch.where(mask[:, :, None], t, torch.tensor(0.0, device=t.device))\n num = t.sum(dim=1)\n den = mask.float().sum(dim=1)\n\n return num / den.clamp(min=1.0)\n\n\n# simple utf-8 tokenizer, since paper went character based\ndef list_str_to_tensor(text: list[str], padding_value=-1) -> int[\"b nt\"]: # noqa: F722\n list_tensors = [torch.tensor([*bytes(t, \"UTF-8\")]) for t in text] # ByT5 style\n text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.maybe_masked_mean","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.maybe_masked_mean#L78-L86","kind":"function","name":"maybe_masked_mean","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":78,"end_line":86,"context_start_line":58,"context_end_line":106,"code":"\ndef mask_from_start_end_indices(seq_len: int[\"b\"], start: int[\"b\"], end: int[\"b\"]): # noqa: F722 F821\n max_seq_len = seq_len.max().item()\n seq = torch.arange(max_seq_len, device=start.device).long()\n start_mask = seq[None, :] >= start[:, None]\n end_mask = seq[None, :] < end[:, None]\n return start_mask & end_mask\n\n\ndef mask_from_frac_lengths(seq_len: int[\"b\"], frac_lengths: float[\"b\"]): # noqa: F722 F821\n lengths = (frac_lengths * seq_len).long()\n max_start = seq_len - lengths\n\n rand = torch.rand_like(frac_lengths)\n start = (max_start * rand).long().clamp(min=0)\n end = start + lengths\n\n return mask_from_start_end_indices(seq_len, start, end)\n\n\ndef maybe_masked_mean(t: float[\"b n d\"], mask: bool[\"b n\"] = None) -> float[\"b d\"]: # noqa: F722\n if not exists(mask):\n return t.mean(dim=1)\n\n t = torch.where(mask[:, :, None], t, torch.tensor(0.0, device=t.device))\n num = t.sum(dim=1)\n den = mask.float().sum(dim=1)\n\n return num / den.clamp(min=1.0)\n\n\n# simple utf-8 tokenizer, since paper went character based\ndef list_str_to_tensor(text: list[str], padding_value=-1) -> int[\"b nt\"]: # noqa: F722\n list_tensors = [torch.tensor([*bytes(t, \"UTF-8\")]) for t in text] # ByT5 style\n text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# char tokenizer, based on custom dataset's extracted .txt file\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n) -> int[\"b nt\"]: # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.list_str_to_tensor","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.list_str_to_tensor#L90-L93","kind":"function","name":"list_str_to_tensor","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":90,"end_line":93,"context_start_line":70,"context_end_line":113,"code":"\n rand = torch.rand_like(frac_lengths)\n start = (max_start * rand).long().clamp(min=0)\n end = start + lengths\n\n return mask_from_start_end_indices(seq_len, start, end)\n\n\ndef maybe_masked_mean(t: float[\"b n d\"], mask: bool[\"b n\"] = None) -> float[\"b d\"]: # noqa: F722\n if not exists(mask):\n return t.mean(dim=1)\n\n t = torch.where(mask[:, :, None], t, torch.tensor(0.0, device=t.device))\n num = t.sum(dim=1)\n den = mask.float().sum(dim=1)\n\n return num / den.clamp(min=1.0)\n\n\n# simple utf-8 tokenizer, since paper went character based\ndef list_str_to_tensor(text: list[str], padding_value=-1) -> int[\"b nt\"]: # noqa: F722\n list_tensors = [torch.tensor([*bytes(t, \"UTF-8\")]) for t in text] # ByT5 style\n text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# char tokenizer, based on custom dataset's extracted .txt file\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n) -> int[\"b nt\"]: # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# Get tokenizer\n\n\ndef get_tokenizer(dataset_name, tokenizer: str = \"pinyin\"):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.list_str_to_idx","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.list_str_to_idx#L97-L104","kind":"function","name":"list_str_to_idx","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":97,"end_line":104,"context_start_line":77,"context_end_line":124,"code":"\ndef maybe_masked_mean(t: float[\"b n d\"], mask: bool[\"b n\"] = None) -> float[\"b d\"]: # noqa: F722\n if not exists(mask):\n return t.mean(dim=1)\n\n t = torch.where(mask[:, :, None], t, torch.tensor(0.0, device=t.device))\n num = t.sum(dim=1)\n den = mask.float().sum(dim=1)\n\n return num / den.clamp(min=1.0)\n\n\n# simple utf-8 tokenizer, since paper went character based\ndef list_str_to_tensor(text: list[str], padding_value=-1) -> int[\"b nt\"]: # noqa: F722\n list_tensors = [torch.tensor([*bytes(t, \"UTF-8\")]) for t in text] # ByT5 style\n text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# char tokenizer, based on custom dataset's extracted .txt file\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n) -> int[\"b nt\"]: # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# Get tokenizer\n\n\ndef get_tokenizer(dataset_name, tokenizer: str = \"pinyin\"):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file\n - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n if tokenizer in [\"pinyin\", \"char\"]:\n tokenizer_path = os.path.join(files(\"f5_tts\").joinpath(\"../../data\"), f\"{dataset_name}_{tokenizer}/vocab.txt\")\n with open(tokenizer_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.get_tokenizer","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.get_tokenizer#L110-L140","kind":"function","name":"get_tokenizer","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":110,"end_line":140,"context_start_line":90,"context_end_line":160,"code":"def list_str_to_tensor(text: list[str], padding_value=-1) -> int[\"b nt\"]: # noqa: F722\n list_tensors = [torch.tensor([*bytes(t, \"UTF-8\")]) for t in text] # ByT5 style\n text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# char tokenizer, based on custom dataset's extracted .txt file\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n) -> int[\"b nt\"]: # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return text\n\n\n# Get tokenizer\n\n\ndef get_tokenizer(dataset_name, tokenizer: str = \"pinyin\"):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file\n - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n if tokenizer in [\"pinyin\", \"char\"]:\n tokenizer_path = os.path.join(files(\"f5_tts\").joinpath(\"../../data\"), f\"{dataset_name}_{tokenizer}/vocab.txt\")\n with open(tokenizer_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n assert vocab_char_map[\" \"] == 0, \"make sure space is of idx 0 in vocab.txt, cuz 0 is used for unknown char\"\n\n elif tokenizer == \"byte\":\n vocab_char_map = None\n vocab_size = 256\n\n elif tokenizer == \"custom\":\n with open(dataset_name, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n\n return vocab_char_map, vocab_size\n\n\n# convert char to pinyin\n\n\ndef convert_char_to_pinyin(text_list, polyphone=True):\n if jieba.dt.initialized is False:\n jieba.default_logger.setLevel(50) # CRITICAL\n jieba.initialize()\n\n final_text_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return (\n \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n )\n","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.convert_char_to_pinyin","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.convert_char_to_pinyin#L146-L187","kind":"function","name":"convert_char_to_pinyin","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":146,"end_line":187,"context_start_line":126,"context_end_line":207,"code":" vocab_size = len(vocab_char_map)\n assert vocab_char_map[\" \"] == 0, \"make sure space is of idx 0 in vocab.txt, cuz 0 is used for unknown char\"\n\n elif tokenizer == \"byte\":\n vocab_char_map = None\n vocab_size = 256\n\n elif tokenizer == \"custom\":\n with open(dataset_name, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n\n return vocab_char_map, vocab_size\n\n\n# convert char to pinyin\n\n\ndef convert_char_to_pinyin(text_list, polyphone=True):\n if jieba.dt.initialized is False:\n jieba.default_logger.setLevel(50) # CRITICAL\n jieba.initialize()\n\n final_text_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return (\n \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n )\n\n for text in text_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)\n elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)\n elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_text_list.append(char_list)\n\n return final_text_list\n\n\n# filter func for dirty data with many repetitions\n\n\ndef repetition_found(text, length=2, tolerance=10):\n pattern_count = defaultdict(int)\n for i in range(len(text) - length + 1):\n pattern = text[i : i + length]\n pattern_count[pattern] += 1\n for pattern, count in pattern_count.items():\n if count > tolerance:\n return True\n return False\n\n\n# get the empirically pruned step for sampling\n\n\ndef get_epss_timesteps(n, device, dtype):","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.repetition_found","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.repetition_found#L193-L201","kind":"function","name":"repetition_found","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":193,"end_line":201,"context_start_line":173,"context_end_line":220,"code":" if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)\n elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_text_list.append(char_list)\n\n return final_text_list\n\n\n# filter func for dirty data with many repetitions\n\n\ndef repetition_found(text, length=2, tolerance=10):\n pattern_count = defaultdict(int)\n for i in range(len(text) - length + 1):\n pattern = text[i : i + length]\n pattern_count[pattern] += 1\n for pattern, count in pattern_count.items():\n if count > tolerance:\n return True\n return False\n\n\n# get the empirically pruned step for sampling\n\n\ndef get_epss_timesteps(n, device, dtype):\n dt = 1 / 32\n predefined_timesteps = {\n 5: [0, 2, 4, 8, 16, 32],\n 6: [0, 2, 4, 6, 8, 16, 32],\n 7: [0, 2, 4, 6, 8, 16, 24, 32],\n 10: [0, 2, 4, 6, 8, 12, 16, 20, 24, 28, 32],\n 12: [0, 2, 4, 6, 8, 10, 12, 14, 16, 20, 24, 28, 32],\n 16: [0, 1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 20, 24, 28, 32],\n }\n t = predefined_timesteps.get(n, [])\n if not t:\n return torch.linspace(0, 1, n + 1, device=device, dtype=dtype)\n return dt * torch.tensor(t, device=device, dtype=dtype)","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.get_epss_timesteps","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.get_epss_timesteps#L207-L220","kind":"function","name":"get_epss_timesteps","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":207,"end_line":220,"context_start_line":187,"context_end_line":220,"code":" return final_text_list\n\n\n# filter func for dirty data with many repetitions\n\n\ndef repetition_found(text, length=2, tolerance=10):\n pattern_count = defaultdict(int)\n for i in range(len(text) - length + 1):\n pattern = text[i : i + length]\n pattern_count[pattern] += 1\n for pattern, count in pattern_count.items():\n if count > tolerance:\n return True\n return False\n\n\n# get the empirically pruned step for sampling\n\n\ndef get_epss_timesteps(n, device, dtype):\n dt = 1 / 32\n predefined_timesteps = {\n 5: [0, 2, 4, 8, 16, 32],\n 6: [0, 2, 4, 6, 8, 16, 32],\n 7: [0, 2, 4, 6, 8, 16, 24, 32],\n 10: [0, 2, 4, 6, 8, 12, 16, 20, 24, 28, 32],\n 12: [0, 2, 4, 6, 8, 10, 12, 14, 16, 20, 24, 28, 32],\n 16: [0, 1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 20, 24, 28, 32],\n }\n t = predefined_timesteps.get(n, [])\n if not t:\n return torch.linspace(0, 1, n + 1, device=device, dtype=dtype)\n return dt * torch.tensor(t, device=device, dtype=dtype)","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.utils.is_chinese","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.utils.is_chinese#L156-L159","kind":"function","name":"is_chinese","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":156,"end_line":159,"context_start_line":136,"context_end_line":179,"code":" for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n\n return vocab_char_map, vocab_size\n\n\n# convert char to pinyin\n\n\ndef convert_char_to_pinyin(text_list, polyphone=True):\n if jieba.dt.initialized is False:\n jieba.default_logger.setLevel(50) # CRITICAL\n jieba.initialize()\n\n final_text_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return (\n \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n )\n\n for text in text_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)\n elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules","uri":"program://DMOSpeech2/module/src.f5_tts.model_new.modules#L1-L784","kind":"module","name":"src.f5_tts.model_new.modules","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":1,"end_line":784,"context_start_line":1,"context_end_line":784,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n# flake8: noqa\n\nfrom __future__ import annotations\n\nimport math\nfrom typing import Optional\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom librosa.filters import mel as librosa_mel_fn\nfrom torch import nn\nfrom x_transformers.x_transformers import apply_rotary_pos_emb\n\nfrom f5_tts.model_new.utils import is_package_available\n\n\n# raw wav to mel spec\n\n\nmel_basis_cache = {}\nhann_window_cache = {}\n\n\ndef get_bigvgan_mel_spectrogram(\n waveform,\n n_fft=1024,\n n_mel_channels=100,\n target_sample_rate=24000,\n hop_length=256,\n win_length=1024,\n fmin=0,\n fmax=None,\n center=False,\n): # Copy from https://github.com/NVIDIA/BigVGAN/tree/main\n device = waveform.device\n key = f\"{n_fft}_{n_mel_channels}_{target_sample_rate}_{hop_length}_{win_length}_{fmin}_{fmax}_{device}\"\n\n if key not in mel_basis_cache:\n mel = librosa_mel_fn(sr=target_sample_rate, n_fft=n_fft, n_mels=n_mel_channels, fmin=fmin, fmax=fmax)\n mel_basis_cache[key] = torch.from_numpy(mel).float().to(device) # TODO: why they need .float()?\n hann_window_cache[key] = torch.hann_window(win_length).to(device)\n\n mel_basis = mel_basis_cache[key]\n hann_window = hann_window_cache[key]\n\n padding = (n_fft - hop_length) // 2\n waveform = torch.nn.functional.pad(waveform.unsqueeze(1), (padding, padding), mode=\"reflect\").squeeze(1)\n\n spec = torch.stft(\n waveform,\n n_fft,\n hop_length=hop_length,\n win_length=win_length,\n window=hann_window,\n center=center,\n pad_mode=\"reflect\",\n normalized=False,\n onesided=True,\n return_complex=True,\n )\n spec = torch.sqrt(torch.view_as_real(spec).pow(2).sum(-1) + 1e-9)\n\n mel_spec = torch.matmul(mel_basis, spec)\n mel_spec = torch.log(torch.clamp(mel_spec, min=1e-5))\n\n return mel_spec\n\n\ndef get_vocos_mel_spectrogram(\n waveform,\n n_fft=1024,\n n_mel_channels=100,\n target_sample_rate=24000,\n hop_length=256,\n win_length=1024,\n):\n mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=target_sample_rate,\n n_fft=n_fft,\n win_length=win_length,\n hop_length=hop_length,\n n_mels=n_mel_channels,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(waveform.device)\n if len(waveform.shape) == 3:\n waveform = waveform.squeeze(1) # 'b 1 nw -> b nw'\n\n assert len(waveform.shape) == 2\n\n mel = mel_stft(waveform)\n mel = mel.clamp(min=1e-5).log()\n return mel\n\n\nclass MelSpec(nn.Module):\n def __init__(\n self,\n n_fft=1024,\n hop_length=256,\n win_length=1024,\n n_mel_channels=100,\n target_sample_rate=24_000,\n mel_spec_type=\"vocos\",\n ):\n super().__init__()\n assert mel_spec_type in [\"vocos\", \"bigvgan\"], print(\"We only support two extract mel backend: vocos or bigvgan\")\n\n self.n_fft = n_fft\n self.hop_length = hop_length\n self.win_length = win_length\n self.n_mel_channels = n_mel_channels\n self.target_sample_rate = target_sample_rate\n\n if mel_spec_type == \"vocos\":\n self.extractor = get_vocos_mel_spectrogram\n elif mel_spec_type == \"bigvgan\":\n self.extractor = get_bigvgan_mel_spectrogram\n\n self.register_buffer(\"dummy\", torch.tensor(0), persistent=False)\n\n def forward(self, wav):\n if self.dummy.device != wav.device:\n self.to(wav.device)\n\n mel = self.extractor(\n waveform=wav,\n n_fft=self.n_fft,\n n_mel_channels=self.n_mel_channels,\n target_sample_rate=self.target_sample_rate,\n hop_length=self.hop_length,\n win_length=self.win_length,\n )\n\n return mel\n\n\n# sinusoidal position embedding\n\n\nclass SinusPositionEmbedding(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.dim = dim\n\n def forward(self, x, scale=1000):\n device = x.device\n half_dim = self.dim // 2\n emb = math.log(10000) / (half_dim - 1)\n emb = torch.exp(torch.arange(half_dim, device=device).float() * -emb)\n emb = scale * x.unsqueeze(1) * emb.unsqueeze(0)\n emb = torch.cat((emb.sin(), emb.cos()), dim=-1)\n return emb\n\n\n# convolutional position embedding\n\n\nclass ConvPositionEmbedding(nn.Module):\n def __init__(self, dim, kernel_size=31, groups=16):\n super().__init__()\n assert kernel_size % 2 != 0\n self.conv1d = nn.Sequential(\n nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),\n nn.Mish(),\n nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),\n nn.Mish(),\n )\n\n def forward(self, x: float[\"b n d\"], mask: bool[\"b n\"] | None = None):\n if mask is not None:\n mask = mask[..., None]\n x = x.masked_fill(~mask, 0.0)\n\n x = x.permute(0, 2, 1)\n x = self.conv1d(x)\n out = x.permute(0, 2, 1)\n\n if mask is not None:\n out = out.masked_fill(~mask, 0.0)\n\n return out\n\n\n# rotary positional embedding related\n\n\ndef precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0):\n # proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning\n # has some connection to NTK literature\n # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/\n # https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py\n theta *= theta_rescale_factor ** (dim / (dim - 2))\n freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))\n t = torch.arange(end, device=freqs.device) # type: ignore\n freqs = torch.outer(t, freqs).float() # type: ignore\n freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef get_pos_embed_indices(start, length, max_pos, scale=1.0):\n # length = length if isinstance(length, int) else length.max()\n scale = scale * torch.ones_like(start, dtype=torch.float32) # in case scale is a scalar\n pos = (\n start.unsqueeze(1)\n + (torch.arange(length, device=start.device, dtype=torch.float32).unsqueeze(0) * scale.unsqueeze(1)).long()\n )\n # avoid extra long error.\n pos = torch.where(pos < max_pos, pos, max_pos - 1)\n return pos\n\n\n# Global Response Normalization layer (Instance Normalization ?)\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.gamma = nn.Parameter(torch.zeros(1, 1, dim))\n self.beta = nn.Parameter(torch.zeros(1, 1, dim))\n\n def forward(self, x):\n Gx = torch.norm(x, p=2, dim=1, keepdim=True)\n Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)\n return self.gamma * (x * Nx) + self.beta + x\n\n\n# ConvNeXt-V2 Block https://github.com/facebookresearch/ConvNeXt-V2/blob/main/models/convnextv2.py\n# ref: https://github.com/bfs18/e2_tts/blob/main/rfwave/modules.py#L108\n\n\nclass ConvNeXtV2Block(nn.Module):\n def __init__(\n self,\n dim: int,\n intermediate_dim: int,\n dilation: int = 1,\n ):\n super().__init__()\n padding = (dilation * (7 - 1)) // 2\n self.dwconv = nn.Conv1d(\n dim, dim, kernel_size=7, padding=padding, groups=dim, dilation=dilation\n ) # depthwise conv\n self.norm = nn.LayerNorm(dim, eps=1e-6)\n self.pwconv1 = nn.Linear(dim, intermediate_dim) # pointwise/1x1 convs, implemented with linear layers\n self.act = nn.GELU()\n self.grn = GRN(intermediate_dim)\n self.pwconv2 = nn.Linear(intermediate_dim, dim)\n\n def forward(self, x: torch.Tensor) -> torch.Tensor:\n residual = x\n x = x.transpose(1, 2) # b n d -> b d n\n x = self.dwconv(x)\n x = x.transpose(1, 2) # b d n -> b n d\n x = self.norm(x)\n x = self.pwconv1(x)\n x = self.act(x)\n x = self.grn(x)\n x = self.pwconv2(x)\n return residual + x\n\n\n# RMSNorm\n\n\nclass RMSNorm(nn.Module):\n def __init__(self, dim: int, eps: float):\n super().__init__()\n self.eps = eps\n self.weight = nn.Parameter(torch.ones(dim))\n self.native_rms_norm = float(torch.__version__[:3]) >= 2.4\n\n def forward(self, x):\n if self.native_rms_norm:\n if self.weight.dtype in [torch.float16, torch.bfloat16]:\n x = x.to(self.weight.dtype)\n x = F.rms_norm(x, normalized_shape=(x.shape[-1],), weight=self.weight, eps=self.eps)\n else:\n variance = x.to(torch.float32).pow(2).mean(-1, keepdim=True)\n x = x * torch.rsqrt(variance + self.eps)\n if self.weight.dtype in [torch.float16, torch.bfloat16]:\n x = x.to(self.weight.dtype)\n x = x * self.weight\n\n return x\n\n\n# AdaLayerNorm\n# return with modulated x for attn input, and params for later mlp modulation\n\n\nclass AdaLayerNorm(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 6)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb=None):\n emb = self.linear(self.silu(emb))\n shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = torch.chunk(emb, 6, dim=1)\n\n x = self.norm(x) * (1 + scale_msa[:, None]) + shift_msa[:, None]\n return x, gate_msa, shift_mlp, scale_mlp, gate_mlp\n\n\n# AdaLayerNorm for final layer\n# return only with modulated x for attn input, cuz no more mlp modulation\n\n\nclass AdaLayerNorm_Final(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 2)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb):\n emb = self.linear(self.silu(emb))\n scale, shift = torch.chunk(emb, 2, dim=1)\n\n x = self.norm(x) * (1 + scale)[:, None, :] + shift[:, None, :]\n return x\n\n\n# FeedForward\n\n\nclass FeedForward(nn.Module):\n def __init__(self, dim, dim_out=None, mult=4, dropout=0.0, approximate: str = \"none\"):\n super().__init__()\n inner_dim = int(dim * mult)\n dim_out = dim_out if dim_out is not None else dim\n\n activation = nn.GELU(approximate=approximate)\n project_in = nn.Sequential(nn.Linear(dim, inner_dim), activation)\n self.ff = nn.Sequential(project_in, nn.Dropout(dropout), nn.Linear(inner_dim, dim_out))\n\n def forward(self, x):\n return self.ff(x)\n\n\n# Attention with possible joint part\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass Attention(nn.Module):\n def __init__(\n self,\n processor: JointAttnProcessor | AttnProcessor,\n dim: int,\n heads: int = 8,\n dim_head: int = 64,\n dropout: float = 0.0,\n context_dim: Optional[int] = None, # if not None -> joint attention\n context_pre_only: bool = False,\n qk_norm: Optional[str] = None,\n ):\n super().__init__()\n\n if not hasattr(F, \"scaled_dot_product_attention\"):\n raise ImportError(\"Attention equires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.\")\n\n self.processor = processor\n\n self.dim = dim\n self.heads = heads\n self.inner_dim = dim_head * heads\n self.dropout = dropout\n\n self.context_dim = context_dim\n self.context_pre_only = context_pre_only\n\n self.to_q = nn.Linear(dim, self.inner_dim)\n self.to_k = nn.Linear(dim, self.inner_dim)\n self.to_v = nn.Linear(dim, self.inner_dim)\n\n if qk_norm is None:\n self.q_norm = None\n self.k_norm = None\n elif qk_norm == \"rms_norm\":\n self.q_norm = RMSNorm(dim_head, eps=1e-6)\n self.k_norm = RMSNorm(dim_head, eps=1e-6)\n else:\n raise ValueError(f\"Unimplemented qk_norm: {qk_norm}\")\n\n if self.context_dim is not None:\n self.to_q_c = nn.Linear(context_dim, self.inner_dim)\n self.to_k_c = nn.Linear(context_dim, self.inner_dim)\n self.to_v_c = nn.Linear(context_dim, self.inner_dim)\n if qk_norm is None:\n self.c_q_norm = None\n self.c_k_norm = None\n elif qk_norm == \"rms_norm\":\n self.c_q_norm = RMSNorm(dim_head, eps=1e-6)\n self.c_k_norm = RMSNorm(dim_head, eps=1e-6)\n\n self.to_out = nn.ModuleList([])\n self.to_out.append(nn.Linear(self.inner_dim, dim))\n self.to_out.append(nn.Dropout(dropout))\n\n if self.context_dim is not None and not self.context_pre_only:\n self.to_out_c = nn.Linear(self.inner_dim, context_dim)\n\n def forward(\n self,\n x: float[\"b n d\"], # noised input x\n c: float[\"b n d\"] = None, # context c\n mask: bool[\"b n\"] | None = None,\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.Tensor:\n if c is not None:\n return self.processor(self, x, c=c, mask=mask, rope=rope, c_rope=c_rope)\n else:\n return self.processor(self, x, mask=mask, rope=rope)\n\n\n# Attention processor\n\nif is_package_available(\"flash_attn\"):\n from flash_attn.bert_padding import pad_input, unpad_input\n from flash_attn import flash_attn_varlen_func, flash_attn_func\n\n\nclass AttnProcessor:\n def __init__(\n self,\n pe_attn_head: int | None = None, # number of attention head to apply rope, None for all\n attn_backend: str = \"torch\", # \"torch\" or \"flash_attn\"\n attn_mask_enabled: bool = True,\n ):\n if attn_backend == \"flash_attn\":\n assert is_package_available(\"flash_attn\"), \"Please install flash-attn first.\"\n\n self.pe_attn_head = pe_attn_head\n self.attn_backend = attn_backend\n self.attn_mask_enabled = attn_mask_enabled\n\n def __call__(\n self,\n attn: Attention,\n x: float[\"b n d\"], # noised input x\n mask: bool[\"b n\"] | None = None,\n rope=None, # rotary position embedding\n ) -> torch.FloatTensor:\n batch_size = x.shape[0]\n\n # `sample` projections\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n\n # attention\n inner_dim = key.shape[-1]\n head_dim = inner_dim // attn.heads\n query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n\n # qk norm\n if attn.q_norm is not None:\n query = attn.q_norm(query)\n if attn.k_norm is not None:\n key = attn.k_norm(key)\n\n # apply rotary position embedding\n if rope is not None:\n freqs, xpos_scale = rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n\n if self.pe_attn_head is not None:\n pn = self.pe_attn_head\n query[:, :pn, :, :] = apply_rotary_pos_emb(query[:, :pn, :, :], freqs, q_xpos_scale)\n key[:, :pn, :, :] = apply_rotary_pos_emb(key[:, :pn, :, :], freqs, k_xpos_scale)\n else:\n query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)\n key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)\n\n if self.attn_backend == \"torch\":\n # mask. e.g. inference got a batch with different target durations, mask out the padding\n if self.attn_mask_enabled and mask is not None:\n attn_mask = mask\n attn_mask = attn_mask.unsqueeze(1).unsqueeze(1) # 'b n -> b 1 1 n'\n attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])\n else:\n attn_mask = None\n x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)\n x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)\n\n elif self.attn_backend == \"flash_attn\":\n query = query.transpose(1, 2) # [b, h, n, d] -> [b, n, h, d]\n key = key.transpose(1, 2)\n value = value.transpose(1, 2)\n if self.attn_mask_enabled and mask is not None:\n query, indices, q_cu_seqlens, q_max_seqlen_in_batch, _ = unpad_input(query, mask)\n key, _, k_cu_seqlens, k_max_seqlen_in_batch, _ = unpad_input(key, mask)\n value, _, _, _, _ = unpad_input(value, mask)\n x = flash_attn_varlen_func(\n query,\n key,\n value,\n q_cu_seqlens,\n k_cu_seqlens,\n q_max_seqlen_in_batch,\n k_max_seqlen_in_batch,\n )\n x = pad_input(x, indices, batch_size, q_max_seqlen_in_batch)\n x = x.reshape(batch_size, -1, attn.heads * head_dim)\n else:\n x = flash_attn_func(query, key, value, dropout_p=0.0, causal=False)\n x = x.reshape(batch_size, -1, attn.heads * head_dim)\n\n x = x.to(query.dtype)\n\n # linear proj\n x = attn.to_out[0](x)\n # dropout\n x = attn.to_out[1](x)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n\n return x\n\n\n# Joint Attention processor for MM-DiT\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass JointAttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn: Attention,\n x: float[\"b n d\"], # noised input x\n c: float[\"b nt d\"] = None, # context c, here text\n mask: bool[\"b n\"] | None = None,\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.FloatTensor:\n residual = x\n\n batch_size = c.shape[0]\n\n # `sample` projections\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n\n # `context` projections\n c_query = attn.to_q_c(c)\n c_key = attn.to_k_c(c)\n c_value = attn.to_v_c(c)\n\n # attention\n inner_dim = key.shape[-1]\n head_dim = inner_dim // attn.heads\n query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n c_query = c_query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n c_key = c_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n c_value = c_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n\n # qk norm\n if attn.q_norm is not None:\n query = attn.q_norm(query)\n if attn.k_norm is not None:\n key = attn.k_norm(key)\n if attn.c_q_norm is not None:\n c_query = attn.c_q_norm(c_query)\n if attn.c_k_norm is not None:\n c_key = attn.c_k_norm(c_key)\n\n # apply rope for context and noised input independently\n if rope is not None:\n freqs, xpos_scale = rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)\n key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)\n if c_rope is not None:\n# ... truncated ...","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.get_bigvgan_mel_spectrogram","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.modules.get_bigvgan_mel_spectrogram#L33-L75","kind":"function","name":"get_bigvgan_mel_spectrogram","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":33,"end_line":75,"context_start_line":13,"context_end_line":95,"code":"import math\nfrom typing import Optional\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom librosa.filters import mel as librosa_mel_fn\nfrom torch import nn\nfrom x_transformers.x_transformers import apply_rotary_pos_emb\n\nfrom f5_tts.model_new.utils import is_package_available\n\n\n# raw wav to mel spec\n\n\nmel_basis_cache = {}\nhann_window_cache = {}\n\n\ndef get_bigvgan_mel_spectrogram(\n waveform,\n n_fft=1024,\n n_mel_channels=100,\n target_sample_rate=24000,\n hop_length=256,\n win_length=1024,\n fmin=0,\n fmax=None,\n center=False,\n): # Copy from https://github.com/NVIDIA/BigVGAN/tree/main\n device = waveform.device\n key = f\"{n_fft}_{n_mel_channels}_{target_sample_rate}_{hop_length}_{win_length}_{fmin}_{fmax}_{device}\"\n\n if key not in mel_basis_cache:\n mel = librosa_mel_fn(sr=target_sample_rate, n_fft=n_fft, n_mels=n_mel_channels, fmin=fmin, fmax=fmax)\n mel_basis_cache[key] = torch.from_numpy(mel).float().to(device) # TODO: why they need .float()?\n hann_window_cache[key] = torch.hann_window(win_length).to(device)\n\n mel_basis = mel_basis_cache[key]\n hann_window = hann_window_cache[key]\n\n padding = (n_fft - hop_length) // 2\n waveform = torch.nn.functional.pad(waveform.unsqueeze(1), (padding, padding), mode=\"reflect\").squeeze(1)\n\n spec = torch.stft(\n waveform,\n n_fft,\n hop_length=hop_length,\n win_length=win_length,\n window=hann_window,\n center=center,\n pad_mode=\"reflect\",\n normalized=False,\n onesided=True,\n return_complex=True,\n )\n spec = torch.sqrt(torch.view_as_real(spec).pow(2).sum(-1) + 1e-9)\n\n mel_spec = torch.matmul(mel_basis, spec)\n mel_spec = torch.log(torch.clamp(mel_spec, min=1e-5))\n\n return mel_spec\n\n\ndef get_vocos_mel_spectrogram(\n waveform,\n n_fft=1024,\n n_mel_channels=100,\n target_sample_rate=24000,\n hop_length=256,\n win_length=1024,\n):\n mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=target_sample_rate,\n n_fft=n_fft,\n win_length=win_length,\n hop_length=hop_length,\n n_mels=n_mel_channels,\n power=1,\n center=True,\n normalized=False,\n norm=None,","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.get_vocos_mel_spectrogram","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.modules.get_vocos_mel_spectrogram#L78-L104","kind":"function","name":"get_vocos_mel_spectrogram","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":78,"end_line":104,"context_start_line":58,"context_end_line":124,"code":" spec = torch.stft(\n waveform,\n n_fft,\n hop_length=hop_length,\n win_length=win_length,\n window=hann_window,\n center=center,\n pad_mode=\"reflect\",\n normalized=False,\n onesided=True,\n return_complex=True,\n )\n spec = torch.sqrt(torch.view_as_real(spec).pow(2).sum(-1) + 1e-9)\n\n mel_spec = torch.matmul(mel_basis, spec)\n mel_spec = torch.log(torch.clamp(mel_spec, min=1e-5))\n\n return mel_spec\n\n\ndef get_vocos_mel_spectrogram(\n waveform,\n n_fft=1024,\n n_mel_channels=100,\n target_sample_rate=24000,\n hop_length=256,\n win_length=1024,\n):\n mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=target_sample_rate,\n n_fft=n_fft,\n win_length=win_length,\n hop_length=hop_length,\n n_mels=n_mel_channels,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(waveform.device)\n if len(waveform.shape) == 3:\n waveform = waveform.squeeze(1) # 'b 1 nw -> b nw'\n\n assert len(waveform.shape) == 2\n\n mel = mel_stft(waveform)\n mel = mel.clamp(min=1e-5).log()\n return mel\n\n\nclass MelSpec(nn.Module):\n def __init__(\n self,\n n_fft=1024,\n hop_length=256,\n win_length=1024,\n n_mel_channels=100,\n target_sample_rate=24_000,\n mel_spec_type=\"vocos\",\n ):\n super().__init__()\n assert mel_spec_type in [\"vocos\", \"bigvgan\"], print(\"We only support two extract mel backend: vocos or bigvgan\")\n\n self.n_fft = n_fft\n self.hop_length = hop_length\n self.win_length = win_length\n self.n_mel_channels = n_mel_channels\n self.target_sample_rate = target_sample_rate","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.MelSpec","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.MelSpec#L107-L146","kind":"class","name":"MelSpec","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":107,"end_line":146,"context_start_line":87,"context_end_line":166,"code":" sample_rate=target_sample_rate,\n n_fft=n_fft,\n win_length=win_length,\n hop_length=hop_length,\n n_mels=n_mel_channels,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(waveform.device)\n if len(waveform.shape) == 3:\n waveform = waveform.squeeze(1) # 'b 1 nw -> b nw'\n\n assert len(waveform.shape) == 2\n\n mel = mel_stft(waveform)\n mel = mel.clamp(min=1e-5).log()\n return mel\n\n\nclass MelSpec(nn.Module):\n def __init__(\n self,\n n_fft=1024,\n hop_length=256,\n win_length=1024,\n n_mel_channels=100,\n target_sample_rate=24_000,\n mel_spec_type=\"vocos\",\n ):\n super().__init__()\n assert mel_spec_type in [\"vocos\", \"bigvgan\"], print(\"We only support two extract mel backend: vocos or bigvgan\")\n\n self.n_fft = n_fft\n self.hop_length = hop_length\n self.win_length = win_length\n self.n_mel_channels = n_mel_channels\n self.target_sample_rate = target_sample_rate\n\n if mel_spec_type == \"vocos\":\n self.extractor = get_vocos_mel_spectrogram\n elif mel_spec_type == \"bigvgan\":\n self.extractor = get_bigvgan_mel_spectrogram\n\n self.register_buffer(\"dummy\", torch.tensor(0), persistent=False)\n\n def forward(self, wav):\n if self.dummy.device != wav.device:\n self.to(wav.device)\n\n mel = self.extractor(\n waveform=wav,\n n_fft=self.n_fft,\n n_mel_channels=self.n_mel_channels,\n target_sample_rate=self.target_sample_rate,\n hop_length=self.hop_length,\n win_length=self.win_length,\n )\n\n return mel\n\n\n# sinusoidal position embedding\n\n\nclass SinusPositionEmbedding(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.dim = dim\n\n def forward(self, x, scale=1000):\n device = x.device\n half_dim = self.dim // 2\n emb = math.log(10000) / (half_dim - 1)\n emb = torch.exp(torch.arange(half_dim, device=device).float() * -emb)\n emb = scale * x.unsqueeze(1) * emb.unsqueeze(0)\n emb = torch.cat((emb.sin(), emb.cos()), dim=-1)\n return emb\n\n","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.SinusPositionEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.SinusPositionEmbedding#L152-L164","kind":"class","name":"SinusPositionEmbedding","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":152,"end_line":164,"context_start_line":132,"context_end_line":184,"code":"\n def forward(self, wav):\n if self.dummy.device != wav.device:\n self.to(wav.device)\n\n mel = self.extractor(\n waveform=wav,\n n_fft=self.n_fft,\n n_mel_channels=self.n_mel_channels,\n target_sample_rate=self.target_sample_rate,\n hop_length=self.hop_length,\n win_length=self.win_length,\n )\n\n return mel\n\n\n# sinusoidal position embedding\n\n\nclass SinusPositionEmbedding(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.dim = dim\n\n def forward(self, x, scale=1000):\n device = x.device\n half_dim = self.dim // 2\n emb = math.log(10000) / (half_dim - 1)\n emb = torch.exp(torch.arange(half_dim, device=device).float() * -emb)\n emb = scale * x.unsqueeze(1) * emb.unsqueeze(0)\n emb = torch.cat((emb.sin(), emb.cos()), dim=-1)\n return emb\n\n\n# convolutional position embedding\n\n\nclass ConvPositionEmbedding(nn.Module):\n def __init__(self, dim, kernel_size=31, groups=16):\n super().__init__()\n assert kernel_size % 2 != 0\n self.conv1d = nn.Sequential(\n nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),\n nn.Mish(),\n nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),\n nn.Mish(),\n )\n\n def forward(self, x: float[\"b n d\"], mask: bool[\"b n\"] | None = None):\n if mask is not None:\n mask = mask[..., None]\n x = x.masked_fill(~mask, 0.0)","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.ConvPositionEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.ConvPositionEmbedding#L170-L193","kind":"class","name":"ConvPositionEmbedding","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":170,"end_line":193,"context_start_line":150,"context_end_line":213,"code":"\n\nclass SinusPositionEmbedding(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.dim = dim\n\n def forward(self, x, scale=1000):\n device = x.device\n half_dim = self.dim // 2\n emb = math.log(10000) / (half_dim - 1)\n emb = torch.exp(torch.arange(half_dim, device=device).float() * -emb)\n emb = scale * x.unsqueeze(1) * emb.unsqueeze(0)\n emb = torch.cat((emb.sin(), emb.cos()), dim=-1)\n return emb\n\n\n# convolutional position embedding\n\n\nclass ConvPositionEmbedding(nn.Module):\n def __init__(self, dim, kernel_size=31, groups=16):\n super().__init__()\n assert kernel_size % 2 != 0\n self.conv1d = nn.Sequential(\n nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),\n nn.Mish(),\n nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),\n nn.Mish(),\n )\n\n def forward(self, x: float[\"b n d\"], mask: bool[\"b n\"] | None = None):\n if mask is not None:\n mask = mask[..., None]\n x = x.masked_fill(~mask, 0.0)\n\n x = x.permute(0, 2, 1)\n x = self.conv1d(x)\n out = x.permute(0, 2, 1)\n\n if mask is not None:\n out = out.masked_fill(~mask, 0.0)\n\n return out\n\n\n# rotary positional embedding related\n\n\ndef precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0):\n # proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning\n # has some connection to NTK literature\n # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/\n # https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py\n theta *= theta_rescale_factor ** (dim / (dim - 2))\n freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))\n t = torch.arange(end, device=freqs.device) # type: ignore\n freqs = torch.outer(t, freqs).float() # type: ignore\n freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef get_pos_embed_indices(start, length, max_pos, scale=1.0):","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.precompute_freqs_cis","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.modules.precompute_freqs_cis#L199-L210","kind":"function","name":"precompute_freqs_cis","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":199,"end_line":210,"context_start_line":179,"context_end_line":230,"code":" )\n\n def forward(self, x: float[\"b n d\"], mask: bool[\"b n\"] | None = None):\n if mask is not None:\n mask = mask[..., None]\n x = x.masked_fill(~mask, 0.0)\n\n x = x.permute(0, 2, 1)\n x = self.conv1d(x)\n out = x.permute(0, 2, 1)\n\n if mask is not None:\n out = out.masked_fill(~mask, 0.0)\n\n return out\n\n\n# rotary positional embedding related\n\n\ndef precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0):\n # proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning\n # has some connection to NTK literature\n # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/\n # https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py\n theta *= theta_rescale_factor ** (dim / (dim - 2))\n freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))\n t = torch.arange(end, device=freqs.device) # type: ignore\n freqs = torch.outer(t, freqs).float() # type: ignore\n freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef get_pos_embed_indices(start, length, max_pos, scale=1.0):\n # length = length if isinstance(length, int) else length.max()\n scale = scale * torch.ones_like(start, dtype=torch.float32) # in case scale is a scalar\n pos = (\n start.unsqueeze(1)\n + (torch.arange(length, device=start.device, dtype=torch.float32).unsqueeze(0) * scale.unsqueeze(1)).long()\n )\n # avoid extra long error.\n pos = torch.where(pos < max_pos, pos, max_pos - 1)\n return pos\n\n\n# Global Response Normalization layer (Instance Normalization ?)\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.get_pos_embed_indices","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.modules.get_pos_embed_indices#L213-L222","kind":"function","name":"get_pos_embed_indices","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":213,"end_line":222,"context_start_line":193,"context_end_line":242,"code":" return out\n\n\n# rotary positional embedding related\n\n\ndef precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0):\n # proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning\n # has some connection to NTK literature\n # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/\n # https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py\n theta *= theta_rescale_factor ** (dim / (dim - 2))\n freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))\n t = torch.arange(end, device=freqs.device) # type: ignore\n freqs = torch.outer(t, freqs).float() # type: ignore\n freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef get_pos_embed_indices(start, length, max_pos, scale=1.0):\n # length = length if isinstance(length, int) else length.max()\n scale = scale * torch.ones_like(start, dtype=torch.float32) # in case scale is a scalar\n pos = (\n start.unsqueeze(1)\n + (torch.arange(length, device=start.device, dtype=torch.float32).unsqueeze(0) * scale.unsqueeze(1)).long()\n )\n # avoid extra long error.\n pos = torch.where(pos < max_pos, pos, max_pos - 1)\n return pos\n\n\n# Global Response Normalization layer (Instance Normalization ?)\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.gamma = nn.Parameter(torch.zeros(1, 1, dim))\n self.beta = nn.Parameter(torch.zeros(1, 1, dim))\n\n def forward(self, x):\n Gx = torch.norm(x, p=2, dim=1, keepdim=True)\n Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)\n return self.gamma * (x * Nx) + self.beta + x\n\n\n# ConvNeXt-V2 Block https://github.com/facebookresearch/ConvNeXt-V2/blob/main/models/convnextv2.py\n# ref: https://github.com/bfs18/e2_tts/blob/main/rfwave/modules.py#L108\n","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.GRN","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.GRN#L228-L237","kind":"class","name":"GRN","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":228,"end_line":237,"context_start_line":208,"context_end_line":257,"code":" freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef get_pos_embed_indices(start, length, max_pos, scale=1.0):\n # length = length if isinstance(length, int) else length.max()\n scale = scale * torch.ones_like(start, dtype=torch.float32) # in case scale is a scalar\n pos = (\n start.unsqueeze(1)\n + (torch.arange(length, device=start.device, dtype=torch.float32).unsqueeze(0) * scale.unsqueeze(1)).long()\n )\n # avoid extra long error.\n pos = torch.where(pos < max_pos, pos, max_pos - 1)\n return pos\n\n\n# Global Response Normalization layer (Instance Normalization ?)\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.gamma = nn.Parameter(torch.zeros(1, 1, dim))\n self.beta = nn.Parameter(torch.zeros(1, 1, dim))\n\n def forward(self, x):\n Gx = torch.norm(x, p=2, dim=1, keepdim=True)\n Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)\n return self.gamma * (x * Nx) + self.beta + x\n\n\n# ConvNeXt-V2 Block https://github.com/facebookresearch/ConvNeXt-V2/blob/main/models/convnextv2.py\n# ref: https://github.com/bfs18/e2_tts/blob/main/rfwave/modules.py#L108\n\n\nclass ConvNeXtV2Block(nn.Module):\n def __init__(\n self,\n dim: int,\n intermediate_dim: int,\n dilation: int = 1,\n ):\n super().__init__()\n padding = (dilation * (7 - 1)) // 2\n self.dwconv = nn.Conv1d(\n dim, dim, kernel_size=7, padding=padding, groups=dim, dilation=dilation\n ) # depthwise conv\n self.norm = nn.LayerNorm(dim, eps=1e-6)\n self.pwconv1 = nn.Linear(dim, intermediate_dim) # pointwise/1x1 convs, implemented with linear layers","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.ConvNeXtV2Block","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.ConvNeXtV2Block#L244-L272","kind":"class","name":"ConvNeXtV2Block","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":244,"end_line":272,"context_start_line":224,"context_end_line":292,"code":"\n# Global Response Normalization layer (Instance Normalization ?)\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.gamma = nn.Parameter(torch.zeros(1, 1, dim))\n self.beta = nn.Parameter(torch.zeros(1, 1, dim))\n\n def forward(self, x):\n Gx = torch.norm(x, p=2, dim=1, keepdim=True)\n Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)\n return self.gamma * (x * Nx) + self.beta + x\n\n\n# ConvNeXt-V2 Block https://github.com/facebookresearch/ConvNeXt-V2/blob/main/models/convnextv2.py\n# ref: https://github.com/bfs18/e2_tts/blob/main/rfwave/modules.py#L108\n\n\nclass ConvNeXtV2Block(nn.Module):\n def __init__(\n self,\n dim: int,\n intermediate_dim: int,\n dilation: int = 1,\n ):\n super().__init__()\n padding = (dilation * (7 - 1)) // 2\n self.dwconv = nn.Conv1d(\n dim, dim, kernel_size=7, padding=padding, groups=dim, dilation=dilation\n ) # depthwise conv\n self.norm = nn.LayerNorm(dim, eps=1e-6)\n self.pwconv1 = nn.Linear(dim, intermediate_dim) # pointwise/1x1 convs, implemented with linear layers\n self.act = nn.GELU()\n self.grn = GRN(intermediate_dim)\n self.pwconv2 = nn.Linear(intermediate_dim, dim)\n\n def forward(self, x: torch.Tensor) -> torch.Tensor:\n residual = x\n x = x.transpose(1, 2) # b n d -> b d n\n x = self.dwconv(x)\n x = x.transpose(1, 2) # b d n -> b n d\n x = self.norm(x)\n x = self.pwconv1(x)\n x = self.act(x)\n x = self.grn(x)\n x = self.pwconv2(x)\n return residual + x\n\n\n# RMSNorm\n\n\nclass RMSNorm(nn.Module):\n def __init__(self, dim: int, eps: float):\n super().__init__()\n self.eps = eps\n self.weight = nn.Parameter(torch.ones(dim))\n self.native_rms_norm = float(torch.__version__[:3]) >= 2.4\n\n def forward(self, x):\n if self.native_rms_norm:\n if self.weight.dtype in [torch.float16, torch.bfloat16]:\n x = x.to(self.weight.dtype)\n x = F.rms_norm(x, normalized_shape=(x.shape[-1],), weight=self.weight, eps=self.eps)\n else:\n variance = x.to(torch.float32).pow(2).mean(-1, keepdim=True)\n x = x * torch.rsqrt(variance + self.eps)","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.RMSNorm","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.RMSNorm#L278-L297","kind":"class","name":"RMSNorm","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":278,"end_line":297,"context_start_line":258,"context_end_line":317,"code":" self.act = nn.GELU()\n self.grn = GRN(intermediate_dim)\n self.pwconv2 = nn.Linear(intermediate_dim, dim)\n\n def forward(self, x: torch.Tensor) -> torch.Tensor:\n residual = x\n x = x.transpose(1, 2) # b n d -> b d n\n x = self.dwconv(x)\n x = x.transpose(1, 2) # b d n -> b n d\n x = self.norm(x)\n x = self.pwconv1(x)\n x = self.act(x)\n x = self.grn(x)\n x = self.pwconv2(x)\n return residual + x\n\n\n# RMSNorm\n\n\nclass RMSNorm(nn.Module):\n def __init__(self, dim: int, eps: float):\n super().__init__()\n self.eps = eps\n self.weight = nn.Parameter(torch.ones(dim))\n self.native_rms_norm = float(torch.__version__[:3]) >= 2.4\n\n def forward(self, x):\n if self.native_rms_norm:\n if self.weight.dtype in [torch.float16, torch.bfloat16]:\n x = x.to(self.weight.dtype)\n x = F.rms_norm(x, normalized_shape=(x.shape[-1],), weight=self.weight, eps=self.eps)\n else:\n variance = x.to(torch.float32).pow(2).mean(-1, keepdim=True)\n x = x * torch.rsqrt(variance + self.eps)\n if self.weight.dtype in [torch.float16, torch.bfloat16]:\n x = x.to(self.weight.dtype)\n x = x * self.weight\n\n return x\n\n\n# AdaLayerNorm\n# return with modulated x for attn input, and params for later mlp modulation\n\n\nclass AdaLayerNorm(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 6)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb=None):\n emb = self.linear(self.silu(emb))\n shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = torch.chunk(emb, 6, dim=1)\n\n x = self.norm(x) * (1 + scale_msa[:, None]) + shift_msa[:, None]","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.AdaLayerNorm","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.AdaLayerNorm#L304-L318","kind":"class","name":"AdaLayerNorm","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":304,"end_line":318,"context_start_line":284,"context_end_line":338,"code":"\n def forward(self, x):\n if self.native_rms_norm:\n if self.weight.dtype in [torch.float16, torch.bfloat16]:\n x = x.to(self.weight.dtype)\n x = F.rms_norm(x, normalized_shape=(x.shape[-1],), weight=self.weight, eps=self.eps)\n else:\n variance = x.to(torch.float32).pow(2).mean(-1, keepdim=True)\n x = x * torch.rsqrt(variance + self.eps)\n if self.weight.dtype in [torch.float16, torch.bfloat16]:\n x = x.to(self.weight.dtype)\n x = x * self.weight\n\n return x\n\n\n# AdaLayerNorm\n# return with modulated x for attn input, and params for later mlp modulation\n\n\nclass AdaLayerNorm(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 6)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb=None):\n emb = self.linear(self.silu(emb))\n shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = torch.chunk(emb, 6, dim=1)\n\n x = self.norm(x) * (1 + scale_msa[:, None]) + shift_msa[:, None]\n return x, gate_msa, shift_mlp, scale_mlp, gate_mlp\n\n\n# AdaLayerNorm for final layer\n# return only with modulated x for attn input, cuz no more mlp modulation\n\n\nclass AdaLayerNorm_Final(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 2)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb):\n emb = self.linear(self.silu(emb))\n scale, shift = torch.chunk(emb, 2, dim=1)\n\n x = self.norm(x) * (1 + scale)[:, None, :] + shift[:, None, :]","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.AdaLayerNorm_Final","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.AdaLayerNorm_Final#L325-L339","kind":"class","name":"AdaLayerNorm_Final","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":325,"end_line":339,"context_start_line":305,"context_end_line":359,"code":" def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 6)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb=None):\n emb = self.linear(self.silu(emb))\n shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = torch.chunk(emb, 6, dim=1)\n\n x = self.norm(x) * (1 + scale_msa[:, None]) + shift_msa[:, None]\n return x, gate_msa, shift_mlp, scale_mlp, gate_mlp\n\n\n# AdaLayerNorm for final layer\n# return only with modulated x for attn input, cuz no more mlp modulation\n\n\nclass AdaLayerNorm_Final(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 2)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb):\n emb = self.linear(self.silu(emb))\n scale, shift = torch.chunk(emb, 2, dim=1)\n\n x = self.norm(x) * (1 + scale)[:, None, :] + shift[:, None, :]\n return x\n\n\n# FeedForward\n\n\nclass FeedForward(nn.Module):\n def __init__(self, dim, dim_out=None, mult=4, dropout=0.0, approximate: str = \"none\"):\n super().__init__()\n inner_dim = int(dim * mult)\n dim_out = dim_out if dim_out is not None else dim\n\n activation = nn.GELU(approximate=approximate)\n project_in = nn.Sequential(nn.Linear(dim, inner_dim), activation)\n self.ff = nn.Sequential(project_in, nn.Dropout(dropout), nn.Linear(inner_dim, dim_out))\n\n def forward(self, x):\n return self.ff(x)\n\n\n# Attention with possible joint part","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.FeedForward","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.FeedForward#L345-L356","kind":"class","name":"FeedForward","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":345,"end_line":356,"context_start_line":325,"context_end_line":376,"code":"class AdaLayerNorm_Final(nn.Module):\n def __init__(self, dim):\n super().__init__()\n\n self.silu = nn.SiLU()\n self.linear = nn.Linear(dim, dim * 2)\n\n self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb):\n emb = self.linear(self.silu(emb))\n scale, shift = torch.chunk(emb, 2, dim=1)\n\n x = self.norm(x) * (1 + scale)[:, None, :] + shift[:, None, :]\n return x\n\n\n# FeedForward\n\n\nclass FeedForward(nn.Module):\n def __init__(self, dim, dim_out=None, mult=4, dropout=0.0, approximate: str = \"none\"):\n super().__init__()\n inner_dim = int(dim * mult)\n dim_out = dim_out if dim_out is not None else dim\n\n activation = nn.GELU(approximate=approximate)\n project_in = nn.Sequential(nn.Linear(dim, inner_dim), activation)\n self.ff = nn.Sequential(project_in, nn.Dropout(dropout), nn.Linear(inner_dim, dim_out))\n\n def forward(self, x):\n return self.ff(x)\n\n\n# Attention with possible joint part\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass Attention(nn.Module):\n def __init__(\n self,\n processor: JointAttnProcessor | AttnProcessor,\n dim: int,\n heads: int = 8,\n dim_head: int = 64,\n dropout: float = 0.0,\n context_dim: Optional[int] = None, # if not None -> joint attention\n context_pre_only: bool = False,\n qk_norm: Optional[str] = None,\n ):\n super().__init__()\n","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.Attention","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.Attention#L363-L432","kind":"class","name":"Attention","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":363,"end_line":432,"context_start_line":343,"context_end_line":452,"code":"\n\nclass FeedForward(nn.Module):\n def __init__(self, dim, dim_out=None, mult=4, dropout=0.0, approximate: str = \"none\"):\n super().__init__()\n inner_dim = int(dim * mult)\n dim_out = dim_out if dim_out is not None else dim\n\n activation = nn.GELU(approximate=approximate)\n project_in = nn.Sequential(nn.Linear(dim, inner_dim), activation)\n self.ff = nn.Sequential(project_in, nn.Dropout(dropout), nn.Linear(inner_dim, dim_out))\n\n def forward(self, x):\n return self.ff(x)\n\n\n# Attention with possible joint part\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass Attention(nn.Module):\n def __init__(\n self,\n processor: JointAttnProcessor | AttnProcessor,\n dim: int,\n heads: int = 8,\n dim_head: int = 64,\n dropout: float = 0.0,\n context_dim: Optional[int] = None, # if not None -> joint attention\n context_pre_only: bool = False,\n qk_norm: Optional[str] = None,\n ):\n super().__init__()\n\n if not hasattr(F, \"scaled_dot_product_attention\"):\n raise ImportError(\"Attention equires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.\")\n\n self.processor = processor\n\n self.dim = dim\n self.heads = heads\n self.inner_dim = dim_head * heads\n self.dropout = dropout\n\n self.context_dim = context_dim\n self.context_pre_only = context_pre_only\n\n self.to_q = nn.Linear(dim, self.inner_dim)\n self.to_k = nn.Linear(dim, self.inner_dim)\n self.to_v = nn.Linear(dim, self.inner_dim)\n\n if qk_norm is None:\n self.q_norm = None\n self.k_norm = None\n elif qk_norm == \"rms_norm\":\n self.q_norm = RMSNorm(dim_head, eps=1e-6)\n self.k_norm = RMSNorm(dim_head, eps=1e-6)\n else:\n raise ValueError(f\"Unimplemented qk_norm: {qk_norm}\")\n\n if self.context_dim is not None:\n self.to_q_c = nn.Linear(context_dim, self.inner_dim)\n self.to_k_c = nn.Linear(context_dim, self.inner_dim)\n self.to_v_c = nn.Linear(context_dim, self.inner_dim)\n if qk_norm is None:\n self.c_q_norm = None\n self.c_k_norm = None\n elif qk_norm == \"rms_norm\":\n self.c_q_norm = RMSNorm(dim_head, eps=1e-6)\n self.c_k_norm = RMSNorm(dim_head, eps=1e-6)\n\n self.to_out = nn.ModuleList([])\n self.to_out.append(nn.Linear(self.inner_dim, dim))\n self.to_out.append(nn.Dropout(dropout))\n\n if self.context_dim is not None and not self.context_pre_only:\n self.to_out_c = nn.Linear(self.inner_dim, context_dim)\n\n def forward(\n self,\n x: float[\"b n d\"], # noised input x\n c: float[\"b n d\"] = None, # context c\n mask: bool[\"b n\"] | None = None,\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.Tensor:\n if c is not None:\n return self.processor(self, x, c=c, mask=mask, rope=rope, c_rope=c_rope)\n else:\n return self.processor(self, x, mask=mask, rope=rope)\n\n\n# Attention processor\n\nif is_package_available(\"flash_attn\"):\n from flash_attn.bert_padding import pad_input, unpad_input\n from flash_attn import flash_attn_varlen_func, flash_attn_func\n\n\nclass AttnProcessor:\n def __init__(\n self,\n pe_attn_head: int | None = None, # number of attention head to apply rope, None for all\n attn_backend: str = \"torch\", # \"torch\" or \"flash_attn\"\n attn_mask_enabled: bool = True,\n ):\n if attn_backend == \"flash_attn\":\n assert is_package_available(\"flash_attn\"), \"Please install flash-attn first.\"\n\n self.pe_attn_head = pe_attn_head","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.AttnProcessor","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.AttnProcessor#L442-L541","kind":"class","name":"AttnProcessor","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":442,"end_line":541,"context_start_line":422,"context_end_line":561,"code":" self,\n x: float[\"b n d\"], # noised input x\n c: float[\"b n d\"] = None, # context c\n mask: bool[\"b n\"] | None = None,\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.Tensor:\n if c is not None:\n return self.processor(self, x, c=c, mask=mask, rope=rope, c_rope=c_rope)\n else:\n return self.processor(self, x, mask=mask, rope=rope)\n\n\n# Attention processor\n\nif is_package_available(\"flash_attn\"):\n from flash_attn.bert_padding import pad_input, unpad_input\n from flash_attn import flash_attn_varlen_func, flash_attn_func\n\n\nclass AttnProcessor:\n def __init__(\n self,\n pe_attn_head: int | None = None, # number of attention head to apply rope, None for all\n attn_backend: str = \"torch\", # \"torch\" or \"flash_attn\"\n attn_mask_enabled: bool = True,\n ):\n if attn_backend == \"flash_attn\":\n assert is_package_available(\"flash_attn\"), \"Please install flash-attn first.\"\n\n self.pe_attn_head = pe_attn_head\n self.attn_backend = attn_backend\n self.attn_mask_enabled = attn_mask_enabled\n\n def __call__(\n self,\n attn: Attention,\n x: float[\"b n d\"], # noised input x\n mask: bool[\"b n\"] | None = None,\n rope=None, # rotary position embedding\n ) -> torch.FloatTensor:\n batch_size = x.shape[0]\n\n # `sample` projections\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n\n # attention\n inner_dim = key.shape[-1]\n head_dim = inner_dim // attn.heads\n query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n\n # qk norm\n if attn.q_norm is not None:\n query = attn.q_norm(query)\n if attn.k_norm is not None:\n key = attn.k_norm(key)\n\n # apply rotary position embedding\n if rope is not None:\n freqs, xpos_scale = rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n\n if self.pe_attn_head is not None:\n pn = self.pe_attn_head\n query[:, :pn, :, :] = apply_rotary_pos_emb(query[:, :pn, :, :], freqs, q_xpos_scale)\n key[:, :pn, :, :] = apply_rotary_pos_emb(key[:, :pn, :, :], freqs, k_xpos_scale)\n else:\n query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)\n key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)\n\n if self.attn_backend == \"torch\":\n # mask. e.g. inference got a batch with different target durations, mask out the padding\n if self.attn_mask_enabled and mask is not None:\n attn_mask = mask\n attn_mask = attn_mask.unsqueeze(1).unsqueeze(1) # 'b n -> b 1 1 n'\n attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])\n else:\n attn_mask = None\n x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)\n x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)\n\n elif self.attn_backend == \"flash_attn\":\n query = query.transpose(1, 2) # [b, h, n, d] -> [b, n, h, d]\n key = key.transpose(1, 2)\n value = value.transpose(1, 2)\n if self.attn_mask_enabled and mask is not None:\n query, indices, q_cu_seqlens, q_max_seqlen_in_batch, _ = unpad_input(query, mask)\n key, _, k_cu_seqlens, k_max_seqlen_in_batch, _ = unpad_input(key, mask)\n value, _, _, _, _ = unpad_input(value, mask)\n x = flash_attn_varlen_func(\n query,\n key,\n value,\n q_cu_seqlens,\n k_cu_seqlens,\n q_max_seqlen_in_batch,\n k_max_seqlen_in_batch,\n )\n x = pad_input(x, indices, batch_size, q_max_seqlen_in_batch)\n x = x.reshape(batch_size, -1, attn.heads * head_dim)\n else:\n x = flash_attn_func(query, key, value, dropout_p=0.0, causal=False)\n x = x.reshape(batch_size, -1, attn.heads * head_dim)\n\n x = x.to(query.dtype)\n\n # linear proj\n x = attn.to_out[0](x)\n # dropout\n x = attn.to_out[1](x)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n\n return x\n\n\n# Joint Attention processor for MM-DiT\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass JointAttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn: Attention,\n x: float[\"b n d\"], # noised input x\n c: float[\"b nt d\"] = None, # context c, here text\n mask: bool[\"b n\"] | None = None,\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.FloatTensor:\n residual = x","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.JointAttnProcessor","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.JointAttnProcessor#L548-L642","kind":"class","name":"JointAttnProcessor","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":548,"end_line":642,"context_start_line":528,"context_end_line":662,"code":" x = x.reshape(batch_size, -1, attn.heads * head_dim)\n\n x = x.to(query.dtype)\n\n # linear proj\n x = attn.to_out[0](x)\n # dropout\n x = attn.to_out[1](x)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n\n return x\n\n\n# Joint Attention processor for MM-DiT\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass JointAttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn: Attention,\n x: float[\"b n d\"], # noised input x\n c: float[\"b nt d\"] = None, # context c, here text\n mask: bool[\"b n\"] | None = None,\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.FloatTensor:\n residual = x\n\n batch_size = c.shape[0]\n\n # `sample` projections\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n\n # `context` projections\n c_query = attn.to_q_c(c)\n c_key = attn.to_k_c(c)\n c_value = attn.to_v_c(c)\n\n # attention\n inner_dim = key.shape[-1]\n head_dim = inner_dim // attn.heads\n query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n c_query = c_query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n c_key = c_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n c_value = c_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n\n # qk norm\n if attn.q_norm is not None:\n query = attn.q_norm(query)\n if attn.k_norm is not None:\n key = attn.k_norm(key)\n if attn.c_q_norm is not None:\n c_query = attn.c_q_norm(c_query)\n if attn.c_k_norm is not None:\n c_key = attn.c_k_norm(c_key)\n\n # apply rope for context and noised input independently\n if rope is not None:\n freqs, xpos_scale = rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)\n key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)\n if c_rope is not None:\n freqs, xpos_scale = c_rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n c_query = apply_rotary_pos_emb(c_query, freqs, q_xpos_scale)\n c_key = apply_rotary_pos_emb(c_key, freqs, k_xpos_scale)\n\n # joint attention\n query = torch.cat([query, c_query], dim=2)\n key = torch.cat([key, c_key], dim=2)\n value = torch.cat([value, c_value], dim=2)\n\n # mask. e.g. inference got a batch with different target durations, mask out the padding\n if mask is not None:\n attn_mask = F.pad(mask, (0, c.shape[1]), value=True) # no mask for c (text)\n attn_mask = attn_mask.unsqueeze(1).unsqueeze(1) # 'b n -> b 1 1 n'\n attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])\n else:\n attn_mask = None\n\n x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)\n x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)\n x = x.to(query.dtype)\n\n # Split the attention outputs.\n x, c = (\n x[:, : residual.shape[1]],\n x[:, residual.shape[1] :],\n )\n\n # linear proj\n x = attn.to_out[0](x)\n # dropout\n x = attn.to_out[1](x)\n if not attn.context_pre_only:\n c = attn.to_out_c(c)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n # c = c.masked_fill(~mask, 0.) # no mask for c (text)\n\n return x, c\n\n\n# DiT Block\n\n\nclass DiTBlock(nn.Module):\n def __init__(\n self,\n dim,\n heads,\n dim_head,\n ff_mult=4,\n dropout=0.1,\n qk_norm=None,\n pe_attn_head=None,\n attn_backend=\"torch\", # \"torch\" or \"flash_attn\"\n attn_mask_enabled=True,\n ):\n super().__init__()\n","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.DiTBlock","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.DiTBlock#L648-L694","kind":"class","name":"DiTBlock","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":648,"end_line":694,"context_start_line":628,"context_end_line":714,"code":" )\n\n # linear proj\n x = attn.to_out[0](x)\n # dropout\n x = attn.to_out[1](x)\n if not attn.context_pre_only:\n c = attn.to_out_c(c)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n # c = c.masked_fill(~mask, 0.) # no mask for c (text)\n\n return x, c\n\n\n# DiT Block\n\n\nclass DiTBlock(nn.Module):\n def __init__(\n self,\n dim,\n heads,\n dim_head,\n ff_mult=4,\n dropout=0.1,\n qk_norm=None,\n pe_attn_head=None,\n attn_backend=\"torch\", # \"torch\" or \"flash_attn\"\n attn_mask_enabled=True,\n ):\n super().__init__()\n\n self.attn_norm = AdaLayerNorm(dim)\n self.attn = Attention(\n processor=AttnProcessor(\n pe_attn_head=pe_attn_head,\n attn_backend=attn_backend,\n attn_mask_enabled=attn_mask_enabled,\n ),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n qk_norm=qk_norm,\n )\n\n self.ff_norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n self.ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n\n def forward(self, x, t, mask=None, rope=None): # x: noised input, t: time embedding\n # pre-norm & modulation for attention input\n norm, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.attn_norm(x, emb=t)\n\n # attention\n attn_output = self.attn(x=norm, mask=mask, rope=rope)\n\n # process attention output for input x\n x = x + gate_msa.unsqueeze(1) * attn_output\n\n norm = self.ff_norm(x) * (1 + scale_mlp[:, None]) + shift_mlp[:, None]\n ff_output = self.ff(norm)\n x = x + gate_mlp.unsqueeze(1) * ff_output\n\n return x\n\n\n# MMDiT Block https://arxiv.org/abs/2403.03206\n\n\nclass MMDiTBlock(nn.Module):\n r\"\"\"\n modified from diffusers/src/diffusers/models/attention.py\n\n notes.\n _c: context related. text, cond, etc. (left part in sd3 fig2.b)\n _x: noised input related. (right part)\n context_pre_only: last layer only do prenorm + modulation cuz no more ffn\n \"\"\"\n\n def __init__(\n self, dim, heads, dim_head, ff_mult=4, dropout=0.1, context_dim=None, context_pre_only=False, qk_norm=None\n ):\n super().__init__()\n if context_dim is None:","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.MMDiTBlock","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.MMDiTBlock#L700-L768","kind":"class","name":"MMDiTBlock","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":700,"end_line":768,"context_start_line":680,"context_end_line":784,"code":" def forward(self, x, t, mask=None, rope=None): # x: noised input, t: time embedding\n # pre-norm & modulation for attention input\n norm, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.attn_norm(x, emb=t)\n\n # attention\n attn_output = self.attn(x=norm, mask=mask, rope=rope)\n\n # process attention output for input x\n x = x + gate_msa.unsqueeze(1) * attn_output\n\n norm = self.ff_norm(x) * (1 + scale_mlp[:, None]) + shift_mlp[:, None]\n ff_output = self.ff(norm)\n x = x + gate_mlp.unsqueeze(1) * ff_output\n\n return x\n\n\n# MMDiT Block https://arxiv.org/abs/2403.03206\n\n\nclass MMDiTBlock(nn.Module):\n r\"\"\"\n modified from diffusers/src/diffusers/models/attention.py\n\n notes.\n _c: context related. text, cond, etc. (left part in sd3 fig2.b)\n _x: noised input related. (right part)\n context_pre_only: last layer only do prenorm + modulation cuz no more ffn\n \"\"\"\n\n def __init__(\n self, dim, heads, dim_head, ff_mult=4, dropout=0.1, context_dim=None, context_pre_only=False, qk_norm=None\n ):\n super().__init__()\n if context_dim is None:\n context_dim = dim\n self.context_pre_only = context_pre_only\n\n self.attn_norm_c = AdaLayerNorm_Final(context_dim) if context_pre_only else AdaLayerNorm(context_dim)\n self.attn_norm_x = AdaLayerNorm(dim)\n self.attn = Attention(\n processor=JointAttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n context_dim=context_dim,\n context_pre_only=context_pre_only,\n qk_norm=qk_norm,\n )\n\n if not context_pre_only:\n self.ff_norm_c = nn.LayerNorm(context_dim, elementwise_affine=False, eps=1e-6)\n self.ff_c = FeedForward(dim=context_dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n else:\n self.ff_norm_c = None\n self.ff_c = None\n self.ff_norm_x = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n self.ff_x = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n\n def forward(self, x, c, t, mask=None, rope=None, c_rope=None): # x: noised input, c: context, t: time embedding\n # pre-norm & modulation for attention input\n if self.context_pre_only:\n norm_c = self.attn_norm_c(c, t)\n else:\n norm_c, c_gate_msa, c_shift_mlp, c_scale_mlp, c_gate_mlp = self.attn_norm_c(c, emb=t)\n norm_x, x_gate_msa, x_shift_mlp, x_scale_mlp, x_gate_mlp = self.attn_norm_x(x, emb=t)\n\n # attention\n x_attn_output, c_attn_output = self.attn(x=norm_x, c=norm_c, mask=mask, rope=rope, c_rope=c_rope)\n\n # process attention output for context c\n if self.context_pre_only:\n c = None\n else: # if not last layer\n c = c + c_gate_msa.unsqueeze(1) * c_attn_output\n\n norm_c = self.ff_norm_c(c) * (1 + c_scale_mlp[:, None]) + c_shift_mlp[:, None]\n c_ff_output = self.ff_c(norm_c)\n c = c + c_gate_mlp.unsqueeze(1) * c_ff_output\n\n # process attention output for input x\n x = x + x_gate_msa.unsqueeze(1) * x_attn_output\n\n norm_x = self.ff_norm_x(x) * (1 + x_scale_mlp[:, None]) + x_shift_mlp[:, None]\n x_ff_output = self.ff_x(norm_x)\n x = x + x_gate_mlp.unsqueeze(1) * x_ff_output\n\n return c, x\n\n\n# time step conditioning embedding\n\n\nclass TimestepEmbedding(nn.Module):\n def __init__(self, dim, freq_embed_dim=256):\n super().__init__()\n self.time_embed = SinusPositionEmbedding(freq_embed_dim)\n self.time_mlp = nn.Sequential(nn.Linear(freq_embed_dim, dim), nn.SiLU(), nn.Linear(dim, dim))\n\n def forward(self, timestep: float[\"b\"]):\n time_hidden = self.time_embed(timestep)\n time_hidden = time_hidden.to(timestep.dtype)\n time = self.time_mlp(time_hidden) # b d\n return time","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.TimestepEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.modules.TimestepEmbedding#L774-L784","kind":"class","name":"TimestepEmbedding","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":774,"end_line":784,"context_start_line":754,"context_end_line":784,"code":" else: # if not last layer\n c = c + c_gate_msa.unsqueeze(1) * c_attn_output\n\n norm_c = self.ff_norm_c(c) * (1 + c_scale_mlp[:, None]) + c_shift_mlp[:, None]\n c_ff_output = self.ff_c(norm_c)\n c = c + c_gate_mlp.unsqueeze(1) * c_ff_output\n\n # process attention output for input x\n x = x + x_gate_msa.unsqueeze(1) * x_attn_output\n\n norm_x = self.ff_norm_x(x) * (1 + x_scale_mlp[:, None]) + x_shift_mlp[:, None]\n x_ff_output = self.ff_x(norm_x)\n x = x + x_gate_mlp.unsqueeze(1) * x_ff_output\n\n return c, x\n\n\n# time step conditioning embedding\n\n\nclass TimestepEmbedding(nn.Module):\n def __init__(self, dim, freq_embed_dim=256):\n super().__init__()\n self.time_embed = SinusPositionEmbedding(freq_embed_dim)\n self.time_mlp = nn.Sequential(nn.Linear(freq_embed_dim, dim), nn.SiLU(), nn.Linear(dim, dim))\n\n def forward(self, timestep: float[\"b\"]):\n time_hidden = self.time_embed(timestep)\n time_hidden = time_hidden.to(timestep.dtype)\n time = self.time_mlp(time_hidden) # b d\n return time","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.modules.__init__#L775-L778","kind":"function","name":"__init__","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":775,"end_line":778,"context_start_line":755,"context_end_line":784,"code":" c = c + c_gate_msa.unsqueeze(1) * c_attn_output\n\n norm_c = self.ff_norm_c(c) * (1 + c_scale_mlp[:, None]) + c_shift_mlp[:, None]\n c_ff_output = self.ff_c(norm_c)\n c = c + c_gate_mlp.unsqueeze(1) * c_ff_output\n\n # process attention output for input x\n x = x + x_gate_msa.unsqueeze(1) * x_attn_output\n\n norm_x = self.ff_norm_x(x) * (1 + x_scale_mlp[:, None]) + x_shift_mlp[:, None]\n x_ff_output = self.ff_x(norm_x)\n x = x + x_gate_mlp.unsqueeze(1) * x_ff_output\n\n return c, x\n\n\n# time step conditioning embedding\n\n\nclass TimestepEmbedding(nn.Module):\n def __init__(self, dim, freq_embed_dim=256):\n super().__init__()\n self.time_embed = SinusPositionEmbedding(freq_embed_dim)\n self.time_mlp = nn.Sequential(nn.Linear(freq_embed_dim, dim), nn.SiLU(), nn.Linear(dim, dim))\n\n def forward(self, timestep: float[\"b\"]):\n time_hidden = self.time_embed(timestep)\n time_hidden = time_hidden.to(timestep.dtype)\n time = self.time_mlp(time_hidden) # b d\n return time","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.forward","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.modules.forward#L780-L784","kind":"function","name":"forward","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":780,"end_line":784,"context_start_line":760,"context_end_line":784,"code":"\n # process attention output for input x\n x = x + x_gate_msa.unsqueeze(1) * x_attn_output\n\n norm_x = self.ff_norm_x(x) * (1 + x_scale_mlp[:, None]) + x_shift_mlp[:, None]\n x_ff_output = self.ff_x(norm_x)\n x = x + x_gate_mlp.unsqueeze(1) * x_ff_output\n\n return c, x\n\n\n# time step conditioning embedding\n\n\nclass TimestepEmbedding(nn.Module):\n def __init__(self, dim, freq_embed_dim=256):\n super().__init__()\n self.time_embed = SinusPositionEmbedding(freq_embed_dim)\n self.time_mlp = nn.Sequential(nn.Linear(freq_embed_dim, dim), nn.SiLU(), nn.Linear(dim, dim))\n\n def forward(self, timestep: float[\"b\"]):\n time_hidden = self.time_embed(timestep)\n time_hidden = time_hidden.to(timestep.dtype)\n time = self.time_mlp(time_hidden) # b d\n return time","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.modules.__call__","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.modules.__call__#L552-L642","kind":"function","name":"__call__","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":552,"end_line":642,"context_start_line":532,"context_end_line":662,"code":" # linear proj\n x = attn.to_out[0](x)\n # dropout\n x = attn.to_out[1](x)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n\n return x\n\n\n# Joint Attention processor for MM-DiT\n# modified from diffusers/src/diffusers/models/attention_processor.py\n\n\nclass JointAttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn: Attention,\n x: float[\"b n d\"], # noised input x\n c: float[\"b nt d\"] = None, # context c, here text\n mask: bool[\"b n\"] | None = None,\n rope=None, # rotary position embedding for x\n c_rope=None, # rotary position embedding for c\n ) -> torch.FloatTensor:\n residual = x\n\n batch_size = c.shape[0]\n\n # `sample` projections\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n\n # `context` projections\n c_query = attn.to_q_c(c)\n c_key = attn.to_k_c(c)\n c_value = attn.to_v_c(c)\n\n # attention\n inner_dim = key.shape[-1]\n head_dim = inner_dim // attn.heads\n query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n c_query = c_query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n c_key = c_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n c_value = c_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)\n\n # qk norm\n if attn.q_norm is not None:\n query = attn.q_norm(query)\n if attn.k_norm is not None:\n key = attn.k_norm(key)\n if attn.c_q_norm is not None:\n c_query = attn.c_q_norm(c_query)\n if attn.c_k_norm is not None:\n c_key = attn.c_k_norm(c_key)\n\n # apply rope for context and noised input independently\n if rope is not None:\n freqs, xpos_scale = rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)\n key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)\n if c_rope is not None:\n freqs, xpos_scale = c_rope\n q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)\n c_query = apply_rotary_pos_emb(c_query, freqs, q_xpos_scale)\n c_key = apply_rotary_pos_emb(c_key, freqs, k_xpos_scale)\n\n # joint attention\n query = torch.cat([query, c_query], dim=2)\n key = torch.cat([key, c_key], dim=2)\n value = torch.cat([value, c_value], dim=2)\n\n # mask. e.g. inference got a batch with different target durations, mask out the padding\n if mask is not None:\n attn_mask = F.pad(mask, (0, c.shape[1]), value=True) # no mask for c (text)\n attn_mask = attn_mask.unsqueeze(1).unsqueeze(1) # 'b n -> b 1 1 n'\n attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])\n else:\n attn_mask = None\n\n x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)\n x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)\n x = x.to(query.dtype)\n\n # Split the attention outputs.\n x, c = (\n x[:, : residual.shape[1]],\n x[:, residual.shape[1] :],\n )\n\n # linear proj\n x = attn.to_out[0](x)\n # dropout\n x = attn.to_out[1](x)\n if not attn.context_pre_only:\n c = attn.to_out_c(c)\n\n if mask is not None:\n mask = mask.unsqueeze(-1)\n x = x.masked_fill(~mask, 0.0)\n # c = c.masked_fill(~mask, 0.) # no mask for c (text)\n\n return x, c\n\n\n# DiT Block\n\n\nclass DiTBlock(nn.Module):\n def __init__(\n self,\n dim,\n heads,\n dim_head,\n ff_mult=4,\n dropout=0.1,\n qk_norm=None,\n pe_attn_head=None,\n attn_backend=\"torch\", # \"torch\" or \"flash_attn\"\n attn_mask_enabled=True,\n ):\n super().__init__()\n","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.dit","uri":"program://DMOSpeech2/module/src.f5_tts.model_new.backbones.dit#L1-L259","kind":"module","name":"src.f5_tts.model_new.backbones.dit","path":"src/f5_tts/model_new/backbones/dit.py","language":"python","start_line":1,"end_line":259,"context_start_line":1,"context_end_line":259,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nimport torch\nimport torch.nn.functional as F\nfrom torch import nn\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model_new.modules import (\n AdaLayerNorm_Final,\n ConvNeXtV2Block,\n ConvPositionEmbedding,\n DiTBlock,\n TimestepEmbedding,\n get_pos_embed_indices,\n precompute_freqs_cis,\n)\n\n\n# Text embedding\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, text_num_embeds, text_dim, mask_padding=True, conv_layers=0, conv_mult=2):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim) # use 0 as filler token\n\n self.mask_padding = mask_padding # mask filler and batch padding tokens or not\n\n if conv_layers > 0:\n self.extra_modeling = True\n self.precompute_max_pos = 4096 # ~44s of 24khz audio\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(text_dim, self.precompute_max_pos), persistent=False)\n self.text_blocks = nn.Sequential(\n *[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)]\n )\n else:\n self.extra_modeling = False\n\n def forward(self, text: int[\"b nt\"], seq_len, drop_text=False): # noqa: F722\n text = text + 1 # use 0 as filler token. preprocess of batch pad -1, see list_str_to_idx()\n text = text[:, :seq_len] # curtail if character tokens are more than the mel spec tokens\n batch, text_len = text.shape[0], text.shape[1]\n text = F.pad(text, (0, seq_len - text_len), value=0)\n if self.mask_padding:\n text_mask = text == 0\n\n if drop_text: # cfg for text\n text = torch.zeros_like(text)\n\n text = self.text_embed(text) # b n -> b n d\n\n # possible extra modeling\n if self.extra_modeling:\n # sinus pos emb\n batch_start = torch.zeros((batch,), dtype=torch.long)\n pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n text = text + text_pos_embed\n\n # convnextv2 blocks\n if self.mask_padding:\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n for block in self.text_blocks:\n text = block(text)\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n else:\n text = self.text_blocks(text)\n\n return text\n\n\n# noised input audio and context mixing embedding\n\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using DiT blocks\n\n\nclass DiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n text_mask_padding=True,\n qk_norm=None,\n conv_layers=0,\n pe_attn_head=None,\n attn_backend=\"torch\", # \"torch\" | \"flash_attn\"\n attn_mask_enabled=False,\n long_skip_connection=False,\n checkpoint_activations=False,\n ):\n super().__init__()\n\n self.time_embed = TimestepEmbedding(dim)\n if text_dim is None:\n text_dim = mel_dim\n self.text_embed = TextEmbedding(\n text_num_embeds, text_dim, mask_padding=text_mask_padding, conv_layers=conv_layers\n )\n self.text_cond, self.text_uncond = None, None # text cache\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n self.dim = dim\n self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [\n DiTBlock(\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n ff_mult=ff_mult,\n dropout=dropout,\n qk_norm=qk_norm,\n pe_attn_head=pe_attn_head,\n attn_backend=attn_backend,\n attn_mask_enabled=attn_mask_enabled,\n )\n for _ in range(depth)\n ]\n )\n self.long_skip_connection = nn.Linear(dim * 2, dim, bias=False) if long_skip_connection else None\n\n self.norm_out = AdaLayerNorm_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.checkpoint_activations = checkpoint_activations\n\n self.initialize_weights()\n\n def initialize_weights(self):\n # Zero-out AdaLN layers in DiT blocks:\n for block in self.transformer_blocks:\n nn.init.constant_(block.attn_norm.linear.weight, 0)\n nn.init.constant_(block.attn_norm.linear.bias, 0)\n\n # Zero-out output layers:\n nn.init.constant_(self.norm_out.linear.weight, 0)\n nn.init.constant_(self.norm_out.linear.bias, 0)\n nn.init.constant_(self.proj_out.weight, 0)\n nn.init.constant_(self.proj_out.bias, 0)\n\n def ckpt_wrapper(self, module):\n # https://github.com/chuanyangjin/fast-DiT/blob/main/models.py\n def ckpt_forward(*inputs):\n outputs = module(*inputs)\n return outputs\n\n return ckpt_forward\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n seq_len = x.shape[1]\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, seq_len, drop_text=True)\n text_embed = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, seq_len, drop_text=False)\n text_embed = self.text_cond\n else:\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n return x\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, text: text, x: noised audio + cond audio + text\n t = self.time_embed(time)\n if cfg_infer: # pack cond & uncond forward: b n d -> 2b n d\n x_cond = self.get_input_embed(x, cond, text, drop_audio_cond=False, drop_text=False, cache=cache)\n x_uncond = self.get_input_embed(x, cond, text, drop_audio_cond=True, drop_text=True, cache=cache)\n x = torch.cat((x_cond, x_uncond), dim=0)\n t = torch.cat((t, t), dim=0)\n mask = torch.cat((mask, mask), dim=0) if mask is not None else None\n else:\n x = self.get_input_embed(x, cond, text, drop_audio_cond=drop_audio_cond, drop_text=drop_text, cache=cache)\n\n rope = self.rotary_embed.forward_from_seq_len(seq_len)\n\n if self.long_skip_connection is not None:\n residual = x\n\n for block in self.transformer_blocks:\n if self.checkpoint_activations:\n # https://pytorch.org/docs/stable/checkpoint.html#torch.utils.checkpoint.checkpoint\n x = torch.utils.checkpoint.checkpoint(self.ckpt_wrapper(block), x, t, mask, rope, use_reentrant=False)\n else:\n x = block(x, t, mask=mask, rope=rope)\n\n if self.long_skip_connection is not None:\n x = self.long_skip_connection(torch.cat((x, residual), dim=-1))\n\n x = self.norm_out(x, t)\n output = self.proj_out(x)\n\n return output","source_hash":"f6bfffce2c45a7042aa1d60f709b909db7dafdd95ac4425bf7fc2ee4eb75840d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.dit.TextEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.backbones.dit.TextEmbedding#L31-L78","kind":"class","name":"TextEmbedding","path":"src/f5_tts/model_new/backbones/dit.py","language":"python","start_line":31,"end_line":78,"context_start_line":11,"context_end_line":98,"code":"\nimport torch\nimport torch.nn.functional as F\nfrom torch import nn\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model_new.modules import (\n AdaLayerNorm_Final,\n ConvNeXtV2Block,\n ConvPositionEmbedding,\n DiTBlock,\n TimestepEmbedding,\n get_pos_embed_indices,\n precompute_freqs_cis,\n)\n\n\n# Text embedding\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, text_num_embeds, text_dim, mask_padding=True, conv_layers=0, conv_mult=2):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim) # use 0 as filler token\n\n self.mask_padding = mask_padding # mask filler and batch padding tokens or not\n\n if conv_layers > 0:\n self.extra_modeling = True\n self.precompute_max_pos = 4096 # ~44s of 24khz audio\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(text_dim, self.precompute_max_pos), persistent=False)\n self.text_blocks = nn.Sequential(\n *[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)]\n )\n else:\n self.extra_modeling = False\n\n def forward(self, text: int[\"b nt\"], seq_len, drop_text=False): # noqa: F722\n text = text + 1 # use 0 as filler token. preprocess of batch pad -1, see list_str_to_idx()\n text = text[:, :seq_len] # curtail if character tokens are more than the mel spec tokens\n batch, text_len = text.shape[0], text.shape[1]\n text = F.pad(text, (0, seq_len - text_len), value=0)\n if self.mask_padding:\n text_mask = text == 0\n\n if drop_text: # cfg for text\n text = torch.zeros_like(text)\n\n text = self.text_embed(text) # b n -> b n d\n\n # possible extra modeling\n if self.extra_modeling:\n # sinus pos emb\n batch_start = torch.zeros((batch,), dtype=torch.long)\n pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n text = text + text_pos_embed\n\n # convnextv2 blocks\n if self.mask_padding:\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n for block in self.text_blocks:\n text = block(text)\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n else:\n text = self.text_blocks(text)\n\n return text\n\n\n# noised input audio and context mixing embedding\n\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n","source_hash":"f6bfffce2c45a7042aa1d60f709b909db7dafdd95ac4425bf7fc2ee4eb75840d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.dit.InputEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.backbones.dit.InputEmbedding#L84-L96","kind":"class","name":"InputEmbedding","path":"src/f5_tts/model_new/backbones/dit.py","language":"python","start_line":84,"end_line":96,"context_start_line":64,"context_end_line":116,"code":" batch_start = torch.zeros((batch,), dtype=torch.long)\n pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n text = text + text_pos_embed\n\n # convnextv2 blocks\n if self.mask_padding:\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n for block in self.text_blocks:\n text = block(text)\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n else:\n text = self.text_blocks(text)\n\n return text\n\n\n# noised input audio and context mixing embedding\n\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using DiT blocks\n\n\nclass DiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n text_mask_padding=True,\n qk_norm=None,","source_hash":"f6bfffce2c45a7042aa1d60f709b909db7dafdd95ac4425bf7fc2ee4eb75840d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.dit.DiT","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.backbones.dit.DiT#L102-L259","kind":"class","name":"DiT","path":"src/f5_tts/model_new/backbones/dit.py","language":"python","start_line":102,"end_line":259,"context_start_line":82,"context_end_line":259,"code":"\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using DiT blocks\n\n\nclass DiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n text_mask_padding=True,\n qk_norm=None,\n conv_layers=0,\n pe_attn_head=None,\n attn_backend=\"torch\", # \"torch\" | \"flash_attn\"\n attn_mask_enabled=False,\n long_skip_connection=False,\n checkpoint_activations=False,\n ):\n super().__init__()\n\n self.time_embed = TimestepEmbedding(dim)\n if text_dim is None:\n text_dim = mel_dim\n self.text_embed = TextEmbedding(\n text_num_embeds, text_dim, mask_padding=text_mask_padding, conv_layers=conv_layers\n )\n self.text_cond, self.text_uncond = None, None # text cache\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n self.dim = dim\n self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [\n DiTBlock(\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n ff_mult=ff_mult,\n dropout=dropout,\n qk_norm=qk_norm,\n pe_attn_head=pe_attn_head,\n attn_backend=attn_backend,\n attn_mask_enabled=attn_mask_enabled,\n )\n for _ in range(depth)\n ]\n )\n self.long_skip_connection = nn.Linear(dim * 2, dim, bias=False) if long_skip_connection else None\n\n self.norm_out = AdaLayerNorm_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.checkpoint_activations = checkpoint_activations\n\n self.initialize_weights()\n\n def initialize_weights(self):\n # Zero-out AdaLN layers in DiT blocks:\n for block in self.transformer_blocks:\n nn.init.constant_(block.attn_norm.linear.weight, 0)\n nn.init.constant_(block.attn_norm.linear.bias, 0)\n\n # Zero-out output layers:\n nn.init.constant_(self.norm_out.linear.weight, 0)\n nn.init.constant_(self.norm_out.linear.bias, 0)\n nn.init.constant_(self.proj_out.weight, 0)\n nn.init.constant_(self.proj_out.bias, 0)\n\n def ckpt_wrapper(self, module):\n # https://github.com/chuanyangjin/fast-DiT/blob/main/models.py\n def ckpt_forward(*inputs):\n outputs = module(*inputs)\n return outputs\n\n return ckpt_forward\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n seq_len = x.shape[1]\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, seq_len, drop_text=True)\n text_embed = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, seq_len, drop_text=False)\n text_embed = self.text_cond\n else:\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n return x\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, text: text, x: noised audio + cond audio + text\n t = self.time_embed(time)\n if cfg_infer: # pack cond & uncond forward: b n d -> 2b n d\n x_cond = self.get_input_embed(x, cond, text, drop_audio_cond=False, drop_text=False, cache=cache)\n x_uncond = self.get_input_embed(x, cond, text, drop_audio_cond=True, drop_text=True, cache=cache)\n x = torch.cat((x_cond, x_uncond), dim=0)\n t = torch.cat((t, t), dim=0)\n mask = torch.cat((mask, mask), dim=0) if mask is not None else None\n else:\n x = self.get_input_embed(x, cond, text, drop_audio_cond=drop_audio_cond, drop_text=drop_text, cache=cache)\n\n rope = self.rotary_embed.forward_from_seq_len(seq_len)\n\n if self.long_skip_connection is not None:\n residual = x\n\n for block in self.transformer_blocks:\n if self.checkpoint_activations:\n # https://pytorch.org/docs/stable/checkpoint.html#torch.utils.checkpoint.checkpoint\n x = torch.utils.checkpoint.checkpoint(self.ckpt_wrapper(block), x, t, mask, rope, use_reentrant=False)\n else:\n x = block(x, t, mask=mask, rope=rope)\n\n if self.long_skip_connection is not None:\n x = self.long_skip_connection(torch.cat((x, residual), dim=-1))\n\n x = self.norm_out(x, t)\n output = self.proj_out(x)\n\n return output","source_hash":"f6bfffce2c45a7042aa1d60f709b909db7dafdd95ac4425bf7fc2ee4eb75840d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.dit.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.dit.__init__#L103-L163","kind":"function","name":"__init__","path":"src/f5_tts/model_new/backbones/dit.py","language":"python","start_line":103,"end_line":163,"context_start_line":83,"context_end_line":183,"code":"\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using DiT blocks\n\n\nclass DiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n text_mask_padding=True,\n qk_norm=None,\n conv_layers=0,\n pe_attn_head=None,\n attn_backend=\"torch\", # \"torch\" | \"flash_attn\"\n attn_mask_enabled=False,\n long_skip_connection=False,\n checkpoint_activations=False,\n ):\n super().__init__()\n\n self.time_embed = TimestepEmbedding(dim)\n if text_dim is None:\n text_dim = mel_dim\n self.text_embed = TextEmbedding(\n text_num_embeds, text_dim, mask_padding=text_mask_padding, conv_layers=conv_layers\n )\n self.text_cond, self.text_uncond = None, None # text cache\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n self.dim = dim\n self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [\n DiTBlock(\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n ff_mult=ff_mult,\n dropout=dropout,\n qk_norm=qk_norm,\n pe_attn_head=pe_attn_head,\n attn_backend=attn_backend,\n attn_mask_enabled=attn_mask_enabled,\n )\n for _ in range(depth)\n ]\n )\n self.long_skip_connection = nn.Linear(dim * 2, dim, bias=False) if long_skip_connection else None\n\n self.norm_out = AdaLayerNorm_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.checkpoint_activations = checkpoint_activations\n\n self.initialize_weights()\n\n def initialize_weights(self):\n # Zero-out AdaLN layers in DiT blocks:\n for block in self.transformer_blocks:\n nn.init.constant_(block.attn_norm.linear.weight, 0)\n nn.init.constant_(block.attn_norm.linear.bias, 0)\n\n # Zero-out output layers:\n nn.init.constant_(self.norm_out.linear.weight, 0)\n nn.init.constant_(self.norm_out.linear.bias, 0)\n nn.init.constant_(self.proj_out.weight, 0)\n nn.init.constant_(self.proj_out.bias, 0)\n\n def ckpt_wrapper(self, module):\n # https://github.com/chuanyangjin/fast-DiT/blob/main/models.py\n def ckpt_forward(*inputs):\n outputs = module(*inputs)\n return outputs\n\n return ckpt_forward","source_hash":"f6bfffce2c45a7042aa1d60f709b909db7dafdd95ac4425bf7fc2ee4eb75840d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.dit.forward","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.dit.forward#L214-L259","kind":"function","name":"forward","path":"src/f5_tts/model_new/backbones/dit.py","language":"python","start_line":214,"end_line":259,"context_start_line":194,"context_end_line":259,"code":" seq_len = x.shape[1]\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, seq_len, drop_text=True)\n text_embed = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, seq_len, drop_text=False)\n text_embed = self.text_cond\n else:\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n return x\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, text: text, x: noised audio + cond audio + text\n t = self.time_embed(time)\n if cfg_infer: # pack cond & uncond forward: b n d -> 2b n d\n x_cond = self.get_input_embed(x, cond, text, drop_audio_cond=False, drop_text=False, cache=cache)\n x_uncond = self.get_input_embed(x, cond, text, drop_audio_cond=True, drop_text=True, cache=cache)\n x = torch.cat((x_cond, x_uncond), dim=0)\n t = torch.cat((t, t), dim=0)\n mask = torch.cat((mask, mask), dim=0) if mask is not None else None\n else:\n x = self.get_input_embed(x, cond, text, drop_audio_cond=drop_audio_cond, drop_text=drop_text, cache=cache)\n\n rope = self.rotary_embed.forward_from_seq_len(seq_len)\n\n if self.long_skip_connection is not None:\n residual = x\n\n for block in self.transformer_blocks:\n if self.checkpoint_activations:\n # https://pytorch.org/docs/stable/checkpoint.html#torch.utils.checkpoint.checkpoint\n x = torch.utils.checkpoint.checkpoint(self.ckpt_wrapper(block), x, t, mask, rope, use_reentrant=False)\n else:\n x = block(x, t, mask=mask, rope=rope)\n\n if self.long_skip_connection is not None:\n x = self.long_skip_connection(torch.cat((x, residual), dim=-1))\n\n x = self.norm_out(x, t)\n output = self.proj_out(x)\n\n return output","source_hash":"f6bfffce2c45a7042aa1d60f709b909db7dafdd95ac4425bf7fc2ee4eb75840d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.dit.initialize_weights","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.dit.initialize_weights#L165-L175","kind":"function","name":"initialize_weights","path":"src/f5_tts/model_new/backbones/dit.py","language":"python","start_line":165,"end_line":175,"context_start_line":145,"context_end_line":195,"code":" dim_head=dim_head,\n ff_mult=ff_mult,\n dropout=dropout,\n qk_norm=qk_norm,\n pe_attn_head=pe_attn_head,\n attn_backend=attn_backend,\n attn_mask_enabled=attn_mask_enabled,\n )\n for _ in range(depth)\n ]\n )\n self.long_skip_connection = nn.Linear(dim * 2, dim, bias=False) if long_skip_connection else None\n\n self.norm_out = AdaLayerNorm_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.checkpoint_activations = checkpoint_activations\n\n self.initialize_weights()\n\n def initialize_weights(self):\n # Zero-out AdaLN layers in DiT blocks:\n for block in self.transformer_blocks:\n nn.init.constant_(block.attn_norm.linear.weight, 0)\n nn.init.constant_(block.attn_norm.linear.bias, 0)\n\n # Zero-out output layers:\n nn.init.constant_(self.norm_out.linear.weight, 0)\n nn.init.constant_(self.norm_out.linear.bias, 0)\n nn.init.constant_(self.proj_out.weight, 0)\n nn.init.constant_(self.proj_out.bias, 0)\n\n def ckpt_wrapper(self, module):\n # https://github.com/chuanyangjin/fast-DiT/blob/main/models.py\n def ckpt_forward(*inputs):\n outputs = module(*inputs)\n return outputs\n\n return ckpt_forward\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n seq_len = x.shape[1]\n if cache:","source_hash":"f6bfffce2c45a7042aa1d60f709b909db7dafdd95ac4425bf7fc2ee4eb75840d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.dit.ckpt_wrapper","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.dit.ckpt_wrapper#L177-L183","kind":"function","name":"ckpt_wrapper","path":"src/f5_tts/model_new/backbones/dit.py","language":"python","start_line":177,"end_line":183,"context_start_line":157,"context_end_line":203,"code":"\n self.norm_out = AdaLayerNorm_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.checkpoint_activations = checkpoint_activations\n\n self.initialize_weights()\n\n def initialize_weights(self):\n # Zero-out AdaLN layers in DiT blocks:\n for block in self.transformer_blocks:\n nn.init.constant_(block.attn_norm.linear.weight, 0)\n nn.init.constant_(block.attn_norm.linear.bias, 0)\n\n # Zero-out output layers:\n nn.init.constant_(self.norm_out.linear.weight, 0)\n nn.init.constant_(self.norm_out.linear.bias, 0)\n nn.init.constant_(self.proj_out.weight, 0)\n nn.init.constant_(self.proj_out.bias, 0)\n\n def ckpt_wrapper(self, module):\n # https://github.com/chuanyangjin/fast-DiT/blob/main/models.py\n def ckpt_forward(*inputs):\n outputs = module(*inputs)\n return outputs\n\n return ckpt_forward\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n seq_len = x.shape[1]\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, seq_len, drop_text=True)\n text_embed = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, seq_len, drop_text=False)\n text_embed = self.text_cond","source_hash":"f6bfffce2c45a7042aa1d60f709b909db7dafdd95ac4425bf7fc2ee4eb75840d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.dit.get_input_embed","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.dit.get_input_embed#L185-L209","kind":"function","name":"get_input_embed","path":"src/f5_tts/model_new/backbones/dit.py","language":"python","start_line":185,"end_line":209,"context_start_line":165,"context_end_line":229,"code":" def initialize_weights(self):\n # Zero-out AdaLN layers in DiT blocks:\n for block in self.transformer_blocks:\n nn.init.constant_(block.attn_norm.linear.weight, 0)\n nn.init.constant_(block.attn_norm.linear.bias, 0)\n\n # Zero-out output layers:\n nn.init.constant_(self.norm_out.linear.weight, 0)\n nn.init.constant_(self.norm_out.linear.bias, 0)\n nn.init.constant_(self.proj_out.weight, 0)\n nn.init.constant_(self.proj_out.bias, 0)\n\n def ckpt_wrapper(self, module):\n # https://github.com/chuanyangjin/fast-DiT/blob/main/models.py\n def ckpt_forward(*inputs):\n outputs = module(*inputs)\n return outputs\n\n return ckpt_forward\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n seq_len = x.shape[1]\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, seq_len, drop_text=True)\n text_embed = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, seq_len, drop_text=False)\n text_embed = self.text_cond\n else:\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n return x\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n","source_hash":"f6bfffce2c45a7042aa1d60f709b909db7dafdd95ac4425bf7fc2ee4eb75840d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.dit.clear_cache","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.dit.clear_cache#L211-L212","kind":"function","name":"clear_cache","path":"src/f5_tts/model_new/backbones/dit.py","language":"python","start_line":211,"end_line":212,"context_start_line":191,"context_end_line":232,"code":" drop_text: bool = False,\n cache: bool = True,\n ):\n seq_len = x.shape[1]\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, seq_len, drop_text=True)\n text_embed = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, seq_len, drop_text=False)\n text_embed = self.text_cond\n else:\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n return x\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, text: text, x: noised audio + cond audio + text\n t = self.time_embed(time)\n if cfg_infer: # pack cond & uncond forward: b n d -> 2b n d","source_hash":"f6bfffce2c45a7042aa1d60f709b909db7dafdd95ac4425bf7fc2ee4eb75840d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.dit.ckpt_forward","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.dit.ckpt_forward#L179-L181","kind":"function","name":"ckpt_forward","path":"src/f5_tts/model_new/backbones/dit.py","language":"python","start_line":179,"end_line":181,"context_start_line":159,"context_end_line":201,"code":" self.proj_out = nn.Linear(dim, mel_dim)\n\n self.checkpoint_activations = checkpoint_activations\n\n self.initialize_weights()\n\n def initialize_weights(self):\n # Zero-out AdaLN layers in DiT blocks:\n for block in self.transformer_blocks:\n nn.init.constant_(block.attn_norm.linear.weight, 0)\n nn.init.constant_(block.attn_norm.linear.bias, 0)\n\n # Zero-out output layers:\n nn.init.constant_(self.norm_out.linear.weight, 0)\n nn.init.constant_(self.norm_out.linear.bias, 0)\n nn.init.constant_(self.proj_out.weight, 0)\n nn.init.constant_(self.proj_out.bias, 0)\n\n def ckpt_wrapper(self, module):\n # https://github.com/chuanyangjin/fast-DiT/blob/main/models.py\n def ckpt_forward(*inputs):\n outputs = module(*inputs)\n return outputs\n\n return ckpt_forward\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n seq_len = x.shape[1]\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, seq_len, drop_text=True)\n text_embed = self.text_uncond\n else:\n if self.text_cond is None:","source_hash":"f6bfffce2c45a7042aa1d60f709b909db7dafdd95ac4425bf7fc2ee4eb75840d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.mmdit","uri":"program://DMOSpeech2/module/src.f5_tts.model_new.backbones.mmdit#L1-L212","kind":"module","name":"src.f5_tts.model_new.backbones.mmdit","path":"src/f5_tts/model_new/backbones/mmdit.py","language":"python","start_line":1,"end_line":212,"context_start_line":1,"context_end_line":212,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nimport torch\nfrom torch import nn\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model_new.modules import (\n AdaLayerNorm_Final,\n ConvPositionEmbedding,\n MMDiTBlock,\n TimestepEmbedding,\n get_pos_embed_indices,\n precompute_freqs_cis,\n)\n\n\n# text embedding\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, out_dim, text_num_embeds, mask_padding=True):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, out_dim) # will use 0 as filler token\n\n self.mask_padding = mask_padding # mask filler and batch padding tokens or not\n\n self.precompute_max_pos = 1024\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(out_dim, self.precompute_max_pos), persistent=False)\n\n def forward(self, text: int[\"b nt\"], drop_text=False) -> int[\"b nt d\"]: # noqa: F722\n text = text + 1 # use 0 as filler token. preprocess of batch pad -1, see list_str_to_idx()\n if self.mask_padding:\n text_mask = text == 0\n\n if drop_text: # cfg for text\n text = torch.zeros_like(text)\n\n text = self.text_embed(text) # b nt -> b nt d\n\n # sinus pos emb\n batch_start = torch.zeros((text.shape[0],), dtype=torch.long)\n batch_text_len = text.shape[1]\n pos_idx = get_pos_embed_indices(batch_start, batch_text_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n\n text = text + text_pos_embed\n\n if self.mask_padding:\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n\n return text\n\n\n# noised input & masked cond audio embedding\n\n\nclass AudioEmbedding(nn.Module):\n def __init__(self, in_dim, out_dim):\n super().__init__()\n self.linear = nn.Linear(2 * in_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond:\n cond = torch.zeros_like(cond)\n x = torch.cat((x, cond), dim=-1)\n x = self.linear(x)\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using MM-DiT blocks\n\n\nclass MMDiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_mask_padding=True,\n qk_norm=None,\n ):\n super().__init__()\n\n self.time_embed = TimestepEmbedding(dim)\n self.text_embed = TextEmbedding(dim, text_num_embeds, mask_padding=text_mask_padding)\n self.text_cond, self.text_uncond = None, None # text cache\n self.audio_embed = AudioEmbedding(mel_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n self.dim = dim\n self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [\n MMDiTBlock(\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n ff_mult=ff_mult,\n context_pre_only=i == depth - 1,\n qk_norm=qk_norm,\n )\n for i in range(depth)\n ]\n )\n self.norm_out = AdaLayerNorm_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.initialize_weights()\n\n def initialize_weights(self):\n # Zero-out AdaLN layers in MMDiT blocks:\n for block in self.transformer_blocks:\n nn.init.constant_(block.attn_norm_x.linear.weight, 0)\n nn.init.constant_(block.attn_norm_x.linear.bias, 0)\n nn.init.constant_(block.attn_norm_c.linear.weight, 0)\n nn.init.constant_(block.attn_norm_c.linear.bias, 0)\n\n # Zero-out output layers:\n nn.init.constant_(self.norm_out.linear.weight, 0)\n nn.init.constant_(self.norm_out.linear.bias, 0)\n nn.init.constant_(self.proj_out.weight, 0)\n nn.init.constant_(self.proj_out.bias, 0)\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, drop_text=True)\n c = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, drop_text=False)\n c = self.text_cond\n else:\n c = self.text_embed(text, drop_text=drop_text)\n x = self.audio_embed(x, cond, drop_audio_cond=drop_audio_cond)\n\n return x, c\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch = x.shape[0]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning (time), c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if cfg_infer: # pack cond & uncond forward: b n d -> 2b n d\n x_cond, c_cond = self.get_input_embed(x, cond, text, drop_audio_cond=False, drop_text=False, cache=cache)\n x_uncond, c_uncond = self.get_input_embed(x, cond, text, drop_audio_cond=True, drop_text=True, cache=cache)\n x = torch.cat((x_cond, x_uncond), dim=0)\n c = torch.cat((c_cond, c_uncond), dim=0)\n t = torch.cat((t, t), dim=0)\n mask = torch.cat((mask, mask), dim=0) if mask is not None else None\n else:\n x, c = self.get_input_embed(\n x, cond, text, drop_audio_cond=drop_audio_cond, drop_text=drop_text, cache=cache\n )\n\n seq_len = x.shape[1]\n text_len = text.shape[1]\n rope_audio = self.rotary_embed.forward_from_seq_len(seq_len)\n rope_text = self.rotary_embed.forward_from_seq_len(text_len)\n\n for block in self.transformer_blocks:\n c, x = block(x, c, t, mask=mask, rope=rope_audio, c_rope=rope_text)\n\n x = self.norm_out(x, t)\n output = self.proj_out(x)\n\n return output","source_hash":"95a35a246150d327f79a341648e833a6dbfecf112134ffbaaff6cefa97322e94","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.mmdit.TextEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.backbones.mmdit.TextEmbedding#L29-L60","kind":"class","name":"TextEmbedding","path":"src/f5_tts/model_new/backbones/mmdit.py","language":"python","start_line":29,"end_line":60,"context_start_line":9,"context_end_line":80,"code":"\nfrom __future__ import annotations\n\nimport torch\nfrom torch import nn\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model_new.modules import (\n AdaLayerNorm_Final,\n ConvPositionEmbedding,\n MMDiTBlock,\n TimestepEmbedding,\n get_pos_embed_indices,\n precompute_freqs_cis,\n)\n\n\n# text embedding\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, out_dim, text_num_embeds, mask_padding=True):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, out_dim) # will use 0 as filler token\n\n self.mask_padding = mask_padding # mask filler and batch padding tokens or not\n\n self.precompute_max_pos = 1024\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(out_dim, self.precompute_max_pos), persistent=False)\n\n def forward(self, text: int[\"b nt\"], drop_text=False) -> int[\"b nt d\"]: # noqa: F722\n text = text + 1 # use 0 as filler token. preprocess of batch pad -1, see list_str_to_idx()\n if self.mask_padding:\n text_mask = text == 0\n\n if drop_text: # cfg for text\n text = torch.zeros_like(text)\n\n text = self.text_embed(text) # b nt -> b nt d\n\n # sinus pos emb\n batch_start = torch.zeros((text.shape[0],), dtype=torch.long)\n batch_text_len = text.shape[1]\n pos_idx = get_pos_embed_indices(batch_start, batch_text_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n\n text = text + text_pos_embed\n\n if self.mask_padding:\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n\n return text\n\n\n# noised input & masked cond audio embedding\n\n\nclass AudioEmbedding(nn.Module):\n def __init__(self, in_dim, out_dim):\n super().__init__()\n self.linear = nn.Linear(2 * in_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond:\n cond = torch.zeros_like(cond)\n x = torch.cat((x, cond), dim=-1)\n x = self.linear(x)\n x = self.conv_pos_embed(x) + x\n return x\n\n","source_hash":"95a35a246150d327f79a341648e833a6dbfecf112134ffbaaff6cefa97322e94","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.mmdit.AudioEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.backbones.mmdit.AudioEmbedding#L66-L78","kind":"class","name":"AudioEmbedding","path":"src/f5_tts/model_new/backbones/mmdit.py","language":"python","start_line":66,"end_line":78,"context_start_line":46,"context_end_line":98,"code":"\n text = self.text_embed(text) # b nt -> b nt d\n\n # sinus pos emb\n batch_start = torch.zeros((text.shape[0],), dtype=torch.long)\n batch_text_len = text.shape[1]\n pos_idx = get_pos_embed_indices(batch_start, batch_text_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n\n text = text + text_pos_embed\n\n if self.mask_padding:\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n\n return text\n\n\n# noised input & masked cond audio embedding\n\n\nclass AudioEmbedding(nn.Module):\n def __init__(self, in_dim, out_dim):\n super().__init__()\n self.linear = nn.Linear(2 * in_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond:\n cond = torch.zeros_like(cond)\n x = torch.cat((x, cond), dim=-1)\n x = self.linear(x)\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using MM-DiT blocks\n\n\nclass MMDiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_mask_padding=True,\n qk_norm=None,\n ):","source_hash":"95a35a246150d327f79a341648e833a6dbfecf112134ffbaaff6cefa97322e94","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.mmdit.MMDiT","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.backbones.mmdit.MMDiT#L84-L212","kind":"class","name":"MMDiT","path":"src/f5_tts/model_new/backbones/mmdit.py","language":"python","start_line":84,"end_line":212,"context_start_line":64,"context_end_line":212,"code":"\n\nclass AudioEmbedding(nn.Module):\n def __init__(self, in_dim, out_dim):\n super().__init__()\n self.linear = nn.Linear(2 * in_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond:\n cond = torch.zeros_like(cond)\n x = torch.cat((x, cond), dim=-1)\n x = self.linear(x)\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using MM-DiT blocks\n\n\nclass MMDiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_mask_padding=True,\n qk_norm=None,\n ):\n super().__init__()\n\n self.time_embed = TimestepEmbedding(dim)\n self.text_embed = TextEmbedding(dim, text_num_embeds, mask_padding=text_mask_padding)\n self.text_cond, self.text_uncond = None, None # text cache\n self.audio_embed = AudioEmbedding(mel_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n self.dim = dim\n self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [\n MMDiTBlock(\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n ff_mult=ff_mult,\n context_pre_only=i == depth - 1,\n qk_norm=qk_norm,\n )\n for i in range(depth)\n ]\n )\n self.norm_out = AdaLayerNorm_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.initialize_weights()\n\n def initialize_weights(self):\n # Zero-out AdaLN layers in MMDiT blocks:\n for block in self.transformer_blocks:\n nn.init.constant_(block.attn_norm_x.linear.weight, 0)\n nn.init.constant_(block.attn_norm_x.linear.bias, 0)\n nn.init.constant_(block.attn_norm_c.linear.weight, 0)\n nn.init.constant_(block.attn_norm_c.linear.bias, 0)\n\n # Zero-out output layers:\n nn.init.constant_(self.norm_out.linear.weight, 0)\n nn.init.constant_(self.norm_out.linear.bias, 0)\n nn.init.constant_(self.proj_out.weight, 0)\n nn.init.constant_(self.proj_out.bias, 0)\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, drop_text=True)\n c = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, drop_text=False)\n c = self.text_cond\n else:\n c = self.text_embed(text, drop_text=drop_text)\n x = self.audio_embed(x, cond, drop_audio_cond=drop_audio_cond)\n\n return x, c\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch = x.shape[0]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning (time), c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if cfg_infer: # pack cond & uncond forward: b n d -> 2b n d\n x_cond, c_cond = self.get_input_embed(x, cond, text, drop_audio_cond=False, drop_text=False, cache=cache)\n x_uncond, c_uncond = self.get_input_embed(x, cond, text, drop_audio_cond=True, drop_text=True, cache=cache)\n x = torch.cat((x_cond, x_uncond), dim=0)\n c = torch.cat((c_cond, c_uncond), dim=0)\n t = torch.cat((t, t), dim=0)\n mask = torch.cat((mask, mask), dim=0) if mask is not None else None\n else:\n x, c = self.get_input_embed(\n x, cond, text, drop_audio_cond=drop_audio_cond, drop_text=drop_text, cache=cache\n )\n\n seq_len = x.shape[1]\n text_len = text.shape[1]\n rope_audio = self.rotary_embed.forward_from_seq_len(seq_len)\n rope_text = self.rotary_embed.forward_from_seq_len(text_len)\n\n for block in self.transformer_blocks:\n c, x = block(x, c, t, mask=mask, rope=rope_audio, c_rope=rope_text)\n\n x = self.norm_out(x, t)\n output = self.proj_out(x)\n\n return output","source_hash":"95a35a246150d327f79a341648e833a6dbfecf112134ffbaaff6cefa97322e94","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.mmdit.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.mmdit.__init__#L85-L128","kind":"function","name":"__init__","path":"src/f5_tts/model_new/backbones/mmdit.py","language":"python","start_line":85,"end_line":128,"context_start_line":65,"context_end_line":148,"code":"\nclass AudioEmbedding(nn.Module):\n def __init__(self, in_dim, out_dim):\n super().__init__()\n self.linear = nn.Linear(2 * in_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond:\n cond = torch.zeros_like(cond)\n x = torch.cat((x, cond), dim=-1)\n x = self.linear(x)\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Transformer backbone using MM-DiT blocks\n\n\nclass MMDiT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_mask_padding=True,\n qk_norm=None,\n ):\n super().__init__()\n\n self.time_embed = TimestepEmbedding(dim)\n self.text_embed = TextEmbedding(dim, text_num_embeds, mask_padding=text_mask_padding)\n self.text_cond, self.text_uncond = None, None # text cache\n self.audio_embed = AudioEmbedding(mel_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n self.dim = dim\n self.depth = depth\n\n self.transformer_blocks = nn.ModuleList(\n [\n MMDiTBlock(\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n ff_mult=ff_mult,\n context_pre_only=i == depth - 1,\n qk_norm=qk_norm,\n )\n for i in range(depth)\n ]\n )\n self.norm_out = AdaLayerNorm_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.initialize_weights()\n\n def initialize_weights(self):\n # Zero-out AdaLN layers in MMDiT blocks:\n for block in self.transformer_blocks:\n nn.init.constant_(block.attn_norm_x.linear.weight, 0)\n nn.init.constant_(block.attn_norm_x.linear.bias, 0)\n nn.init.constant_(block.attn_norm_c.linear.weight, 0)\n nn.init.constant_(block.attn_norm_c.linear.bias, 0)\n\n # Zero-out output layers:\n nn.init.constant_(self.norm_out.linear.weight, 0)\n nn.init.constant_(self.norm_out.linear.bias, 0)\n nn.init.constant_(self.proj_out.weight, 0)\n nn.init.constant_(self.proj_out.bias, 0)\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt","source_hash":"95a35a246150d327f79a341648e833a6dbfecf112134ffbaaff6cefa97322e94","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.mmdit.forward","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.mmdit.forward#L171-L212","kind":"function","name":"forward","path":"src/f5_tts/model_new/backbones/mmdit.py","language":"python","start_line":171,"end_line":212,"context_start_line":151,"context_end_line":212,"code":" cache: bool = True,\n ):\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, drop_text=True)\n c = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, drop_text=False)\n c = self.text_cond\n else:\n c = self.text_embed(text, drop_text=drop_text)\n x = self.audio_embed(x, cond, drop_audio_cond=drop_audio_cond)\n\n return x, c\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch = x.shape[0]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning (time), c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if cfg_infer: # pack cond & uncond forward: b n d -> 2b n d\n x_cond, c_cond = self.get_input_embed(x, cond, text, drop_audio_cond=False, drop_text=False, cache=cache)\n x_uncond, c_uncond = self.get_input_embed(x, cond, text, drop_audio_cond=True, drop_text=True, cache=cache)\n x = torch.cat((x_cond, x_uncond), dim=0)\n c = torch.cat((c_cond, c_uncond), dim=0)\n t = torch.cat((t, t), dim=0)\n mask = torch.cat((mask, mask), dim=0) if mask is not None else None\n else:\n x, c = self.get_input_embed(\n x, cond, text, drop_audio_cond=drop_audio_cond, drop_text=drop_text, cache=cache\n )\n\n seq_len = x.shape[1]\n text_len = text.shape[1]\n rope_audio = self.rotary_embed.forward_from_seq_len(seq_len)\n rope_text = self.rotary_embed.forward_from_seq_len(text_len)\n\n for block in self.transformer_blocks:\n c, x = block(x, c, t, mask=mask, rope=rope_audio, c_rope=rope_text)\n\n x = self.norm_out(x, t)\n output = self.proj_out(x)\n\n return output","source_hash":"95a35a246150d327f79a341648e833a6dbfecf112134ffbaaff6cefa97322e94","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.mmdit.initialize_weights","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.mmdit.initialize_weights#L130-L142","kind":"function","name":"initialize_weights","path":"src/f5_tts/model_new/backbones/mmdit.py","language":"python","start_line":130,"end_line":142,"context_start_line":110,"context_end_line":162,"code":"\n self.transformer_blocks = nn.ModuleList(\n [\n MMDiTBlock(\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n ff_mult=ff_mult,\n context_pre_only=i == depth - 1,\n qk_norm=qk_norm,\n )\n for i in range(depth)\n ]\n )\n self.norm_out = AdaLayerNorm_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.initialize_weights()\n\n def initialize_weights(self):\n # Zero-out AdaLN layers in MMDiT blocks:\n for block in self.transformer_blocks:\n nn.init.constant_(block.attn_norm_x.linear.weight, 0)\n nn.init.constant_(block.attn_norm_x.linear.bias, 0)\n nn.init.constant_(block.attn_norm_c.linear.weight, 0)\n nn.init.constant_(block.attn_norm_c.linear.bias, 0)\n\n # Zero-out output layers:\n nn.init.constant_(self.norm_out.linear.weight, 0)\n nn.init.constant_(self.norm_out.linear.bias, 0)\n nn.init.constant_(self.proj_out.weight, 0)\n nn.init.constant_(self.proj_out.bias, 0)\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, drop_text=True)\n c = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, drop_text=False)\n c = self.text_cond\n else:","source_hash":"95a35a246150d327f79a341648e833a6dbfecf112134ffbaaff6cefa97322e94","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.mmdit.get_input_embed","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.mmdit.get_input_embed#L144-L166","kind":"function","name":"get_input_embed","path":"src/f5_tts/model_new/backbones/mmdit.py","language":"python","start_line":144,"end_line":166,"context_start_line":124,"context_end_line":186,"code":" )\n self.norm_out = AdaLayerNorm_Final(dim) # final modulation\n self.proj_out = nn.Linear(dim, mel_dim)\n\n self.initialize_weights()\n\n def initialize_weights(self):\n # Zero-out AdaLN layers in MMDiT blocks:\n for block in self.transformer_blocks:\n nn.init.constant_(block.attn_norm_x.linear.weight, 0)\n nn.init.constant_(block.attn_norm_x.linear.bias, 0)\n nn.init.constant_(block.attn_norm_c.linear.weight, 0)\n nn.init.constant_(block.attn_norm_c.linear.bias, 0)\n\n # Zero-out output layers:\n nn.init.constant_(self.norm_out.linear.weight, 0)\n nn.init.constant_(self.norm_out.linear.bias, 0)\n nn.init.constant_(self.proj_out.weight, 0)\n nn.init.constant_(self.proj_out.bias, 0)\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, drop_text=True)\n c = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, drop_text=False)\n c = self.text_cond\n else:\n c = self.text_embed(text, drop_text=drop_text)\n x = self.audio_embed(x, cond, drop_audio_cond=drop_audio_cond)\n\n return x, c\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch = x.shape[0]\n if time.ndim == 0:\n time = time.repeat(batch)\n","source_hash":"95a35a246150d327f79a341648e833a6dbfecf112134ffbaaff6cefa97322e94","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.mmdit.clear_cache","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.mmdit.clear_cache#L168-L169","kind":"function","name":"clear_cache","path":"src/f5_tts/model_new/backbones/mmdit.py","language":"python","start_line":168,"end_line":169,"context_start_line":148,"context_end_line":189,"code":" text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, drop_text=True)\n c = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, drop_text=False)\n c = self.text_cond\n else:\n c = self.text_embed(text, drop_text=drop_text)\n x = self.audio_embed(x, cond, drop_audio_cond=drop_audio_cond)\n\n return x, c\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch = x.shape[0]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning (time), c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if cfg_infer: # pack cond & uncond forward: b n d -> 2b n d","source_hash":"95a35a246150d327f79a341648e833a6dbfecf112134ffbaaff6cefa97322e94","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.unett","uri":"program://DMOSpeech2/module/src.f5_tts.model_new.backbones.unett#L1-L273","kind":"module","name":"src.f5_tts.model_new.backbones.unett","path":"src/f5_tts/model_new/backbones/unett.py","language":"python","start_line":1,"end_line":273,"context_start_line":1,"context_end_line":273,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nfrom typing import Literal\n\nimport torch\nimport torch.nn.functional as F\nfrom torch import nn\nfrom x_transformers import RMSNorm\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model_new.modules import (\n Attention,\n AttnProcessor,\n ConvNeXtV2Block,\n ConvPositionEmbedding,\n FeedForward,\n TimestepEmbedding,\n get_pos_embed_indices,\n precompute_freqs_cis,\n)\n\n\n# Text embedding\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, text_num_embeds, text_dim, mask_padding=True, conv_layers=0, conv_mult=2):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim) # use 0 as filler token\n\n self.mask_padding = mask_padding # mask filler and batch padding tokens or not\n\n if conv_layers > 0:\n self.extra_modeling = True\n self.precompute_max_pos = 4096 # ~44s of 24khz audio\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(text_dim, self.precompute_max_pos), persistent=False)\n self.text_blocks = nn.Sequential(\n *[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)]\n )\n else:\n self.extra_modeling = False\n\n def forward(self, text: int[\"b nt\"], seq_len, drop_text=False): # noqa: F722\n text = text + 1 # use 0 as filler token. preprocess of batch pad -1, see list_str_to_idx()\n text = text[:, :seq_len] # curtail if character tokens are more than the mel spec tokens\n batch, text_len = text.shape[0], text.shape[1]\n text = F.pad(text, (0, seq_len - text_len), value=0)\n if self.mask_padding:\n text_mask = text == 0\n\n if drop_text: # cfg for text\n text = torch.zeros_like(text)\n\n text = self.text_embed(text) # b n -> b n d\n\n # possible extra modeling\n if self.extra_modeling:\n # sinus pos emb\n batch_start = torch.zeros((batch,), dtype=torch.long)\n pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n text = text + text_pos_embed\n\n # convnextv2 blocks\n if self.mask_padding:\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n for block in self.text_blocks:\n text = block(text)\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n else:\n text = self.text_blocks(text)\n\n return text\n\n\n# noised input audio and context mixing embedding\n\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Flat UNet Transformer backbone\n\n\nclass UNetT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n text_mask_padding=True,\n qk_norm=None,\n conv_layers=0,\n pe_attn_head=None,\n skip_connect_type: Literal[\"add\", \"concat\", \"none\"] = \"concat\",\n ):\n super().__init__()\n assert depth % 2 == 0, \"UNet-Transformer's depth should be even.\"\n\n self.time_embed = TimestepEmbedding(dim)\n if text_dim is None:\n text_dim = mel_dim\n self.text_embed = TextEmbedding(\n text_num_embeds, text_dim, mask_padding=text_mask_padding, conv_layers=conv_layers\n )\n self.text_cond, self.text_uncond = None, None # text cache\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n # transformer layers & skip connections\n\n self.dim = dim\n self.skip_connect_type = skip_connect_type\n needs_skip_proj = skip_connect_type == \"concat\"\n\n self.depth = depth\n self.layers = nn.ModuleList([])\n\n for idx in range(depth):\n is_later_half = idx >= (depth // 2)\n\n attn_norm = RMSNorm(dim)\n attn = Attention(\n processor=AttnProcessor(pe_attn_head=pe_attn_head),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n qk_norm=qk_norm,\n )\n\n ff_norm = RMSNorm(dim)\n ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n\n skip_proj = nn.Linear(dim * 2, dim, bias=False) if needs_skip_proj and is_later_half else None\n\n self.layers.append(\n nn.ModuleList(\n [\n skip_proj,\n attn_norm,\n attn,\n ff_norm,\n ff,\n ]\n )\n )\n\n self.norm_out = RMSNorm(dim)\n self.proj_out = nn.Linear(dim, mel_dim)\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n seq_len = x.shape[1]\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, seq_len, drop_text=True)\n text_embed = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, seq_len, drop_text=False)\n text_embed = self.text_cond\n else:\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n return x\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if cfg_infer: # pack cond & uncond forward: b n d -> 2b n d\n x_cond = self.get_input_embed(x, cond, text, drop_audio_cond=False, drop_text=False, cache=cache)\n x_uncond = self.get_input_embed(x, cond, text, drop_audio_cond=True, drop_text=True, cache=cache)\n x = torch.cat((x_cond, x_uncond), dim=0)\n t = torch.cat((t, t), dim=0)\n mask = torch.cat((mask, mask), dim=0) if mask is not None else None\n else:\n x = self.get_input_embed(x, cond, text, drop_audio_cond=drop_audio_cond, drop_text=drop_text, cache=cache)\n\n # postfix time t to input x, [b n d] -> [b n+1 d]\n x = torch.cat([t.unsqueeze(1), x], dim=1) # pack t to x\n if mask is not None:\n mask = F.pad(mask, (1, 0), value=1)\n\n rope = self.rotary_embed.forward_from_seq_len(seq_len + 1)\n\n # flat unet transformer\n skip_connect_type = self.skip_connect_type\n skips = []\n for idx, (maybe_skip_proj, attn_norm, attn, ff_norm, ff) in enumerate(self.layers):\n layer = idx + 1\n\n # skip connection logic\n is_first_half = layer <= (self.depth // 2)\n is_later_half = not is_first_half\n\n if is_first_half:\n skips.append(x)\n\n if is_later_half:\n skip = skips.pop()\n if skip_connect_type == \"concat\":\n x = torch.cat((x, skip), dim=-1)\n x = maybe_skip_proj(x)\n elif skip_connect_type == \"add\":\n x = x + skip\n\n # attention and feedforward blocks\n x = attn(attn_norm(x), rope=rope, mask=mask) + x\n x = ff(ff_norm(x)) + x\n\n assert len(skips) == 0\n\n x = self.norm_out(x)[:, 1:, :] # unpack t from x\n\n return self.proj_out(x)","source_hash":"108eedf2ce95cd9d8d8bdce990e98db27e089a47f9e147fb7ee3c620ae41ea50","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.unett.TextEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.backbones.unett.TextEmbedding#L35-L82","kind":"class","name":"TextEmbedding","path":"src/f5_tts/model_new/backbones/unett.py","language":"python","start_line":35,"end_line":82,"context_start_line":15,"context_end_line":102,"code":"import torch.nn.functional as F\nfrom torch import nn\nfrom x_transformers import RMSNorm\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model_new.modules import (\n Attention,\n AttnProcessor,\n ConvNeXtV2Block,\n ConvPositionEmbedding,\n FeedForward,\n TimestepEmbedding,\n get_pos_embed_indices,\n precompute_freqs_cis,\n)\n\n\n# Text embedding\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, text_num_embeds, text_dim, mask_padding=True, conv_layers=0, conv_mult=2):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim) # use 0 as filler token\n\n self.mask_padding = mask_padding # mask filler and batch padding tokens or not\n\n if conv_layers > 0:\n self.extra_modeling = True\n self.precompute_max_pos = 4096 # ~44s of 24khz audio\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(text_dim, self.precompute_max_pos), persistent=False)\n self.text_blocks = nn.Sequential(\n *[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)]\n )\n else:\n self.extra_modeling = False\n\n def forward(self, text: int[\"b nt\"], seq_len, drop_text=False): # noqa: F722\n text = text + 1 # use 0 as filler token. preprocess of batch pad -1, see list_str_to_idx()\n text = text[:, :seq_len] # curtail if character tokens are more than the mel spec tokens\n batch, text_len = text.shape[0], text.shape[1]\n text = F.pad(text, (0, seq_len - text_len), value=0)\n if self.mask_padding:\n text_mask = text == 0\n\n if drop_text: # cfg for text\n text = torch.zeros_like(text)\n\n text = self.text_embed(text) # b n -> b n d\n\n # possible extra modeling\n if self.extra_modeling:\n # sinus pos emb\n batch_start = torch.zeros((batch,), dtype=torch.long)\n pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n text = text + text_pos_embed\n\n # convnextv2 blocks\n if self.mask_padding:\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n for block in self.text_blocks:\n text = block(text)\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n else:\n text = self.text_blocks(text)\n\n return text\n\n\n# noised input audio and context mixing embedding\n\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n","source_hash":"108eedf2ce95cd9d8d8bdce990e98db27e089a47f9e147fb7ee3c620ae41ea50","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.unett.InputEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.backbones.unett.InputEmbedding#L88-L100","kind":"class","name":"InputEmbedding","path":"src/f5_tts/model_new/backbones/unett.py","language":"python","start_line":88,"end_line":100,"context_start_line":68,"context_end_line":120,"code":" batch_start = torch.zeros((batch,), dtype=torch.long)\n pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)\n text_pos_embed = self.freqs_cis[pos_idx]\n text = text + text_pos_embed\n\n # convnextv2 blocks\n if self.mask_padding:\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n for block in self.text_blocks:\n text = block(text)\n text = text.masked_fill(text_mask.unsqueeze(-1).expand(-1, -1, text.size(-1)), 0.0)\n else:\n text = self.text_blocks(text)\n\n return text\n\n\n# noised input audio and context mixing embedding\n\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Flat UNet Transformer backbone\n\n\nclass UNetT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n text_mask_padding=True,\n qk_norm=None,","source_hash":"108eedf2ce95cd9d8d8bdce990e98db27e089a47f9e147fb7ee3c620ae41ea50","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.unett.UNetT","uri":"program://DMOSpeech2/class/src.f5_tts.model_new.backbones.unett.UNetT#L106-L273","kind":"class","name":"UNetT","path":"src/f5_tts/model_new/backbones/unett.py","language":"python","start_line":106,"end_line":273,"context_start_line":86,"context_end_line":273,"code":"\n\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Flat UNet Transformer backbone\n\n\nclass UNetT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n text_mask_padding=True,\n qk_norm=None,\n conv_layers=0,\n pe_attn_head=None,\n skip_connect_type: Literal[\"add\", \"concat\", \"none\"] = \"concat\",\n ):\n super().__init__()\n assert depth % 2 == 0, \"UNet-Transformer's depth should be even.\"\n\n self.time_embed = TimestepEmbedding(dim)\n if text_dim is None:\n text_dim = mel_dim\n self.text_embed = TextEmbedding(\n text_num_embeds, text_dim, mask_padding=text_mask_padding, conv_layers=conv_layers\n )\n self.text_cond, self.text_uncond = None, None # text cache\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n # transformer layers & skip connections\n\n self.dim = dim\n self.skip_connect_type = skip_connect_type\n needs_skip_proj = skip_connect_type == \"concat\"\n\n self.depth = depth\n self.layers = nn.ModuleList([])\n\n for idx in range(depth):\n is_later_half = idx >= (depth // 2)\n\n attn_norm = RMSNorm(dim)\n attn = Attention(\n processor=AttnProcessor(pe_attn_head=pe_attn_head),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n qk_norm=qk_norm,\n )\n\n ff_norm = RMSNorm(dim)\n ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n\n skip_proj = nn.Linear(dim * 2, dim, bias=False) if needs_skip_proj and is_later_half else None\n\n self.layers.append(\n nn.ModuleList(\n [\n skip_proj,\n attn_norm,\n attn,\n ff_norm,\n ff,\n ]\n )\n )\n\n self.norm_out = RMSNorm(dim)\n self.proj_out = nn.Linear(dim, mel_dim)\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n seq_len = x.shape[1]\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, seq_len, drop_text=True)\n text_embed = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, seq_len, drop_text=False)\n text_embed = self.text_cond\n else:\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n return x\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if cfg_infer: # pack cond & uncond forward: b n d -> 2b n d\n x_cond = self.get_input_embed(x, cond, text, drop_audio_cond=False, drop_text=False, cache=cache)\n x_uncond = self.get_input_embed(x, cond, text, drop_audio_cond=True, drop_text=True, cache=cache)\n x = torch.cat((x_cond, x_uncond), dim=0)\n t = torch.cat((t, t), dim=0)\n mask = torch.cat((mask, mask), dim=0) if mask is not None else None\n else:\n x = self.get_input_embed(x, cond, text, drop_audio_cond=drop_audio_cond, drop_text=drop_text, cache=cache)\n\n # postfix time t to input x, [b n d] -> [b n+1 d]\n x = torch.cat([t.unsqueeze(1), x], dim=1) # pack t to x\n if mask is not None:\n mask = F.pad(mask, (1, 0), value=1)\n\n rope = self.rotary_embed.forward_from_seq_len(seq_len + 1)\n\n # flat unet transformer\n skip_connect_type = self.skip_connect_type\n skips = []\n for idx, (maybe_skip_proj, attn_norm, attn, ff_norm, ff) in enumerate(self.layers):\n layer = idx + 1\n\n # skip connection logic\n is_first_half = layer <= (self.depth // 2)\n is_later_half = not is_first_half\n\n if is_first_half:\n skips.append(x)\n\n if is_later_half:\n skip = skips.pop()\n if skip_connect_type == \"concat\":\n x = torch.cat((x, skip), dim=-1)\n x = maybe_skip_proj(x)\n elif skip_connect_type == \"add\":\n x = x + skip\n\n # attention and feedforward blocks\n x = attn(attn_norm(x), rope=rope, mask=mask) + x\n x = ff(ff_norm(x)) + x\n\n assert len(skips) == 0\n\n x = self.norm_out(x)[:, 1:, :] # unpack t from x\n\n return self.proj_out(x)","source_hash":"108eedf2ce95cd9d8d8bdce990e98db27e089a47f9e147fb7ee3c620ae41ea50","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.unett.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.unett.__init__#L107-L179","kind":"function","name":"__init__","path":"src/f5_tts/model_new/backbones/unett.py","language":"python","start_line":107,"end_line":179,"context_start_line":87,"context_end_line":199,"code":"\nclass InputEmbedding(nn.Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x: float[\"b n d\"], cond: float[\"b n d\"], text_embed: float[\"b n d\"], drop_audio_cond=False): # noqa: F722\n if drop_audio_cond: # cfg for cond audio\n cond = torch.zeros_like(cond)\n\n x = self.proj(torch.cat((x, cond, text_embed), dim=-1))\n x = self.conv_pos_embed(x) + x\n return x\n\n\n# Flat UNet Transformer backbone\n\n\nclass UNetT(nn.Module):\n def __init__(\n self,\n *,\n dim,\n depth=8,\n heads=8,\n dim_head=64,\n dropout=0.1,\n ff_mult=4,\n mel_dim=100,\n text_num_embeds=256,\n text_dim=None,\n text_mask_padding=True,\n qk_norm=None,\n conv_layers=0,\n pe_attn_head=None,\n skip_connect_type: Literal[\"add\", \"concat\", \"none\"] = \"concat\",\n ):\n super().__init__()\n assert depth % 2 == 0, \"UNet-Transformer's depth should be even.\"\n\n self.time_embed = TimestepEmbedding(dim)\n if text_dim is None:\n text_dim = mel_dim\n self.text_embed = TextEmbedding(\n text_num_embeds, text_dim, mask_padding=text_mask_padding, conv_layers=conv_layers\n )\n self.text_cond, self.text_uncond = None, None # text cache\n self.input_embed = InputEmbedding(mel_dim, text_dim, dim)\n\n self.rotary_embed = RotaryEmbedding(dim_head)\n\n # transformer layers & skip connections\n\n self.dim = dim\n self.skip_connect_type = skip_connect_type\n needs_skip_proj = skip_connect_type == \"concat\"\n\n self.depth = depth\n self.layers = nn.ModuleList([])\n\n for idx in range(depth):\n is_later_half = idx >= (depth // 2)\n\n attn_norm = RMSNorm(dim)\n attn = Attention(\n processor=AttnProcessor(pe_attn_head=pe_attn_head),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n qk_norm=qk_norm,\n )\n\n ff_norm = RMSNorm(dim)\n ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n\n skip_proj = nn.Linear(dim * 2, dim, bias=False) if needs_skip_proj and is_later_half else None\n\n self.layers.append(\n nn.ModuleList(\n [\n skip_proj,\n attn_norm,\n attn,\n ff_norm,\n ff,\n ]\n )\n )\n\n self.norm_out = RMSNorm(dim)\n self.proj_out = nn.Linear(dim, mel_dim)\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n seq_len = x.shape[1]\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, seq_len, drop_text=True)\n text_embed = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, seq_len, drop_text=False)\n text_embed = self.text_cond","source_hash":"108eedf2ce95cd9d8d8bdce990e98db27e089a47f9e147fb7ee3c620ae41ea50","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.unett.forward","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.unett.forward#L210-L273","kind":"function","name":"forward","path":"src/f5_tts/model_new/backbones/unett.py","language":"python","start_line":210,"end_line":273,"context_start_line":190,"context_end_line":273,"code":" seq_len = x.shape[1]\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, seq_len, drop_text=True)\n text_embed = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, seq_len, drop_text=False)\n text_embed = self.text_cond\n else:\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n return x\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if cfg_infer: # pack cond & uncond forward: b n d -> 2b n d\n x_cond = self.get_input_embed(x, cond, text, drop_audio_cond=False, drop_text=False, cache=cache)\n x_uncond = self.get_input_embed(x, cond, text, drop_audio_cond=True, drop_text=True, cache=cache)\n x = torch.cat((x_cond, x_uncond), dim=0)\n t = torch.cat((t, t), dim=0)\n mask = torch.cat((mask, mask), dim=0) if mask is not None else None\n else:\n x = self.get_input_embed(x, cond, text, drop_audio_cond=drop_audio_cond, drop_text=drop_text, cache=cache)\n\n # postfix time t to input x, [b n d] -> [b n+1 d]\n x = torch.cat([t.unsqueeze(1), x], dim=1) # pack t to x\n if mask is not None:\n mask = F.pad(mask, (1, 0), value=1)\n\n rope = self.rotary_embed.forward_from_seq_len(seq_len + 1)\n\n # flat unet transformer\n skip_connect_type = self.skip_connect_type\n skips = []\n for idx, (maybe_skip_proj, attn_norm, attn, ff_norm, ff) in enumerate(self.layers):\n layer = idx + 1\n\n # skip connection logic\n is_first_half = layer <= (self.depth // 2)\n is_later_half = not is_first_half\n\n if is_first_half:\n skips.append(x)\n\n if is_later_half:\n skip = skips.pop()\n if skip_connect_type == \"concat\":\n x = torch.cat((x, skip), dim=-1)\n x = maybe_skip_proj(x)\n elif skip_connect_type == \"add\":\n x = x + skip\n\n # attention and feedforward blocks\n x = attn(attn_norm(x), rope=rope, mask=mask) + x\n x = ff(ff_norm(x)) + x\n\n assert len(skips) == 0\n\n x = self.norm_out(x)[:, 1:, :] # unpack t from x\n\n return self.proj_out(x)","source_hash":"108eedf2ce95cd9d8d8bdce990e98db27e089a47f9e147fb7ee3c620ae41ea50","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.unett.get_input_embed","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.unett.get_input_embed#L181-L205","kind":"function","name":"get_input_embed","path":"src/f5_tts/model_new/backbones/unett.py","language":"python","start_line":181,"end_line":205,"context_start_line":161,"context_end_line":225,"code":" ff_norm = RMSNorm(dim)\n ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate=\"tanh\")\n\n skip_proj = nn.Linear(dim * 2, dim, bias=False) if needs_skip_proj and is_later_half else None\n\n self.layers.append(\n nn.ModuleList(\n [\n skip_proj,\n attn_norm,\n attn,\n ff_norm,\n ff,\n ]\n )\n )\n\n self.norm_out = RMSNorm(dim)\n self.proj_out = nn.Linear(dim, mel_dim)\n\n def get_input_embed(\n self,\n x, # b n d\n cond, # b n d\n text, # b nt\n drop_audio_cond: bool = False,\n drop_text: bool = False,\n cache: bool = True,\n ):\n seq_len = x.shape[1]\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, seq_len, drop_text=True)\n text_embed = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, seq_len, drop_text=False)\n text_embed = self.text_cond\n else:\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n return x\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n","source_hash":"108eedf2ce95cd9d8d8bdce990e98db27e089a47f9e147fb7ee3c620ae41ea50","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.model_new.backbones.unett.clear_cache","uri":"program://DMOSpeech2/function/src.f5_tts.model_new.backbones.unett.clear_cache#L207-L208","kind":"function","name":"clear_cache","path":"src/f5_tts/model_new/backbones/unett.py","language":"python","start_line":207,"end_line":208,"context_start_line":187,"context_end_line":228,"code":" drop_text: bool = False,\n cache: bool = True,\n ):\n seq_len = x.shape[1]\n if cache:\n if drop_text:\n if self.text_uncond is None:\n self.text_uncond = self.text_embed(text, seq_len, drop_text=True)\n text_embed = self.text_uncond\n else:\n if self.text_cond is None:\n self.text_cond = self.text_embed(text, seq_len, drop_text=False)\n text_embed = self.text_cond\n else:\n text_embed = self.text_embed(text, seq_len, drop_text=drop_text)\n\n x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)\n\n return x\n\n def clear_cache(self):\n self.text_cond, self.text_uncond = None, None\n\n def forward(\n self,\n x: float[\"b n d\"], # nosied input audio # noqa: F722\n cond: float[\"b n d\"], # masked cond audio # noqa: F722\n text: int[\"b nt\"], # text # noqa: F722\n time: float[\"b\"] | float[\"\"], # time step # noqa: F821 F722\n mask: bool[\"b n\"] | None = None, # noqa: F722\n drop_audio_cond: bool = False, # cfg for cond audio\n drop_text: bool = False, # cfg for text\n cfg_infer: bool = False, # cfg inference, pack cond & uncond forward\n cache: bool = False,\n ):\n batch, seq_len = x.shape[0], x.shape[1]\n if time.ndim == 0:\n time = time.repeat(batch)\n\n # t: conditioning time, c: context (text + masked cond audio), x: noised input audio\n t = self.time_embed(time)\n if cfg_infer: # pack cond & uncond forward: b n d -> 2b n d","source_hash":"108eedf2ce95cd9d8d8bdce990e98db27e089a47f9e147fb7ee3c620ae41ea50","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.utils_eval","uri":"program://DMOSpeech2/module/src.f5_tts.eval.utils_eval#L1-L419","kind":"module","name":"src.f5_tts.eval.utils_eval","path":"src/f5_tts/eval/utils_eval.py","language":"python","start_line":1,"end_line":419,"context_start_line":1,"context_end_line":419,"code":"import math\nimport os\nimport random\nimport string\nfrom pathlib import Path\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom tqdm import tqdm\n\nfrom f5_tts.eval.ecapa_tdnn import ECAPA_TDNN_SMALL\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import convert_char_to_pinyin\n\n\n# seedtts testset metainfo: utt, prompt_text, prompt_wav, gt_text, gt_wav\ndef get_seedtts_testset_metainfo(metalst):\n f = open(metalst)\n lines = f.readlines()\n f.close()\n metainfo = []\n for line in lines:\n if len(line.strip().split(\"|\")) == 5:\n utt, prompt_text, prompt_wav, gt_text, gt_wav = line.strip().split(\"|\")\n elif len(line.strip().split(\"|\")) == 4:\n utt, prompt_text, prompt_wav, gt_text = line.strip().split(\"|\")\n gt_wav = os.path.join(os.path.dirname(metalst), \"wavs\", utt + \".wav\")\n if not os.path.isabs(prompt_wav):\n prompt_wav = os.path.join(os.path.dirname(metalst), prompt_wav)\n metainfo.append((utt, prompt_text, prompt_wav, gt_text, gt_wav))\n return metainfo\n\n\n# librispeech test-clean metainfo: gen_utt, ref_txt, ref_wav, gen_txt, gen_wav\ndef get_librispeech_test_clean_metainfo(metalst, librispeech_test_clean_path):\n f = open(metalst)\n lines = f.readlines()\n f.close()\n metainfo = []\n for line in lines:\n ref_utt, ref_dur, ref_txt, gen_utt, gen_dur, gen_txt = line.strip().split(\"\\t\")\n\n # ref_txt = ref_txt[0] + ref_txt[1:].lower() + '.' # if use librispeech test-clean (no-pc)\n ref_spk_id, ref_chaptr_id, _ = ref_utt.split(\"-\")\n ref_wav = os.path.join(librispeech_test_clean_path, ref_spk_id, ref_chaptr_id, ref_utt + \".flac\")\n\n # gen_txt = gen_txt[0] + gen_txt[1:].lower() + '.' # if use librispeech test-clean (no-pc)\n gen_spk_id, gen_chaptr_id, _ = gen_utt.split(\"-\")\n gen_wav = os.path.join(librispeech_test_clean_path, gen_spk_id, gen_chaptr_id, gen_utt + \".flac\")\n\n metainfo.append((gen_utt, ref_txt, ref_wav, \" \" + gen_txt, gen_wav))\n\n return metainfo\n\n\n# padded to max length mel batch\ndef padded_mel_batch(ref_mels):\n max_mel_length = torch.LongTensor([mel.shape[-1] for mel in ref_mels]).amax()\n padded_ref_mels = []\n for mel in ref_mels:\n padded_ref_mel = F.pad(mel, (0, max_mel_length - mel.shape[-1]), value=0)\n padded_ref_mels.append(padded_ref_mel)\n padded_ref_mels = torch.stack(padded_ref_mels)\n padded_ref_mels = padded_ref_mels.permute(0, 2, 1)\n return padded_ref_mels\n\n\n# get prompts from metainfo containing: utt, prompt_text, prompt_wav, gt_text, gt_wav\n\n\ndef get_inference_prompt(\n metainfo,\n speed=1.0,\n tokenizer=\"pinyin\",\n polyphone=True,\n target_sample_rate=24000,\n n_fft=1024,\n win_length=1024,\n n_mel_channels=100,\n hop_length=256,\n mel_spec_type=\"vocos\",\n target_rms=0.1,\n use_truth_duration=False,\n infer_batch_size=1,\n num_buckets=200,\n min_secs=3,\n max_secs=40,\n):\n prompts_all = []\n\n min_tokens = min_secs * target_sample_rate // hop_length\n max_tokens = max_secs * target_sample_rate // hop_length\n\n batch_accum = [0] * num_buckets\n utts, ref_rms_list, ref_mels, ref_mel_lens, total_mel_lens, final_text_list = (\n [[] for _ in range(num_buckets)] for _ in range(6)\n )\n\n mel_spectrogram = MelSpec(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n )\n\n for utt, prompt_text, prompt_wav, gt_text, gt_wav in tqdm(metainfo, desc=\"Processing prompts...\"):\n # Audio\n ref_audio, ref_sr = torchaudio.load(prompt_wav)\n ref_rms = torch.sqrt(torch.mean(torch.square(ref_audio)))\n if ref_rms < target_rms:\n ref_audio = ref_audio * target_rms / ref_rms\n assert ref_audio.shape[-1] > 5000, f\"Empty prompt wav: {prompt_wav}, or torchaudio backend issue.\"\n if ref_sr != target_sample_rate:\n resampler = torchaudio.transforms.Resample(ref_sr, target_sample_rate)\n ref_audio = resampler(ref_audio)\n\n # Text\n if len(prompt_text[-1].encode(\"utf-8\")) == 1:\n prompt_text = prompt_text + \" \"\n text = [prompt_text + gt_text]\n if tokenizer == \"pinyin\":\n text_list = convert_char_to_pinyin(text, polyphone=polyphone)\n else:\n text_list = text\n\n # to mel spectrogram\n ref_mel = mel_spectrogram(ref_audio)\n ref_mel = ref_mel.squeeze(0)\n\n # Duration, mel frame length\n ref_mel_len = ref_mel.shape[-1]\n\n if use_truth_duration:\n gt_audio, gt_sr = torchaudio.load(gt_wav)\n if gt_sr != target_sample_rate:\n resampler = torchaudio.transforms.Resample(gt_sr, target_sample_rate)\n gt_audio = resampler(gt_audio)\n total_mel_len = ref_mel_len + int(gt_audio.shape[-1] / hop_length / speed)\n\n # # test vocoder resynthesis\n # ref_audio = gt_audio\n else:\n ref_text_len = len(prompt_text.encode(\"utf-8\"))\n gen_text_len = len(gt_text.encode(\"utf-8\"))\n total_mel_len = ref_mel_len + int(ref_mel_len / ref_text_len * gen_text_len / speed)\n\n # deal with batch\n assert infer_batch_size > 0, \"infer_batch_size should be greater than 0.\"\n assert min_tokens <= total_mel_len <= max_tokens, (\n f\"Audio {utt} has duration {total_mel_len * hop_length // target_sample_rate}s out of range [{min_secs}, {max_secs}].\"\n )\n bucket_i = math.floor((total_mel_len - min_tokens) / (max_tokens - min_tokens + 1) * num_buckets)\n\n utts[bucket_i].append(utt)\n ref_rms_list[bucket_i].append(ref_rms)\n ref_mels[bucket_i].append(ref_mel)\n ref_mel_lens[bucket_i].append(ref_mel_len)\n total_mel_lens[bucket_i].append(total_mel_len)\n final_text_list[bucket_i].extend(text_list)\n\n batch_accum[bucket_i] += total_mel_len\n\n if batch_accum[bucket_i] >= infer_batch_size:\n # print(f\"\\n{len(ref_mels[bucket_i][0][0])}\\n{ref_mel_lens[bucket_i]}\\n{total_mel_lens[bucket_i]}\")\n prompts_all.append(\n (\n utts[bucket_i],\n ref_rms_list[bucket_i],\n padded_mel_batch(ref_mels[bucket_i]),\n ref_mel_lens[bucket_i],\n total_mel_lens[bucket_i],\n final_text_list[bucket_i],\n )\n )\n batch_accum[bucket_i] = 0\n (\n utts[bucket_i],\n ref_rms_list[bucket_i],\n ref_mels[bucket_i],\n ref_mel_lens[bucket_i],\n total_mel_lens[bucket_i],\n final_text_list[bucket_i],\n ) = [], [], [], [], [], []\n\n # add residual\n for bucket_i, bucket_frames in enumerate(batch_accum):\n if bucket_frames > 0:\n prompts_all.append(\n (\n utts[bucket_i],\n ref_rms_list[bucket_i],\n padded_mel_batch(ref_mels[bucket_i]),\n ref_mel_lens[bucket_i],\n total_mel_lens[bucket_i],\n final_text_list[bucket_i],\n )\n )\n # not only leave easy work for last workers\n random.seed(666)\n random.shuffle(prompts_all)\n\n return prompts_all\n\n\n# get wav_res_ref_text of seed-tts test metalst\n# https://github.com/BytedanceSpeech/seed-tts-eval\n\n\ndef get_seed_tts_test(metalst, gen_wav_dir, gpus):\n f = open(metalst)\n lines = f.readlines()\n f.close()\n\n test_set_ = []\n for line in tqdm(lines):\n if len(line.strip().split(\"|\")) == 5:\n utt, prompt_text, prompt_wav, gt_text, gt_wav = line.strip().split(\"|\")\n elif len(line.strip().split(\"|\")) == 4:\n utt, prompt_text, prompt_wav, gt_text = line.strip().split(\"|\")\n\n if not os.path.exists(os.path.join(gen_wav_dir, utt + \".wav\")):\n continue\n gen_wav = os.path.join(gen_wav_dir, utt + \".wav\")\n if not os.path.isabs(prompt_wav):\n prompt_wav = os.path.join(os.path.dirname(metalst), prompt_wav)\n\n test_set_.append((gen_wav, prompt_wav, gt_text))\n\n num_jobs = len(gpus)\n if num_jobs == 1:\n return [(gpus[0], test_set_)]\n\n wav_per_job = len(test_set_) // num_jobs + 1\n test_set = []\n for i in range(num_jobs):\n test_set.append((gpus[i], test_set_[i * wav_per_job : (i + 1) * wav_per_job]))\n\n return test_set\n\n\n# get librispeech test-clean cross sentence test\n\n\ndef get_librispeech_test(metalst, gen_wav_dir, gpus, librispeech_test_clean_path, eval_ground_truth=False):\n f = open(metalst)\n lines = f.readlines()\n f.close()\n\n test_set_ = []\n for line in tqdm(lines):\n ref_utt, ref_dur, ref_txt, gen_utt, gen_dur, gen_txt = line.strip().split(\"\\t\")\n\n if eval_ground_truth:\n gen_spk_id, gen_chaptr_id, _ = gen_utt.split(\"-\")\n gen_wav = os.path.join(librispeech_test_clean_path, gen_spk_id, gen_chaptr_id, gen_utt + \".flac\")\n else:\n if not os.path.exists(os.path.join(gen_wav_dir, gen_utt + \".wav\")):\n raise FileNotFoundError(f\"Generated wav not found: {gen_utt}\")\n gen_wav = os.path.join(gen_wav_dir, gen_utt + \".wav\")\n\n ref_spk_id, ref_chaptr_id, _ = ref_utt.split(\"-\")\n ref_wav = os.path.join(librispeech_test_clean_path, ref_spk_id, ref_chaptr_id, ref_utt + \".flac\")\n\n test_set_.append((gen_wav, ref_wav, gen_txt))\n\n num_jobs = len(gpus)\n if num_jobs == 1:\n return [(gpus[0], test_set_)]\n\n wav_per_job = len(test_set_) // num_jobs + 1\n test_set = []\n for i in range(num_jobs):\n test_set.append((gpus[i], test_set_[i * wav_per_job : (i + 1) * wav_per_job]))\n\n return test_set\n\n\n# load asr model\n\n\ndef load_asr_model(lang, ckpt_dir=\"\"):\n if lang == \"zh\":\n from funasr import AutoModel\n\n model = AutoModel(\n model=os.path.join(ckpt_dir, \"paraformer-zh\"),\n # vad_model = os.path.join(ckpt_dir, \"fsmn-vad\"),\n # punc_model = os.path.join(ckpt_dir, \"ct-punc\"),\n # spk_model = os.path.join(ckpt_dir, \"cam++\"),\n disable_update=True,\n ) # following seed-tts setting\n elif lang == \"en\":\n from faster_whisper import WhisperModel\n\n model_size = \"large-v3\" if ckpt_dir == \"\" else ckpt_dir\n model = WhisperModel(model_size, device=\"cuda\", compute_type=\"float16\")\n return model\n\n\n# WER Evaluation, the way Seed-TTS does\n\n\ndef run_asr_wer(args):\n rank, lang, test_set, ckpt_dir = args\n\n if lang == \"zh\":\n import zhconv\n\n torch.cuda.set_device(rank)\n elif lang == \"en\":\n os.environ[\"CUDA_VISIBLE_DEVICES\"] = str(rank)\n else:\n raise NotImplementedError(\n \"lang support only 'zh' (funasr paraformer-zh), 'en' (faster-whisper-large-v3), for now.\"\n )\n\n asr_model = load_asr_model(lang, ckpt_dir=ckpt_dir)\n\n from zhon.hanzi import punctuation\n\n punctuation_all = punctuation + string.punctuation\n wer_results = []\n\n from jiwer import compute_measures\n\n for gen_wav, prompt_wav, truth in tqdm(test_set):\n if lang == \"zh\":\n res = asr_model.generate(input=gen_wav, batch_size_s=300, disable_pbar=True)\n hypo = res[0][\"text\"]\n hypo = zhconv.convert(hypo, \"zh-cn\")\n elif lang == \"en\":\n segments, _ = asr_model.transcribe(gen_wav, beam_size=5, language=\"en\")\n hypo = \"\"\n for segment in segments:\n hypo = hypo + \" \" + segment.text\n\n raw_truth = truth\n raw_hypo = hypo\n\n for x in punctuation_all:\n truth = truth.replace(x, \"\")\n hypo = hypo.replace(x, \"\")\n\n truth = truth.replace(\" \", \" \")\n hypo = hypo.replace(\" \", \" \")\n\n if lang == \"zh\":\n truth = \" \".join([x for x in truth])\n hypo = \" \".join([x for x in hypo])\n elif lang == \"en\":\n truth = truth.lower()\n hypo = hypo.lower()\n\n measures = compute_measures(truth, hypo)\n wer = measures[\"wer\"]\n\n # ref_list = truth.split(\" \")\n # subs = measures[\"substitutions\"] / len(ref_list)\n # dele = measures[\"deletions\"] / len(ref_list)\n # inse = measures[\"insertions\"] / len(ref_list)\n\n wer_results.append(\n {\n \"wav\": Path(gen_wav).stem,\n \"truth\": raw_truth,\n \"hypo\": raw_hypo,\n \"wer\": wer,\n }\n )\n\n return wer_results\n\n\n# SIM Evaluation\n\n\ndef run_sim(args):\n rank, test_set, ckpt_dir = args\n device = f\"cuda:{rank}\"\n\n model = ECAPA_TDNN_SMALL(feat_dim=1024, feat_type=\"wavlm_large\", config_path=None)\n state_dict = torch.load(ckpt_dir, weights_only=True, map_location=lambda storage, loc: storage)\n model.load_state_dict(state_dict[\"model\"], strict=False)\n\n use_gpu = True if torch.cuda.is_available() else False\n if use_gpu:\n model = model.cuda(device)\n model.eval()\n\n sim_results = []\n for gen_wav, prompt_wav, truth in tqdm(test_set):\n wav1, sr1 = torchaudio.load(gen_wav)\n wav2, sr2 = torchaudio.load(prompt_wav)\n\n resample1 = torchaudio.transforms.Resample(orig_freq=sr1, new_freq=16000)\n resample2 = torchaudio.transforms.Resample(orig_freq=sr2, new_freq=16000)\n wav1 = resample1(wav1)\n wav2 = resample2(wav2)\n\n if use_gpu:\n wav1 = wav1.cuda(device)\n wav2 = wav2.cuda(device)\n with torch.no_grad():\n emb1 = model(wav1)\n emb2 = model(wav2)\n\n sim = F.cosine_similarity(emb1, emb2)[0].item()\n # print(f\"VSim score between two audios: {sim:.4f} (-1.0, 1.0).\")\n sim_results.append(\n {\n \"wav\": Path(gen_wav).stem,\n \"sim\": sim,\n }\n )\n\n return sim_results","source_hash":"5ca3e0c70d7143d3bf5f23182499b82ddd4bba026d76b9a99d003a65136d7b68","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.utils_eval.get_seedtts_testset_metainfo","uri":"program://DMOSpeech2/function/src.f5_tts.eval.utils_eval.get_seedtts_testset_metainfo#L18-L32","kind":"function","name":"get_seedtts_testset_metainfo","path":"src/f5_tts/eval/utils_eval.py","language":"python","start_line":18,"end_line":32,"context_start_line":1,"context_end_line":52,"code":"import math\nimport os\nimport random\nimport string\nfrom pathlib import Path\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom tqdm import tqdm\n\nfrom f5_tts.eval.ecapa_tdnn import ECAPA_TDNN_SMALL\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import convert_char_to_pinyin\n\n\n# seedtts testset metainfo: utt, prompt_text, prompt_wav, gt_text, gt_wav\ndef get_seedtts_testset_metainfo(metalst):\n f = open(metalst)\n lines = f.readlines()\n f.close()\n metainfo = []\n for line in lines:\n if len(line.strip().split(\"|\")) == 5:\n utt, prompt_text, prompt_wav, gt_text, gt_wav = line.strip().split(\"|\")\n elif len(line.strip().split(\"|\")) == 4:\n utt, prompt_text, prompt_wav, gt_text = line.strip().split(\"|\")\n gt_wav = os.path.join(os.path.dirname(metalst), \"wavs\", utt + \".wav\")\n if not os.path.isabs(prompt_wav):\n prompt_wav = os.path.join(os.path.dirname(metalst), prompt_wav)\n metainfo.append((utt, prompt_text, prompt_wav, gt_text, gt_wav))\n return metainfo\n\n\n# librispeech test-clean metainfo: gen_utt, ref_txt, ref_wav, gen_txt, gen_wav\ndef get_librispeech_test_clean_metainfo(metalst, librispeech_test_clean_path):\n f = open(metalst)\n lines = f.readlines()\n f.close()\n metainfo = []\n for line in lines:\n ref_utt, ref_dur, ref_txt, gen_utt, gen_dur, gen_txt = line.strip().split(\"\\t\")\n\n # ref_txt = ref_txt[0] + ref_txt[1:].lower() + '.' # if use librispeech test-clean (no-pc)\n ref_spk_id, ref_chaptr_id, _ = ref_utt.split(\"-\")\n ref_wav = os.path.join(librispeech_test_clean_path, ref_spk_id, ref_chaptr_id, ref_utt + \".flac\")\n\n # gen_txt = gen_txt[0] + gen_txt[1:].lower() + '.' # if use librispeech test-clean (no-pc)\n gen_spk_id, gen_chaptr_id, _ = gen_utt.split(\"-\")\n gen_wav = os.path.join(librispeech_test_clean_path, gen_spk_id, gen_chaptr_id, gen_utt + \".flac\")\n\n metainfo.append((gen_utt, ref_txt, ref_wav, \" \" + gen_txt, gen_wav))","source_hash":"5ca3e0c70d7143d3bf5f23182499b82ddd4bba026d76b9a99d003a65136d7b68","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.utils_eval.get_librispeech_test_clean_metainfo","uri":"program://DMOSpeech2/function/src.f5_tts.eval.utils_eval.get_librispeech_test_clean_metainfo#L36-L54","kind":"function","name":"get_librispeech_test_clean_metainfo","path":"src/f5_tts/eval/utils_eval.py","language":"python","start_line":36,"end_line":54,"context_start_line":16,"context_end_line":74,"code":"\n# seedtts testset metainfo: utt, prompt_text, prompt_wav, gt_text, gt_wav\ndef get_seedtts_testset_metainfo(metalst):\n f = open(metalst)\n lines = f.readlines()\n f.close()\n metainfo = []\n for line in lines:\n if len(line.strip().split(\"|\")) == 5:\n utt, prompt_text, prompt_wav, gt_text, gt_wav = line.strip().split(\"|\")\n elif len(line.strip().split(\"|\")) == 4:\n utt, prompt_text, prompt_wav, gt_text = line.strip().split(\"|\")\n gt_wav = os.path.join(os.path.dirname(metalst), \"wavs\", utt + \".wav\")\n if not os.path.isabs(prompt_wav):\n prompt_wav = os.path.join(os.path.dirname(metalst), prompt_wav)\n metainfo.append((utt, prompt_text, prompt_wav, gt_text, gt_wav))\n return metainfo\n\n\n# librispeech test-clean metainfo: gen_utt, ref_txt, ref_wav, gen_txt, gen_wav\ndef get_librispeech_test_clean_metainfo(metalst, librispeech_test_clean_path):\n f = open(metalst)\n lines = f.readlines()\n f.close()\n metainfo = []\n for line in lines:\n ref_utt, ref_dur, ref_txt, gen_utt, gen_dur, gen_txt = line.strip().split(\"\\t\")\n\n # ref_txt = ref_txt[0] + ref_txt[1:].lower() + '.' # if use librispeech test-clean (no-pc)\n ref_spk_id, ref_chaptr_id, _ = ref_utt.split(\"-\")\n ref_wav = os.path.join(librispeech_test_clean_path, ref_spk_id, ref_chaptr_id, ref_utt + \".flac\")\n\n # gen_txt = gen_txt[0] + gen_txt[1:].lower() + '.' # if use librispeech test-clean (no-pc)\n gen_spk_id, gen_chaptr_id, _ = gen_utt.split(\"-\")\n gen_wav = os.path.join(librispeech_test_clean_path, gen_spk_id, gen_chaptr_id, gen_utt + \".flac\")\n\n metainfo.append((gen_utt, ref_txt, ref_wav, \" \" + gen_txt, gen_wav))\n\n return metainfo\n\n\n# padded to max length mel batch\ndef padded_mel_batch(ref_mels):\n max_mel_length = torch.LongTensor([mel.shape[-1] for mel in ref_mels]).amax()\n padded_ref_mels = []\n for mel in ref_mels:\n padded_ref_mel = F.pad(mel, (0, max_mel_length - mel.shape[-1]), value=0)\n padded_ref_mels.append(padded_ref_mel)\n padded_ref_mels = torch.stack(padded_ref_mels)\n padded_ref_mels = padded_ref_mels.permute(0, 2, 1)\n return padded_ref_mels\n\n\n# get prompts from metainfo containing: utt, prompt_text, prompt_wav, gt_text, gt_wav\n\n\ndef get_inference_prompt(\n metainfo,\n speed=1.0,","source_hash":"5ca3e0c70d7143d3bf5f23182499b82ddd4bba026d76b9a99d003a65136d7b68","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.utils_eval.padded_mel_batch","uri":"program://DMOSpeech2/function/src.f5_tts.eval.utils_eval.padded_mel_batch#L58-L66","kind":"function","name":"padded_mel_batch","path":"src/f5_tts/eval/utils_eval.py","language":"python","start_line":58,"end_line":66,"context_start_line":38,"context_end_line":86,"code":" lines = f.readlines()\n f.close()\n metainfo = []\n for line in lines:\n ref_utt, ref_dur, ref_txt, gen_utt, gen_dur, gen_txt = line.strip().split(\"\\t\")\n\n # ref_txt = ref_txt[0] + ref_txt[1:].lower() + '.' # if use librispeech test-clean (no-pc)\n ref_spk_id, ref_chaptr_id, _ = ref_utt.split(\"-\")\n ref_wav = os.path.join(librispeech_test_clean_path, ref_spk_id, ref_chaptr_id, ref_utt + \".flac\")\n\n # gen_txt = gen_txt[0] + gen_txt[1:].lower() + '.' # if use librispeech test-clean (no-pc)\n gen_spk_id, gen_chaptr_id, _ = gen_utt.split(\"-\")\n gen_wav = os.path.join(librispeech_test_clean_path, gen_spk_id, gen_chaptr_id, gen_utt + \".flac\")\n\n metainfo.append((gen_utt, ref_txt, ref_wav, \" \" + gen_txt, gen_wav))\n\n return metainfo\n\n\n# padded to max length mel batch\ndef padded_mel_batch(ref_mels):\n max_mel_length = torch.LongTensor([mel.shape[-1] for mel in ref_mels]).amax()\n padded_ref_mels = []\n for mel in ref_mels:\n padded_ref_mel = F.pad(mel, (0, max_mel_length - mel.shape[-1]), value=0)\n padded_ref_mels.append(padded_ref_mel)\n padded_ref_mels = torch.stack(padded_ref_mels)\n padded_ref_mels = padded_ref_mels.permute(0, 2, 1)\n return padded_ref_mels\n\n\n# get prompts from metainfo containing: utt, prompt_text, prompt_wav, gt_text, gt_wav\n\n\ndef get_inference_prompt(\n metainfo,\n speed=1.0,\n tokenizer=\"pinyin\",\n polyphone=True,\n target_sample_rate=24000,\n n_fft=1024,\n win_length=1024,\n n_mel_channels=100,\n hop_length=256,\n mel_spec_type=\"vocos\",\n target_rms=0.1,\n use_truth_duration=False,\n infer_batch_size=1,\n num_buckets=200,","source_hash":"5ca3e0c70d7143d3bf5f23182499b82ddd4bba026d76b9a99d003a65136d7b68","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.utils_eval.get_inference_prompt","uri":"program://DMOSpeech2/function/src.f5_tts.eval.utils_eval.get_inference_prompt#L72-L205","kind":"function","name":"get_inference_prompt","path":"src/f5_tts/eval/utils_eval.py","language":"python","start_line":72,"end_line":205,"context_start_line":52,"context_end_line":225,"code":" metainfo.append((gen_utt, ref_txt, ref_wav, \" \" + gen_txt, gen_wav))\n\n return metainfo\n\n\n# padded to max length mel batch\ndef padded_mel_batch(ref_mels):\n max_mel_length = torch.LongTensor([mel.shape[-1] for mel in ref_mels]).amax()\n padded_ref_mels = []\n for mel in ref_mels:\n padded_ref_mel = F.pad(mel, (0, max_mel_length - mel.shape[-1]), value=0)\n padded_ref_mels.append(padded_ref_mel)\n padded_ref_mels = torch.stack(padded_ref_mels)\n padded_ref_mels = padded_ref_mels.permute(0, 2, 1)\n return padded_ref_mels\n\n\n# get prompts from metainfo containing: utt, prompt_text, prompt_wav, gt_text, gt_wav\n\n\ndef get_inference_prompt(\n metainfo,\n speed=1.0,\n tokenizer=\"pinyin\",\n polyphone=True,\n target_sample_rate=24000,\n n_fft=1024,\n win_length=1024,\n n_mel_channels=100,\n hop_length=256,\n mel_spec_type=\"vocos\",\n target_rms=0.1,\n use_truth_duration=False,\n infer_batch_size=1,\n num_buckets=200,\n min_secs=3,\n max_secs=40,\n):\n prompts_all = []\n\n min_tokens = min_secs * target_sample_rate // hop_length\n max_tokens = max_secs * target_sample_rate // hop_length\n\n batch_accum = [0] * num_buckets\n utts, ref_rms_list, ref_mels, ref_mel_lens, total_mel_lens, final_text_list = (\n [[] for _ in range(num_buckets)] for _ in range(6)\n )\n\n mel_spectrogram = MelSpec(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n )\n\n for utt, prompt_text, prompt_wav, gt_text, gt_wav in tqdm(metainfo, desc=\"Processing prompts...\"):\n # Audio\n ref_audio, ref_sr = torchaudio.load(prompt_wav)\n ref_rms = torch.sqrt(torch.mean(torch.square(ref_audio)))\n if ref_rms < target_rms:\n ref_audio = ref_audio * target_rms / ref_rms\n assert ref_audio.shape[-1] > 5000, f\"Empty prompt wav: {prompt_wav}, or torchaudio backend issue.\"\n if ref_sr != target_sample_rate:\n resampler = torchaudio.transforms.Resample(ref_sr, target_sample_rate)\n ref_audio = resampler(ref_audio)\n\n # Text\n if len(prompt_text[-1].encode(\"utf-8\")) == 1:\n prompt_text = prompt_text + \" \"\n text = [prompt_text + gt_text]\n if tokenizer == \"pinyin\":\n text_list = convert_char_to_pinyin(text, polyphone=polyphone)\n else:\n text_list = text\n\n # to mel spectrogram\n ref_mel = mel_spectrogram(ref_audio)\n ref_mel = ref_mel.squeeze(0)\n\n # Duration, mel frame length\n ref_mel_len = ref_mel.shape[-1]\n\n if use_truth_duration:\n gt_audio, gt_sr = torchaudio.load(gt_wav)\n if gt_sr != target_sample_rate:\n resampler = torchaudio.transforms.Resample(gt_sr, target_sample_rate)\n gt_audio = resampler(gt_audio)\n total_mel_len = ref_mel_len + int(gt_audio.shape[-1] / hop_length / speed)\n\n # # test vocoder resynthesis\n # ref_audio = gt_audio\n else:\n ref_text_len = len(prompt_text.encode(\"utf-8\"))\n gen_text_len = len(gt_text.encode(\"utf-8\"))\n total_mel_len = ref_mel_len + int(ref_mel_len / ref_text_len * gen_text_len / speed)\n\n # deal with batch\n assert infer_batch_size > 0, \"infer_batch_size should be greater than 0.\"\n assert min_tokens <= total_mel_len <= max_tokens, (\n f\"Audio {utt} has duration {total_mel_len * hop_length // target_sample_rate}s out of range [{min_secs}, {max_secs}].\"\n )\n bucket_i = math.floor((total_mel_len - min_tokens) / (max_tokens - min_tokens + 1) * num_buckets)\n\n utts[bucket_i].append(utt)\n ref_rms_list[bucket_i].append(ref_rms)\n ref_mels[bucket_i].append(ref_mel)\n ref_mel_lens[bucket_i].append(ref_mel_len)\n total_mel_lens[bucket_i].append(total_mel_len)\n final_text_list[bucket_i].extend(text_list)\n\n batch_accum[bucket_i] += total_mel_len\n\n if batch_accum[bucket_i] >= infer_batch_size:\n # print(f\"\\n{len(ref_mels[bucket_i][0][0])}\\n{ref_mel_lens[bucket_i]}\\n{total_mel_lens[bucket_i]}\")\n prompts_all.append(\n (\n utts[bucket_i],\n ref_rms_list[bucket_i],\n padded_mel_batch(ref_mels[bucket_i]),\n ref_mel_lens[bucket_i],\n total_mel_lens[bucket_i],\n final_text_list[bucket_i],\n )\n )\n batch_accum[bucket_i] = 0\n (\n utts[bucket_i],\n ref_rms_list[bucket_i],\n ref_mels[bucket_i],\n ref_mel_lens[bucket_i],\n total_mel_lens[bucket_i],\n final_text_list[bucket_i],\n ) = [], [], [], [], [], []\n\n # add residual\n for bucket_i, bucket_frames in enumerate(batch_accum):\n if bucket_frames > 0:\n prompts_all.append(\n (\n utts[bucket_i],\n ref_rms_list[bucket_i],\n padded_mel_batch(ref_mels[bucket_i]),\n ref_mel_lens[bucket_i],\n total_mel_lens[bucket_i],\n final_text_list[bucket_i],\n )\n )\n # not only leave easy work for last workers\n random.seed(666)\n random.shuffle(prompts_all)\n\n return prompts_all\n\n\n# get wav_res_ref_text of seed-tts test metalst\n# https://github.com/BytedanceSpeech/seed-tts-eval\n\n\ndef get_seed_tts_test(metalst, gen_wav_dir, gpus):\n f = open(metalst)\n lines = f.readlines()\n f.close()\n\n test_set_ = []\n for line in tqdm(lines):\n if len(line.strip().split(\"|\")) == 5:\n utt, prompt_text, prompt_wav, gt_text, gt_wav = line.strip().split(\"|\")\n elif len(line.strip().split(\"|\")) == 4:\n utt, prompt_text, prompt_wav, gt_text = line.strip().split(\"|\")\n\n if not os.path.exists(os.path.join(gen_wav_dir, utt + \".wav\")):\n continue","source_hash":"5ca3e0c70d7143d3bf5f23182499b82ddd4bba026d76b9a99d003a65136d7b68","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.utils_eval.get_seed_tts_test","uri":"program://DMOSpeech2/function/src.f5_tts.eval.utils_eval.get_seed_tts_test#L212-L241","kind":"function","name":"get_seed_tts_test","path":"src/f5_tts/eval/utils_eval.py","language":"python","start_line":212,"end_line":241,"context_start_line":192,"context_end_line":261,"code":" (\n utts[bucket_i],\n ref_rms_list[bucket_i],\n padded_mel_batch(ref_mels[bucket_i]),\n ref_mel_lens[bucket_i],\n total_mel_lens[bucket_i],\n final_text_list[bucket_i],\n )\n )\n # not only leave easy work for last workers\n random.seed(666)\n random.shuffle(prompts_all)\n\n return prompts_all\n\n\n# get wav_res_ref_text of seed-tts test metalst\n# https://github.com/BytedanceSpeech/seed-tts-eval\n\n\ndef get_seed_tts_test(metalst, gen_wav_dir, gpus):\n f = open(metalst)\n lines = f.readlines()\n f.close()\n\n test_set_ = []\n for line in tqdm(lines):\n if len(line.strip().split(\"|\")) == 5:\n utt, prompt_text, prompt_wav, gt_text, gt_wav = line.strip().split(\"|\")\n elif len(line.strip().split(\"|\")) == 4:\n utt, prompt_text, prompt_wav, gt_text = line.strip().split(\"|\")\n\n if not os.path.exists(os.path.join(gen_wav_dir, utt + \".wav\")):\n continue\n gen_wav = os.path.join(gen_wav_dir, utt + \".wav\")\n if not os.path.isabs(prompt_wav):\n prompt_wav = os.path.join(os.path.dirname(metalst), prompt_wav)\n\n test_set_.append((gen_wav, prompt_wav, gt_text))\n\n num_jobs = len(gpus)\n if num_jobs == 1:\n return [(gpus[0], test_set_)]\n\n wav_per_job = len(test_set_) // num_jobs + 1\n test_set = []\n for i in range(num_jobs):\n test_set.append((gpus[i], test_set_[i * wav_per_job : (i + 1) * wav_per_job]))\n\n return test_set\n\n\n# get librispeech test-clean cross sentence test\n\n\ndef get_librispeech_test(metalst, gen_wav_dir, gpus, librispeech_test_clean_path, eval_ground_truth=False):\n f = open(metalst)\n lines = f.readlines()\n f.close()\n\n test_set_ = []\n for line in tqdm(lines):\n ref_utt, ref_dur, ref_txt, gen_utt, gen_dur, gen_txt = line.strip().split(\"\\t\")\n\n if eval_ground_truth:\n gen_spk_id, gen_chaptr_id, _ = gen_utt.split(\"-\")\n gen_wav = os.path.join(librispeech_test_clean_path, gen_spk_id, gen_chaptr_id, gen_utt + \".flac\")\n else:\n if not os.path.exists(os.path.join(gen_wav_dir, gen_utt + \".wav\")):\n raise FileNotFoundError(f\"Generated wav not found: {gen_utt}\")","source_hash":"5ca3e0c70d7143d3bf5f23182499b82ddd4bba026d76b9a99d003a65136d7b68","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.utils_eval.get_librispeech_test","uri":"program://DMOSpeech2/function/src.f5_tts.eval.utils_eval.get_librispeech_test#L247-L278","kind":"function","name":"get_librispeech_test","path":"src/f5_tts/eval/utils_eval.py","language":"python","start_line":247,"end_line":278,"context_start_line":227,"context_end_line":298,"code":" if not os.path.isabs(prompt_wav):\n prompt_wav = os.path.join(os.path.dirname(metalst), prompt_wav)\n\n test_set_.append((gen_wav, prompt_wav, gt_text))\n\n num_jobs = len(gpus)\n if num_jobs == 1:\n return [(gpus[0], test_set_)]\n\n wav_per_job = len(test_set_) // num_jobs + 1\n test_set = []\n for i in range(num_jobs):\n test_set.append((gpus[i], test_set_[i * wav_per_job : (i + 1) * wav_per_job]))\n\n return test_set\n\n\n# get librispeech test-clean cross sentence test\n\n\ndef get_librispeech_test(metalst, gen_wav_dir, gpus, librispeech_test_clean_path, eval_ground_truth=False):\n f = open(metalst)\n lines = f.readlines()\n f.close()\n\n test_set_ = []\n for line in tqdm(lines):\n ref_utt, ref_dur, ref_txt, gen_utt, gen_dur, gen_txt = line.strip().split(\"\\t\")\n\n if eval_ground_truth:\n gen_spk_id, gen_chaptr_id, _ = gen_utt.split(\"-\")\n gen_wav = os.path.join(librispeech_test_clean_path, gen_spk_id, gen_chaptr_id, gen_utt + \".flac\")\n else:\n if not os.path.exists(os.path.join(gen_wav_dir, gen_utt + \".wav\")):\n raise FileNotFoundError(f\"Generated wav not found: {gen_utt}\")\n gen_wav = os.path.join(gen_wav_dir, gen_utt + \".wav\")\n\n ref_spk_id, ref_chaptr_id, _ = ref_utt.split(\"-\")\n ref_wav = os.path.join(librispeech_test_clean_path, ref_spk_id, ref_chaptr_id, ref_utt + \".flac\")\n\n test_set_.append((gen_wav, ref_wav, gen_txt))\n\n num_jobs = len(gpus)\n if num_jobs == 1:\n return [(gpus[0], test_set_)]\n\n wav_per_job = len(test_set_) // num_jobs + 1\n test_set = []\n for i in range(num_jobs):\n test_set.append((gpus[i], test_set_[i * wav_per_job : (i + 1) * wav_per_job]))\n\n return test_set\n\n\n# load asr model\n\n\ndef load_asr_model(lang, ckpt_dir=\"\"):\n if lang == \"zh\":\n from funasr import AutoModel\n\n model = AutoModel(\n model=os.path.join(ckpt_dir, \"paraformer-zh\"),\n # vad_model = os.path.join(ckpt_dir, \"fsmn-vad\"),\n # punc_model = os.path.join(ckpt_dir, \"ct-punc\"),\n # spk_model = os.path.join(ckpt_dir, \"cam++\"),\n disable_update=True,\n ) # following seed-tts setting\n elif lang == \"en\":\n from faster_whisper import WhisperModel\n\n model_size = \"large-v3\" if ckpt_dir == \"\" else ckpt_dir","source_hash":"5ca3e0c70d7143d3bf5f23182499b82ddd4bba026d76b9a99d003a65136d7b68","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.utils_eval.load_asr_model","uri":"program://DMOSpeech2/function/src.f5_tts.eval.utils_eval.load_asr_model#L284-L300","kind":"function","name":"load_asr_model","path":"src/f5_tts/eval/utils_eval.py","language":"python","start_line":284,"end_line":300,"context_start_line":264,"context_end_line":320,"code":" ref_spk_id, ref_chaptr_id, _ = ref_utt.split(\"-\")\n ref_wav = os.path.join(librispeech_test_clean_path, ref_spk_id, ref_chaptr_id, ref_utt + \".flac\")\n\n test_set_.append((gen_wav, ref_wav, gen_txt))\n\n num_jobs = len(gpus)\n if num_jobs == 1:\n return [(gpus[0], test_set_)]\n\n wav_per_job = len(test_set_) // num_jobs + 1\n test_set = []\n for i in range(num_jobs):\n test_set.append((gpus[i], test_set_[i * wav_per_job : (i + 1) * wav_per_job]))\n\n return test_set\n\n\n# load asr model\n\n\ndef load_asr_model(lang, ckpt_dir=\"\"):\n if lang == \"zh\":\n from funasr import AutoModel\n\n model = AutoModel(\n model=os.path.join(ckpt_dir, \"paraformer-zh\"),\n # vad_model = os.path.join(ckpt_dir, \"fsmn-vad\"),\n # punc_model = os.path.join(ckpt_dir, \"ct-punc\"),\n # spk_model = os.path.join(ckpt_dir, \"cam++\"),\n disable_update=True,\n ) # following seed-tts setting\n elif lang == \"en\":\n from faster_whisper import WhisperModel\n\n model_size = \"large-v3\" if ckpt_dir == \"\" else ckpt_dir\n model = WhisperModel(model_size, device=\"cuda\", compute_type=\"float16\")\n return model\n\n\n# WER Evaluation, the way Seed-TTS does\n\n\ndef run_asr_wer(args):\n rank, lang, test_set, ckpt_dir = args\n\n if lang == \"zh\":\n import zhconv\n\n torch.cuda.set_device(rank)\n elif lang == \"en\":\n os.environ[\"CUDA_VISIBLE_DEVICES\"] = str(rank)\n else:\n raise NotImplementedError(\n \"lang support only 'zh' (funasr paraformer-zh), 'en' (faster-whisper-large-v3), for now.\"\n )\n\n asr_model = load_asr_model(lang, ckpt_dir=ckpt_dir)","source_hash":"5ca3e0c70d7143d3bf5f23182499b82ddd4bba026d76b9a99d003a65136d7b68","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.utils_eval.run_asr_wer","uri":"program://DMOSpeech2/function/src.f5_tts.eval.utils_eval.run_asr_wer#L306-L374","kind":"function","name":"run_asr_wer","path":"src/f5_tts/eval/utils_eval.py","language":"python","start_line":306,"end_line":374,"context_start_line":286,"context_end_line":394,"code":" from funasr import AutoModel\n\n model = AutoModel(\n model=os.path.join(ckpt_dir, \"paraformer-zh\"),\n # vad_model = os.path.join(ckpt_dir, \"fsmn-vad\"),\n # punc_model = os.path.join(ckpt_dir, \"ct-punc\"),\n # spk_model = os.path.join(ckpt_dir, \"cam++\"),\n disable_update=True,\n ) # following seed-tts setting\n elif lang == \"en\":\n from faster_whisper import WhisperModel\n\n model_size = \"large-v3\" if ckpt_dir == \"\" else ckpt_dir\n model = WhisperModel(model_size, device=\"cuda\", compute_type=\"float16\")\n return model\n\n\n# WER Evaluation, the way Seed-TTS does\n\n\ndef run_asr_wer(args):\n rank, lang, test_set, ckpt_dir = args\n\n if lang == \"zh\":\n import zhconv\n\n torch.cuda.set_device(rank)\n elif lang == \"en\":\n os.environ[\"CUDA_VISIBLE_DEVICES\"] = str(rank)\n else:\n raise NotImplementedError(\n \"lang support only 'zh' (funasr paraformer-zh), 'en' (faster-whisper-large-v3), for now.\"\n )\n\n asr_model = load_asr_model(lang, ckpt_dir=ckpt_dir)\n\n from zhon.hanzi import punctuation\n\n punctuation_all = punctuation + string.punctuation\n wer_results = []\n\n from jiwer import compute_measures\n\n for gen_wav, prompt_wav, truth in tqdm(test_set):\n if lang == \"zh\":\n res = asr_model.generate(input=gen_wav, batch_size_s=300, disable_pbar=True)\n hypo = res[0][\"text\"]\n hypo = zhconv.convert(hypo, \"zh-cn\")\n elif lang == \"en\":\n segments, _ = asr_model.transcribe(gen_wav, beam_size=5, language=\"en\")\n hypo = \"\"\n for segment in segments:\n hypo = hypo + \" \" + segment.text\n\n raw_truth = truth\n raw_hypo = hypo\n\n for x in punctuation_all:\n truth = truth.replace(x, \"\")\n hypo = hypo.replace(x, \"\")\n\n truth = truth.replace(\" \", \" \")\n hypo = hypo.replace(\" \", \" \")\n\n if lang == \"zh\":\n truth = \" \".join([x for x in truth])\n hypo = \" \".join([x for x in hypo])\n elif lang == \"en\":\n truth = truth.lower()\n hypo = hypo.lower()\n\n measures = compute_measures(truth, hypo)\n wer = measures[\"wer\"]\n\n # ref_list = truth.split(\" \")\n # subs = measures[\"substitutions\"] / len(ref_list)\n # dele = measures[\"deletions\"] / len(ref_list)\n # inse = measures[\"insertions\"] / len(ref_list)\n\n wer_results.append(\n {\n \"wav\": Path(gen_wav).stem,\n \"truth\": raw_truth,\n \"hypo\": raw_hypo,\n \"wer\": wer,\n }\n )\n\n return wer_results\n\n\n# SIM Evaluation\n\n\ndef run_sim(args):\n rank, test_set, ckpt_dir = args\n device = f\"cuda:{rank}\"\n\n model = ECAPA_TDNN_SMALL(feat_dim=1024, feat_type=\"wavlm_large\", config_path=None)\n state_dict = torch.load(ckpt_dir, weights_only=True, map_location=lambda storage, loc: storage)\n model.load_state_dict(state_dict[\"model\"], strict=False)\n\n use_gpu = True if torch.cuda.is_available() else False\n if use_gpu:\n model = model.cuda(device)\n model.eval()\n\n sim_results = []\n for gen_wav, prompt_wav, truth in tqdm(test_set):","source_hash":"5ca3e0c70d7143d3bf5f23182499b82ddd4bba026d76b9a99d003a65136d7b68","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.utils_eval.run_sim","uri":"program://DMOSpeech2/function/src.f5_tts.eval.utils_eval.run_sim#L380-L419","kind":"function","name":"run_sim","path":"src/f5_tts/eval/utils_eval.py","language":"python","start_line":380,"end_line":419,"context_start_line":360,"context_end_line":419,"code":" # ref_list = truth.split(\" \")\n # subs = measures[\"substitutions\"] / len(ref_list)\n # dele = measures[\"deletions\"] / len(ref_list)\n # inse = measures[\"insertions\"] / len(ref_list)\n\n wer_results.append(\n {\n \"wav\": Path(gen_wav).stem,\n \"truth\": raw_truth,\n \"hypo\": raw_hypo,\n \"wer\": wer,\n }\n )\n\n return wer_results\n\n\n# SIM Evaluation\n\n\ndef run_sim(args):\n rank, test_set, ckpt_dir = args\n device = f\"cuda:{rank}\"\n\n model = ECAPA_TDNN_SMALL(feat_dim=1024, feat_type=\"wavlm_large\", config_path=None)\n state_dict = torch.load(ckpt_dir, weights_only=True, map_location=lambda storage, loc: storage)\n model.load_state_dict(state_dict[\"model\"], strict=False)\n\n use_gpu = True if torch.cuda.is_available() else False\n if use_gpu:\n model = model.cuda(device)\n model.eval()\n\n sim_results = []\n for gen_wav, prompt_wav, truth in tqdm(test_set):\n wav1, sr1 = torchaudio.load(gen_wav)\n wav2, sr2 = torchaudio.load(prompt_wav)\n\n resample1 = torchaudio.transforms.Resample(orig_freq=sr1, new_freq=16000)\n resample2 = torchaudio.transforms.Resample(orig_freq=sr2, new_freq=16000)\n wav1 = resample1(wav1)\n wav2 = resample2(wav2)\n\n if use_gpu:\n wav1 = wav1.cuda(device)\n wav2 = wav2.cuda(device)\n with torch.no_grad():\n emb1 = model(wav1)\n emb2 = model(wav2)\n\n sim = F.cosine_similarity(emb1, emb2)[0].item()\n # print(f\"VSim score between two audios: {sim:.4f} (-1.0, 1.0).\")\n sim_results.append(\n {\n \"wav\": Path(gen_wav).stem,\n \"sim\": sim,\n }\n )\n\n return sim_results","source_hash":"5ca3e0c70d7143d3bf5f23182499b82ddd4bba026d76b9a99d003a65136d7b68","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.eval_infer_batch","uri":"program://DMOSpeech2/module/src.f5_tts.eval.eval_infer_batch#L1-L210","kind":"module","name":"src.f5_tts.eval.eval_infer_batch","path":"src/f5_tts/eval/eval_infer_batch.py","language":"python","start_line":1,"end_line":210,"context_start_line":1,"context_end_line":210,"code":"import os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport argparse\nimport time\nfrom importlib.resources import files\n\nimport torch\nimport torchaudio\nfrom accelerate import Accelerator\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\nfrom tqdm import tqdm\n\nfrom f5_tts.eval.utils_eval import (\n get_inference_prompt,\n get_librispeech_test_clean_metainfo,\n get_seedtts_testset_metainfo,\n)\nfrom f5_tts.infer.utils_infer import load_checkpoint, load_vocoder\nfrom f5_tts.model import CFM\nfrom f5_tts.model.utils import get_tokenizer\n\n\naccelerator = Accelerator()\ndevice = f\"cuda:{accelerator.process_index}\"\n\n\nuse_ema = True\ntarget_rms = 0.1\n\n\nrel_path = str(files(\"f5_tts\").joinpath(\"../../\"))\n\n\ndef main():\n parser = argparse.ArgumentParser(description=\"batch inference\")\n\n parser.add_argument(\"-s\", \"--seed\", default=None, type=int)\n parser.add_argument(\"-n\", \"--expname\", required=True)\n parser.add_argument(\"-c\", \"--ckptstep\", default=1250000, type=int)\n\n parser.add_argument(\"-nfe\", \"--nfestep\", default=32, type=int)\n parser.add_argument(\"-o\", \"--odemethod\", default=\"euler\")\n parser.add_argument(\"-ss\", \"--swaysampling\", default=-1, type=float)\n\n parser.add_argument(\"-t\", \"--testset\", required=True)\n\n args = parser.parse_args()\n\n seed = args.seed\n exp_name = args.expname\n ckpt_step = args.ckptstep\n\n nfe_step = args.nfestep\n ode_method = args.odemethod\n sway_sampling_coef = args.swaysampling\n\n testset = args.testset\n\n infer_batch_size = 1 # max frames. 1 for ddp single inference (recommended)\n cfg_strength = 2.0\n speed = 1.0\n use_truth_duration = False\n no_ref_audio = False\n\n model_cfg = OmegaConf.load(str(files(\"f5_tts\").joinpath(f\"configs/{exp_name}.yaml\")))\n model_cls = get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n model_arc = model_cfg.model.arch\n\n dataset_name = model_cfg.datasets.name\n tokenizer = model_cfg.model.tokenizer\n\n mel_spec_type = model_cfg.model.mel_spec.mel_spec_type\n target_sample_rate = model_cfg.model.mel_spec.target_sample_rate\n n_mel_channels = model_cfg.model.mel_spec.n_mel_channels\n hop_length = model_cfg.model.mel_spec.hop_length\n win_length = model_cfg.model.mel_spec.win_length\n n_fft = model_cfg.model.mel_spec.n_fft\n\n if testset == \"ls_pc_test_clean\":\n metalst = rel_path + \"/data/librispeech_pc_test_clean_cross_sentence.lst\"\n librispeech_test_clean_path = \"/LibriSpeech/test-clean\" # test-clean path\n metainfo = get_librispeech_test_clean_metainfo(metalst, librispeech_test_clean_path)\n\n elif testset == \"seedtts_test_zh\":\n metalst = rel_path + \"/data/seedtts_testset/zh/meta.lst\"\n metainfo = get_seedtts_testset_metainfo(metalst)\n\n elif testset == \"seedtts_test_en\":\n metalst = rel_path + \"/data/seedtts_testset/en/meta.lst\"\n metainfo = get_seedtts_testset_metainfo(metalst)\n\n # path to save genereted wavs\n output_dir = (\n f\"{rel_path}/\"\n f\"results/{exp_name}_{ckpt_step}/{testset}/\"\n f\"seed{seed}_{ode_method}_nfe{nfe_step}_{mel_spec_type}\"\n f\"{f'_ss{sway_sampling_coef}' if sway_sampling_coef else ''}\"\n f\"_cfg{cfg_strength}_speed{speed}\"\n f\"{'_gt-dur' if use_truth_duration else ''}\"\n f\"{'_no-ref-audio' if no_ref_audio else ''}\"\n )\n\n # -------------------------------------------------#\n\n prompts_all = get_inference_prompt(\n metainfo,\n speed=speed,\n tokenizer=tokenizer,\n target_sample_rate=target_sample_rate,\n n_mel_channels=n_mel_channels,\n hop_length=hop_length,\n mel_spec_type=mel_spec_type,\n target_rms=target_rms,\n use_truth_duration=use_truth_duration,\n infer_batch_size=infer_batch_size,\n )\n\n # Vocoder model\n local = False\n if mel_spec_type == \"vocos\":\n vocoder_local_path = \"../checkpoints/charactr/vocos-mel-24khz\"\n elif mel_spec_type == \"bigvgan\":\n vocoder_local_path = \"../checkpoints/bigvgan_v2_24khz_100band_256x\"\n vocoder = load_vocoder(vocoder_name=mel_spec_type, is_local=local, local_path=vocoder_local_path)\n\n # Tokenizer\n vocab_char_map, vocab_size = get_tokenizer(dataset_name, tokenizer)\n\n # Model\n model = CFM(\n transformer=model_cls(**model_arc, text_num_embeds=vocab_size, mel_dim=n_mel_channels),\n mel_spec_kwargs=dict(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n ),\n odeint_kwargs=dict(\n method=ode_method,\n ),\n vocab_char_map=vocab_char_map,\n ).to(device)\n\n ckpt_prefix = rel_path + f\"/ckpts/{exp_name}/model_{ckpt_step}\"\n if os.path.exists(ckpt_prefix + \".pt\"):\n ckpt_path = ckpt_prefix + \".pt\"\n elif os.path.exists(ckpt_prefix + \".safetensors\"):\n ckpt_path = ckpt_prefix + \".safetensors\"\n else:\n print(\"Loading from self-organized training checkpoints rather than released pretrained.\")\n ckpt_path = rel_path + f\"/{model_cfg.ckpts.save_dir}/model_{ckpt_step}.pt\"\n\n dtype = torch.float32 if mel_spec_type == \"bigvgan\" else None\n model = load_checkpoint(model, ckpt_path, device, dtype=dtype, use_ema=use_ema)\n\n if not os.path.exists(output_dir) and accelerator.is_main_process:\n os.makedirs(output_dir)\n\n # start batch inference\n accelerator.wait_for_everyone()\n start = time.time()\n\n with accelerator.split_between_processes(prompts_all) as prompts:\n for prompt in tqdm(prompts, disable=not accelerator.is_local_main_process):\n utts, ref_rms_list, ref_mels, ref_mel_lens, total_mel_lens, final_text_list = prompt\n ref_mels = ref_mels.to(device)\n ref_mel_lens = torch.tensor(ref_mel_lens, dtype=torch.long).to(device)\n total_mel_lens = torch.tensor(total_mel_lens, dtype=torch.long).to(device)\n\n # Inference\n with torch.inference_mode():\n generated, _ = model.sample(\n cond=ref_mels,\n text=final_text_list,\n duration=total_mel_lens,\n lens=ref_mel_lens,\n steps=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n no_ref_audio=no_ref_audio,\n seed=seed,\n )\n # Final result\n for i, gen in enumerate(generated):\n gen = gen[ref_mel_lens[i] : total_mel_lens[i], :].unsqueeze(0)\n gen_mel_spec = gen.permute(0, 2, 1).to(torch.float32)\n if mel_spec_type == \"vocos\":\n generated_wave = vocoder.decode(gen_mel_spec).cpu()\n elif mel_spec_type == \"bigvgan\":\n generated_wave = vocoder(gen_mel_spec).squeeze(0).cpu()\n\n if ref_rms_list[i] < target_rms:\n generated_wave = generated_wave * ref_rms_list[i] / target_rms\n torchaudio.save(f\"{output_dir}/{utts[i]}.wav\", generated_wave, target_sample_rate)\n\n accelerator.wait_for_everyone()\n if accelerator.is_main_process:\n timediff = time.time() - start\n print(f\"Done batch inference in {timediff / 60:.2f} minutes.\")\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"6d5a9ac66798e903d73dc92e658312aa783796fc5cc79bac32bf9d0bc450f353","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.eval_infer_batch.main","uri":"program://DMOSpeech2/function/src.f5_tts.eval.eval_infer_batch.main#L39-L206","kind":"function","name":"main","path":"src/f5_tts/eval/eval_infer_batch.py","language":"python","start_line":39,"end_line":206,"context_start_line":19,"context_end_line":210,"code":" get_inference_prompt,\n get_librispeech_test_clean_metainfo,\n get_seedtts_testset_metainfo,\n)\nfrom f5_tts.infer.utils_infer import load_checkpoint, load_vocoder\nfrom f5_tts.model import CFM\nfrom f5_tts.model.utils import get_tokenizer\n\n\naccelerator = Accelerator()\ndevice = f\"cuda:{accelerator.process_index}\"\n\n\nuse_ema = True\ntarget_rms = 0.1\n\n\nrel_path = str(files(\"f5_tts\").joinpath(\"../../\"))\n\n\ndef main():\n parser = argparse.ArgumentParser(description=\"batch inference\")\n\n parser.add_argument(\"-s\", \"--seed\", default=None, type=int)\n parser.add_argument(\"-n\", \"--expname\", required=True)\n parser.add_argument(\"-c\", \"--ckptstep\", default=1250000, type=int)\n\n parser.add_argument(\"-nfe\", \"--nfestep\", default=32, type=int)\n parser.add_argument(\"-o\", \"--odemethod\", default=\"euler\")\n parser.add_argument(\"-ss\", \"--swaysampling\", default=-1, type=float)\n\n parser.add_argument(\"-t\", \"--testset\", required=True)\n\n args = parser.parse_args()\n\n seed = args.seed\n exp_name = args.expname\n ckpt_step = args.ckptstep\n\n nfe_step = args.nfestep\n ode_method = args.odemethod\n sway_sampling_coef = args.swaysampling\n\n testset = args.testset\n\n infer_batch_size = 1 # max frames. 1 for ddp single inference (recommended)\n cfg_strength = 2.0\n speed = 1.0\n use_truth_duration = False\n no_ref_audio = False\n\n model_cfg = OmegaConf.load(str(files(\"f5_tts\").joinpath(f\"configs/{exp_name}.yaml\")))\n model_cls = get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n model_arc = model_cfg.model.arch\n\n dataset_name = model_cfg.datasets.name\n tokenizer = model_cfg.model.tokenizer\n\n mel_spec_type = model_cfg.model.mel_spec.mel_spec_type\n target_sample_rate = model_cfg.model.mel_spec.target_sample_rate\n n_mel_channels = model_cfg.model.mel_spec.n_mel_channels\n hop_length = model_cfg.model.mel_spec.hop_length\n win_length = model_cfg.model.mel_spec.win_length\n n_fft = model_cfg.model.mel_spec.n_fft\n\n if testset == \"ls_pc_test_clean\":\n metalst = rel_path + \"/data/librispeech_pc_test_clean_cross_sentence.lst\"\n librispeech_test_clean_path = \"/LibriSpeech/test-clean\" # test-clean path\n metainfo = get_librispeech_test_clean_metainfo(metalst, librispeech_test_clean_path)\n\n elif testset == \"seedtts_test_zh\":\n metalst = rel_path + \"/data/seedtts_testset/zh/meta.lst\"\n metainfo = get_seedtts_testset_metainfo(metalst)\n\n elif testset == \"seedtts_test_en\":\n metalst = rel_path + \"/data/seedtts_testset/en/meta.lst\"\n metainfo = get_seedtts_testset_metainfo(metalst)\n\n # path to save genereted wavs\n output_dir = (\n f\"{rel_path}/\"\n f\"results/{exp_name}_{ckpt_step}/{testset}/\"\n f\"seed{seed}_{ode_method}_nfe{nfe_step}_{mel_spec_type}\"\n f\"{f'_ss{sway_sampling_coef}' if sway_sampling_coef else ''}\"\n f\"_cfg{cfg_strength}_speed{speed}\"\n f\"{'_gt-dur' if use_truth_duration else ''}\"\n f\"{'_no-ref-audio' if no_ref_audio else ''}\"\n )\n\n # -------------------------------------------------#\n\n prompts_all = get_inference_prompt(\n metainfo,\n speed=speed,\n tokenizer=tokenizer,\n target_sample_rate=target_sample_rate,\n n_mel_channels=n_mel_channels,\n hop_length=hop_length,\n mel_spec_type=mel_spec_type,\n target_rms=target_rms,\n use_truth_duration=use_truth_duration,\n infer_batch_size=infer_batch_size,\n )\n\n # Vocoder model\n local = False\n if mel_spec_type == \"vocos\":\n vocoder_local_path = \"../checkpoints/charactr/vocos-mel-24khz\"\n elif mel_spec_type == \"bigvgan\":\n vocoder_local_path = \"../checkpoints/bigvgan_v2_24khz_100band_256x\"\n vocoder = load_vocoder(vocoder_name=mel_spec_type, is_local=local, local_path=vocoder_local_path)\n\n # Tokenizer\n vocab_char_map, vocab_size = get_tokenizer(dataset_name, tokenizer)\n\n # Model\n model = CFM(\n transformer=model_cls(**model_arc, text_num_embeds=vocab_size, mel_dim=n_mel_channels),\n mel_spec_kwargs=dict(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n ),\n odeint_kwargs=dict(\n method=ode_method,\n ),\n vocab_char_map=vocab_char_map,\n ).to(device)\n\n ckpt_prefix = rel_path + f\"/ckpts/{exp_name}/model_{ckpt_step}\"\n if os.path.exists(ckpt_prefix + \".pt\"):\n ckpt_path = ckpt_prefix + \".pt\"\n elif os.path.exists(ckpt_prefix + \".safetensors\"):\n ckpt_path = ckpt_prefix + \".safetensors\"\n else:\n print(\"Loading from self-organized training checkpoints rather than released pretrained.\")\n ckpt_path = rel_path + f\"/{model_cfg.ckpts.save_dir}/model_{ckpt_step}.pt\"\n\n dtype = torch.float32 if mel_spec_type == \"bigvgan\" else None\n model = load_checkpoint(model, ckpt_path, device, dtype=dtype, use_ema=use_ema)\n\n if not os.path.exists(output_dir) and accelerator.is_main_process:\n os.makedirs(output_dir)\n\n # start batch inference\n accelerator.wait_for_everyone()\n start = time.time()\n\n with accelerator.split_between_processes(prompts_all) as prompts:\n for prompt in tqdm(prompts, disable=not accelerator.is_local_main_process):\n utts, ref_rms_list, ref_mels, ref_mel_lens, total_mel_lens, final_text_list = prompt\n ref_mels = ref_mels.to(device)\n ref_mel_lens = torch.tensor(ref_mel_lens, dtype=torch.long).to(device)\n total_mel_lens = torch.tensor(total_mel_lens, dtype=torch.long).to(device)\n\n # Inference\n with torch.inference_mode():\n generated, _ = model.sample(\n cond=ref_mels,\n text=final_text_list,\n duration=total_mel_lens,\n lens=ref_mel_lens,\n steps=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n no_ref_audio=no_ref_audio,\n seed=seed,\n )\n # Final result\n for i, gen in enumerate(generated):\n gen = gen[ref_mel_lens[i] : total_mel_lens[i], :].unsqueeze(0)\n gen_mel_spec = gen.permute(0, 2, 1).to(torch.float32)\n if mel_spec_type == \"vocos\":\n generated_wave = vocoder.decode(gen_mel_spec).cpu()\n elif mel_spec_type == \"bigvgan\":\n generated_wave = vocoder(gen_mel_spec).squeeze(0).cpu()\n\n if ref_rms_list[i] < target_rms:\n generated_wave = generated_wave * ref_rms_list[i] / target_rms\n torchaudio.save(f\"{output_dir}/{utts[i]}.wav\", generated_wave, target_sample_rate)\n\n accelerator.wait_for_everyone()\n if accelerator.is_main_process:\n timediff = time.time() - start\n print(f\"Done batch inference in {timediff / 60:.2f} minutes.\")\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"6d5a9ac66798e903d73dc92e658312aa783796fc5cc79bac32bf9d0bc450f353","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.ecapa_tdnn","uri":"program://DMOSpeech2/module/src.f5_tts.eval.ecapa_tdnn#L1-L331","kind":"module","name":"src.f5_tts.eval.ecapa_tdnn","path":"src/f5_tts/eval/ecapa_tdnn.py","language":"python","start_line":1,"end_line":331,"context_start_line":1,"context_end_line":331,"code":"# just for speaker similarity evaluation, third-party code\n\n# From https://github.com/microsoft/UniSpeech/blob/main/downstreams/speaker_verification/models/\n# part of the code is borrowed from https://github.com/lawlict/ECAPA-TDNN\n\nimport os\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\n\n\n\"\"\" Res2Conv1d + BatchNorm1d + ReLU\n\"\"\"\n\n\nclass Res2Conv1dReluBn(nn.Module):\n \"\"\"\n in_channels == out_channels == channels\n \"\"\"\n\n def __init__(self, channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True, scale=4):\n super().__init__()\n assert channels % scale == 0, \"{} % {} != 0\".format(channels, scale)\n self.scale = scale\n self.width = channels // scale\n self.nums = scale if scale == 1 else scale - 1\n\n self.convs = []\n self.bns = []\n for i in range(self.nums):\n self.convs.append(nn.Conv1d(self.width, self.width, kernel_size, stride, padding, dilation, bias=bias))\n self.bns.append(nn.BatchNorm1d(self.width))\n self.convs = nn.ModuleList(self.convs)\n self.bns = nn.ModuleList(self.bns)\n\n def forward(self, x):\n out = []\n spx = torch.split(x, self.width, 1)\n for i in range(self.nums):\n if i == 0:\n sp = spx[i]\n else:\n sp = sp + spx[i]\n # Order: conv -> relu -> bn\n sp = self.convs[i](sp)\n sp = self.bns[i](F.relu(sp))\n out.append(sp)\n if self.scale != 1:\n out.append(spx[self.nums])\n out = torch.cat(out, dim=1)\n\n return out\n\n\n\"\"\" Conv1d + BatchNorm1d + ReLU\n\"\"\"\n\n\nclass Conv1dReluBn(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True):\n super().__init__()\n self.conv = nn.Conv1d(in_channels, out_channels, kernel_size, stride, padding, dilation, bias=bias)\n self.bn = nn.BatchNorm1d(out_channels)\n\n def forward(self, x):\n return self.bn(F.relu(self.conv(x)))\n\n\n\"\"\" The SE connection of 1D case.\n\"\"\"\n\n\nclass SE_Connect(nn.Module):\n def __init__(self, channels, se_bottleneck_dim=128):\n super().__init__()\n self.linear1 = nn.Linear(channels, se_bottleneck_dim)\n self.linear2 = nn.Linear(se_bottleneck_dim, channels)\n\n def forward(self, x):\n out = x.mean(dim=2)\n out = F.relu(self.linear1(out))\n out = torch.sigmoid(self.linear2(out))\n out = x * out.unsqueeze(2)\n\n return out\n\n\n\"\"\" SE-Res2Block of the ECAPA-TDNN architecture.\n\"\"\"\n\n# def SE_Res2Block(channels, kernel_size, stride, padding, dilation, scale):\n# return nn.Sequential(\n# Conv1dReluBn(channels, 512, kernel_size=1, stride=1, padding=0),\n# Res2Conv1dReluBn(512, kernel_size, stride, padding, dilation, scale=scale),\n# Conv1dReluBn(512, channels, kernel_size=1, stride=1, padding=0),\n# SE_Connect(channels)\n# )\n\n\nclass SE_Res2Block(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size, stride, padding, dilation, scale, se_bottleneck_dim):\n super().__init__()\n self.Conv1dReluBn1 = Conv1dReluBn(in_channels, out_channels, kernel_size=1, stride=1, padding=0)\n self.Res2Conv1dReluBn = Res2Conv1dReluBn(out_channels, kernel_size, stride, padding, dilation, scale=scale)\n self.Conv1dReluBn2 = Conv1dReluBn(out_channels, out_channels, kernel_size=1, stride=1, padding=0)\n self.SE_Connect = SE_Connect(out_channels, se_bottleneck_dim)\n\n self.shortcut = None\n if in_channels != out_channels:\n self.shortcut = nn.Conv1d(\n in_channels=in_channels,\n out_channels=out_channels,\n kernel_size=1,\n )\n\n def forward(self, x):\n residual = x\n if self.shortcut:\n residual = self.shortcut(x)\n\n x = self.Conv1dReluBn1(x)\n x = self.Res2Conv1dReluBn(x)\n x = self.Conv1dReluBn2(x)\n x = self.SE_Connect(x)\n\n return x + residual\n\n\n\"\"\" Attentive weighted mean and standard deviation pooling.\n\"\"\"\n\n\nclass AttentiveStatsPool(nn.Module):\n def __init__(self, in_dim, attention_channels=128, global_context_att=False):\n super().__init__()\n self.global_context_att = global_context_att\n\n # Use Conv1d with stride == 1 rather than Linear, then we don't need to transpose inputs.\n if global_context_att:\n self.linear1 = nn.Conv1d(in_dim * 3, attention_channels, kernel_size=1) # equals W and b in the paper\n else:\n self.linear1 = nn.Conv1d(in_dim, attention_channels, kernel_size=1) # equals W and b in the paper\n self.linear2 = nn.Conv1d(attention_channels, in_dim, kernel_size=1) # equals V and k in the paper\n\n def forward(self, x):\n if self.global_context_att:\n context_mean = torch.mean(x, dim=-1, keepdim=True).expand_as(x)\n context_std = torch.sqrt(torch.var(x, dim=-1, keepdim=True) + 1e-10).expand_as(x)\n x_in = torch.cat((x, context_mean, context_std), dim=1)\n else:\n x_in = x\n\n # DON'T use ReLU here! In experiments, I find ReLU hard to converge.\n alpha = torch.tanh(self.linear1(x_in))\n # alpha = F.relu(self.linear1(x_in))\n alpha = torch.softmax(self.linear2(alpha), dim=2)\n mean = torch.sum(alpha * x, dim=2)\n residuals = torch.sum(alpha * (x**2), dim=2) - mean**2\n std = torch.sqrt(residuals.clamp(min=1e-9))\n return torch.cat([mean, std], dim=1)\n\n\nclass ECAPA_TDNN(nn.Module):\n def __init__(\n self,\n feat_dim=80,\n channels=512,\n emb_dim=192,\n global_context_att=False,\n feat_type=\"wavlm_large\",\n sr=16000,\n feature_selection=\"hidden_states\",\n update_extract=False,\n config_path=None,\n ):\n super().__init__()\n\n self.feat_type = feat_type\n self.feature_selection = feature_selection\n self.update_extract = update_extract\n self.sr = sr\n\n torch.hub._validate_not_a_forked_repo = lambda a, b, c: True\n try:\n local_s3prl_path = os.path.expanduser(\"~/.cache/torch/hub/s3prl_s3prl_main\")\n self.feature_extract = torch.hub.load(local_s3prl_path, feat_type, source=\"local\", config_path=config_path)\n except: # noqa: E722\n self.feature_extract = torch.hub.load(\"s3prl/s3prl\", feat_type)\n\n if len(self.feature_extract.model.encoder.layers) == 24 and hasattr(\n self.feature_extract.model.encoder.layers[23].self_attn, \"fp32_attention\"\n ):\n self.feature_extract.model.encoder.layers[23].self_attn.fp32_attention = False\n if len(self.feature_extract.model.encoder.layers) == 24 and hasattr(\n self.feature_extract.model.encoder.layers[11].self_attn, \"fp32_attention\"\n ):\n self.feature_extract.model.encoder.layers[11].self_attn.fp32_attention = False\n\n self.feat_num = self.get_feat_num()\n self.feature_weight = nn.Parameter(torch.zeros(self.feat_num))\n\n if feat_type != \"fbank\" and feat_type != \"mfcc\":\n freeze_list = [\"final_proj\", \"label_embs_concat\", \"mask_emb\", \"project_q\", \"quantizer\"]\n for name, param in self.feature_extract.named_parameters():\n for freeze_val in freeze_list:\n if freeze_val in name:\n param.requires_grad = False\n break\n\n if not self.update_extract:\n for param in self.feature_extract.parameters():\n param.requires_grad = False\n\n self.instance_norm = nn.InstanceNorm1d(feat_dim)\n # self.channels = [channels] * 4 + [channels * 3]\n self.channels = [channels] * 4 + [1536]\n\n self.layer1 = Conv1dReluBn(feat_dim, self.channels[0], kernel_size=5, padding=2)\n self.layer2 = SE_Res2Block(\n self.channels[0],\n self.channels[1],\n kernel_size=3,\n stride=1,\n padding=2,\n dilation=2,\n scale=8,\n se_bottleneck_dim=128,\n )\n self.layer3 = SE_Res2Block(\n self.channels[1],\n self.channels[2],\n kernel_size=3,\n stride=1,\n padding=3,\n dilation=3,\n scale=8,\n se_bottleneck_dim=128,\n )\n self.layer4 = SE_Res2Block(\n self.channels[2],\n self.channels[3],\n kernel_size=3,\n stride=1,\n padding=4,\n dilation=4,\n scale=8,\n se_bottleneck_dim=128,\n )\n\n # self.conv = nn.Conv1d(self.channels[-1], self.channels[-1], kernel_size=1)\n cat_channels = channels * 3\n self.conv = nn.Conv1d(cat_channels, self.channels[-1], kernel_size=1)\n self.pooling = AttentiveStatsPool(\n self.channels[-1], attention_channels=128, global_context_att=global_context_att\n )\n self.bn = nn.BatchNorm1d(self.channels[-1] * 2)\n self.linear = nn.Linear(self.channels[-1] * 2, emb_dim)\n\n def get_feat_num(self):\n self.feature_extract.eval()\n wav = [torch.randn(self.sr).to(next(self.feature_extract.parameters()).device)]\n with torch.no_grad():\n features = self.feature_extract(wav)\n select_feature = features[self.feature_selection]\n if isinstance(select_feature, (list, tuple)):\n return len(select_feature)\n else:\n return 1\n\n def get_feat(self, x):\n if self.update_extract:\n x = self.feature_extract([sample for sample in x])\n else:\n with torch.no_grad():\n if self.feat_type == \"fbank\" or self.feat_type == \"mfcc\":\n x = self.feature_extract(x) + 1e-6 # B x feat_dim x time_len\n else:\n x = self.feature_extract([sample for sample in x])\n\n if self.feat_type == \"fbank\":\n x = x.log()\n\n if self.feat_type != \"fbank\" and self.feat_type != \"mfcc\":\n x = x[self.feature_selection]\n if isinstance(x, (list, tuple)):\n x = torch.stack(x, dim=0)\n else:\n x = x.unsqueeze(0)\n norm_weights = F.softmax(self.feature_weight, dim=-1).unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)\n x = (norm_weights * x).sum(dim=0)\n x = torch.transpose(x, 1, 2) + 1e-6\n\n x = self.instance_norm(x)\n return x\n\n def forward(self, x):\n x = self.get_feat(x)\n\n out1 = self.layer1(x)\n out2 = self.layer2(out1)\n out3 = self.layer3(out2)\n out4 = self.layer4(out3)\n\n out = torch.cat([out2, out3, out4], dim=1)\n out = F.relu(self.conv(out))\n out = self.bn(self.pooling(out))\n out = self.linear(out)\n\n return out\n\n\ndef ECAPA_TDNN_SMALL(\n feat_dim,\n emb_dim=256,\n feat_type=\"wavlm_large\",\n sr=16000,\n feature_selection=\"hidden_states\",\n update_extract=False,\n config_path=None,\n):\n return ECAPA_TDNN(\n feat_dim=feat_dim,\n channels=512,\n emb_dim=emb_dim,\n feat_type=feat_type,\n sr=sr,\n feature_selection=feature_selection,\n update_extract=update_extract,\n config_path=config_path,\n )","source_hash":"3d934824932b71e6ba3824c817367328c5f099335106dc03e6ba68a143da6a0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.ecapa_tdnn.Res2Conv1dReluBn","uri":"program://DMOSpeech2/class/src.f5_tts.eval.ecapa_tdnn.Res2Conv1dReluBn#L17-L53","kind":"class","name":"Res2Conv1dReluBn","path":"src/f5_tts/eval/ecapa_tdnn.py","language":"python","start_line":17,"end_line":53,"context_start_line":1,"context_end_line":73,"code":"# just for speaker similarity evaluation, third-party code\n\n# From https://github.com/microsoft/UniSpeech/blob/main/downstreams/speaker_verification/models/\n# part of the code is borrowed from https://github.com/lawlict/ECAPA-TDNN\n\nimport os\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\n\n\n\"\"\" Res2Conv1d + BatchNorm1d + ReLU\n\"\"\"\n\n\nclass Res2Conv1dReluBn(nn.Module):\n \"\"\"\n in_channels == out_channels == channels\n \"\"\"\n\n def __init__(self, channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True, scale=4):\n super().__init__()\n assert channels % scale == 0, \"{} % {} != 0\".format(channels, scale)\n self.scale = scale\n self.width = channels // scale\n self.nums = scale if scale == 1 else scale - 1\n\n self.convs = []\n self.bns = []\n for i in range(self.nums):\n self.convs.append(nn.Conv1d(self.width, self.width, kernel_size, stride, padding, dilation, bias=bias))\n self.bns.append(nn.BatchNorm1d(self.width))\n self.convs = nn.ModuleList(self.convs)\n self.bns = nn.ModuleList(self.bns)\n\n def forward(self, x):\n out = []\n spx = torch.split(x, self.width, 1)\n for i in range(self.nums):\n if i == 0:\n sp = spx[i]\n else:\n sp = sp + spx[i]\n # Order: conv -> relu -> bn\n sp = self.convs[i](sp)\n sp = self.bns[i](F.relu(sp))\n out.append(sp)\n if self.scale != 1:\n out.append(spx[self.nums])\n out = torch.cat(out, dim=1)\n\n return out\n\n\n\"\"\" Conv1d + BatchNorm1d + ReLU\n\"\"\"\n\n\nclass Conv1dReluBn(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True):\n super().__init__()\n self.conv = nn.Conv1d(in_channels, out_channels, kernel_size, stride, padding, dilation, bias=bias)\n self.bn = nn.BatchNorm1d(out_channels)\n\n def forward(self, x):\n return self.bn(F.relu(self.conv(x)))\n\n\n\"\"\" The SE connection of 1D case.\n\"\"\"\n\n","source_hash":"3d934824932b71e6ba3824c817367328c5f099335106dc03e6ba68a143da6a0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.ecapa_tdnn.Conv1dReluBn","uri":"program://DMOSpeech2/class/src.f5_tts.eval.ecapa_tdnn.Conv1dReluBn#L60-L67","kind":"class","name":"Conv1dReluBn","path":"src/f5_tts/eval/ecapa_tdnn.py","language":"python","start_line":60,"end_line":67,"context_start_line":40,"context_end_line":87,"code":" for i in range(self.nums):\n if i == 0:\n sp = spx[i]\n else:\n sp = sp + spx[i]\n # Order: conv -> relu -> bn\n sp = self.convs[i](sp)\n sp = self.bns[i](F.relu(sp))\n out.append(sp)\n if self.scale != 1:\n out.append(spx[self.nums])\n out = torch.cat(out, dim=1)\n\n return out\n\n\n\"\"\" Conv1d + BatchNorm1d + ReLU\n\"\"\"\n\n\nclass Conv1dReluBn(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True):\n super().__init__()\n self.conv = nn.Conv1d(in_channels, out_channels, kernel_size, stride, padding, dilation, bias=bias)\n self.bn = nn.BatchNorm1d(out_channels)\n\n def forward(self, x):\n return self.bn(F.relu(self.conv(x)))\n\n\n\"\"\" The SE connection of 1D case.\n\"\"\"\n\n\nclass SE_Connect(nn.Module):\n def __init__(self, channels, se_bottleneck_dim=128):\n super().__init__()\n self.linear1 = nn.Linear(channels, se_bottleneck_dim)\n self.linear2 = nn.Linear(se_bottleneck_dim, channels)\n\n def forward(self, x):\n out = x.mean(dim=2)\n out = F.relu(self.linear1(out))\n out = torch.sigmoid(self.linear2(out))\n out = x * out.unsqueeze(2)\n\n return out\n","source_hash":"3d934824932b71e6ba3824c817367328c5f099335106dc03e6ba68a143da6a0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.ecapa_tdnn.SE_Connect","uri":"program://DMOSpeech2/class/src.f5_tts.eval.ecapa_tdnn.SE_Connect#L74-L86","kind":"class","name":"SE_Connect","path":"src/f5_tts/eval/ecapa_tdnn.py","language":"python","start_line":74,"end_line":86,"context_start_line":54,"context_end_line":106,"code":"\n\n\"\"\" Conv1d + BatchNorm1d + ReLU\n\"\"\"\n\n\nclass Conv1dReluBn(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True):\n super().__init__()\n self.conv = nn.Conv1d(in_channels, out_channels, kernel_size, stride, padding, dilation, bias=bias)\n self.bn = nn.BatchNorm1d(out_channels)\n\n def forward(self, x):\n return self.bn(F.relu(self.conv(x)))\n\n\n\"\"\" The SE connection of 1D case.\n\"\"\"\n\n\nclass SE_Connect(nn.Module):\n def __init__(self, channels, se_bottleneck_dim=128):\n super().__init__()\n self.linear1 = nn.Linear(channels, se_bottleneck_dim)\n self.linear2 = nn.Linear(se_bottleneck_dim, channels)\n\n def forward(self, x):\n out = x.mean(dim=2)\n out = F.relu(self.linear1(out))\n out = torch.sigmoid(self.linear2(out))\n out = x * out.unsqueeze(2)\n\n return out\n\n\n\"\"\" SE-Res2Block of the ECAPA-TDNN architecture.\n\"\"\"\n\n# def SE_Res2Block(channels, kernel_size, stride, padding, dilation, scale):\n# return nn.Sequential(\n# Conv1dReluBn(channels, 512, kernel_size=1, stride=1, padding=0),\n# Res2Conv1dReluBn(512, kernel_size, stride, padding, dilation, scale=scale),\n# Conv1dReluBn(512, channels, kernel_size=1, stride=1, padding=0),\n# SE_Connect(channels)\n# )\n\n\nclass SE_Res2Block(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size, stride, padding, dilation, scale, se_bottleneck_dim):\n super().__init__()\n self.Conv1dReluBn1 = Conv1dReluBn(in_channels, out_channels, kernel_size=1, stride=1, padding=0)\n self.Res2Conv1dReluBn = Res2Conv1dReluBn(out_channels, kernel_size, stride, padding, dilation, scale=scale)\n self.Conv1dReluBn2 = Conv1dReluBn(out_channels, out_channels, kernel_size=1, stride=1, padding=0)","source_hash":"3d934824932b71e6ba3824c817367328c5f099335106dc03e6ba68a143da6a0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.ecapa_tdnn.SE_Res2Block","uri":"program://DMOSpeech2/class/src.f5_tts.eval.ecapa_tdnn.SE_Res2Block#L101-L127","kind":"class","name":"SE_Res2Block","path":"src/f5_tts/eval/ecapa_tdnn.py","language":"python","start_line":101,"end_line":127,"context_start_line":81,"context_end_line":147,"code":" out = x.mean(dim=2)\n out = F.relu(self.linear1(out))\n out = torch.sigmoid(self.linear2(out))\n out = x * out.unsqueeze(2)\n\n return out\n\n\n\"\"\" SE-Res2Block of the ECAPA-TDNN architecture.\n\"\"\"\n\n# def SE_Res2Block(channels, kernel_size, stride, padding, dilation, scale):\n# return nn.Sequential(\n# Conv1dReluBn(channels, 512, kernel_size=1, stride=1, padding=0),\n# Res2Conv1dReluBn(512, kernel_size, stride, padding, dilation, scale=scale),\n# Conv1dReluBn(512, channels, kernel_size=1, stride=1, padding=0),\n# SE_Connect(channels)\n# )\n\n\nclass SE_Res2Block(nn.Module):\n def __init__(self, in_channels, out_channels, kernel_size, stride, padding, dilation, scale, se_bottleneck_dim):\n super().__init__()\n self.Conv1dReluBn1 = Conv1dReluBn(in_channels, out_channels, kernel_size=1, stride=1, padding=0)\n self.Res2Conv1dReluBn = Res2Conv1dReluBn(out_channels, kernel_size, stride, padding, dilation, scale=scale)\n self.Conv1dReluBn2 = Conv1dReluBn(out_channels, out_channels, kernel_size=1, stride=1, padding=0)\n self.SE_Connect = SE_Connect(out_channels, se_bottleneck_dim)\n\n self.shortcut = None\n if in_channels != out_channels:\n self.shortcut = nn.Conv1d(\n in_channels=in_channels,\n out_channels=out_channels,\n kernel_size=1,\n )\n\n def forward(self, x):\n residual = x\n if self.shortcut:\n residual = self.shortcut(x)\n\n x = self.Conv1dReluBn1(x)\n x = self.Res2Conv1dReluBn(x)\n x = self.Conv1dReluBn2(x)\n x = self.SE_Connect(x)\n\n return x + residual\n\n\n\"\"\" Attentive weighted mean and standard deviation pooling.\n\"\"\"\n\n\nclass AttentiveStatsPool(nn.Module):\n def __init__(self, in_dim, attention_channels=128, global_context_att=False):\n super().__init__()\n self.global_context_att = global_context_att\n\n # Use Conv1d with stride == 1 rather than Linear, then we don't need to transpose inputs.\n if global_context_att:\n self.linear1 = nn.Conv1d(in_dim * 3, attention_channels, kernel_size=1) # equals W and b in the paper\n else:\n self.linear1 = nn.Conv1d(in_dim, attention_channels, kernel_size=1) # equals W and b in the paper\n self.linear2 = nn.Conv1d(attention_channels, in_dim, kernel_size=1) # equals V and k in the paper\n\n def forward(self, x):\n if self.global_context_att:","source_hash":"3d934824932b71e6ba3824c817367328c5f099335106dc03e6ba68a143da6a0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.ecapa_tdnn.AttentiveStatsPool","uri":"program://DMOSpeech2/class/src.f5_tts.eval.ecapa_tdnn.AttentiveStatsPool#L134-L161","kind":"class","name":"AttentiveStatsPool","path":"src/f5_tts/eval/ecapa_tdnn.py","language":"python","start_line":134,"end_line":161,"context_start_line":114,"context_end_line":181,"code":" kernel_size=1,\n )\n\n def forward(self, x):\n residual = x\n if self.shortcut:\n residual = self.shortcut(x)\n\n x = self.Conv1dReluBn1(x)\n x = self.Res2Conv1dReluBn(x)\n x = self.Conv1dReluBn2(x)\n x = self.SE_Connect(x)\n\n return x + residual\n\n\n\"\"\" Attentive weighted mean and standard deviation pooling.\n\"\"\"\n\n\nclass AttentiveStatsPool(nn.Module):\n def __init__(self, in_dim, attention_channels=128, global_context_att=False):\n super().__init__()\n self.global_context_att = global_context_att\n\n # Use Conv1d with stride == 1 rather than Linear, then we don't need to transpose inputs.\n if global_context_att:\n self.linear1 = nn.Conv1d(in_dim * 3, attention_channels, kernel_size=1) # equals W and b in the paper\n else:\n self.linear1 = nn.Conv1d(in_dim, attention_channels, kernel_size=1) # equals W and b in the paper\n self.linear2 = nn.Conv1d(attention_channels, in_dim, kernel_size=1) # equals V and k in the paper\n\n def forward(self, x):\n if self.global_context_att:\n context_mean = torch.mean(x, dim=-1, keepdim=True).expand_as(x)\n context_std = torch.sqrt(torch.var(x, dim=-1, keepdim=True) + 1e-10).expand_as(x)\n x_in = torch.cat((x, context_mean, context_std), dim=1)\n else:\n x_in = x\n\n # DON'T use ReLU here! In experiments, I find ReLU hard to converge.\n alpha = torch.tanh(self.linear1(x_in))\n # alpha = F.relu(self.linear1(x_in))\n alpha = torch.softmax(self.linear2(alpha), dim=2)\n mean = torch.sum(alpha * x, dim=2)\n residuals = torch.sum(alpha * (x**2), dim=2) - mean**2\n std = torch.sqrt(residuals.clamp(min=1e-9))\n return torch.cat([mean, std], dim=1)\n\n\nclass ECAPA_TDNN(nn.Module):\n def __init__(\n self,\n feat_dim=80,\n channels=512,\n emb_dim=192,\n global_context_att=False,\n feat_type=\"wavlm_large\",\n sr=16000,\n feature_selection=\"hidden_states\",\n update_extract=False,\n config_path=None,\n ):\n super().__init__()\n\n self.feat_type = feat_type\n self.feature_selection = feature_selection\n self.update_extract = update_extract","source_hash":"3d934824932b71e6ba3824c817367328c5f099335106dc03e6ba68a143da6a0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.ecapa_tdnn.ECAPA_TDNN","uri":"program://DMOSpeech2/class/src.f5_tts.eval.ecapa_tdnn.ECAPA_TDNN#L164-L310","kind":"class","name":"ECAPA_TDNN","path":"src/f5_tts/eval/ecapa_tdnn.py","language":"python","start_line":164,"end_line":310,"context_start_line":144,"context_end_line":330,"code":" self.linear2 = nn.Conv1d(attention_channels, in_dim, kernel_size=1) # equals V and k in the paper\n\n def forward(self, x):\n if self.global_context_att:\n context_mean = torch.mean(x, dim=-1, keepdim=True).expand_as(x)\n context_std = torch.sqrt(torch.var(x, dim=-1, keepdim=True) + 1e-10).expand_as(x)\n x_in = torch.cat((x, context_mean, context_std), dim=1)\n else:\n x_in = x\n\n # DON'T use ReLU here! In experiments, I find ReLU hard to converge.\n alpha = torch.tanh(self.linear1(x_in))\n # alpha = F.relu(self.linear1(x_in))\n alpha = torch.softmax(self.linear2(alpha), dim=2)\n mean = torch.sum(alpha * x, dim=2)\n residuals = torch.sum(alpha * (x**2), dim=2) - mean**2\n std = torch.sqrt(residuals.clamp(min=1e-9))\n return torch.cat([mean, std], dim=1)\n\n\nclass ECAPA_TDNN(nn.Module):\n def __init__(\n self,\n feat_dim=80,\n channels=512,\n emb_dim=192,\n global_context_att=False,\n feat_type=\"wavlm_large\",\n sr=16000,\n feature_selection=\"hidden_states\",\n update_extract=False,\n config_path=None,\n ):\n super().__init__()\n\n self.feat_type = feat_type\n self.feature_selection = feature_selection\n self.update_extract = update_extract\n self.sr = sr\n\n torch.hub._validate_not_a_forked_repo = lambda a, b, c: True\n try:\n local_s3prl_path = os.path.expanduser(\"~/.cache/torch/hub/s3prl_s3prl_main\")\n self.feature_extract = torch.hub.load(local_s3prl_path, feat_type, source=\"local\", config_path=config_path)\n except: # noqa: E722\n self.feature_extract = torch.hub.load(\"s3prl/s3prl\", feat_type)\n\n if len(self.feature_extract.model.encoder.layers) == 24 and hasattr(\n self.feature_extract.model.encoder.layers[23].self_attn, \"fp32_attention\"\n ):\n self.feature_extract.model.encoder.layers[23].self_attn.fp32_attention = False\n if len(self.feature_extract.model.encoder.layers) == 24 and hasattr(\n self.feature_extract.model.encoder.layers[11].self_attn, \"fp32_attention\"\n ):\n self.feature_extract.model.encoder.layers[11].self_attn.fp32_attention = False\n\n self.feat_num = self.get_feat_num()\n self.feature_weight = nn.Parameter(torch.zeros(self.feat_num))\n\n if feat_type != \"fbank\" and feat_type != \"mfcc\":\n freeze_list = [\"final_proj\", \"label_embs_concat\", \"mask_emb\", \"project_q\", \"quantizer\"]\n for name, param in self.feature_extract.named_parameters():\n for freeze_val in freeze_list:\n if freeze_val in name:\n param.requires_grad = False\n break\n\n if not self.update_extract:\n for param in self.feature_extract.parameters():\n param.requires_grad = False\n\n self.instance_norm = nn.InstanceNorm1d(feat_dim)\n # self.channels = [channels] * 4 + [channels * 3]\n self.channels = [channels] * 4 + [1536]\n\n self.layer1 = Conv1dReluBn(feat_dim, self.channels[0], kernel_size=5, padding=2)\n self.layer2 = SE_Res2Block(\n self.channels[0],\n self.channels[1],\n kernel_size=3,\n stride=1,\n padding=2,\n dilation=2,\n scale=8,\n se_bottleneck_dim=128,\n )\n self.layer3 = SE_Res2Block(\n self.channels[1],\n self.channels[2],\n kernel_size=3,\n stride=1,\n padding=3,\n dilation=3,\n scale=8,\n se_bottleneck_dim=128,\n )\n self.layer4 = SE_Res2Block(\n self.channels[2],\n self.channels[3],\n kernel_size=3,\n stride=1,\n padding=4,\n dilation=4,\n scale=8,\n se_bottleneck_dim=128,\n )\n\n # self.conv = nn.Conv1d(self.channels[-1], self.channels[-1], kernel_size=1)\n cat_channels = channels * 3\n self.conv = nn.Conv1d(cat_channels, self.channels[-1], kernel_size=1)\n self.pooling = AttentiveStatsPool(\n self.channels[-1], attention_channels=128, global_context_att=global_context_att\n )\n self.bn = nn.BatchNorm1d(self.channels[-1] * 2)\n self.linear = nn.Linear(self.channels[-1] * 2, emb_dim)\n\n def get_feat_num(self):\n self.feature_extract.eval()\n wav = [torch.randn(self.sr).to(next(self.feature_extract.parameters()).device)]\n with torch.no_grad():\n features = self.feature_extract(wav)\n select_feature = features[self.feature_selection]\n if isinstance(select_feature, (list, tuple)):\n return len(select_feature)\n else:\n return 1\n\n def get_feat(self, x):\n if self.update_extract:\n x = self.feature_extract([sample for sample in x])\n else:\n with torch.no_grad():\n if self.feat_type == \"fbank\" or self.feat_type == \"mfcc\":\n x = self.feature_extract(x) + 1e-6 # B x feat_dim x time_len\n else:\n x = self.feature_extract([sample for sample in x])\n\n if self.feat_type == \"fbank\":\n x = x.log()\n\n if self.feat_type != \"fbank\" and self.feat_type != \"mfcc\":\n x = x[self.feature_selection]\n if isinstance(x, (list, tuple)):\n x = torch.stack(x, dim=0)\n else:\n x = x.unsqueeze(0)\n norm_weights = F.softmax(self.feature_weight, dim=-1).unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)\n x = (norm_weights * x).sum(dim=0)\n x = torch.transpose(x, 1, 2) + 1e-6\n\n x = self.instance_norm(x)\n return x\n\n def forward(self, x):\n x = self.get_feat(x)\n\n out1 = self.layer1(x)\n out2 = self.layer2(out1)\n out3 = self.layer3(out2)\n out4 = self.layer4(out3)\n\n out = torch.cat([out2, out3, out4], dim=1)\n out = F.relu(self.conv(out))\n out = self.bn(self.pooling(out))\n out = self.linear(out)\n\n return out\n\n\ndef ECAPA_TDNN_SMALL(\n feat_dim,\n emb_dim=256,\n feat_type=\"wavlm_large\",\n sr=16000,\n feature_selection=\"hidden_states\",\n update_extract=False,\n config_path=None,\n):\n return ECAPA_TDNN(\n feat_dim=feat_dim,\n channels=512,\n emb_dim=emb_dim,\n feat_type=feat_type,\n sr=sr,\n feature_selection=feature_selection,\n update_extract=update_extract,\n config_path=config_path,","source_hash":"3d934824932b71e6ba3824c817367328c5f099335106dc03e6ba68a143da6a0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.ecapa_tdnn.ECAPA_TDNN_SMALL","uri":"program://DMOSpeech2/function/src.f5_tts.eval.ecapa_tdnn.ECAPA_TDNN_SMALL#L313-L331","kind":"function","name":"ECAPA_TDNN_SMALL","path":"src/f5_tts/eval/ecapa_tdnn.py","language":"python","start_line":313,"end_line":331,"context_start_line":293,"context_end_line":331,"code":"\n x = self.instance_norm(x)\n return x\n\n def forward(self, x):\n x = self.get_feat(x)\n\n out1 = self.layer1(x)\n out2 = self.layer2(out1)\n out3 = self.layer3(out2)\n out4 = self.layer4(out3)\n\n out = torch.cat([out2, out3, out4], dim=1)\n out = F.relu(self.conv(out))\n out = self.bn(self.pooling(out))\n out = self.linear(out)\n\n return out\n\n\ndef ECAPA_TDNN_SMALL(\n feat_dim,\n emb_dim=256,\n feat_type=\"wavlm_large\",\n sr=16000,\n feature_selection=\"hidden_states\",\n update_extract=False,\n config_path=None,\n):\n return ECAPA_TDNN(\n feat_dim=feat_dim,\n channels=512,\n emb_dim=emb_dim,\n feat_type=feat_type,\n sr=sr,\n feature_selection=feature_selection,\n update_extract=update_extract,\n config_path=config_path,\n )","source_hash":"3d934824932b71e6ba3824c817367328c5f099335106dc03e6ba68a143da6a0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.ecapa_tdnn.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.eval.ecapa_tdnn.__init__#L165-L258","kind":"function","name":"__init__","path":"src/f5_tts/eval/ecapa_tdnn.py","language":"python","start_line":165,"end_line":258,"context_start_line":145,"context_end_line":278,"code":"\n def forward(self, x):\n if self.global_context_att:\n context_mean = torch.mean(x, dim=-1, keepdim=True).expand_as(x)\n context_std = torch.sqrt(torch.var(x, dim=-1, keepdim=True) + 1e-10).expand_as(x)\n x_in = torch.cat((x, context_mean, context_std), dim=1)\n else:\n x_in = x\n\n # DON'T use ReLU here! In experiments, I find ReLU hard to converge.\n alpha = torch.tanh(self.linear1(x_in))\n # alpha = F.relu(self.linear1(x_in))\n alpha = torch.softmax(self.linear2(alpha), dim=2)\n mean = torch.sum(alpha * x, dim=2)\n residuals = torch.sum(alpha * (x**2), dim=2) - mean**2\n std = torch.sqrt(residuals.clamp(min=1e-9))\n return torch.cat([mean, std], dim=1)\n\n\nclass ECAPA_TDNN(nn.Module):\n def __init__(\n self,\n feat_dim=80,\n channels=512,\n emb_dim=192,\n global_context_att=False,\n feat_type=\"wavlm_large\",\n sr=16000,\n feature_selection=\"hidden_states\",\n update_extract=False,\n config_path=None,\n ):\n super().__init__()\n\n self.feat_type = feat_type\n self.feature_selection = feature_selection\n self.update_extract = update_extract\n self.sr = sr\n\n torch.hub._validate_not_a_forked_repo = lambda a, b, c: True\n try:\n local_s3prl_path = os.path.expanduser(\"~/.cache/torch/hub/s3prl_s3prl_main\")\n self.feature_extract = torch.hub.load(local_s3prl_path, feat_type, source=\"local\", config_path=config_path)\n except: # noqa: E722\n self.feature_extract = torch.hub.load(\"s3prl/s3prl\", feat_type)\n\n if len(self.feature_extract.model.encoder.layers) == 24 and hasattr(\n self.feature_extract.model.encoder.layers[23].self_attn, \"fp32_attention\"\n ):\n self.feature_extract.model.encoder.layers[23].self_attn.fp32_attention = False\n if len(self.feature_extract.model.encoder.layers) == 24 and hasattr(\n self.feature_extract.model.encoder.layers[11].self_attn, \"fp32_attention\"\n ):\n self.feature_extract.model.encoder.layers[11].self_attn.fp32_attention = False\n\n self.feat_num = self.get_feat_num()\n self.feature_weight = nn.Parameter(torch.zeros(self.feat_num))\n\n if feat_type != \"fbank\" and feat_type != \"mfcc\":\n freeze_list = [\"final_proj\", \"label_embs_concat\", \"mask_emb\", \"project_q\", \"quantizer\"]\n for name, param in self.feature_extract.named_parameters():\n for freeze_val in freeze_list:\n if freeze_val in name:\n param.requires_grad = False\n break\n\n if not self.update_extract:\n for param in self.feature_extract.parameters():\n param.requires_grad = False\n\n self.instance_norm = nn.InstanceNorm1d(feat_dim)\n # self.channels = [channels] * 4 + [channels * 3]\n self.channels = [channels] * 4 + [1536]\n\n self.layer1 = Conv1dReluBn(feat_dim, self.channels[0], kernel_size=5, padding=2)\n self.layer2 = SE_Res2Block(\n self.channels[0],\n self.channels[1],\n kernel_size=3,\n stride=1,\n padding=2,\n dilation=2,\n scale=8,\n se_bottleneck_dim=128,\n )\n self.layer3 = SE_Res2Block(\n self.channels[1],\n self.channels[2],\n kernel_size=3,\n stride=1,\n padding=3,\n dilation=3,\n scale=8,\n se_bottleneck_dim=128,\n )\n self.layer4 = SE_Res2Block(\n self.channels[2],\n self.channels[3],\n kernel_size=3,\n stride=1,\n padding=4,\n dilation=4,\n scale=8,\n se_bottleneck_dim=128,\n )\n\n # self.conv = nn.Conv1d(self.channels[-1], self.channels[-1], kernel_size=1)\n cat_channels = channels * 3\n self.conv = nn.Conv1d(cat_channels, self.channels[-1], kernel_size=1)\n self.pooling = AttentiveStatsPool(\n self.channels[-1], attention_channels=128, global_context_att=global_context_att\n )\n self.bn = nn.BatchNorm1d(self.channels[-1] * 2)\n self.linear = nn.Linear(self.channels[-1] * 2, emb_dim)\n\n def get_feat_num(self):\n self.feature_extract.eval()\n wav = [torch.randn(self.sr).to(next(self.feature_extract.parameters()).device)]\n with torch.no_grad():\n features = self.feature_extract(wav)\n select_feature = features[self.feature_selection]\n if isinstance(select_feature, (list, tuple)):\n return len(select_feature)\n else:\n return 1\n\n def get_feat(self, x):\n if self.update_extract:\n x = self.feature_extract([sample for sample in x])\n else:\n with torch.no_grad():\n if self.feat_type == \"fbank\" or self.feat_type == \"mfcc\":\n x = self.feature_extract(x) + 1e-6 # B x feat_dim x time_len\n else:","source_hash":"3d934824932b71e6ba3824c817367328c5f099335106dc03e6ba68a143da6a0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.ecapa_tdnn.forward","uri":"program://DMOSpeech2/function/src.f5_tts.eval.ecapa_tdnn.forward#L297-L310","kind":"function","name":"forward","path":"src/f5_tts/eval/ecapa_tdnn.py","language":"python","start_line":297,"end_line":310,"context_start_line":277,"context_end_line":330,"code":" x = self.feature_extract(x) + 1e-6 # B x feat_dim x time_len\n else:\n x = self.feature_extract([sample for sample in x])\n\n if self.feat_type == \"fbank\":\n x = x.log()\n\n if self.feat_type != \"fbank\" and self.feat_type != \"mfcc\":\n x = x[self.feature_selection]\n if isinstance(x, (list, tuple)):\n x = torch.stack(x, dim=0)\n else:\n x = x.unsqueeze(0)\n norm_weights = F.softmax(self.feature_weight, dim=-1).unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)\n x = (norm_weights * x).sum(dim=0)\n x = torch.transpose(x, 1, 2) + 1e-6\n\n x = self.instance_norm(x)\n return x\n\n def forward(self, x):\n x = self.get_feat(x)\n\n out1 = self.layer1(x)\n out2 = self.layer2(out1)\n out3 = self.layer3(out2)\n out4 = self.layer4(out3)\n\n out = torch.cat([out2, out3, out4], dim=1)\n out = F.relu(self.conv(out))\n out = self.bn(self.pooling(out))\n out = self.linear(out)\n\n return out\n\n\ndef ECAPA_TDNN_SMALL(\n feat_dim,\n emb_dim=256,\n feat_type=\"wavlm_large\",\n sr=16000,\n feature_selection=\"hidden_states\",\n update_extract=False,\n config_path=None,\n):\n return ECAPA_TDNN(\n feat_dim=feat_dim,\n channels=512,\n emb_dim=emb_dim,\n feat_type=feat_type,\n sr=sr,\n feature_selection=feature_selection,\n update_extract=update_extract,\n config_path=config_path,","source_hash":"3d934824932b71e6ba3824c817367328c5f099335106dc03e6ba68a143da6a0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.ecapa_tdnn.get_feat_num","uri":"program://DMOSpeech2/function/src.f5_tts.eval.ecapa_tdnn.get_feat_num#L260-L269","kind":"function","name":"get_feat_num","path":"src/f5_tts/eval/ecapa_tdnn.py","language":"python","start_line":260,"end_line":269,"context_start_line":240,"context_end_line":289,"code":" self.layer4 = SE_Res2Block(\n self.channels[2],\n self.channels[3],\n kernel_size=3,\n stride=1,\n padding=4,\n dilation=4,\n scale=8,\n se_bottleneck_dim=128,\n )\n\n # self.conv = nn.Conv1d(self.channels[-1], self.channels[-1], kernel_size=1)\n cat_channels = channels * 3\n self.conv = nn.Conv1d(cat_channels, self.channels[-1], kernel_size=1)\n self.pooling = AttentiveStatsPool(\n self.channels[-1], attention_channels=128, global_context_att=global_context_att\n )\n self.bn = nn.BatchNorm1d(self.channels[-1] * 2)\n self.linear = nn.Linear(self.channels[-1] * 2, emb_dim)\n\n def get_feat_num(self):\n self.feature_extract.eval()\n wav = [torch.randn(self.sr).to(next(self.feature_extract.parameters()).device)]\n with torch.no_grad():\n features = self.feature_extract(wav)\n select_feature = features[self.feature_selection]\n if isinstance(select_feature, (list, tuple)):\n return len(select_feature)\n else:\n return 1\n\n def get_feat(self, x):\n if self.update_extract:\n x = self.feature_extract([sample for sample in x])\n else:\n with torch.no_grad():\n if self.feat_type == \"fbank\" or self.feat_type == \"mfcc\":\n x = self.feature_extract(x) + 1e-6 # B x feat_dim x time_len\n else:\n x = self.feature_extract([sample for sample in x])\n\n if self.feat_type == \"fbank\":\n x = x.log()\n\n if self.feat_type != \"fbank\" and self.feat_type != \"mfcc\":\n x = x[self.feature_selection]\n if isinstance(x, (list, tuple)):\n x = torch.stack(x, dim=0)\n else:\n x = x.unsqueeze(0)","source_hash":"3d934824932b71e6ba3824c817367328c5f099335106dc03e6ba68a143da6a0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.ecapa_tdnn.get_feat","uri":"program://DMOSpeech2/function/src.f5_tts.eval.ecapa_tdnn.get_feat#L271-L295","kind":"function","name":"get_feat","path":"src/f5_tts/eval/ecapa_tdnn.py","language":"python","start_line":271,"end_line":295,"context_start_line":251,"context_end_line":315,"code":" # self.conv = nn.Conv1d(self.channels[-1], self.channels[-1], kernel_size=1)\n cat_channels = channels * 3\n self.conv = nn.Conv1d(cat_channels, self.channels[-1], kernel_size=1)\n self.pooling = AttentiveStatsPool(\n self.channels[-1], attention_channels=128, global_context_att=global_context_att\n )\n self.bn = nn.BatchNorm1d(self.channels[-1] * 2)\n self.linear = nn.Linear(self.channels[-1] * 2, emb_dim)\n\n def get_feat_num(self):\n self.feature_extract.eval()\n wav = [torch.randn(self.sr).to(next(self.feature_extract.parameters()).device)]\n with torch.no_grad():\n features = self.feature_extract(wav)\n select_feature = features[self.feature_selection]\n if isinstance(select_feature, (list, tuple)):\n return len(select_feature)\n else:\n return 1\n\n def get_feat(self, x):\n if self.update_extract:\n x = self.feature_extract([sample for sample in x])\n else:\n with torch.no_grad():\n if self.feat_type == \"fbank\" or self.feat_type == \"mfcc\":\n x = self.feature_extract(x) + 1e-6 # B x feat_dim x time_len\n else:\n x = self.feature_extract([sample for sample in x])\n\n if self.feat_type == \"fbank\":\n x = x.log()\n\n if self.feat_type != \"fbank\" and self.feat_type != \"mfcc\":\n x = x[self.feature_selection]\n if isinstance(x, (list, tuple)):\n x = torch.stack(x, dim=0)\n else:\n x = x.unsqueeze(0)\n norm_weights = F.softmax(self.feature_weight, dim=-1).unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)\n x = (norm_weights * x).sum(dim=0)\n x = torch.transpose(x, 1, 2) + 1e-6\n\n x = self.instance_norm(x)\n return x\n\n def forward(self, x):\n x = self.get_feat(x)\n\n out1 = self.layer1(x)\n out2 = self.layer2(out1)\n out3 = self.layer3(out2)\n out4 = self.layer4(out3)\n\n out = torch.cat([out2, out3, out4], dim=1)\n out = F.relu(self.conv(out))\n out = self.bn(self.pooling(out))\n out = self.linear(out)\n\n return out\n\n\ndef ECAPA_TDNN_SMALL(\n feat_dim,\n emb_dim=256,","source_hash":"3d934824932b71e6ba3824c817367328c5f099335106dc03e6ba68a143da6a0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.eval_seedtts_testset","uri":"program://DMOSpeech2/module/src.f5_tts.eval.eval_seedtts_testset#L1-L88","kind":"module","name":"src.f5_tts.eval.eval_seedtts_testset","path":"src/f5_tts/eval/eval_seedtts_testset.py","language":"python","start_line":1,"end_line":88,"context_start_line":1,"context_end_line":88,"code":"# Evaluate with Seed-TTS testset\n\nimport argparse\nimport json\nimport os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport multiprocessing as mp\nfrom importlib.resources import files\n\nimport numpy as np\n\nfrom f5_tts.eval.utils_eval import get_seed_tts_test, run_asr_wer, run_sim\n\n\nrel_path = str(files(\"f5_tts\").joinpath(\"../../\"))\n\n\ndef get_args():\n parser = argparse.ArgumentParser()\n parser.add_argument(\"-e\", \"--eval_task\", type=str, default=\"wer\", choices=[\"sim\", \"wer\"])\n parser.add_argument(\"-l\", \"--lang\", type=str, default=\"en\", choices=[\"zh\", \"en\"])\n parser.add_argument(\"-g\", \"--gen_wav_dir\", type=str, required=True)\n parser.add_argument(\"-n\", \"--gpu_nums\", type=int, default=8, help=\"Number of GPUs to use\")\n parser.add_argument(\"--local\", action=\"store_true\", help=\"Use local custom checkpoint directory\")\n return parser.parse_args()\n\n\ndef main():\n args = get_args()\n eval_task = args.eval_task\n lang = args.lang\n gen_wav_dir = args.gen_wav_dir\n metalst = rel_path + f\"/data/seedtts_testset/{lang}/meta.lst\" # seed-tts testset\n\n # NOTE. paraformer-zh result will be slightly different according to the number of gpus, cuz batchsize is different\n # zh 1.254 seems a result of 4 workers wer_seed_tts\n gpus = list(range(args.gpu_nums))\n test_set = get_seed_tts_test(metalst, gen_wav_dir, gpus)\n\n local = args.local\n if local: # use local custom checkpoint dir\n if lang == \"zh\":\n asr_ckpt_dir = \"../checkpoints/funasr\" # paraformer-zh dir under funasr\n elif lang == \"en\":\n asr_ckpt_dir = \"../checkpoints/Systran/faster-whisper-large-v3\"\n else:\n asr_ckpt_dir = \"\" # auto download to cache dir\n wavlm_ckpt_dir = \"../checkpoints/UniSpeech/wavlm_large_finetune.pth\"\n\n # --------------------------------------------------------------------------\n\n full_results = []\n metrics = []\n\n if eval_task == \"wer\":\n with mp.Pool(processes=len(gpus)) as pool:\n args = [(rank, lang, sub_test_set, asr_ckpt_dir) for (rank, sub_test_set) in test_set]\n results = pool.map(run_asr_wer, args)\n for r in results:\n full_results.extend(r)\n elif eval_task == \"sim\":\n with mp.Pool(processes=len(gpus)) as pool:\n args = [(rank, sub_test_set, wavlm_ckpt_dir) for (rank, sub_test_set) in test_set]\n results = pool.map(run_sim, args)\n for r in results:\n full_results.extend(r)\n else:\n raise ValueError(f\"Unknown metric type: {eval_task}\")\n\n result_path = f\"{gen_wav_dir}/_{eval_task}_results.jsonl\"\n with open(result_path, \"w\") as f:\n for line in full_results:\n metrics.append(line[eval_task])\n f.write(json.dumps(line, ensure_ascii=False) + \"\\n\")\n metric = round(np.mean(metrics), 5)\n f.write(f\"\\n{eval_task.upper()}: {metric}\\n\")\n\n print(f\"\\nTotal {len(metrics)} samples\")\n print(f\"{eval_task.upper()}: {metric}\")\n print(f\"{eval_task.upper()} results saved to {result_path}\")\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"25fd3663207e4448631d3366eec65dae50c3d77def18467925fabd6db2b7a5b0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.eval_seedtts_testset.get_args","uri":"program://DMOSpeech2/function/src.f5_tts.eval.eval_seedtts_testset.get_args#L22-L29","kind":"function","name":"get_args","path":"src/f5_tts/eval/eval_seedtts_testset.py","language":"python","start_line":22,"end_line":29,"context_start_line":2,"context_end_line":49,"code":"\nimport argparse\nimport json\nimport os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport multiprocessing as mp\nfrom importlib.resources import files\n\nimport numpy as np\n\nfrom f5_tts.eval.utils_eval import get_seed_tts_test, run_asr_wer, run_sim\n\n\nrel_path = str(files(\"f5_tts\").joinpath(\"../../\"))\n\n\ndef get_args():\n parser = argparse.ArgumentParser()\n parser.add_argument(\"-e\", \"--eval_task\", type=str, default=\"wer\", choices=[\"sim\", \"wer\"])\n parser.add_argument(\"-l\", \"--lang\", type=str, default=\"en\", choices=[\"zh\", \"en\"])\n parser.add_argument(\"-g\", \"--gen_wav_dir\", type=str, required=True)\n parser.add_argument(\"-n\", \"--gpu_nums\", type=int, default=8, help=\"Number of GPUs to use\")\n parser.add_argument(\"--local\", action=\"store_true\", help=\"Use local custom checkpoint directory\")\n return parser.parse_args()\n\n\ndef main():\n args = get_args()\n eval_task = args.eval_task\n lang = args.lang\n gen_wav_dir = args.gen_wav_dir\n metalst = rel_path + f\"/data/seedtts_testset/{lang}/meta.lst\" # seed-tts testset\n\n # NOTE. paraformer-zh result will be slightly different according to the number of gpus, cuz batchsize is different\n # zh 1.254 seems a result of 4 workers wer_seed_tts\n gpus = list(range(args.gpu_nums))\n test_set = get_seed_tts_test(metalst, gen_wav_dir, gpus)\n\n local = args.local\n if local: # use local custom checkpoint dir\n if lang == \"zh\":\n asr_ckpt_dir = \"../checkpoints/funasr\" # paraformer-zh dir under funasr\n elif lang == \"en\":\n asr_ckpt_dir = \"../checkpoints/Systran/faster-whisper-large-v3\"","source_hash":"25fd3663207e4448631d3366eec65dae50c3d77def18467925fabd6db2b7a5b0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.eval_seedtts_testset.main","uri":"program://DMOSpeech2/function/src.f5_tts.eval.eval_seedtts_testset.main#L32-L84","kind":"function","name":"main","path":"src/f5_tts/eval/eval_seedtts_testset.py","language":"python","start_line":32,"end_line":84,"context_start_line":12,"context_end_line":88,"code":"from importlib.resources import files\n\nimport numpy as np\n\nfrom f5_tts.eval.utils_eval import get_seed_tts_test, run_asr_wer, run_sim\n\n\nrel_path = str(files(\"f5_tts\").joinpath(\"../../\"))\n\n\ndef get_args():\n parser = argparse.ArgumentParser()\n parser.add_argument(\"-e\", \"--eval_task\", type=str, default=\"wer\", choices=[\"sim\", \"wer\"])\n parser.add_argument(\"-l\", \"--lang\", type=str, default=\"en\", choices=[\"zh\", \"en\"])\n parser.add_argument(\"-g\", \"--gen_wav_dir\", type=str, required=True)\n parser.add_argument(\"-n\", \"--gpu_nums\", type=int, default=8, help=\"Number of GPUs to use\")\n parser.add_argument(\"--local\", action=\"store_true\", help=\"Use local custom checkpoint directory\")\n return parser.parse_args()\n\n\ndef main():\n args = get_args()\n eval_task = args.eval_task\n lang = args.lang\n gen_wav_dir = args.gen_wav_dir\n metalst = rel_path + f\"/data/seedtts_testset/{lang}/meta.lst\" # seed-tts testset\n\n # NOTE. paraformer-zh result will be slightly different according to the number of gpus, cuz batchsize is different\n # zh 1.254 seems a result of 4 workers wer_seed_tts\n gpus = list(range(args.gpu_nums))\n test_set = get_seed_tts_test(metalst, gen_wav_dir, gpus)\n\n local = args.local\n if local: # use local custom checkpoint dir\n if lang == \"zh\":\n asr_ckpt_dir = \"../checkpoints/funasr\" # paraformer-zh dir under funasr\n elif lang == \"en\":\n asr_ckpt_dir = \"../checkpoints/Systran/faster-whisper-large-v3\"\n else:\n asr_ckpt_dir = \"\" # auto download to cache dir\n wavlm_ckpt_dir = \"../checkpoints/UniSpeech/wavlm_large_finetune.pth\"\n\n # --------------------------------------------------------------------------\n\n full_results = []\n metrics = []\n\n if eval_task == \"wer\":\n with mp.Pool(processes=len(gpus)) as pool:\n args = [(rank, lang, sub_test_set, asr_ckpt_dir) for (rank, sub_test_set) in test_set]\n results = pool.map(run_asr_wer, args)\n for r in results:\n full_results.extend(r)\n elif eval_task == \"sim\":\n with mp.Pool(processes=len(gpus)) as pool:\n args = [(rank, sub_test_set, wavlm_ckpt_dir) for (rank, sub_test_set) in test_set]\n results = pool.map(run_sim, args)\n for r in results:\n full_results.extend(r)\n else:\n raise ValueError(f\"Unknown metric type: {eval_task}\")\n\n result_path = f\"{gen_wav_dir}/_{eval_task}_results.jsonl\"\n with open(result_path, \"w\") as f:\n for line in full_results:\n metrics.append(line[eval_task])\n f.write(json.dumps(line, ensure_ascii=False) + \"\\n\")\n metric = round(np.mean(metrics), 5)\n f.write(f\"\\n{eval_task.upper()}: {metric}\\n\")\n\n print(f\"\\nTotal {len(metrics)} samples\")\n print(f\"{eval_task.upper()}: {metric}\")\n print(f\"{eval_task.upper()} results saved to {result_path}\")\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"25fd3663207e4448631d3366eec65dae50c3d77def18467925fabd6db2b7a5b0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.eval_librispeech_test_clean","uri":"program://DMOSpeech2/module/src.f5_tts.eval.eval_librispeech_test_clean#L1-L89","kind":"module","name":"src.f5_tts.eval.eval_librispeech_test_clean","path":"src/f5_tts/eval/eval_librispeech_test_clean.py","language":"python","start_line":1,"end_line":89,"context_start_line":1,"context_end_line":89,"code":"# Evaluate with Librispeech test-clean, ~3s prompt to generate 4-10s audio (the way of valle/voicebox evaluation)\n\nimport argparse\nimport json\nimport os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport multiprocessing as mp\nfrom importlib.resources import files\n\nimport numpy as np\n\nfrom f5_tts.eval.utils_eval import get_librispeech_test, run_asr_wer, run_sim\n\n\nrel_path = str(files(\"f5_tts\").joinpath(\"../../\"))\n\n\ndef get_args():\n parser = argparse.ArgumentParser()\n parser.add_argument(\"-e\", \"--eval_task\", type=str, default=\"wer\", choices=[\"sim\", \"wer\"])\n parser.add_argument(\"-l\", \"--lang\", type=str, default=\"en\")\n parser.add_argument(\"-g\", \"--gen_wav_dir\", type=str, required=True)\n parser.add_argument(\"-p\", \"--librispeech_test_clean_path\", type=str, required=True)\n parser.add_argument(\"-n\", \"--gpu_nums\", type=int, default=8, help=\"Number of GPUs to use\")\n parser.add_argument(\"--local\", action=\"store_true\", help=\"Use local custom checkpoint directory\")\n return parser.parse_args()\n\n\ndef main():\n args = get_args()\n eval_task = args.eval_task\n lang = args.lang\n librispeech_test_clean_path = args.librispeech_test_clean_path # test-clean path\n gen_wav_dir = args.gen_wav_dir\n metalst = rel_path + \"/data/librispeech_pc_test_clean_cross_sentence.lst\"\n\n gpus = list(range(args.gpu_nums))\n test_set = get_librispeech_test(metalst, gen_wav_dir, gpus, librispeech_test_clean_path)\n\n ## In LibriSpeech, some speakers utilized varying voice characteristics for different characters in the book,\n ## leading to a low similarity for the ground truth in some cases.\n # test_set = get_librispeech_test(metalst, gen_wav_dir, gpus, librispeech_test_clean_path, eval_ground_truth = True) # eval ground truth\n\n local = args.local\n if local: # use local custom checkpoint dir\n asr_ckpt_dir = \"../checkpoints/Systran/faster-whisper-large-v3\"\n else:\n asr_ckpt_dir = \"\" # auto download to cache dir\n wavlm_ckpt_dir = \"../checkpoints/UniSpeech/wavlm_large_finetune.pth\"\n\n # --------------------------------------------------------------------------\n\n full_results = []\n metrics = []\n\n if eval_task == \"wer\":\n with mp.Pool(processes=len(gpus)) as pool:\n args = [(rank, lang, sub_test_set, asr_ckpt_dir) for (rank, sub_test_set) in test_set]\n results = pool.map(run_asr_wer, args)\n for r in results:\n full_results.extend(r)\n elif eval_task == \"sim\":\n with mp.Pool(processes=len(gpus)) as pool:\n args = [(rank, sub_test_set, wavlm_ckpt_dir) for (rank, sub_test_set) in test_set]\n results = pool.map(run_sim, args)\n for r in results:\n full_results.extend(r)\n else:\n raise ValueError(f\"Unknown metric type: {eval_task}\")\n\n result_path = f\"{gen_wav_dir}/_{eval_task}_results.jsonl\"\n with open(result_path, \"w\") as f:\n for line in full_results:\n metrics.append(line[eval_task])\n f.write(json.dumps(line, ensure_ascii=False) + \"\\n\")\n metric = round(np.mean(metrics), 5)\n f.write(f\"\\n{eval_task.upper()}: {metric}\\n\")\n\n print(f\"\\nTotal {len(metrics)} samples\")\n print(f\"{eval_task.upper()}: {metric}\")\n print(f\"{eval_task.upper()} results saved to {result_path}\")\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"9f5b078ca2dbd7ba9c1f0cbfa70a2ce84f990a8e2db8669ad6aef0af0c4f7199","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.eval_librispeech_test_clean.get_args","uri":"program://DMOSpeech2/function/src.f5_tts.eval.eval_librispeech_test_clean.get_args#L22-L30","kind":"function","name":"get_args","path":"src/f5_tts/eval/eval_librispeech_test_clean.py","language":"python","start_line":22,"end_line":30,"context_start_line":2,"context_end_line":50,"code":"\nimport argparse\nimport json\nimport os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport multiprocessing as mp\nfrom importlib.resources import files\n\nimport numpy as np\n\nfrom f5_tts.eval.utils_eval import get_librispeech_test, run_asr_wer, run_sim\n\n\nrel_path = str(files(\"f5_tts\").joinpath(\"../../\"))\n\n\ndef get_args():\n parser = argparse.ArgumentParser()\n parser.add_argument(\"-e\", \"--eval_task\", type=str, default=\"wer\", choices=[\"sim\", \"wer\"])\n parser.add_argument(\"-l\", \"--lang\", type=str, default=\"en\")\n parser.add_argument(\"-g\", \"--gen_wav_dir\", type=str, required=True)\n parser.add_argument(\"-p\", \"--librispeech_test_clean_path\", type=str, required=True)\n parser.add_argument(\"-n\", \"--gpu_nums\", type=int, default=8, help=\"Number of GPUs to use\")\n parser.add_argument(\"--local\", action=\"store_true\", help=\"Use local custom checkpoint directory\")\n return parser.parse_args()\n\n\ndef main():\n args = get_args()\n eval_task = args.eval_task\n lang = args.lang\n librispeech_test_clean_path = args.librispeech_test_clean_path # test-clean path\n gen_wav_dir = args.gen_wav_dir\n metalst = rel_path + \"/data/librispeech_pc_test_clean_cross_sentence.lst\"\n\n gpus = list(range(args.gpu_nums))\n test_set = get_librispeech_test(metalst, gen_wav_dir, gpus, librispeech_test_clean_path)\n\n ## In LibriSpeech, some speakers utilized varying voice characteristics for different characters in the book,\n ## leading to a low similarity for the ground truth in some cases.\n # test_set = get_librispeech_test(metalst, gen_wav_dir, gpus, librispeech_test_clean_path, eval_ground_truth = True) # eval ground truth\n\n local = args.local\n if local: # use local custom checkpoint dir\n asr_ckpt_dir = \"../checkpoints/Systran/faster-whisper-large-v3\"","source_hash":"9f5b078ca2dbd7ba9c1f0cbfa70a2ce84f990a8e2db8669ad6aef0af0c4f7199","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.eval_librispeech_test_clean.main","uri":"program://DMOSpeech2/function/src.f5_tts.eval.eval_librispeech_test_clean.main#L33-L85","kind":"function","name":"main","path":"src/f5_tts/eval/eval_librispeech_test_clean.py","language":"python","start_line":33,"end_line":85,"context_start_line":13,"context_end_line":89,"code":"\nimport numpy as np\n\nfrom f5_tts.eval.utils_eval import get_librispeech_test, run_asr_wer, run_sim\n\n\nrel_path = str(files(\"f5_tts\").joinpath(\"../../\"))\n\n\ndef get_args():\n parser = argparse.ArgumentParser()\n parser.add_argument(\"-e\", \"--eval_task\", type=str, default=\"wer\", choices=[\"sim\", \"wer\"])\n parser.add_argument(\"-l\", \"--lang\", type=str, default=\"en\")\n parser.add_argument(\"-g\", \"--gen_wav_dir\", type=str, required=True)\n parser.add_argument(\"-p\", \"--librispeech_test_clean_path\", type=str, required=True)\n parser.add_argument(\"-n\", \"--gpu_nums\", type=int, default=8, help=\"Number of GPUs to use\")\n parser.add_argument(\"--local\", action=\"store_true\", help=\"Use local custom checkpoint directory\")\n return parser.parse_args()\n\n\ndef main():\n args = get_args()\n eval_task = args.eval_task\n lang = args.lang\n librispeech_test_clean_path = args.librispeech_test_clean_path # test-clean path\n gen_wav_dir = args.gen_wav_dir\n metalst = rel_path + \"/data/librispeech_pc_test_clean_cross_sentence.lst\"\n\n gpus = list(range(args.gpu_nums))\n test_set = get_librispeech_test(metalst, gen_wav_dir, gpus, librispeech_test_clean_path)\n\n ## In LibriSpeech, some speakers utilized varying voice characteristics for different characters in the book,\n ## leading to a low similarity for the ground truth in some cases.\n # test_set = get_librispeech_test(metalst, gen_wav_dir, gpus, librispeech_test_clean_path, eval_ground_truth = True) # eval ground truth\n\n local = args.local\n if local: # use local custom checkpoint dir\n asr_ckpt_dir = \"../checkpoints/Systran/faster-whisper-large-v3\"\n else:\n asr_ckpt_dir = \"\" # auto download to cache dir\n wavlm_ckpt_dir = \"../checkpoints/UniSpeech/wavlm_large_finetune.pth\"\n\n # --------------------------------------------------------------------------\n\n full_results = []\n metrics = []\n\n if eval_task == \"wer\":\n with mp.Pool(processes=len(gpus)) as pool:\n args = [(rank, lang, sub_test_set, asr_ckpt_dir) for (rank, sub_test_set) in test_set]\n results = pool.map(run_asr_wer, args)\n for r in results:\n full_results.extend(r)\n elif eval_task == \"sim\":\n with mp.Pool(processes=len(gpus)) as pool:\n args = [(rank, sub_test_set, wavlm_ckpt_dir) for (rank, sub_test_set) in test_set]\n results = pool.map(run_sim, args)\n for r in results:\n full_results.extend(r)\n else:\n raise ValueError(f\"Unknown metric type: {eval_task}\")\n\n result_path = f\"{gen_wav_dir}/_{eval_task}_results.jsonl\"\n with open(result_path, \"w\") as f:\n for line in full_results:\n metrics.append(line[eval_task])\n f.write(json.dumps(line, ensure_ascii=False) + \"\\n\")\n metric = round(np.mean(metrics), 5)\n f.write(f\"\\n{eval_task.upper()}: {metric}\\n\")\n\n print(f\"\\nTotal {len(metrics)} samples\")\n print(f\"{eval_task.upper()}: {metric}\")\n print(f\"{eval_task.upper()} results saved to {result_path}\")\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"9f5b078ca2dbd7ba9c1f0cbfa70a2ce84f990a8e2db8669ad6aef0af0c4f7199","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.eval_utmos","uri":"program://DMOSpeech2/module/src.f5_tts.eval.eval_utmos#L1-L42","kind":"module","name":"src.f5_tts.eval.eval_utmos","path":"src/f5_tts/eval/eval_utmos.py","language":"python","start_line":1,"end_line":42,"context_start_line":1,"context_end_line":42,"code":"import argparse\nimport json\nfrom pathlib import Path\n\nimport librosa\nimport torch\nfrom tqdm import tqdm\n\n\ndef main():\n parser = argparse.ArgumentParser(description=\"UTMOS Evaluation\")\n parser.add_argument(\"--audio_dir\", type=str, required=True, help=\"Audio file path.\")\n parser.add_argument(\"--ext\", type=str, default=\"wav\", help=\"Audio extension.\")\n args = parser.parse_args()\n\n device = \"cuda\" if torch.cuda.is_available() else \"xpu\" if torch.xpu.is_available() else \"cpu\"\n\n predictor = torch.hub.load(\"tarepan/SpeechMOS:v1.2.0\", \"utmos22_strong\", trust_repo=True)\n predictor = predictor.to(device)\n\n audio_paths = list(Path(args.audio_dir).rglob(f\"*.{args.ext}\"))\n utmos_score = 0\n\n utmos_result_path = Path(args.audio_dir) / \"_utmos_results.jsonl\"\n with open(utmos_result_path, \"w\", encoding=\"utf-8\") as f:\n for audio_path in tqdm(audio_paths, desc=\"Processing\"):\n wav, sr = librosa.load(audio_path, sr=None, mono=True)\n wav_tensor = torch.from_numpy(wav).to(device).unsqueeze(0)\n score = predictor(wav_tensor, sr)\n line = {}\n line[\"wav\"], line[\"utmos\"] = str(audio_path.stem), score.item()\n utmos_score += score.item()\n f.write(json.dumps(line, ensure_ascii=False) + \"\\n\")\n avg_score = utmos_score / len(audio_paths) if len(audio_paths) > 0 else 0\n f.write(f\"\\nUTMOS: {avg_score:.4f}\\n\")\n\n print(f\"UTMOS: {avg_score:.4f}\")\n print(f\"UTMOS results saved to {utmos_result_path}\")\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"852c55bf0b26006faf1ab2a3f74695dd071058944723bdde2b378f85e3a5b6df","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.eval.eval_utmos.main","uri":"program://DMOSpeech2/function/src.f5_tts.eval.eval_utmos.main#L10-L38","kind":"function","name":"main","path":"src/f5_tts/eval/eval_utmos.py","language":"python","start_line":10,"end_line":38,"context_start_line":1,"context_end_line":42,"code":"import argparse\nimport json\nfrom pathlib import Path\n\nimport librosa\nimport torch\nfrom tqdm import tqdm\n\n\ndef main():\n parser = argparse.ArgumentParser(description=\"UTMOS Evaluation\")\n parser.add_argument(\"--audio_dir\", type=str, required=True, help=\"Audio file path.\")\n parser.add_argument(\"--ext\", type=str, default=\"wav\", help=\"Audio extension.\")\n args = parser.parse_args()\n\n device = \"cuda\" if torch.cuda.is_available() else \"xpu\" if torch.xpu.is_available() else \"cpu\"\n\n predictor = torch.hub.load(\"tarepan/SpeechMOS:v1.2.0\", \"utmos22_strong\", trust_repo=True)\n predictor = predictor.to(device)\n\n audio_paths = list(Path(args.audio_dir).rglob(f\"*.{args.ext}\"))\n utmos_score = 0\n\n utmos_result_path = Path(args.audio_dir) / \"_utmos_results.jsonl\"\n with open(utmos_result_path, \"w\", encoding=\"utf-8\") as f:\n for audio_path in tqdm(audio_paths, desc=\"Processing\"):\n wav, sr = librosa.load(audio_path, sr=None, mono=True)\n wav_tensor = torch.from_numpy(wav).to(device).unsqueeze(0)\n score = predictor(wav_tensor, sr)\n line = {}\n line[\"wav\"], line[\"utmos\"] = str(audio_path.stem), score.item()\n utmos_score += score.item()\n f.write(json.dumps(line, ensure_ascii=False) + \"\\n\")\n avg_score = utmos_score / len(audio_paths) if len(audio_paths) > 0 else 0\n f.write(f\"\\nUTMOS: {avg_score:.4f}\\n\")\n\n print(f\"UTMOS: {avg_score:.4f}\")\n print(f\"UTMOS results saved to {utmos_result_path}\")\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"852c55bf0b26006faf1ab2a3f74695dd071058944723bdde2b378f85e3a5b6df","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark","uri":"program://DMOSpeech2/module/src.f5_tts.runtime.triton_trtllm.benchmark#L1-L560","kind":"module","name":"src.f5_tts.runtime.triton_trtllm.benchmark","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":1,"end_line":560,"context_start_line":1,"context_end_line":560,"code":"# Copyright (c) 2024 Tsinghua Univ. (authors: Xingchen Song)\n# 2025 （authors: Yuekai Zhang）\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# Modified from https://github.com/xingchensong/S3Tokenizer/blob/main/s3tokenizer/cli.py\n\"\"\" Example Usage\ntorchrun --nproc_per_node=1 \\\nbenchmark.py --output-dir $log_dir \\\n--batch-size $batch_size \\\n--enable-warmup \\\n--split-name $split_name \\\n--model-path $F5_TTS_HF_DOWNLOAD_PATH/$model/model_1200000.pt \\\n--vocab-file $F5_TTS_HF_DOWNLOAD_PATH/$model/vocab.txt \\\n--vocoder-trt-engine-path $vocoder_trt_engine_path \\\n--backend-type $backend_type \\\n--tllm-model-dir $F5_TTS_TRT_LLM_ENGINE_PATH || exit 1\n\"\"\"\n\nimport argparse\nimport json\nimport os\nimport time\nfrom typing import Dict, List, Union\n\nimport datasets\nimport jieba\nimport tensorrt as trt\nimport torch\nimport torch.distributed as dist\nimport torch.nn.functional as F\nimport torchaudio\nfrom datasets import load_dataset\nfrom f5_tts_trtllm import F5TTS\nfrom huggingface_hub import hf_hub_download\nfrom pypinyin import Style, lazy_pinyin\nfrom tensorrt_llm._utils import trt_dtype_to_torch\nfrom tensorrt_llm.logger import logger\nfrom tensorrt_llm.runtime.session import Session, TensorInfo\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torch.utils.data import DataLoader, DistributedSampler\nfrom tqdm import tqdm\nfrom vocos import Vocos\n\n\ntorch.manual_seed(0)\n\n\ndef get_args():\n parser = argparse.ArgumentParser(description=\"extract speech code\")\n parser.add_argument(\n \"--split-name\",\n type=str,\n default=\"wenetspeech4tts\",\n choices=[\"wenetspeech4tts\", \"test_zh\", \"test_en\", \"test_hard\"],\n help=\"huggingface dataset split name\",\n )\n parser.add_argument(\"--output-dir\", required=True, type=str, help=\"dir to save result\")\n parser.add_argument(\n \"--vocab-file\",\n required=True,\n type=str,\n help=\"vocab file\",\n )\n parser.add_argument(\n \"--model-path\",\n required=True,\n type=str,\n help=\"model path, to load text embedding\",\n )\n parser.add_argument(\n \"--tllm-model-dir\",\n required=True,\n type=str,\n help=\"tllm model dir\",\n )\n parser.add_argument(\n \"--batch-size\",\n required=True,\n type=int,\n help=\"batch size (per-device) for inference\",\n )\n parser.add_argument(\"--num-workers\", type=int, default=0, help=\"workers for dataloader\")\n parser.add_argument(\"--prefetch\", type=int, default=None, help=\"prefetch for dataloader\")\n parser.add_argument(\n \"--vocoder\",\n default=\"vocos\",\n type=str,\n help=\"vocoder name\",\n )\n parser.add_argument(\n \"--vocoder-trt-engine-path\",\n default=None,\n type=str,\n help=\"vocoder trt engine path\",\n )\n parser.add_argument(\"--enable-warmup\", action=\"store_true\")\n parser.add_argument(\"--remove-input-padding\", action=\"store_true\")\n parser.add_argument(\"--use-perf\", action=\"store_true\", help=\"use nvtx to record performance\")\n parser.add_argument(\"--backend-type\", type=str, default=\"triton\", choices=[\"trt\", \"pytorch\"], help=\"backend type\")\n args = parser.parse_args()\n return args\n\n\ndef padded_mel_batch(ref_mels, max_seq_len):\n padded_ref_mels = []\n for mel in ref_mels:\n # pad along the last dimension\n padded_ref_mel = F.pad(mel, (0, 0, 0, max_seq_len - mel.shape[0]), value=0)\n padded_ref_mels.append(padded_ref_mel)\n padded_ref_mels = torch.stack(padded_ref_mels)\n return padded_ref_mels\n\n\ndef data_collator(batch, vocab_char_map, device=\"cuda\", use_perf=False):\n if use_perf:\n torch.cuda.nvtx.range_push(\"data_collator\")\n target_sample_rate = 24000\n target_rms = 0.1\n ids, ref_mel_list, ref_mel_len_list, estimated_reference_target_mel_len, reference_target_texts_list = (\n [],\n [],\n [],\n [],\n [],\n )\n for i, item in enumerate(batch):\n item_id, prompt_text, target_text = (\n item[\"id\"],\n item[\"prompt_text\"],\n item[\"target_text\"],\n )\n ids.append(item_id)\n reference_target_texts_list.append(prompt_text + target_text)\n\n ref_audio_org, ref_sr = (\n item[\"prompt_audio\"][\"array\"],\n item[\"prompt_audio\"][\"sampling_rate\"],\n )\n ref_audio_org = torch.from_numpy(ref_audio_org).unsqueeze(0).float()\n ref_rms = torch.sqrt(torch.mean(torch.square(ref_audio_org)))\n if ref_rms < target_rms:\n ref_audio_org = ref_audio_org * target_rms / ref_rms\n\n if ref_sr != target_sample_rate:\n resampler = torchaudio.transforms.Resample(ref_sr, target_sample_rate)\n ref_audio = resampler(ref_audio_org)\n else:\n ref_audio = ref_audio_org\n\n if use_perf:\n torch.cuda.nvtx.range_push(f\"mel_spectrogram {i}\")\n ref_mel = mel_spectrogram(ref_audio, vocoder=\"vocos\", device=\"cuda\")\n if use_perf:\n torch.cuda.nvtx.range_pop()\n ref_mel = ref_mel.squeeze()\n ref_mel_len = ref_mel.shape[0]\n assert ref_mel.shape[1] == 100\n\n ref_mel_list.append(ref_mel)\n ref_mel_len_list.append(ref_mel_len)\n\n estimated_reference_target_mel_len.append(\n int(ref_mel.shape[0] * (1 + len(target_text.encode(\"utf-8\")) / len(prompt_text.encode(\"utf-8\"))))\n )\n\n max_seq_len = max(estimated_reference_target_mel_len)\n ref_mel_batch = padded_mel_batch(ref_mel_list, max_seq_len)\n ref_mel_len_batch = torch.LongTensor(ref_mel_len_list)\n\n pinyin_list = convert_char_to_pinyin(reference_target_texts_list, polyphone=True)\n text_pad_sequence = list_str_to_idx(pinyin_list, vocab_char_map)\n\n for i, item in enumerate(text_pad_sequence):\n text_pad_sequence[i] = F.pad(\n item, (0, estimated_reference_target_mel_len[i] - len(item)), mode=\"constant\", value=-1\n )\n text_pad_sequence[i] += 1 # WAR: 0 is reserved for padding token, hard coding in F5-TTS\n text_pad_sequence = pad_sequence(text_pad_sequence, padding_value=-1, batch_first=True).to(device)\n text_pad_sequence = F.pad(\n text_pad_sequence, (0, max_seq_len - text_pad_sequence.shape[1]), mode=\"constant\", value=-1\n )\n if use_perf:\n torch.cuda.nvtx.range_pop()\n return {\n \"ids\": ids,\n \"ref_mel_batch\": ref_mel_batch,\n \"ref_mel_len_batch\": ref_mel_len_batch,\n \"text_pad_sequence\": text_pad_sequence,\n \"estimated_reference_target_mel_len\": estimated_reference_target_mel_len,\n }\n\n\ndef init_distributed():\n world_size = int(os.environ.get(\"WORLD_SIZE\", 1))\n local_rank = int(os.environ.get(\"LOCAL_RANK\", 0))\n rank = int(os.environ.get(\"RANK\", 0))\n print(\n \"Inference on multiple gpus, this gpu {}\".format(local_rank)\n + \", rank {}, world_size {}\".format(rank, world_size)\n )\n torch.cuda.set_device(local_rank)\n # Initialize process group with explicit device IDs\n dist.init_process_group(\n \"nccl\",\n )\n return world_size, local_rank, rank\n\n\ndef get_tokenizer(vocab_file_path: str):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file\n - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n with open(vocab_file_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n return vocab_char_map, vocab_size\n\n\ndef convert_char_to_pinyin(reference_target_texts_list, polyphone=True):\n final_reference_target_texts_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n\n for text in reference_target_texts_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)\n elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)\n elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_reference_target_texts_list.append(char_list)\n\n return final_reference_target_texts_list\n\n\ndef list_str_to_idx(\n text: Union[List[str], List[List[str]]],\n vocab_char_map: Dict[str, int], # {char: idx}\n padding_value=-1,\n):\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n # text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return list_idx_tensors\n\n\ndef load_vocoder(\n vocoder_name=\"vocos\", is_local=False, local_path=\"\", device=\"cuda\", hf_cache_dir=None, vocoder_trt_engine_path=None\n):\n if vocoder_name == \"vocos\":\n if vocoder_trt_engine_path is not None:\n vocoder = VocosTensorRT(engine_path=vocoder_trt_engine_path)\n else:\n # vocoder = Vocos.from_pretrained(\"charactr/vocos-mel-24khz\").to(device)\n if is_local:\n print(f\"Load vocos from local path {local_path}\")\n config_path = f\"{local_path}/config.yaml\"\n model_path = f\"{local_path}/pytorch_model.bin\"\n else:\n print(\"Download Vocos from huggingface charactr/vocos-mel-24khz\")\n repo_id = \"charactr/vocos-mel-24khz\"\n config_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"config.yaml\")\n model_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"pytorch_model.bin\")\n vocoder = Vocos.from_hparams(config_path)\n state_dict = torch.load(model_path, map_location=\"cpu\", weights_only=True)\n from vocos.feature_extractors import EncodecFeatures\n\n if isinstance(vocoder.feature_extractor, EncodecFeatures):\n encodec_parameters = {\n \"feature_extractor.encodec.\" + key: value\n for key, value in vocoder.feature_extractor.encodec.state_dict().items()\n }\n state_dict.update(encodec_parameters)\n vocoder.load_state_dict(state_dict)\n vocoder = vocoder.eval().to(device)\n elif vocoder_name == \"bigvgan\":\n raise NotImplementedError(\"BigVGAN is not implemented yet\")\n return vocoder\n\n\ndef mel_spectrogram(waveform, vocoder=\"vocos\", device=\"cuda\"):\n if vocoder == \"vocos\":\n mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=24000,\n n_fft=1024,\n win_length=1024,\n hop_length=256,\n n_mels=100,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(device)\n mel = mel_stft(waveform.to(device))\n mel = mel.clamp(min=1e-5).log()\n return mel.transpose(1, 2)\n\n\nclass VocosTensorRT:\n def __init__(self, engine_path=\"./vocos_vocoder.plan\", stream=None):\n TRT_LOGGER = trt.Logger(trt.Logger.WARNING)\n trt.init_libnvinfer_plugins(TRT_LOGGER, namespace=\"\")\n logger.info(f\"Loading vae engine from {engine_path}\")\n self.engine_path = engine_path\n with open(engine_path, \"rb\") as f:\n engine_buffer = f.read()\n self.session = Session.from_serialized_engine(engine_buffer)\n self.stream = stream if stream is not None else torch.cuda.current_stream().cuda_stream\n\n def decode(self, mels):\n mels = mels.contiguous()\n inputs = {\"mel\": mels}\n output_info = self.session.infer_shapes([TensorInfo(\"mel\", trt.DataType.FLOAT, mels.shape)])\n outputs = {\n t.name: torch.empty(tuple(t.shape), dtype=trt_dtype_to_torch(t.dtype), device=\"cuda\") for t in output_info\n }\n ok = self.session.run(inputs, outputs, self.stream)\n\n assert ok, \"Runtime execution failed for vae session\"\n\n samples = outputs[\"waveform\"]\n return samples\n\n\ndef main():\n args = get_args()\n os.makedirs(args.output_dir, exist_ok=True)\n\n assert torch.cuda.is_available()\n world_size, local_rank, rank = init_distributed()\n device = torch.device(f\"cuda:{local_rank}\")\n\n vocab_char_map, vocab_size = get_tokenizer(args.vocab_file)\n\n tllm_model_dir = args.tllm_model_dir\n config_file = os.path.join(tllm_model_dir, \"config.json\")\n with open(config_file) as f:\n config = json.load(f)\n if args.backend_type == \"trt\":\n model = F5TTS(\n config, debug_mode=False, tllm_model_dir=tllm_model_dir, model_path=args.model_path, vocab_size=vocab_size\n )\n elif args.backend_type == \"pytorch\":\n import sys\n\n sys.path.append(f\"{os.path.dirname(os.path.abspath(__file__))}/../../../../src/\")\n from f5_tts.infer.utils_infer import load_model\n from f5_tts.model import DiT\n\n F5TTS_model_cfg = dict(\n dim=1024,\n depth=22,\n heads=16,\n ff_mult=2,\n text_dim=512,\n conv_layers=4,\n pe_attn_head=1,\n text_mask_padding=False,\n )\n model = load_model(DiT, F5TTS_model_cfg, args.model_path)\n\n vocoder = load_vocoder(\n vocoder_name=args.vocoder, device=device, vocoder_trt_engine_path=args.vocoder_trt_engine_path\n )\n\n dataset = load_dataset(\n \"yuekai/seed_tts\",\n split=args.split_name,\n trust_remote_code=True,\n )\n\n def add_estimated_duration(example):\n prompt_audio_len = example[\"prompt_audio\"][\"array\"].shape[0]\n scale_factor = 1 + len(example[\"target_text\"]) / len(example[\"prompt_text\"])\n estimated_duration = prompt_audio_len * scale_factor\n example[\"estimated_duration\"] = estimated_duration / example[\"prompt_audio\"][\"sampling_rate\"]\n return example\n\n dataset = dataset.map(add_estimated_duration)\n dataset = dataset.sort(\"estimated_duration\", reverse=True)\n if args.use_perf:\n # dataset_list = [dataset.select(range(1)) for i in range(16)] # seq_len 1000\n dataset_list_short = [dataset.select([24]) for i in range(8)] # seq_len 719\n # dataset_list_long = [dataset.select([23]) for i in range(8)] # seq_len 2002\n # dataset = datasets.concatenate_datasets(dataset_list_short + dataset_list_long)\n dataset = datasets.concatenate_datasets(dataset_list_short)\n if world_size > 1:\n sampler = DistributedSampler(dataset, num_replicas=world_size, rank=rank)\n else:\n # This would disable shuffling\n sampler = None\n\n dataloader = DataLoader(\n dataset,\n batch_size=args.batch_size,\n sampler=sampler,\n shuffle=False,\n num_workers=args.num_workers,\n prefetch_factor=args.prefetch,\n collate_fn=lambda x: data_collator(x, vocab_char_map, use_perf=args.use_perf),\n )\n\n total_steps = len(dataset)\n\n if args.enable_warmup:\n for batch in dataloader:\n ref_mels, ref_mel_lens = batch[\"ref_mel_batch\"].to(device), batch[\"ref_mel_len_batch\"].to(device)\n text_pad_seq = batch[\"text_pad_sequence\"].to(device)\n total_mel_lens = batch[\"estimated_reference_target_mel_len\"]\n if args.backend_type == \"trt\":\n _ = model.sample(\n text_pad_seq, ref_mels, ref_mel_lens, total_mel_lens, remove_input_padding=args.remove_input_padding\n )\n elif args.backend_type == \"pytorch\":\n with torch.inference_mode():\n text_pad_seq -= 1\n text_pad_seq[text_pad_seq == -2] = -1\n total_mel_lens = torch.tensor(total_mel_lens, device=device)\n generated, _ = model.sample(\n cond=ref_mels,\n text=text_pad_seq,\n duration=total_mel_lens,\n steps=16,\n cfg_strength=2.0,\n sway_sampling_coef=-1,\n )\n\n if rank == 0:\n progress_bar = tqdm(total=total_steps, desc=\"Processing\", unit=\"wavs\")\n\n decoding_time = 0\n vocoder_time = 0\n total_duration = 0\n if args.use_perf:\n torch.cuda.cudart().cudaProfilerStart()\n total_decoding_time = time.time()\n for batch in dataloader:\n if args.use_perf:\n torch.cuda.nvtx.range_push(\"data sample\")\n ref_mels, ref_mel_lens = batch[\"ref_mel_batch\"].to(device), batch[\"ref_mel_len_batch\"].to(device)\n text_pad_seq = batch[\"text_pad_sequence\"].to(device)\n total_mel_lens = batch[\"estimated_reference_target_mel_len\"]\n\n if args.use_perf:\n torch.cuda.nvtx.range_pop()\n if args.backend_type == \"trt\":\n generated, cost_time = model.sample(\n text_pad_seq,\n ref_mels,\n ref_mel_lens,\n total_mel_lens,\n remove_input_padding=args.remove_input_padding,\n use_perf=args.use_perf,\n )\n elif args.backend_type == \"pytorch\":\n total_mel_lens = torch.tensor(total_mel_lens, device=device)\n with torch.inference_mode():\n start_time = time.time()\n text_pad_seq -= 1\n text_pad_seq[text_pad_seq == -2] = -1\n generated, _ = model.sample(\n cond=ref_mels,\n text=text_pad_seq,\n duration=total_mel_lens,\n lens=ref_mel_lens,\n steps=16,\n cfg_strength=2.0,\n sway_sampling_coef=-1,\n )\n cost_time = time.time() - start_time\n decoding_time += cost_time\n vocoder_start_time = time.time()\n for i, gen in enumerate(generated):\n gen = gen[ref_mel_lens[i] : total_mel_lens[i], :].unsqueeze(0)\n gen_mel_spec = gen.permute(0, 2, 1).to(torch.float32)\n if args.vocoder == \"vocos\":\n if args.use_perf:\n torch.cuda.nvtx.range_push(\"vocoder decode\")\n gen\n# ... truncated ...","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.get_args","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.get_args#L58-L111","kind":"function","name":"get_args","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":58,"end_line":111,"context_start_line":38,"context_end_line":131,"code":"import torch\nimport torch.distributed as dist\nimport torch.nn.functional as F\nimport torchaudio\nfrom datasets import load_dataset\nfrom f5_tts_trtllm import F5TTS\nfrom huggingface_hub import hf_hub_download\nfrom pypinyin import Style, lazy_pinyin\nfrom tensorrt_llm._utils import trt_dtype_to_torch\nfrom tensorrt_llm.logger import logger\nfrom tensorrt_llm.runtime.session import Session, TensorInfo\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torch.utils.data import DataLoader, DistributedSampler\nfrom tqdm import tqdm\nfrom vocos import Vocos\n\n\ntorch.manual_seed(0)\n\n\ndef get_args():\n parser = argparse.ArgumentParser(description=\"extract speech code\")\n parser.add_argument(\n \"--split-name\",\n type=str,\n default=\"wenetspeech4tts\",\n choices=[\"wenetspeech4tts\", \"test_zh\", \"test_en\", \"test_hard\"],\n help=\"huggingface dataset split name\",\n )\n parser.add_argument(\"--output-dir\", required=True, type=str, help=\"dir to save result\")\n parser.add_argument(\n \"--vocab-file\",\n required=True,\n type=str,\n help=\"vocab file\",\n )\n parser.add_argument(\n \"--model-path\",\n required=True,\n type=str,\n help=\"model path, to load text embedding\",\n )\n parser.add_argument(\n \"--tllm-model-dir\",\n required=True,\n type=str,\n help=\"tllm model dir\",\n )\n parser.add_argument(\n \"--batch-size\",\n required=True,\n type=int,\n help=\"batch size (per-device) for inference\",\n )\n parser.add_argument(\"--num-workers\", type=int, default=0, help=\"workers for dataloader\")\n parser.add_argument(\"--prefetch\", type=int, default=None, help=\"prefetch for dataloader\")\n parser.add_argument(\n \"--vocoder\",\n default=\"vocos\",\n type=str,\n help=\"vocoder name\",\n )\n parser.add_argument(\n \"--vocoder-trt-engine-path\",\n default=None,\n type=str,\n help=\"vocoder trt engine path\",\n )\n parser.add_argument(\"--enable-warmup\", action=\"store_true\")\n parser.add_argument(\"--remove-input-padding\", action=\"store_true\")\n parser.add_argument(\"--use-perf\", action=\"store_true\", help=\"use nvtx to record performance\")\n parser.add_argument(\"--backend-type\", type=str, default=\"triton\", choices=[\"trt\", \"pytorch\"], help=\"backend type\")\n args = parser.parse_args()\n return args\n\n\ndef padded_mel_batch(ref_mels, max_seq_len):\n padded_ref_mels = []\n for mel in ref_mels:\n # pad along the last dimension\n padded_ref_mel = F.pad(mel, (0, 0, 0, max_seq_len - mel.shape[0]), value=0)\n padded_ref_mels.append(padded_ref_mel)\n padded_ref_mels = torch.stack(padded_ref_mels)\n return padded_ref_mels\n\n\ndef data_collator(batch, vocab_char_map, device=\"cuda\", use_perf=False):\n if use_perf:\n torch.cuda.nvtx.range_push(\"data_collator\")\n target_sample_rate = 24000\n target_rms = 0.1\n ids, ref_mel_list, ref_mel_len_list, estimated_reference_target_mel_len, reference_target_texts_list = (\n [],\n [],","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.padded_mel_batch","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.padded_mel_batch#L114-L121","kind":"function","name":"padded_mel_batch","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":114,"end_line":121,"context_start_line":94,"context_end_line":141,"code":" parser.add_argument(\n \"--vocoder\",\n default=\"vocos\",\n type=str,\n help=\"vocoder name\",\n )\n parser.add_argument(\n \"--vocoder-trt-engine-path\",\n default=None,\n type=str,\n help=\"vocoder trt engine path\",\n )\n parser.add_argument(\"--enable-warmup\", action=\"store_true\")\n parser.add_argument(\"--remove-input-padding\", action=\"store_true\")\n parser.add_argument(\"--use-perf\", action=\"store_true\", help=\"use nvtx to record performance\")\n parser.add_argument(\"--backend-type\", type=str, default=\"triton\", choices=[\"trt\", \"pytorch\"], help=\"backend type\")\n args = parser.parse_args()\n return args\n\n\ndef padded_mel_batch(ref_mels, max_seq_len):\n padded_ref_mels = []\n for mel in ref_mels:\n # pad along the last dimension\n padded_ref_mel = F.pad(mel, (0, 0, 0, max_seq_len - mel.shape[0]), value=0)\n padded_ref_mels.append(padded_ref_mel)\n padded_ref_mels = torch.stack(padded_ref_mels)\n return padded_ref_mels\n\n\ndef data_collator(batch, vocab_char_map, device=\"cuda\", use_perf=False):\n if use_perf:\n torch.cuda.nvtx.range_push(\"data_collator\")\n target_sample_rate = 24000\n target_rms = 0.1\n ids, ref_mel_list, ref_mel_len_list, estimated_reference_target_mel_len, reference_target_texts_list = (\n [],\n [],\n [],\n [],\n [],\n )\n for i, item in enumerate(batch):\n item_id, prompt_text, target_text = (\n item[\"id\"],\n item[\"prompt_text\"],\n item[\"target_text\"],\n )","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.data_collator","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.data_collator#L124-L200","kind":"function","name":"data_collator","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":124,"end_line":200,"context_start_line":104,"context_end_line":220,"code":" help=\"vocoder trt engine path\",\n )\n parser.add_argument(\"--enable-warmup\", action=\"store_true\")\n parser.add_argument(\"--remove-input-padding\", action=\"store_true\")\n parser.add_argument(\"--use-perf\", action=\"store_true\", help=\"use nvtx to record performance\")\n parser.add_argument(\"--backend-type\", type=str, default=\"triton\", choices=[\"trt\", \"pytorch\"], help=\"backend type\")\n args = parser.parse_args()\n return args\n\n\ndef padded_mel_batch(ref_mels, max_seq_len):\n padded_ref_mels = []\n for mel in ref_mels:\n # pad along the last dimension\n padded_ref_mel = F.pad(mel, (0, 0, 0, max_seq_len - mel.shape[0]), value=0)\n padded_ref_mels.append(padded_ref_mel)\n padded_ref_mels = torch.stack(padded_ref_mels)\n return padded_ref_mels\n\n\ndef data_collator(batch, vocab_char_map, device=\"cuda\", use_perf=False):\n if use_perf:\n torch.cuda.nvtx.range_push(\"data_collator\")\n target_sample_rate = 24000\n target_rms = 0.1\n ids, ref_mel_list, ref_mel_len_list, estimated_reference_target_mel_len, reference_target_texts_list = (\n [],\n [],\n [],\n [],\n [],\n )\n for i, item in enumerate(batch):\n item_id, prompt_text, target_text = (\n item[\"id\"],\n item[\"prompt_text\"],\n item[\"target_text\"],\n )\n ids.append(item_id)\n reference_target_texts_list.append(prompt_text + target_text)\n\n ref_audio_org, ref_sr = (\n item[\"prompt_audio\"][\"array\"],\n item[\"prompt_audio\"][\"sampling_rate\"],\n )\n ref_audio_org = torch.from_numpy(ref_audio_org).unsqueeze(0).float()\n ref_rms = torch.sqrt(torch.mean(torch.square(ref_audio_org)))\n if ref_rms < target_rms:\n ref_audio_org = ref_audio_org * target_rms / ref_rms\n\n if ref_sr != target_sample_rate:\n resampler = torchaudio.transforms.Resample(ref_sr, target_sample_rate)\n ref_audio = resampler(ref_audio_org)\n else:\n ref_audio = ref_audio_org\n\n if use_perf:\n torch.cuda.nvtx.range_push(f\"mel_spectrogram {i}\")\n ref_mel = mel_spectrogram(ref_audio, vocoder=\"vocos\", device=\"cuda\")\n if use_perf:\n torch.cuda.nvtx.range_pop()\n ref_mel = ref_mel.squeeze()\n ref_mel_len = ref_mel.shape[0]\n assert ref_mel.shape[1] == 100\n\n ref_mel_list.append(ref_mel)\n ref_mel_len_list.append(ref_mel_len)\n\n estimated_reference_target_mel_len.append(\n int(ref_mel.shape[0] * (1 + len(target_text.encode(\"utf-8\")) / len(prompt_text.encode(\"utf-8\"))))\n )\n\n max_seq_len = max(estimated_reference_target_mel_len)\n ref_mel_batch = padded_mel_batch(ref_mel_list, max_seq_len)\n ref_mel_len_batch = torch.LongTensor(ref_mel_len_list)\n\n pinyin_list = convert_char_to_pinyin(reference_target_texts_list, polyphone=True)\n text_pad_sequence = list_str_to_idx(pinyin_list, vocab_char_map)\n\n for i, item in enumerate(text_pad_sequence):\n text_pad_sequence[i] = F.pad(\n item, (0, estimated_reference_target_mel_len[i] - len(item)), mode=\"constant\", value=-1\n )\n text_pad_sequence[i] += 1 # WAR: 0 is reserved for padding token, hard coding in F5-TTS\n text_pad_sequence = pad_sequence(text_pad_sequence, padding_value=-1, batch_first=True).to(device)\n text_pad_sequence = F.pad(\n text_pad_sequence, (0, max_seq_len - text_pad_sequence.shape[1]), mode=\"constant\", value=-1\n )\n if use_perf:\n torch.cuda.nvtx.range_pop()\n return {\n \"ids\": ids,\n \"ref_mel_batch\": ref_mel_batch,\n \"ref_mel_len_batch\": ref_mel_len_batch,\n \"text_pad_sequence\": text_pad_sequence,\n \"estimated_reference_target_mel_len\": estimated_reference_target_mel_len,\n }\n\n\ndef init_distributed():\n world_size = int(os.environ.get(\"WORLD_SIZE\", 1))\n local_rank = int(os.environ.get(\"LOCAL_RANK\", 0))\n rank = int(os.environ.get(\"RANK\", 0))\n print(\n \"Inference on multiple gpus, this gpu {}\".format(local_rank)\n + \", rank {}, world_size {}\".format(rank, world_size)\n )\n torch.cuda.set_device(local_rank)\n # Initialize process group with explicit device IDs\n dist.init_process_group(\n \"nccl\",\n )\n return world_size, local_rank, rank\n\n\ndef get_tokenizer(vocab_file_path: str):\n \"\"\"","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.init_distributed","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.init_distributed#L203-L216","kind":"function","name":"init_distributed","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":203,"end_line":216,"context_start_line":183,"context_end_line":236,"code":" for i, item in enumerate(text_pad_sequence):\n text_pad_sequence[i] = F.pad(\n item, (0, estimated_reference_target_mel_len[i] - len(item)), mode=\"constant\", value=-1\n )\n text_pad_sequence[i] += 1 # WAR: 0 is reserved for padding token, hard coding in F5-TTS\n text_pad_sequence = pad_sequence(text_pad_sequence, padding_value=-1, batch_first=True).to(device)\n text_pad_sequence = F.pad(\n text_pad_sequence, (0, max_seq_len - text_pad_sequence.shape[1]), mode=\"constant\", value=-1\n )\n if use_perf:\n torch.cuda.nvtx.range_pop()\n return {\n \"ids\": ids,\n \"ref_mel_batch\": ref_mel_batch,\n \"ref_mel_len_batch\": ref_mel_len_batch,\n \"text_pad_sequence\": text_pad_sequence,\n \"estimated_reference_target_mel_len\": estimated_reference_target_mel_len,\n }\n\n\ndef init_distributed():\n world_size = int(os.environ.get(\"WORLD_SIZE\", 1))\n local_rank = int(os.environ.get(\"LOCAL_RANK\", 0))\n rank = int(os.environ.get(\"RANK\", 0))\n print(\n \"Inference on multiple gpus, this gpu {}\".format(local_rank)\n + \", rank {}, world_size {}\".format(rank, world_size)\n )\n torch.cuda.set_device(local_rank)\n # Initialize process group with explicit device IDs\n dist.init_process_group(\n \"nccl\",\n )\n return world_size, local_rank, rank\n\n\ndef get_tokenizer(vocab_file_path: str):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file\n - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n with open(vocab_file_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n return vocab_char_map, vocab_size\n\n","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.get_tokenizer","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.get_tokenizer#L219-L234","kind":"function","name":"get_tokenizer","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":219,"end_line":234,"context_start_line":199,"context_end_line":254,"code":" \"estimated_reference_target_mel_len\": estimated_reference_target_mel_len,\n }\n\n\ndef init_distributed():\n world_size = int(os.environ.get(\"WORLD_SIZE\", 1))\n local_rank = int(os.environ.get(\"LOCAL_RANK\", 0))\n rank = int(os.environ.get(\"RANK\", 0))\n print(\n \"Inference on multiple gpus, this gpu {}\".format(local_rank)\n + \", rank {}, world_size {}\".format(rank, world_size)\n )\n torch.cuda.set_device(local_rank)\n # Initialize process group with explicit device IDs\n dist.init_process_group(\n \"nccl\",\n )\n return world_size, local_rank, rank\n\n\ndef get_tokenizer(vocab_file_path: str):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file\n - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n with open(vocab_file_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n return vocab_char_map, vocab_size\n\n\ndef convert_char_to_pinyin(reference_target_texts_list, polyphone=True):\n final_reference_target_texts_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n\n for text in reference_target_texts_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.convert_char_to_pinyin","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.convert_char_to_pinyin#L237-L272","kind":"function","name":"convert_char_to_pinyin","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":237,"end_line":272,"context_start_line":217,"context_end_line":292,"code":"\n\ndef get_tokenizer(vocab_file_path: str):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file\n - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n with open(vocab_file_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n return vocab_char_map, vocab_size\n\n\ndef convert_char_to_pinyin(reference_target_texts_list, polyphone=True):\n final_reference_target_texts_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n\n for text in reference_target_texts_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)\n elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)\n elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_reference_target_texts_list.append(char_list)\n\n return final_reference_target_texts_list\n\n\ndef list_str_to_idx(\n text: Union[List[str], List[List[str]]],\n vocab_char_map: Dict[str, int], # {char: idx}\n padding_value=-1,\n):\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n # text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return list_idx_tensors\n\n\ndef load_vocoder(\n vocoder_name=\"vocos\", is_local=False, local_path=\"\", device=\"cuda\", hf_cache_dir=None, vocoder_trt_engine_path=None\n):\n if vocoder_name == \"vocos\":\n if vocoder_trt_engine_path is not None:\n vocoder = VocosTensorRT(engine_path=vocoder_trt_engine_path)\n else:\n # vocoder = Vocos.from_pretrained(\"charactr/vocos-mel-24khz\").to(device)","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.list_str_to_idx","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.list_str_to_idx#L275-L282","kind":"function","name":"list_str_to_idx","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":275,"end_line":282,"context_start_line":255,"context_end_line":302,"code":" elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)\n elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_reference_target_texts_list.append(char_list)\n\n return final_reference_target_texts_list\n\n\ndef list_str_to_idx(\n text: Union[List[str], List[List[str]]],\n vocab_char_map: Dict[str, int], # {char: idx}\n padding_value=-1,\n):\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n # text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return list_idx_tensors\n\n\ndef load_vocoder(\n vocoder_name=\"vocos\", is_local=False, local_path=\"\", device=\"cuda\", hf_cache_dir=None, vocoder_trt_engine_path=None\n):\n if vocoder_name == \"vocos\":\n if vocoder_trt_engine_path is not None:\n vocoder = VocosTensorRT(engine_path=vocoder_trt_engine_path)\n else:\n # vocoder = Vocos.from_pretrained(\"charactr/vocos-mel-24khz\").to(device)\n if is_local:\n print(f\"Load vocos from local path {local_path}\")\n config_path = f\"{local_path}/config.yaml\"\n model_path = f\"{local_path}/pytorch_model.bin\"\n else:\n print(\"Download Vocos from huggingface charactr/vocos-mel-24khz\")\n repo_id = \"charactr/vocos-mel-24khz\"\n config_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"config.yaml\")\n model_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"pytorch_model.bin\")\n vocoder = Vocos.from_hparams(config_path)","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.load_vocoder","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.load_vocoder#L285-L316","kind":"function","name":"load_vocoder","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":285,"end_line":316,"context_start_line":265,"context_end_line":336,"code":" elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_reference_target_texts_list.append(char_list)\n\n return final_reference_target_texts_list\n\n\ndef list_str_to_idx(\n text: Union[List[str], List[List[str]]],\n vocab_char_map: Dict[str, int], # {char: idx}\n padding_value=-1,\n):\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n # text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)\n return list_idx_tensors\n\n\ndef load_vocoder(\n vocoder_name=\"vocos\", is_local=False, local_path=\"\", device=\"cuda\", hf_cache_dir=None, vocoder_trt_engine_path=None\n):\n if vocoder_name == \"vocos\":\n if vocoder_trt_engine_path is not None:\n vocoder = VocosTensorRT(engine_path=vocoder_trt_engine_path)\n else:\n # vocoder = Vocos.from_pretrained(\"charactr/vocos-mel-24khz\").to(device)\n if is_local:\n print(f\"Load vocos from local path {local_path}\")\n config_path = f\"{local_path}/config.yaml\"\n model_path = f\"{local_path}/pytorch_model.bin\"\n else:\n print(\"Download Vocos from huggingface charactr/vocos-mel-24khz\")\n repo_id = \"charactr/vocos-mel-24khz\"\n config_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"config.yaml\")\n model_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"pytorch_model.bin\")\n vocoder = Vocos.from_hparams(config_path)\n state_dict = torch.load(model_path, map_location=\"cpu\", weights_only=True)\n from vocos.feature_extractors import EncodecFeatures\n\n if isinstance(vocoder.feature_extractor, EncodecFeatures):\n encodec_parameters = {\n \"feature_extractor.encodec.\" + key: value\n for key, value in vocoder.feature_extractor.encodec.state_dict().items()\n }\n state_dict.update(encodec_parameters)\n vocoder.load_state_dict(state_dict)\n vocoder = vocoder.eval().to(device)\n elif vocoder_name == \"bigvgan\":\n raise NotImplementedError(\"BigVGAN is not implemented yet\")\n return vocoder\n\n\ndef mel_spectrogram(waveform, vocoder=\"vocos\", device=\"cuda\"):\n if vocoder == \"vocos\":\n mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=24000,\n n_fft=1024,\n win_length=1024,\n hop_length=256,\n n_mels=100,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(device)\n mel = mel_stft(waveform.to(device))\n mel = mel.clamp(min=1e-5).log()\n return mel.transpose(1, 2)\n\n","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.mel_spectrogram","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.mel_spectrogram#L319-L334","kind":"function","name":"mel_spectrogram","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":319,"end_line":334,"context_start_line":299,"context_end_line":354,"code":" repo_id = \"charactr/vocos-mel-24khz\"\n config_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"config.yaml\")\n model_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"pytorch_model.bin\")\n vocoder = Vocos.from_hparams(config_path)\n state_dict = torch.load(model_path, map_location=\"cpu\", weights_only=True)\n from vocos.feature_extractors import EncodecFeatures\n\n if isinstance(vocoder.feature_extractor, EncodecFeatures):\n encodec_parameters = {\n \"feature_extractor.encodec.\" + key: value\n for key, value in vocoder.feature_extractor.encodec.state_dict().items()\n }\n state_dict.update(encodec_parameters)\n vocoder.load_state_dict(state_dict)\n vocoder = vocoder.eval().to(device)\n elif vocoder_name == \"bigvgan\":\n raise NotImplementedError(\"BigVGAN is not implemented yet\")\n return vocoder\n\n\ndef mel_spectrogram(waveform, vocoder=\"vocos\", device=\"cuda\"):\n if vocoder == \"vocos\":\n mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=24000,\n n_fft=1024,\n win_length=1024,\n hop_length=256,\n n_mels=100,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(device)\n mel = mel_stft(waveform.to(device))\n mel = mel.clamp(min=1e-5).log()\n return mel.transpose(1, 2)\n\n\nclass VocosTensorRT:\n def __init__(self, engine_path=\"./vocos_vocoder.plan\", stream=None):\n TRT_LOGGER = trt.Logger(trt.Logger.WARNING)\n trt.init_libnvinfer_plugins(TRT_LOGGER, namespace=\"\")\n logger.info(f\"Loading vae engine from {engine_path}\")\n self.engine_path = engine_path\n with open(engine_path, \"rb\") as f:\n engine_buffer = f.read()\n self.session = Session.from_serialized_engine(engine_buffer)\n self.stream = stream if stream is not None else torch.cuda.current_stream().cuda_stream\n\n def decode(self, mels):\n mels = mels.contiguous()\n inputs = {\"mel\": mels}\n output_info = self.session.infer_shapes([TensorInfo(\"mel\", trt.DataType.FLOAT, mels.shape)])\n outputs = {\n t.name: torch.empty(tuple(t.shape), dtype=trt_dtype_to_torch(t.dtype), device=\"cuda\") for t in output_info\n }","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.VocosTensorRT","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.benchmark.VocosTensorRT#L337-L360","kind":"class","name":"VocosTensorRT","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":337,"end_line":360,"context_start_line":317,"context_end_line":380,"code":"\n\ndef mel_spectrogram(waveform, vocoder=\"vocos\", device=\"cuda\"):\n if vocoder == \"vocos\":\n mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=24000,\n n_fft=1024,\n win_length=1024,\n hop_length=256,\n n_mels=100,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(device)\n mel = mel_stft(waveform.to(device))\n mel = mel.clamp(min=1e-5).log()\n return mel.transpose(1, 2)\n\n\nclass VocosTensorRT:\n def __init__(self, engine_path=\"./vocos_vocoder.plan\", stream=None):\n TRT_LOGGER = trt.Logger(trt.Logger.WARNING)\n trt.init_libnvinfer_plugins(TRT_LOGGER, namespace=\"\")\n logger.info(f\"Loading vae engine from {engine_path}\")\n self.engine_path = engine_path\n with open(engine_path, \"rb\") as f:\n engine_buffer = f.read()\n self.session = Session.from_serialized_engine(engine_buffer)\n self.stream = stream if stream is not None else torch.cuda.current_stream().cuda_stream\n\n def decode(self, mels):\n mels = mels.contiguous()\n inputs = {\"mel\": mels}\n output_info = self.session.infer_shapes([TensorInfo(\"mel\", trt.DataType.FLOAT, mels.shape)])\n outputs = {\n t.name: torch.empty(tuple(t.shape), dtype=trt_dtype_to_torch(t.dtype), device=\"cuda\") for t in output_info\n }\n ok = self.session.run(inputs, outputs, self.stream)\n\n assert ok, \"Runtime execution failed for vae session\"\n\n samples = outputs[\"waveform\"]\n return samples\n\n\ndef main():\n args = get_args()\n os.makedirs(args.output_dir, exist_ok=True)\n\n assert torch.cuda.is_available()\n world_size, local_rank, rank = init_distributed()\n device = torch.device(f\"cuda:{local_rank}\")\n\n vocab_char_map, vocab_size = get_tokenizer(args.vocab_file)\n\n tllm_model_dir = args.tllm_model_dir\n config_file = os.path.join(tllm_model_dir, \"config.json\")\n with open(config_file) as f:\n config = json.load(f)\n if args.backend_type == \"trt\":\n model = F5TTS(\n config, debug_mode=False, tllm_model_dir=tllm_model_dir, model_path=args.model_path, vocab_size=vocab_size\n )","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.main","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.main#L363-L556","kind":"function","name":"main","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":363,"end_line":556,"context_start_line":343,"context_end_line":560,"code":" with open(engine_path, \"rb\") as f:\n engine_buffer = f.read()\n self.session = Session.from_serialized_engine(engine_buffer)\n self.stream = stream if stream is not None else torch.cuda.current_stream().cuda_stream\n\n def decode(self, mels):\n mels = mels.contiguous()\n inputs = {\"mel\": mels}\n output_info = self.session.infer_shapes([TensorInfo(\"mel\", trt.DataType.FLOAT, mels.shape)])\n outputs = {\n t.name: torch.empty(tuple(t.shape), dtype=trt_dtype_to_torch(t.dtype), device=\"cuda\") for t in output_info\n }\n ok = self.session.run(inputs, outputs, self.stream)\n\n assert ok, \"Runtime execution failed for vae session\"\n\n samples = outputs[\"waveform\"]\n return samples\n\n\ndef main():\n args = get_args()\n os.makedirs(args.output_dir, exist_ok=True)\n\n assert torch.cuda.is_available()\n world_size, local_rank, rank = init_distributed()\n device = torch.device(f\"cuda:{local_rank}\")\n\n vocab_char_map, vocab_size = get_tokenizer(args.vocab_file)\n\n tllm_model_dir = args.tllm_model_dir\n config_file = os.path.join(tllm_model_dir, \"config.json\")\n with open(config_file) as f:\n config = json.load(f)\n if args.backend_type == \"trt\":\n model = F5TTS(\n config, debug_mode=False, tllm_model_dir=tllm_model_dir, model_path=args.model_path, vocab_size=vocab_size\n )\n elif args.backend_type == \"pytorch\":\n import sys\n\n sys.path.append(f\"{os.path.dirname(os.path.abspath(__file__))}/../../../../src/\")\n from f5_tts.infer.utils_infer import load_model\n from f5_tts.model import DiT\n\n F5TTS_model_cfg = dict(\n dim=1024,\n depth=22,\n heads=16,\n ff_mult=2,\n text_dim=512,\n conv_layers=4,\n pe_attn_head=1,\n text_mask_padding=False,\n )\n model = load_model(DiT, F5TTS_model_cfg, args.model_path)\n\n vocoder = load_vocoder(\n vocoder_name=args.vocoder, device=device, vocoder_trt_engine_path=args.vocoder_trt_engine_path\n )\n\n dataset = load_dataset(\n \"yuekai/seed_tts\",\n split=args.split_name,\n trust_remote_code=True,\n )\n\n def add_estimated_duration(example):\n prompt_audio_len = example[\"prompt_audio\"][\"array\"].shape[0]\n scale_factor = 1 + len(example[\"target_text\"]) / len(example[\"prompt_text\"])\n estimated_duration = prompt_audio_len * scale_factor\n example[\"estimated_duration\"] = estimated_duration / example[\"prompt_audio\"][\"sampling_rate\"]\n return example\n\n dataset = dataset.map(add_estimated_duration)\n dataset = dataset.sort(\"estimated_duration\", reverse=True)\n if args.use_perf:\n # dataset_list = [dataset.select(range(1)) for i in range(16)] # seq_len 1000\n dataset_list_short = [dataset.select([24]) for i in range(8)] # seq_len 719\n # dataset_list_long = [dataset.select([23]) for i in range(8)] # seq_len 2002\n # dataset = datasets.concatenate_datasets(dataset_list_short + dataset_list_long)\n dataset = datasets.concatenate_datasets(dataset_list_short)\n if world_size > 1:\n sampler = DistributedSampler(dataset, num_replicas=world_size, rank=rank)\n else:\n # This would disable shuffling\n sampler = None\n\n dataloader = DataLoader(\n dataset,\n batch_size=args.batch_size,\n sampler=sampler,\n shuffle=False,\n num_workers=args.num_workers,\n prefetch_factor=args.prefetch,\n collate_fn=lambda x: data_collator(x, vocab_char_map, use_perf=args.use_perf),\n )\n\n total_steps = len(dataset)\n\n if args.enable_warmup:\n for batch in dataloader:\n ref_mels, ref_mel_lens = batch[\"ref_mel_batch\"].to(device), batch[\"ref_mel_len_batch\"].to(device)\n text_pad_seq = batch[\"text_pad_sequence\"].to(device)\n total_mel_lens = batch[\"estimated_reference_target_mel_len\"]\n if args.backend_type == \"trt\":\n _ = model.sample(\n text_pad_seq, ref_mels, ref_mel_lens, total_mel_lens, remove_input_padding=args.remove_input_padding\n )\n elif args.backend_type == \"pytorch\":\n with torch.inference_mode():\n text_pad_seq -= 1\n text_pad_seq[text_pad_seq == -2] = -1\n total_mel_lens = torch.tensor(total_mel_lens, device=device)\n generated, _ = model.sample(\n cond=ref_mels,\n text=text_pad_seq,\n duration=total_mel_lens,\n steps=16,\n cfg_strength=2.0,\n sway_sampling_coef=-1,\n )\n\n if rank == 0:\n progress_bar = tqdm(total=total_steps, desc=\"Processing\", unit=\"wavs\")\n\n decoding_time = 0\n vocoder_time = 0\n total_duration = 0\n if args.use_perf:\n torch.cuda.cudart().cudaProfilerStart()\n total_decoding_time = time.time()\n for batch in dataloader:\n if args.use_perf:\n torch.cuda.nvtx.range_push(\"data sample\")\n ref_mels, ref_mel_lens = batch[\"ref_mel_batch\"].to(device), batch[\"ref_mel_len_batch\"].to(device)\n text_pad_seq = batch[\"text_pad_sequence\"].to(device)\n total_mel_lens = batch[\"estimated_reference_target_mel_len\"]\n\n if args.use_perf:\n torch.cuda.nvtx.range_pop()\n if args.backend_type == \"trt\":\n generated, cost_time = model.sample(\n text_pad_seq,\n ref_mels,\n ref_mel_lens,\n total_mel_lens,\n remove_input_padding=args.remove_input_padding,\n use_perf=args.use_perf,\n )\n elif args.backend_type == \"pytorch\":\n total_mel_lens = torch.tensor(total_mel_lens, device=device)\n with torch.inference_mode():\n start_time = time.time()\n text_pad_seq -= 1\n text_pad_seq[text_pad_seq == -2] = -1\n generated, _ = model.sample(\n cond=ref_mels,\n text=text_pad_seq,\n duration=total_mel_lens,\n lens=ref_mel_lens,\n steps=16,\n cfg_strength=2.0,\n sway_sampling_coef=-1,\n )\n cost_time = time.time() - start_time\n decoding_time += cost_time\n vocoder_start_time = time.time()\n for i, gen in enumerate(generated):\n gen = gen[ref_mel_lens[i] : total_mel_lens[i], :].unsqueeze(0)\n gen_mel_spec = gen.permute(0, 2, 1).to(torch.float32)\n if args.vocoder == \"vocos\":\n if args.use_perf:\n torch.cuda.nvtx.range_push(\"vocoder decode\")\n generated_wave = vocoder.decode(gen_mel_spec).cpu()\n if args.use_perf:\n torch.cuda.nvtx.range_pop()\n else:\n generated_wave = vocoder(gen_mel_spec).squeeze(0).cpu()\n target_rms = 0.1\n target_sample_rate = 24_000\n # if ref_rms_list[i] < target_rms:\n # generated_wave = generated_wave * ref_rms_list[i] / target_rms\n rms = torch.sqrt(torch.mean(torch.square(generated_wave)))\n if rms < target_rms:\n generated_wave = generated_wave * target_rms / rms\n utt = batch[\"ids\"][i]\n torchaudio.save(\n f\"{args.output_dir}/{utt}.wav\",\n generated_wave,\n target_sample_rate,\n )\n total_duration += generated_wave.shape[1] / target_sample_rate\n vocoder_time += time.time() - vocoder_start_time\n if rank == 0:\n progress_bar.update(world_size * len(batch[\"ids\"]))\n total_decoding_time = time.time() - total_decoding_time\n if rank == 0:\n progress_bar.close()\n rtf = total_decoding_time / total_duration\n s = f\"RTF: {rtf:.4f}\\n\"\n s += f\"total_duration: {total_duration:.3f} seconds\\n\"\n s += f\"({total_duration / 3600:.2f} hours)\\n\"\n s += f\"DiT time: {decoding_time:.3f} seconds ({decoding_time / 3600:.2f} hours)\\n\"\n s += f\"Vocoder time: {vocoder_time:.3f} seconds ({vocoder_time / 3600:.2f} hours)\\n\"\n s += f\"total decoding time: {total_decoding_time:.3f} seconds ({total_decoding_time / 3600:.2f} hours)\\n\"\n s += f\"batch size: {args.batch_size}\\n\"\n print(s)\n\n with open(f\"{args.output_dir}/rtf.txt\", \"w\") as f:\n f.write(s)\n\n dist.barrier()\n dist.destroy_process_group()\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.is_chinese","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.is_chinese#L243-L244","kind":"function","name":"is_chinese","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":243,"end_line":244,"context_start_line":223,"context_end_line":264,"code":" - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n with open(vocab_file_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n return vocab_char_map, vocab_size\n\n\ndef convert_char_to_pinyin(reference_target_texts_list, polyphone=True):\n final_reference_target_texts_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n\n for text in reference_target_texts_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)\n elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.__init__#L338-L346","kind":"function","name":"__init__","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":338,"end_line":346,"context_start_line":318,"context_end_line":366,"code":"\ndef mel_spectrogram(waveform, vocoder=\"vocos\", device=\"cuda\"):\n if vocoder == \"vocos\":\n mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=24000,\n n_fft=1024,\n win_length=1024,\n hop_length=256,\n n_mels=100,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(device)\n mel = mel_stft(waveform.to(device))\n mel = mel.clamp(min=1e-5).log()\n return mel.transpose(1, 2)\n\n\nclass VocosTensorRT:\n def __init__(self, engine_path=\"./vocos_vocoder.plan\", stream=None):\n TRT_LOGGER = trt.Logger(trt.Logger.WARNING)\n trt.init_libnvinfer_plugins(TRT_LOGGER, namespace=\"\")\n logger.info(f\"Loading vae engine from {engine_path}\")\n self.engine_path = engine_path\n with open(engine_path, \"rb\") as f:\n engine_buffer = f.read()\n self.session = Session.from_serialized_engine(engine_buffer)\n self.stream = stream if stream is not None else torch.cuda.current_stream().cuda_stream\n\n def decode(self, mels):\n mels = mels.contiguous()\n inputs = {\"mel\": mels}\n output_info = self.session.infer_shapes([TensorInfo(\"mel\", trt.DataType.FLOAT, mels.shape)])\n outputs = {\n t.name: torch.empty(tuple(t.shape), dtype=trt_dtype_to_torch(t.dtype), device=\"cuda\") for t in output_info\n }\n ok = self.session.run(inputs, outputs, self.stream)\n\n assert ok, \"Runtime execution failed for vae session\"\n\n samples = outputs[\"waveform\"]\n return samples\n\n\ndef main():\n args = get_args()\n os.makedirs(args.output_dir, exist_ok=True)\n","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.decode","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.decode#L348-L360","kind":"function","name":"decode","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":348,"end_line":360,"context_start_line":328,"context_end_line":380,"code":" center=True,\n normalized=False,\n norm=None,\n ).to(device)\n mel = mel_stft(waveform.to(device))\n mel = mel.clamp(min=1e-5).log()\n return mel.transpose(1, 2)\n\n\nclass VocosTensorRT:\n def __init__(self, engine_path=\"./vocos_vocoder.plan\", stream=None):\n TRT_LOGGER = trt.Logger(trt.Logger.WARNING)\n trt.init_libnvinfer_plugins(TRT_LOGGER, namespace=\"\")\n logger.info(f\"Loading vae engine from {engine_path}\")\n self.engine_path = engine_path\n with open(engine_path, \"rb\") as f:\n engine_buffer = f.read()\n self.session = Session.from_serialized_engine(engine_buffer)\n self.stream = stream if stream is not None else torch.cuda.current_stream().cuda_stream\n\n def decode(self, mels):\n mels = mels.contiguous()\n inputs = {\"mel\": mels}\n output_info = self.session.infer_shapes([TensorInfo(\"mel\", trt.DataType.FLOAT, mels.shape)])\n outputs = {\n t.name: torch.empty(tuple(t.shape), dtype=trt_dtype_to_torch(t.dtype), device=\"cuda\") for t in output_info\n }\n ok = self.session.run(inputs, outputs, self.stream)\n\n assert ok, \"Runtime execution failed for vae session\"\n\n samples = outputs[\"waveform\"]\n return samples\n\n\ndef main():\n args = get_args()\n os.makedirs(args.output_dir, exist_ok=True)\n\n assert torch.cuda.is_available()\n world_size, local_rank, rank = init_distributed()\n device = torch.device(f\"cuda:{local_rank}\")\n\n vocab_char_map, vocab_size = get_tokenizer(args.vocab_file)\n\n tllm_model_dir = args.tllm_model_dir\n config_file = os.path.join(tllm_model_dir, \"config.json\")\n with open(config_file) as f:\n config = json.load(f)\n if args.backend_type == \"trt\":\n model = F5TTS(\n config, debug_mode=False, tllm_model_dir=tllm_model_dir, model_path=args.model_path, vocab_size=vocab_size\n )","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.benchmark.add_estimated_duration","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.benchmark.add_estimated_duration#L410-L415","kind":"function","name":"add_estimated_duration","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":410,"end_line":415,"context_start_line":390,"context_end_line":435,"code":" depth=22,\n heads=16,\n ff_mult=2,\n text_dim=512,\n conv_layers=4,\n pe_attn_head=1,\n text_mask_padding=False,\n )\n model = load_model(DiT, F5TTS_model_cfg, args.model_path)\n\n vocoder = load_vocoder(\n vocoder_name=args.vocoder, device=device, vocoder_trt_engine_path=args.vocoder_trt_engine_path\n )\n\n dataset = load_dataset(\n \"yuekai/seed_tts\",\n split=args.split_name,\n trust_remote_code=True,\n )\n\n def add_estimated_duration(example):\n prompt_audio_len = example[\"prompt_audio\"][\"array\"].shape[0]\n scale_factor = 1 + len(example[\"target_text\"]) / len(example[\"prompt_text\"])\n estimated_duration = prompt_audio_len * scale_factor\n example[\"estimated_duration\"] = estimated_duration / example[\"prompt_audio\"][\"sampling_rate\"]\n return example\n\n dataset = dataset.map(add_estimated_duration)\n dataset = dataset.sort(\"estimated_duration\", reverse=True)\n if args.use_perf:\n # dataset_list = [dataset.select(range(1)) for i in range(16)] # seq_len 1000\n dataset_list_short = [dataset.select([24]) for i in range(8)] # seq_len 719\n # dataset_list_long = [dataset.select([23]) for i in range(8)] # seq_len 2002\n # dataset = datasets.concatenate_datasets(dataset_list_short + dataset_list_long)\n dataset = datasets.concatenate_datasets(dataset_list_short)\n if world_size > 1:\n sampler = DistributedSampler(dataset, num_replicas=world_size, rank=rank)\n else:\n # This would disable shuffling\n sampler = None\n\n dataloader = DataLoader(\n dataset,\n batch_size=args.batch_size,\n sampler=sampler,\n shuffle=False,","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.client_grpc","uri":"program://DMOSpeech2/module/src.f5_tts.runtime.triton_trtllm.client_grpc#L1-L470","kind":"module","name":"src.f5_tts.runtime.triton_trtllm.client_grpc","path":"src/f5_tts/runtime/triton_trtllm/client_grpc.py","language":"python","start_line":1,"end_line":470,"context_start_line":1,"context_end_line":470,"code":"#!/usr/bin/env python3\n# Copyright 2022 Xiaomi Corp. (authors: Fangjun Kuang)\n# 2023 Nvidia (authors: Yuekai Zhang)\n# 2023 Recurrent.ai (authors: Songtao Shi)\n# See LICENSE for clarification regarding multiple authors\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\"\"\"\nThis script supports to load dataset from huggingface and sends it to the server\nfor decoding, in parallel.\n\nUsage:\nnum_task=2\n\n# For offline F5-TTS\npython3 client_grpc.py \\\n --server-addr localhost \\\n --model-name f5_tts \\\n --num-tasks $num_task \\\n --huggingface-dataset yuekai/seed_tts \\\n --split-name test_zh \\\n --log-dir ./log_concurrent_tasks_${num_task}\n\n# For offline Spark-TTS-0.5B\npython3 client_grpc.py \\\n --server-addr localhost \\\n --model-name spark_tts \\\n --num-tasks $num_task \\\n --huggingface-dataset yuekai/seed_tts \\\n --split-name wenetspeech4tts \\\n --log-dir ./log_concurrent_tasks_${num_task}\n\"\"\"\n\nimport argparse\nimport asyncio\nimport json\nimport os\nimport time\nimport types\nfrom pathlib import Path\n\nimport numpy as np\nimport soundfile as sf\nimport tritonclient\nimport tritonclient.grpc.aio as grpcclient\nfrom tritonclient.utils import np_to_triton_dtype\n\n\ndef write_triton_stats(stats, summary_file):\n with open(summary_file, \"w\") as summary_f:\n model_stats = stats[\"model_stats\"]\n # write a note, the log is from triton_client.get_inference_statistics(), to better human readability\n summary_f.write(\n \"The log is parsing from triton_client.get_inference_statistics(), to better human readability. \\n\"\n )\n summary_f.write(\"To learn more about the log, please refer to: \\n\")\n summary_f.write(\"1. https://github.com/triton-inference-server/server/blob/main/docs/user_guide/metrics.md \\n\")\n summary_f.write(\"2. https://github.com/triton-inference-server/server/issues/5374 \\n\\n\")\n summary_f.write(\n \"To better improve throughput, we always would like let requests wait in the queue for a while, and then execute them with a larger batch size. \\n\"\n )\n summary_f.write(\n \"However, there is a trade-off between the increased queue time and the increased batch size. \\n\"\n )\n summary_f.write(\n \"You may change 'max_queue_delay_microseconds' and 'preferred_batch_size' in the model configuration file to achieve this. \\n\"\n )\n summary_f.write(\n \"See https://github.com/triton-inference-server/server/blob/main/docs/user_guide/model_configuration.md#delayed-batching for more details. \\n\\n\"\n )\n for model_state in model_stats:\n if \"last_inference\" not in model_state:\n continue\n summary_f.write(f\"model name is {model_state['name']} \\n\")\n model_inference_stats = model_state[\"inference_stats\"]\n total_queue_time_s = int(model_inference_stats[\"queue\"][\"ns\"]) / 1e9\n total_infer_time_s = int(model_inference_stats[\"compute_infer\"][\"ns\"]) / 1e9\n total_input_time_s = int(model_inference_stats[\"compute_input\"][\"ns\"]) / 1e9\n total_output_time_s = int(model_inference_stats[\"compute_output\"][\"ns\"]) / 1e9\n summary_f.write(\n f\"queue time {total_queue_time_s:<5.2f} s, compute infer time {total_infer_time_s:<5.2f} s, compute input time {total_input_time_s:<5.2f} s, compute output time {total_output_time_s:<5.2f} s \\n\" # noqa\n )\n model_batch_stats = model_state[\"batch_stats\"]\n for batch in model_batch_stats:\n batch_size = int(batch[\"batch_size\"])\n compute_input = batch[\"compute_input\"]\n compute_output = batch[\"compute_output\"]\n compute_infer = batch[\"compute_infer\"]\n batch_count = int(compute_infer[\"count\"])\n assert compute_infer[\"count\"] == compute_output[\"count\"] == compute_input[\"count\"]\n compute_infer_time_ms = int(compute_infer[\"ns\"]) / 1e6\n compute_input_time_ms = int(compute_input[\"ns\"]) / 1e6\n compute_output_time_ms = int(compute_output[\"ns\"]) / 1e6\n summary_f.write(\n f\"execuate inference with batch_size {batch_size:<2} total {batch_count:<5} times, total_infer_time {compute_infer_time_ms:<9.2f} ms, avg_infer_time {compute_infer_time_ms:<9.2f}/{batch_count:<5}={compute_infer_time_ms / batch_count:.2f} ms, avg_infer_time_per_sample {compute_infer_time_ms:<9.2f}/{batch_count:<5}/{batch_size}={compute_infer_time_ms / batch_count / batch_size:.2f} ms \\n\" # noqa\n )\n summary_f.write(\n f\"input {compute_input_time_ms:<9.2f} ms, avg {compute_input_time_ms / batch_count:.2f} ms, \" # noqa\n )\n summary_f.write(\n f\"output {compute_output_time_ms:<9.2f} ms, avg {compute_output_time_ms / batch_count:.2f} ms \\n\" # noqa\n )\n\n\ndef get_args():\n parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)\n\n parser.add_argument(\n \"--server-addr\",\n type=str,\n default=\"localhost\",\n help=\"Address of the server\",\n )\n\n parser.add_argument(\n \"--server-port\",\n type=int,\n default=8001,\n help=\"Grpc port of the triton server, default is 8001\",\n )\n\n parser.add_argument(\n \"--reference-audio\",\n type=str,\n default=None,\n help=\"Path to a single audio file. It can't be specified at the same time with --manifest-dir\",\n )\n\n parser.add_argument(\n \"--reference-text\",\n type=str,\n default=\"\",\n help=\"\",\n )\n\n parser.add_argument(\n \"--target-text\",\n type=str,\n default=\"\",\n help=\"\",\n )\n\n parser.add_argument(\n \"--huggingface-dataset\",\n type=str,\n default=\"yuekai/seed_tts\",\n help=\"dataset name in huggingface dataset hub\",\n )\n\n parser.add_argument(\n \"--split-name\",\n type=str,\n default=\"wenetspeech4tts\",\n choices=[\"wenetspeech4tts\", \"test_zh\", \"test_en\", \"test_hard\"],\n help=\"dataset split name, default is 'test'\",\n )\n\n parser.add_argument(\n \"--manifest-path\",\n type=str,\n default=None,\n help=\"Path to the manifest dir which includes wav.scp trans.txt files.\",\n )\n\n parser.add_argument(\n \"--model-name\",\n type=str,\n default=\"f5_tts\",\n choices=[\"f5_tts\", \"spark_tts\"],\n help=\"triton model_repo module name to request: transducer for k2, attention_rescoring for wenet offline, streaming_wenet for wenet streaming, infer_pipeline for paraformer large offline\",\n )\n\n parser.add_argument(\n \"--num-tasks\",\n type=int,\n default=1,\n help=\"Number of concurrent tasks for sending\",\n )\n\n parser.add_argument(\n \"--log-interval\",\n type=int,\n default=5,\n help=\"Controls how frequently we print the log.\",\n )\n\n parser.add_argument(\n \"--compute-wer\",\n action=\"store_true\",\n default=False,\n help=\"\"\"True to compute WER.\n \"\"\",\n )\n\n parser.add_argument(\n \"--log-dir\",\n type=str,\n required=False,\n default=\"./tmp\",\n help=\"log directory\",\n )\n\n parser.add_argument(\n \"--batch-size\",\n type=int,\n default=1,\n help=\"Inference batch_size per request for offline mode.\",\n )\n\n return parser.parse_args()\n\n\ndef load_audio(wav_path, target_sample_rate=24000):\n assert target_sample_rate == 24000, \"hard coding in server\"\n if isinstance(wav_path, dict):\n waveform = wav_path[\"array\"]\n sample_rate = wav_path[\"sampling_rate\"]\n else:\n waveform, sample_rate = sf.read(wav_path)\n if sample_rate != target_sample_rate:\n from scipy.signal import resample\n\n num_samples = int(len(waveform) * (target_sample_rate / sample_rate))\n waveform = resample(waveform, num_samples)\n return waveform, target_sample_rate\n\n\nasync def send(\n manifest_item_list: list,\n name: str,\n triton_client: tritonclient.grpc.aio.InferenceServerClient,\n protocol_client: types.ModuleType,\n log_interval: int,\n model_name: str,\n padding_duration: int = None,\n audio_save_dir: str = \"./\",\n save_sample_rate: int = 24000,\n):\n total_duration = 0.0\n latency_data = []\n task_id = int(name[5:])\n\n print(f\"manifest_item_list: {manifest_item_list}\")\n for i, item in enumerate(manifest_item_list):\n if i % log_interval == 0:\n print(f\"{name}: {i}/{len(manifest_item_list)}\")\n waveform, sample_rate = load_audio(item[\"audio_filepath\"], target_sample_rate=24000)\n duration = len(waveform) / sample_rate\n lengths = np.array([[len(waveform)]], dtype=np.int32)\n\n reference_text, target_text = item[\"reference_text\"], item[\"target_text\"]\n\n estimated_target_duration = duration / len(reference_text) * len(target_text)\n\n if padding_duration:\n # padding to nearset 10 seconds\n samples = np.zeros(\n (\n 1,\n padding_duration\n * sample_rate\n * ((int(estimated_target_duration + duration) // padding_duration) + 1),\n ),\n dtype=np.float32,\n )\n\n samples[0, : len(waveform)] = waveform\n else:\n samples = waveform\n\n samples = samples.reshape(1, -1).astype(np.float32)\n\n inputs = [\n protocol_client.InferInput(\"reference_wav\", samples.shape, np_to_triton_dtype(samples.dtype)),\n protocol_client.InferInput(\"reference_wav_len\", lengths.shape, np_to_triton_dtype(lengths.dtype)),\n protocol_client.InferInput(\"reference_text\", [1, 1], \"BYTES\"),\n protocol_client.InferInput(\"target_text\", [1, 1], \"BYTES\"),\n ]\n inputs[0].set_data_from_numpy(samples)\n inputs[1].set_data_from_numpy(lengths)\n\n input_data_numpy = np.array([reference_text], dtype=object)\n input_data_numpy = input_data_numpy.reshape((1, 1))\n inputs[2].set_data_from_numpy(input_data_numpy)\n\n input_data_numpy = np.array([target_text], dtype=object)\n input_data_numpy = input_data_numpy.reshape((1, 1))\n inputs[3].set_data_from_numpy(input_data_numpy)\n\n outputs = [protocol_client.InferRequestedOutput(\"waveform\")]\n\n sequence_id = 100000000 + i + task_id * 10\n start = time.time()\n response = await triton_client.infer(model_name, inputs, request_id=str(sequence_id), outputs=outputs)\n\n audio = response.as_numpy(\"waveform\").reshape(-1)\n\n end = time.time() - start\n\n audio_save_path = os.path.join(audio_save_dir, f\"{item['target_audio_path']}.wav\")\n sf.write(audio_save_path, audio, save_sample_rate, \"PCM_16\")\n\n actual_duration = len(audio) / save_sample_rate\n latency_data.append((end, actual_duration))\n total_duration += actual_duration\n\n return total_duration, latency_data\n\n\ndef load_manifests(manifest_path):\n with open(manifest_path, \"r\") as f:\n manifest_list = []\n for line in f:\n assert len(line.strip().split(\"|\")) == 4\n utt, prompt_text, prompt_wav, gt_text = line.strip().split(\"|\")\n utt = Path(utt).stem\n # gt_wav = os.path.join(os.path.dirname(manifest_path), \"wavs\", utt + \".wav\")\n if not os.path.isabs(prompt_wav):\n prompt_wav = os.path.join(os.path.dirname(manifest_path), prompt_wav)\n manifest_list.append(\n {\n \"audio_filepath\": prompt_wav,\n \"reference_text\": prompt_text,\n \"target_text\": gt_text,\n \"target_audio_path\": utt,\n }\n )\n return manifest_list\n\n\ndef split_data(data, k):\n n = len(data)\n if n < k:\n print(f\"Warning: the length of the input list ({n}) is less than k ({k}). Setting k to {n}.\")\n k = n\n\n quotient = n // k\n remainder = n % k\n\n result = []\n start = 0\n for i in range(k):\n if i < remainder:\n end = start + quotient + 1\n else:\n end = start + quotient\n\n result.append(data[start:end])\n start = end\n\n return result\n\n\nasync def main():\n args = get_args()\n url = f\"{args.server_addr}:{args.server_port}\"\n\n triton_client = grpcclient.InferenceServerClient(url=url, verbose=False)\n protocol_client = grpcclient\n\n if args.reference_audio:\n args.num_tasks = 1\n args.log_interval = 1\n manifest_item_list = [\n {\n \"reference_text\": args.reference_text,\n \"target_text\": args.target_text,\n \"audio_filepath\": args.reference_audio,\n \"target_audio_path\": \"test\",\n }\n ]\n elif args.huggingface_dataset:\n import datasets\n\n dataset = datasets.load_dataset(\n args.huggingface_dataset,\n split=args.split_name,\n trust_remote_code=True,\n )\n manifest_item_list = []\n for i in range(len(dataset)):\n manifest_item_list.append(\n {\n \"audio_filepath\": dataset[i][\"prompt_audio\"],\n \"reference_text\": dataset[i][\"prompt_text\"],\n \"target_audio_path\": dataset[i][\"id\"],\n \"target_text\": dataset[i][\"target_text\"],\n }\n )\n else:\n manifest_item_list = load_manifests(args.manifest_path)\n\n args.num_tasks = min(args.num_tasks, len(manifest_item_list))\n manifest_item_list = split_data(manifest_item_list, args.num_tasks)\n\n os.makedirs(args.log_dir, exist_ok=True)\n tasks = []\n start_time = time.time()\n for i in range(args.num_tasks):\n task = asyncio.create_task(\n send(\n manifest_item_list[i],\n name=f\"task-{i}\",\n triton_client=triton_client,\n protocol_client=protocol_client,\n log_interval=args.log_interval,\n model_name=args.model_name,\n audio_save_dir=args.log_dir,\n padding_duration=1,\n save_sample_rate=24000,\n )\n )\n tasks.append(task)\n\n ans_list = await asyncio.gather(*tasks)\n\n end_time = time.time()\n elapsed = end_time - start_time\n\n total_duration = 0.0\n latency_data = []\n for ans in ans_list:\n total_duration += ans[0]\n latency_data += ans[1]\n\n rtf = elapsed / total_duration\n\n s = f\"RTF: {rtf:.4f}\\n\"\n s += f\"total_duration: {total_duration:.3f} seconds\\n\"\n s += f\"({total_duration / 3600:.2f} hours)\\n\"\n s += f\"processing time: {elapsed:.3f} seconds ({elapsed / 3600:.2f} hours)\\n\"\n\n latency_list = [chunk_end for (chunk_end, chunk_duration) in latency_data]\n latency_ms = sum(latency_list) / float(len(latency_list)) * 1000.0\n latency_variance = np.var(latency_list, dtype=np.float64) * 1000.0\n s += f\"latency_variance: {latency_variance:.2f}\\n\"\n s += f\"latency_50_percentile_ms: {np.percentile(latency_list, 50) * 1000.0:.2f}\\n\"\n s += f\"latency_90_percentile_ms: {np.percentile(latency_list, 90) * 1000.0:.2f}\\n\"\n s += f\"latency_95_percentile_ms: {np.percentile(latency_list, 95) * 1000.0:.2f}\\n\"\n s += f\"latency_99_percentile_ms: {np.percentile(latency_list, 99) * 1000.0:.2f}\\n\"\n s += f\"average_latency_ms: {latency_ms:.2f}\\n\"\n\n print(s)\n if args.manifest_path:\n name = Path(args.manifest_path).stem\n elif args.split_name:\n name = args.split_name\n with open(f\"{args.log_dir}/rtf-{name}.txt\", \"w\") as f:\n f.write(s)\n\n stats = await triton_client.get_inference_statistics(model_name=\"\", as_json=True)\n write_triton_stats(stats, f\"{args.log_dir}/stats_summary-{name}.txt\")\n\n metadata = await triton_client.get_model_config(model_name=args.model_name, as_json=True)\n with open(f\"{args.log_dir}/model_config-{name}.json\", \"w\") as f:\n json.dump(metadata, f, indent=4)\n\n\nif __name__ == \"__main__\":\n asyncio.run(main())","source_hash":"feddef6c77b73f44f419571ae75dfae5dec9de0f58da03678cf071a7dc93ddc4","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.client_grpc.write_triton_stats","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.client_grpc.write_triton_stats#L59-L112","kind":"function","name":"write_triton_stats","path":"src/f5_tts/runtime/triton_trtllm/client_grpc.py","language":"python","start_line":59,"end_line":112,"context_start_line":39,"context_end_line":132,"code":" --huggingface-dataset yuekai/seed_tts \\\n --split-name wenetspeech4tts \\\n --log-dir ./log_concurrent_tasks_${num_task}\n\"\"\"\n\nimport argparse\nimport asyncio\nimport json\nimport os\nimport time\nimport types\nfrom pathlib import Path\n\nimport numpy as np\nimport soundfile as sf\nimport tritonclient\nimport tritonclient.grpc.aio as grpcclient\nfrom tritonclient.utils import np_to_triton_dtype\n\n\ndef write_triton_stats(stats, summary_file):\n with open(summary_file, \"w\") as summary_f:\n model_stats = stats[\"model_stats\"]\n # write a note, the log is from triton_client.get_inference_statistics(), to better human readability\n summary_f.write(\n \"The log is parsing from triton_client.get_inference_statistics(), to better human readability. \\n\"\n )\n summary_f.write(\"To learn more about the log, please refer to: \\n\")\n summary_f.write(\"1. https://github.com/triton-inference-server/server/blob/main/docs/user_guide/metrics.md \\n\")\n summary_f.write(\"2. https://github.com/triton-inference-server/server/issues/5374 \\n\\n\")\n summary_f.write(\n \"To better improve throughput, we always would like let requests wait in the queue for a while, and then execute them with a larger batch size. \\n\"\n )\n summary_f.write(\n \"However, there is a trade-off between the increased queue time and the increased batch size. \\n\"\n )\n summary_f.write(\n \"You may change 'max_queue_delay_microseconds' and 'preferred_batch_size' in the model configuration file to achieve this. \\n\"\n )\n summary_f.write(\n \"See https://github.com/triton-inference-server/server/blob/main/docs/user_guide/model_configuration.md#delayed-batching for more details. \\n\\n\"\n )\n for model_state in model_stats:\n if \"last_inference\" not in model_state:\n continue\n summary_f.write(f\"model name is {model_state['name']} \\n\")\n model_inference_stats = model_state[\"inference_stats\"]\n total_queue_time_s = int(model_inference_stats[\"queue\"][\"ns\"]) / 1e9\n total_infer_time_s = int(model_inference_stats[\"compute_infer\"][\"ns\"]) / 1e9\n total_input_time_s = int(model_inference_stats[\"compute_input\"][\"ns\"]) / 1e9\n total_output_time_s = int(model_inference_stats[\"compute_output\"][\"ns\"]) / 1e9\n summary_f.write(\n f\"queue time {total_queue_time_s:<5.2f} s, compute infer time {total_infer_time_s:<5.2f} s, compute input time {total_input_time_s:<5.2f} s, compute output time {total_output_time_s:<5.2f} s \\n\" # noqa\n )\n model_batch_stats = model_state[\"batch_stats\"]\n for batch in model_batch_stats:\n batch_size = int(batch[\"batch_size\"])\n compute_input = batch[\"compute_input\"]\n compute_output = batch[\"compute_output\"]\n compute_infer = batch[\"compute_infer\"]\n batch_count = int(compute_infer[\"count\"])\n assert compute_infer[\"count\"] == compute_output[\"count\"] == compute_input[\"count\"]\n compute_infer_time_ms = int(compute_infer[\"ns\"]) / 1e6\n compute_input_time_ms = int(compute_input[\"ns\"]) / 1e6\n compute_output_time_ms = int(compute_output[\"ns\"]) / 1e6\n summary_f.write(\n f\"execuate inference with batch_size {batch_size:<2} total {batch_count:<5} times, total_infer_time {compute_infer_time_ms:<9.2f} ms, avg_infer_time {compute_infer_time_ms:<9.2f}/{batch_count:<5}={compute_infer_time_ms / batch_count:.2f} ms, avg_infer_time_per_sample {compute_infer_time_ms:<9.2f}/{batch_count:<5}/{batch_size}={compute_infer_time_ms / batch_count / batch_size:.2f} ms \\n\" # noqa\n )\n summary_f.write(\n f\"input {compute_input_time_ms:<9.2f} ms, avg {compute_input_time_ms / batch_count:.2f} ms, \" # noqa\n )\n summary_f.write(\n f\"output {compute_output_time_ms:<9.2f} ms, avg {compute_output_time_ms / batch_count:.2f} ms \\n\" # noqa\n )\n\n\ndef get_args():\n parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)\n\n parser.add_argument(\n \"--server-addr\",\n type=str,\n default=\"localhost\",\n help=\"Address of the server\",\n )\n\n parser.add_argument(\n \"--server-port\",\n type=int,\n default=8001,\n help=\"Grpc port of the triton server, default is 8001\",\n )\n\n parser.add_argument(","source_hash":"feddef6c77b73f44f419571ae75dfae5dec9de0f58da03678cf071a7dc93ddc4","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.client_grpc.get_args","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.client_grpc.get_args#L115-L220","kind":"function","name":"get_args","path":"src/f5_tts/runtime/triton_trtllm/client_grpc.py","language":"python","start_line":115,"end_line":220,"context_start_line":95,"context_end_line":240,"code":" batch_size = int(batch[\"batch_size\"])\n compute_input = batch[\"compute_input\"]\n compute_output = batch[\"compute_output\"]\n compute_infer = batch[\"compute_infer\"]\n batch_count = int(compute_infer[\"count\"])\n assert compute_infer[\"count\"] == compute_output[\"count\"] == compute_input[\"count\"]\n compute_infer_time_ms = int(compute_infer[\"ns\"]) / 1e6\n compute_input_time_ms = int(compute_input[\"ns\"]) / 1e6\n compute_output_time_ms = int(compute_output[\"ns\"]) / 1e6\n summary_f.write(\n f\"execuate inference with batch_size {batch_size:<2} total {batch_count:<5} times, total_infer_time {compute_infer_time_ms:<9.2f} ms, avg_infer_time {compute_infer_time_ms:<9.2f}/{batch_count:<5}={compute_infer_time_ms / batch_count:.2f} ms, avg_infer_time_per_sample {compute_infer_time_ms:<9.2f}/{batch_count:<5}/{batch_size}={compute_infer_time_ms / batch_count / batch_size:.2f} ms \\n\" # noqa\n )\n summary_f.write(\n f\"input {compute_input_time_ms:<9.2f} ms, avg {compute_input_time_ms / batch_count:.2f} ms, \" # noqa\n )\n summary_f.write(\n f\"output {compute_output_time_ms:<9.2f} ms, avg {compute_output_time_ms / batch_count:.2f} ms \\n\" # noqa\n )\n\n\ndef get_args():\n parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)\n\n parser.add_argument(\n \"--server-addr\",\n type=str,\n default=\"localhost\",\n help=\"Address of the server\",\n )\n\n parser.add_argument(\n \"--server-port\",\n type=int,\n default=8001,\n help=\"Grpc port of the triton server, default is 8001\",\n )\n\n parser.add_argument(\n \"--reference-audio\",\n type=str,\n default=None,\n help=\"Path to a single audio file. It can't be specified at the same time with --manifest-dir\",\n )\n\n parser.add_argument(\n \"--reference-text\",\n type=str,\n default=\"\",\n help=\"\",\n )\n\n parser.add_argument(\n \"--target-text\",\n type=str,\n default=\"\",\n help=\"\",\n )\n\n parser.add_argument(\n \"--huggingface-dataset\",\n type=str,\n default=\"yuekai/seed_tts\",\n help=\"dataset name in huggingface dataset hub\",\n )\n\n parser.add_argument(\n \"--split-name\",\n type=str,\n default=\"wenetspeech4tts\",\n choices=[\"wenetspeech4tts\", \"test_zh\", \"test_en\", \"test_hard\"],\n help=\"dataset split name, default is 'test'\",\n )\n\n parser.add_argument(\n \"--manifest-path\",\n type=str,\n default=None,\n help=\"Path to the manifest dir which includes wav.scp trans.txt files.\",\n )\n\n parser.add_argument(\n \"--model-name\",\n type=str,\n default=\"f5_tts\",\n choices=[\"f5_tts\", \"spark_tts\"],\n help=\"triton model_repo module name to request: transducer for k2, attention_rescoring for wenet offline, streaming_wenet for wenet streaming, infer_pipeline for paraformer large offline\",\n )\n\n parser.add_argument(\n \"--num-tasks\",\n type=int,\n default=1,\n help=\"Number of concurrent tasks for sending\",\n )\n\n parser.add_argument(\n \"--log-interval\",\n type=int,\n default=5,\n help=\"Controls how frequently we print the log.\",\n )\n\n parser.add_argument(\n \"--compute-wer\",\n action=\"store_true\",\n default=False,\n help=\"\"\"True to compute WER.\n \"\"\",\n )\n\n parser.add_argument(\n \"--log-dir\",\n type=str,\n required=False,\n default=\"./tmp\",\n help=\"log directory\",\n )\n\n parser.add_argument(\n \"--batch-size\",\n type=int,\n default=1,\n help=\"Inference batch_size per request for offline mode.\",\n )\n\n return parser.parse_args()\n\n\ndef load_audio(wav_path, target_sample_rate=24000):\n assert target_sample_rate == 24000, \"hard coding in server\"\n if isinstance(wav_path, dict):\n waveform = wav_path[\"array\"]\n sample_rate = wav_path[\"sampling_rate\"]\n else:\n waveform, sample_rate = sf.read(wav_path)\n if sample_rate != target_sample_rate:\n from scipy.signal import resample\n\n num_samples = int(len(waveform) * (target_sample_rate / sample_rate))\n waveform = resample(waveform, num_samples)\n return waveform, target_sample_rate\n\n\nasync def send(\n manifest_item_list: list,\n name: str,","source_hash":"feddef6c77b73f44f419571ae75dfae5dec9de0f58da03678cf071a7dc93ddc4","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.client_grpc.load_audio","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.client_grpc.load_audio#L223-L235","kind":"function","name":"load_audio","path":"src/f5_tts/runtime/triton_trtllm/client_grpc.py","language":"python","start_line":223,"end_line":235,"context_start_line":203,"context_end_line":255,"code":" )\n\n parser.add_argument(\n \"--log-dir\",\n type=str,\n required=False,\n default=\"./tmp\",\n help=\"log directory\",\n )\n\n parser.add_argument(\n \"--batch-size\",\n type=int,\n default=1,\n help=\"Inference batch_size per request for offline mode.\",\n )\n\n return parser.parse_args()\n\n\ndef load_audio(wav_path, target_sample_rate=24000):\n assert target_sample_rate == 24000, \"hard coding in server\"\n if isinstance(wav_path, dict):\n waveform = wav_path[\"array\"]\n sample_rate = wav_path[\"sampling_rate\"]\n else:\n waveform, sample_rate = sf.read(wav_path)\n if sample_rate != target_sample_rate:\n from scipy.signal import resample\n\n num_samples = int(len(waveform) * (target_sample_rate / sample_rate))\n waveform = resample(waveform, num_samples)\n return waveform, target_sample_rate\n\n\nasync def send(\n manifest_item_list: list,\n name: str,\n triton_client: tritonclient.grpc.aio.InferenceServerClient,\n protocol_client: types.ModuleType,\n log_interval: int,\n model_name: str,\n padding_duration: int = None,\n audio_save_dir: str = \"./\",\n save_sample_rate: int = 24000,\n):\n total_duration = 0.0\n latency_data = []\n task_id = int(name[5:])\n\n print(f\"manifest_item_list: {manifest_item_list}\")\n for i, item in enumerate(manifest_item_list):\n if i % log_interval == 0:","source_hash":"feddef6c77b73f44f419571ae75dfae5dec9de0f58da03678cf071a7dc93ddc4","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.client_grpc.send","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.client_grpc.send#L238-L317","kind":"function","name":"send","path":"src/f5_tts/runtime/triton_trtllm/client_grpc.py","language":"python","start_line":238,"end_line":317,"context_start_line":218,"context_end_line":337,"code":" )\n\n return parser.parse_args()\n\n\ndef load_audio(wav_path, target_sample_rate=24000):\n assert target_sample_rate == 24000, \"hard coding in server\"\n if isinstance(wav_path, dict):\n waveform = wav_path[\"array\"]\n sample_rate = wav_path[\"sampling_rate\"]\n else:\n waveform, sample_rate = sf.read(wav_path)\n if sample_rate != target_sample_rate:\n from scipy.signal import resample\n\n num_samples = int(len(waveform) * (target_sample_rate / sample_rate))\n waveform = resample(waveform, num_samples)\n return waveform, target_sample_rate\n\n\nasync def send(\n manifest_item_list: list,\n name: str,\n triton_client: tritonclient.grpc.aio.InferenceServerClient,\n protocol_client: types.ModuleType,\n log_interval: int,\n model_name: str,\n padding_duration: int = None,\n audio_save_dir: str = \"./\",\n save_sample_rate: int = 24000,\n):\n total_duration = 0.0\n latency_data = []\n task_id = int(name[5:])\n\n print(f\"manifest_item_list: {manifest_item_list}\")\n for i, item in enumerate(manifest_item_list):\n if i % log_interval == 0:\n print(f\"{name}: {i}/{len(manifest_item_list)}\")\n waveform, sample_rate = load_audio(item[\"audio_filepath\"], target_sample_rate=24000)\n duration = len(waveform) / sample_rate\n lengths = np.array([[len(waveform)]], dtype=np.int32)\n\n reference_text, target_text = item[\"reference_text\"], item[\"target_text\"]\n\n estimated_target_duration = duration / len(reference_text) * len(target_text)\n\n if padding_duration:\n # padding to nearset 10 seconds\n samples = np.zeros(\n (\n 1,\n padding_duration\n * sample_rate\n * ((int(estimated_target_duration + duration) // padding_duration) + 1),\n ),\n dtype=np.float32,\n )\n\n samples[0, : len(waveform)] = waveform\n else:\n samples = waveform\n\n samples = samples.reshape(1, -1).astype(np.float32)\n\n inputs = [\n protocol_client.InferInput(\"reference_wav\", samples.shape, np_to_triton_dtype(samples.dtype)),\n protocol_client.InferInput(\"reference_wav_len\", lengths.shape, np_to_triton_dtype(lengths.dtype)),\n protocol_client.InferInput(\"reference_text\", [1, 1], \"BYTES\"),\n protocol_client.InferInput(\"target_text\", [1, 1], \"BYTES\"),\n ]\n inputs[0].set_data_from_numpy(samples)\n inputs[1].set_data_from_numpy(lengths)\n\n input_data_numpy = np.array([reference_text], dtype=object)\n input_data_numpy = input_data_numpy.reshape((1, 1))\n inputs[2].set_data_from_numpy(input_data_numpy)\n\n input_data_numpy = np.array([target_text], dtype=object)\n input_data_numpy = input_data_numpy.reshape((1, 1))\n inputs[3].set_data_from_numpy(input_data_numpy)\n\n outputs = [protocol_client.InferRequestedOutput(\"waveform\")]\n\n sequence_id = 100000000 + i + task_id * 10\n start = time.time()\n response = await triton_client.infer(model_name, inputs, request_id=str(sequence_id), outputs=outputs)\n\n audio = response.as_numpy(\"waveform\").reshape(-1)\n\n end = time.time() - start\n\n audio_save_path = os.path.join(audio_save_dir, f\"{item['target_audio_path']}.wav\")\n sf.write(audio_save_path, audio, save_sample_rate, \"PCM_16\")\n\n actual_duration = len(audio) / save_sample_rate\n latency_data.append((end, actual_duration))\n total_duration += actual_duration\n\n return total_duration, latency_data\n\n\ndef load_manifests(manifest_path):\n with open(manifest_path, \"r\") as f:\n manifest_list = []\n for line in f:\n assert len(line.strip().split(\"|\")) == 4\n utt, prompt_text, prompt_wav, gt_text = line.strip().split(\"|\")\n utt = Path(utt).stem\n # gt_wav = os.path.join(os.path.dirname(manifest_path), \"wavs\", utt + \".wav\")\n if not os.path.isabs(prompt_wav):\n prompt_wav = os.path.join(os.path.dirname(manifest_path), prompt_wav)\n manifest_list.append(\n {\n \"audio_filepath\": prompt_wav,\n \"reference_text\": prompt_text,\n \"target_text\": gt_text,\n \"target_audio_path\": utt,\n }\n )","source_hash":"feddef6c77b73f44f419571ae75dfae5dec9de0f58da03678cf071a7dc93ddc4","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.client_grpc.load_manifests","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.client_grpc.load_manifests#L320-L338","kind":"function","name":"load_manifests","path":"src/f5_tts/runtime/triton_trtllm/client_grpc.py","language":"python","start_line":320,"end_line":338,"context_start_line":300,"context_end_line":358,"code":" outputs = [protocol_client.InferRequestedOutput(\"waveform\")]\n\n sequence_id = 100000000 + i + task_id * 10\n start = time.time()\n response = await triton_client.infer(model_name, inputs, request_id=str(sequence_id), outputs=outputs)\n\n audio = response.as_numpy(\"waveform\").reshape(-1)\n\n end = time.time() - start\n\n audio_save_path = os.path.join(audio_save_dir, f\"{item['target_audio_path']}.wav\")\n sf.write(audio_save_path, audio, save_sample_rate, \"PCM_16\")\n\n actual_duration = len(audio) / save_sample_rate\n latency_data.append((end, actual_duration))\n total_duration += actual_duration\n\n return total_duration, latency_data\n\n\ndef load_manifests(manifest_path):\n with open(manifest_path, \"r\") as f:\n manifest_list = []\n for line in f:\n assert len(line.strip().split(\"|\")) == 4\n utt, prompt_text, prompt_wav, gt_text = line.strip().split(\"|\")\n utt = Path(utt).stem\n # gt_wav = os.path.join(os.path.dirname(manifest_path), \"wavs\", utt + \".wav\")\n if not os.path.isabs(prompt_wav):\n prompt_wav = os.path.join(os.path.dirname(manifest_path), prompt_wav)\n manifest_list.append(\n {\n \"audio_filepath\": prompt_wav,\n \"reference_text\": prompt_text,\n \"target_text\": gt_text,\n \"target_audio_path\": utt,\n }\n )\n return manifest_list\n\n\ndef split_data(data, k):\n n = len(data)\n if n < k:\n print(f\"Warning: the length of the input list ({n}) is less than k ({k}). Setting k to {n}.\")\n k = n\n\n quotient = n // k\n remainder = n % k\n\n result = []\n start = 0\n for i in range(k):\n if i < remainder:\n end = start + quotient + 1\n else:\n end = start + quotient\n\n result.append(data[start:end])","source_hash":"feddef6c77b73f44f419571ae75dfae5dec9de0f58da03678cf071a7dc93ddc4","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.client_grpc.split_data","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.client_grpc.split_data#L341-L361","kind":"function","name":"split_data","path":"src/f5_tts/runtime/triton_trtllm/client_grpc.py","language":"python","start_line":341,"end_line":361,"context_start_line":321,"context_end_line":381,"code":" with open(manifest_path, \"r\") as f:\n manifest_list = []\n for line in f:\n assert len(line.strip().split(\"|\")) == 4\n utt, prompt_text, prompt_wav, gt_text = line.strip().split(\"|\")\n utt = Path(utt).stem\n # gt_wav = os.path.join(os.path.dirname(manifest_path), \"wavs\", utt + \".wav\")\n if not os.path.isabs(prompt_wav):\n prompt_wav = os.path.join(os.path.dirname(manifest_path), prompt_wav)\n manifest_list.append(\n {\n \"audio_filepath\": prompt_wav,\n \"reference_text\": prompt_text,\n \"target_text\": gt_text,\n \"target_audio_path\": utt,\n }\n )\n return manifest_list\n\n\ndef split_data(data, k):\n n = len(data)\n if n < k:\n print(f\"Warning: the length of the input list ({n}) is less than k ({k}). Setting k to {n}.\")\n k = n\n\n quotient = n // k\n remainder = n % k\n\n result = []\n start = 0\n for i in range(k):\n if i < remainder:\n end = start + quotient + 1\n else:\n end = start + quotient\n\n result.append(data[start:end])\n start = end\n\n return result\n\n\nasync def main():\n args = get_args()\n url = f\"{args.server_addr}:{args.server_port}\"\n\n triton_client = grpcclient.InferenceServerClient(url=url, verbose=False)\n protocol_client = grpcclient\n\n if args.reference_audio:\n args.num_tasks = 1\n args.log_interval = 1\n manifest_item_list = [\n {\n \"reference_text\": args.reference_text,\n \"target_text\": args.target_text,\n \"audio_filepath\": args.reference_audio,\n \"target_audio_path\": \"test\",\n }\n ]","source_hash":"feddef6c77b73f44f419571ae75dfae5dec9de0f58da03678cf071a7dc93ddc4","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.client_grpc.main","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.client_grpc.main#L364-L466","kind":"function","name":"main","path":"src/f5_tts/runtime/triton_trtllm/client_grpc.py","language":"python","start_line":364,"end_line":466,"context_start_line":344,"context_end_line":470,"code":" print(f\"Warning: the length of the input list ({n}) is less than k ({k}). Setting k to {n}.\")\n k = n\n\n quotient = n // k\n remainder = n % k\n\n result = []\n start = 0\n for i in range(k):\n if i < remainder:\n end = start + quotient + 1\n else:\n end = start + quotient\n\n result.append(data[start:end])\n start = end\n\n return result\n\n\nasync def main():\n args = get_args()\n url = f\"{args.server_addr}:{args.server_port}\"\n\n triton_client = grpcclient.InferenceServerClient(url=url, verbose=False)\n protocol_client = grpcclient\n\n if args.reference_audio:\n args.num_tasks = 1\n args.log_interval = 1\n manifest_item_list = [\n {\n \"reference_text\": args.reference_text,\n \"target_text\": args.target_text,\n \"audio_filepath\": args.reference_audio,\n \"target_audio_path\": \"test\",\n }\n ]\n elif args.huggingface_dataset:\n import datasets\n\n dataset = datasets.load_dataset(\n args.huggingface_dataset,\n split=args.split_name,\n trust_remote_code=True,\n )\n manifest_item_list = []\n for i in range(len(dataset)):\n manifest_item_list.append(\n {\n \"audio_filepath\": dataset[i][\"prompt_audio\"],\n \"reference_text\": dataset[i][\"prompt_text\"],\n \"target_audio_path\": dataset[i][\"id\"],\n \"target_text\": dataset[i][\"target_text\"],\n }\n )\n else:\n manifest_item_list = load_manifests(args.manifest_path)\n\n args.num_tasks = min(args.num_tasks, len(manifest_item_list))\n manifest_item_list = split_data(manifest_item_list, args.num_tasks)\n\n os.makedirs(args.log_dir, exist_ok=True)\n tasks = []\n start_time = time.time()\n for i in range(args.num_tasks):\n task = asyncio.create_task(\n send(\n manifest_item_list[i],\n name=f\"task-{i}\",\n triton_client=triton_client,\n protocol_client=protocol_client,\n log_interval=args.log_interval,\n model_name=args.model_name,\n audio_save_dir=args.log_dir,\n padding_duration=1,\n save_sample_rate=24000,\n )\n )\n tasks.append(task)\n\n ans_list = await asyncio.gather(*tasks)\n\n end_time = time.time()\n elapsed = end_time - start_time\n\n total_duration = 0.0\n latency_data = []\n for ans in ans_list:\n total_duration += ans[0]\n latency_data += ans[1]\n\n rtf = elapsed / total_duration\n\n s = f\"RTF: {rtf:.4f}\\n\"\n s += f\"total_duration: {total_duration:.3f} seconds\\n\"\n s += f\"({total_duration / 3600:.2f} hours)\\n\"\n s += f\"processing time: {elapsed:.3f} seconds ({elapsed / 3600:.2f} hours)\\n\"\n\n latency_list = [chunk_end for (chunk_end, chunk_duration) in latency_data]\n latency_ms = sum(latency_list) / float(len(latency_list)) * 1000.0\n latency_variance = np.var(latency_list, dtype=np.float64) * 1000.0\n s += f\"latency_variance: {latency_variance:.2f}\\n\"\n s += f\"latency_50_percentile_ms: {np.percentile(latency_list, 50) * 1000.0:.2f}\\n\"\n s += f\"latency_90_percentile_ms: {np.percentile(latency_list, 90) * 1000.0:.2f}\\n\"\n s += f\"latency_95_percentile_ms: {np.percentile(latency_list, 95) * 1000.0:.2f}\\n\"\n s += f\"latency_99_percentile_ms: {np.percentile(latency_list, 99) * 1000.0:.2f}\\n\"\n s += f\"average_latency_ms: {latency_ms:.2f}\\n\"\n\n print(s)\n if args.manifest_path:\n name = Path(args.manifest_path).stem\n elif args.split_name:\n name = args.split_name\n with open(f\"{args.log_dir}/rtf-{name}.txt\", \"w\") as f:\n f.write(s)\n\n stats = await triton_client.get_inference_statistics(model_name=\"\", as_json=True)\n write_triton_stats(stats, f\"{args.log_dir}/stats_summary-{name}.txt\")\n\n metadata = await triton_client.get_model_config(model_name=args.model_name, as_json=True)\n with open(f\"{args.log_dir}/model_config-{name}.json\", \"w\") as f:\n json.dump(metadata, f, indent=4)\n\n\nif __name__ == \"__main__\":\n asyncio.run(main())","source_hash":"feddef6c77b73f44f419571ae75dfae5dec9de0f58da03678cf071a7dc93ddc4","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.client_http","uri":"program://DMOSpeech2/module/src.f5_tts.runtime.triton_trtllm.client_http#L1-L143","kind":"module","name":"src.f5_tts.runtime.triton_trtllm.client_http","path":"src/f5_tts/runtime/triton_trtllm/client_http.py","language":"python","start_line":1,"end_line":143,"context_start_line":1,"context_end_line":143,"code":"# Copyright 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.\n#\n# Redistribution and use in source and binary forms, with or without\n# modification, are permitted provided that the following conditions\n# are met:\n# * Redistributions of source code must retain the above copyright\n# notice, this list of conditions and the following disclaimer.\n# * Redistributions in binary form must reproduce the above copyright\n# notice, this list of conditions and the following disclaimer in the\n# documentation and/or other materials provided with the distribution.\n# * Neither the name of NVIDIA CORPORATION nor the names of its\n# contributors may be used to endorse or promote products derived\n# from this software without specific prior written permission.\n#\n# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY\n# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE\n# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR\n# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR\n# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,\n# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,\n# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR\n# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY\n# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT\n# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE\n# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\nimport argparse\n\nimport numpy as np\nimport requests\nimport soundfile as sf\n\n\ndef get_args():\n parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)\n\n parser.add_argument(\n \"--server-url\",\n type=str,\n default=\"localhost:8000\",\n help=\"Address of the server\",\n )\n\n parser.add_argument(\n \"--reference-audio\",\n type=str,\n default=\"../../infer/examples/basic/basic_ref_en.wav\",\n help=\"Path to a single audio file. It can't be specified at the same time with --manifest-dir\",\n )\n\n parser.add_argument(\n \"--reference-text\",\n type=str,\n default=\"Some call me nature, others call me mother nature.\",\n help=\"\",\n )\n\n parser.add_argument(\n \"--target-text\",\n type=str,\n default=\"I don't really care what you call me. I've been a silent spectator, watching species evolve, empires rise and fall. But always remember, I am mighty and enduring.\",\n help=\"\",\n )\n\n parser.add_argument(\n \"--model-name\",\n type=str,\n default=\"f5_tts\",\n choices=[\"f5_tts\", \"spark_tts\"],\n help=\"triton model_repo module name to request\",\n )\n\n parser.add_argument(\n \"--output-audio\",\n type=str,\n default=\"output.wav\",\n help=\"Path to save the output audio\",\n )\n return parser.parse_args()\n\n\ndef prepare_request(\n samples,\n reference_text,\n target_text,\n sample_rate=24000,\n audio_save_dir: str = \"./\",\n):\n assert len(samples.shape) == 1, \"samples should be 1D\"\n lengths = np.array([[len(samples)]], dtype=np.int32)\n samples = samples.reshape(1, -1).astype(np.float32)\n\n data = {\n \"inputs\": [\n {\"name\": \"reference_wav\", \"shape\": samples.shape, \"datatype\": \"FP32\", \"data\": samples.tolist()},\n {\n \"name\": \"reference_wav_len\",\n \"shape\": lengths.shape,\n \"datatype\": \"INT32\",\n \"data\": lengths.tolist(),\n },\n {\"name\": \"reference_text\", \"shape\": [1, 1], \"datatype\": \"BYTES\", \"data\": [reference_text]},\n {\"name\": \"target_text\", \"shape\": [1, 1], \"datatype\": \"BYTES\", \"data\": [target_text]},\n ]\n }\n\n return data\n\n\ndef load_audio(wav_path, target_sample_rate=24000):\n assert target_sample_rate == 24000, \"hard coding in server\"\n if isinstance(wav_path, dict):\n samples = wav_path[\"array\"]\n sample_rate = wav_path[\"sampling_rate\"]\n else:\n samples, sample_rate = sf.read(wav_path)\n if sample_rate != target_sample_rate:\n from scipy.signal import resample\n\n num_samples = int(len(samples) * (target_sample_rate / sample_rate))\n samples = resample(samples, num_samples)\n return samples, target_sample_rate\n\n\nif __name__ == \"__main__\":\n args = get_args()\n server_url = args.server_url\n if not server_url.startswith((\"http://\", \"https://\")):\n server_url = f\"http://{server_url}\"\n\n url = f\"{server_url}/v2/models/{args.model_name}/infer\"\n samples, sr = load_audio(args.reference_audio)\n assert sr == 24000, \"sample rate hardcoded in server\"\n\n samples = np.array(samples, dtype=np.float32)\n data = prepare_request(samples, args.reference_text, args.target_text)\n\n rsp = requests.post(\n url, headers={\"Content-Type\": \"application/json\"}, json=data, verify=False, params={\"request_id\": \"0\"}\n )\n result = rsp.json()\n audio = result[\"outputs\"][0][\"data\"]\n audio = np.array(audio, dtype=np.float32)\n sf.write(args.output_audio, audio, 24000, \"PCM_16\")","source_hash":"d1af65a41400201a85a18404e74cb456991550d79a17c42c106c37be2a18e39f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.client_http.get_args","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.client_http.get_args#L33-L78","kind":"function","name":"get_args","path":"src/f5_tts/runtime/triton_trtllm/client_http.py","language":"python","start_line":33,"end_line":78,"context_start_line":13,"context_end_line":98,"code":"# from this software without specific prior written permission.\n#\n# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY\n# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE\n# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR\n# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR\n# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,\n# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,\n# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR\n# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY\n# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT\n# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE\n# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\nimport argparse\n\nimport numpy as np\nimport requests\nimport soundfile as sf\n\n\ndef get_args():\n parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)\n\n parser.add_argument(\n \"--server-url\",\n type=str,\n default=\"localhost:8000\",\n help=\"Address of the server\",\n )\n\n parser.add_argument(\n \"--reference-audio\",\n type=str,\n default=\"../../infer/examples/basic/basic_ref_en.wav\",\n help=\"Path to a single audio file. It can't be specified at the same time with --manifest-dir\",\n )\n\n parser.add_argument(\n \"--reference-text\",\n type=str,\n default=\"Some call me nature, others call me mother nature.\",\n help=\"\",\n )\n\n parser.add_argument(\n \"--target-text\",\n type=str,\n default=\"I don't really care what you call me. I've been a silent spectator, watching species evolve, empires rise and fall. But always remember, I am mighty and enduring.\",\n help=\"\",\n )\n\n parser.add_argument(\n \"--model-name\",\n type=str,\n default=\"f5_tts\",\n choices=[\"f5_tts\", \"spark_tts\"],\n help=\"triton model_repo module name to request\",\n )\n\n parser.add_argument(\n \"--output-audio\",\n type=str,\n default=\"output.wav\",\n help=\"Path to save the output audio\",\n )\n return parser.parse_args()\n\n\ndef prepare_request(\n samples,\n reference_text,\n target_text,\n sample_rate=24000,\n audio_save_dir: str = \"./\",\n):\n assert len(samples.shape) == 1, \"samples should be 1D\"\n lengths = np.array([[len(samples)]], dtype=np.int32)\n samples = samples.reshape(1, -1).astype(np.float32)\n\n data = {\n \"inputs\": [\n {\"name\": \"reference_wav\", \"shape\": samples.shape, \"datatype\": \"FP32\", \"data\": samples.tolist()},\n {\n \"name\": \"reference_wav_len\",\n \"shape\": lengths.shape,\n \"datatype\": \"INT32\",","source_hash":"d1af65a41400201a85a18404e74cb456991550d79a17c42c106c37be2a18e39f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.client_http.prepare_request","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.client_http.prepare_request#L81-L106","kind":"function","name":"prepare_request","path":"src/f5_tts/runtime/triton_trtllm/client_http.py","language":"python","start_line":81,"end_line":106,"context_start_line":61,"context_end_line":126,"code":" help=\"\",\n )\n\n parser.add_argument(\n \"--model-name\",\n type=str,\n default=\"f5_tts\",\n choices=[\"f5_tts\", \"spark_tts\"],\n help=\"triton model_repo module name to request\",\n )\n\n parser.add_argument(\n \"--output-audio\",\n type=str,\n default=\"output.wav\",\n help=\"Path to save the output audio\",\n )\n return parser.parse_args()\n\n\ndef prepare_request(\n samples,\n reference_text,\n target_text,\n sample_rate=24000,\n audio_save_dir: str = \"./\",\n):\n assert len(samples.shape) == 1, \"samples should be 1D\"\n lengths = np.array([[len(samples)]], dtype=np.int32)\n samples = samples.reshape(1, -1).astype(np.float32)\n\n data = {\n \"inputs\": [\n {\"name\": \"reference_wav\", \"shape\": samples.shape, \"datatype\": \"FP32\", \"data\": samples.tolist()},\n {\n \"name\": \"reference_wav_len\",\n \"shape\": lengths.shape,\n \"datatype\": \"INT32\",\n \"data\": lengths.tolist(),\n },\n {\"name\": \"reference_text\", \"shape\": [1, 1], \"datatype\": \"BYTES\", \"data\": [reference_text]},\n {\"name\": \"target_text\", \"shape\": [1, 1], \"datatype\": \"BYTES\", \"data\": [target_text]},\n ]\n }\n\n return data\n\n\ndef load_audio(wav_path, target_sample_rate=24000):\n assert target_sample_rate == 24000, \"hard coding in server\"\n if isinstance(wav_path, dict):\n samples = wav_path[\"array\"]\n sample_rate = wav_path[\"sampling_rate\"]\n else:\n samples, sample_rate = sf.read(wav_path)\n if sample_rate != target_sample_rate:\n from scipy.signal import resample\n\n num_samples = int(len(samples) * (target_sample_rate / sample_rate))\n samples = resample(samples, num_samples)\n return samples, target_sample_rate\n\n\nif __name__ == \"__main__\":\n args = get_args()\n server_url = args.server_url","source_hash":"d1af65a41400201a85a18404e74cb456991550d79a17c42c106c37be2a18e39f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.client_http.load_audio","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.client_http.load_audio#L109-L121","kind":"function","name":"load_audio","path":"src/f5_tts/runtime/triton_trtllm/client_http.py","language":"python","start_line":109,"end_line":121,"context_start_line":89,"context_end_line":141,"code":" lengths = np.array([[len(samples)]], dtype=np.int32)\n samples = samples.reshape(1, -1).astype(np.float32)\n\n data = {\n \"inputs\": [\n {\"name\": \"reference_wav\", \"shape\": samples.shape, \"datatype\": \"FP32\", \"data\": samples.tolist()},\n {\n \"name\": \"reference_wav_len\",\n \"shape\": lengths.shape,\n \"datatype\": \"INT32\",\n \"data\": lengths.tolist(),\n },\n {\"name\": \"reference_text\", \"shape\": [1, 1], \"datatype\": \"BYTES\", \"data\": [reference_text]},\n {\"name\": \"target_text\", \"shape\": [1, 1], \"datatype\": \"BYTES\", \"data\": [target_text]},\n ]\n }\n\n return data\n\n\ndef load_audio(wav_path, target_sample_rate=24000):\n assert target_sample_rate == 24000, \"hard coding in server\"\n if isinstance(wav_path, dict):\n samples = wav_path[\"array\"]\n sample_rate = wav_path[\"sampling_rate\"]\n else:\n samples, sample_rate = sf.read(wav_path)\n if sample_rate != target_sample_rate:\n from scipy.signal import resample\n\n num_samples = int(len(samples) * (target_sample_rate / sample_rate))\n samples = resample(samples, num_samples)\n return samples, target_sample_rate\n\n\nif __name__ == \"__main__\":\n args = get_args()\n server_url = args.server_url\n if not server_url.startswith((\"http://\", \"https://\")):\n server_url = f\"http://{server_url}\"\n\n url = f\"{server_url}/v2/models/{args.model_name}/infer\"\n samples, sr = load_audio(args.reference_audio)\n assert sr == 24000, \"sample rate hardcoded in server\"\n\n samples = np.array(samples, dtype=np.float32)\n data = prepare_request(samples, args.reference_text, args.target_text)\n\n rsp = requests.post(\n url, headers={\"Content-Type\": \"application/json\"}, json=data, verify=False, params={\"request_id\": \"0\"}\n )\n result = rsp.json()\n audio = result[\"outputs\"][0][\"data\"]","source_hash":"d1af65a41400201a85a18404e74cb456991550d79a17c42c106c37be2a18e39f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm","uri":"program://DMOSpeech2/module/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm#L1-L430","kind":"module","name":"src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":1,"end_line":430,"context_start_line":1,"context_end_line":430,"code":"import math\nimport os\nimport time\nfrom functools import wraps\nfrom typing import List, Optional\n\nimport tensorrt as trt\nimport tensorrt_llm\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nfrom tensorrt_llm._utils import str_dtype_to_torch, trt_dtype_to_torch\nfrom tensorrt_llm.logger import logger\nfrom tensorrt_llm.runtime.session import Session\n\n\ndef remove_tensor_padding(input_tensor, input_tensor_lengths=None):\n # Audio tensor case: batch, seq_len, feature_len\n # position_ids case: batch, seq_len\n assert input_tensor_lengths is not None, \"input_tensor_lengths must be provided for 3D input_tensor\"\n\n # Initialize a list to collect valid sequences\n valid_sequences = []\n\n for i in range(input_tensor.shape[0]):\n valid_length = input_tensor_lengths[i]\n valid_sequences.append(input_tensor[i, :valid_length])\n\n # Concatenate all valid sequences along the batch dimension\n output_tensor = torch.cat(valid_sequences, dim=0).contiguous()\n return output_tensor\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, text_num_embeds, text_dim, conv_layers=0, conv_mult=2, precompute_max_pos=4096):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim) # use 0 as filler token\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(text_dim, precompute_max_pos), persistent=False)\n self.text_blocks = nn.Sequential(*[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)])\n\n def forward(self, text):\n # only keep tensors with value not -1\n text_mask = text != -1\n text_pad_cut_off_index = text_mask.sum(dim=1).max()\n\n text = text[:, :text_pad_cut_off_index]\n text = self.text_embed(text)\n text = text + self.freqs_cis[: text.shape[1], :]\n for block in self.text_blocks:\n text = block(text)\n # padding text to the original length\n # text shape: B,seq_len,C\n # pad at the second dimension\n text = F.pad(text, (0, 0, 0, text_mask.shape[1] - text.shape[1], 0, 0), value=0)\n return text\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.gamma = nn.Parameter(torch.zeros(1, 1, dim))\n self.beta = nn.Parameter(torch.zeros(1, 1, dim))\n\n def forward(self, x):\n Gx = torch.norm(x, p=2, dim=1, keepdim=True)\n Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)\n return self.gamma * (x * Nx) + self.beta + x\n\n\nclass ConvNeXtV2Block(nn.Module):\n def __init__(\n self,\n dim: int,\n intermediate_dim: int,\n dilation: int = 1,\n ):\n super().__init__()\n padding = (dilation * (7 - 1)) // 2\n self.dwconv = nn.Conv1d(\n dim, dim, kernel_size=7, padding=padding, groups=dim, dilation=dilation\n ) # depthwise conv\n self.norm = nn.LayerNorm(dim, eps=1e-6)\n self.pwconv1 = nn.Linear(dim, intermediate_dim) # pointwise/1x1 convs, implemented with linear layers\n self.act = nn.GELU()\n self.grn = GRN(intermediate_dim)\n self.pwconv2 = nn.Linear(intermediate_dim, dim)\n\n def forward(self, x: torch.Tensor) -> torch.Tensor:\n residual = x\n x = x.transpose(1, 2) # b n d -> b d n\n x = self.dwconv(x)\n x = x.transpose(1, 2) # b d n -> b n d\n x = self.norm(x)\n x = self.pwconv1(x)\n x = self.act(x)\n x = self.grn(x)\n x = self.pwconv2(x)\n return residual + x\n\n\ndef precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0):\n # proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning\n # has some connection to NTK literature\n # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/\n # https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py\n theta *= theta_rescale_factor ** (dim / (dim - 2))\n freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))\n t = torch.arange(end, device=freqs.device) # type: ignore\n freqs = torch.outer(t, freqs).float() # type: ignore\n freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef load_checkpoint(ckpt_path, use_ema=True):\n checkpoint = torch.load(ckpt_path, weights_only=True)\n if use_ema:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n dict_state = checkpoint[\"model_state_dict\"]\n text_embed_dict = {}\n for key in dict_state.keys():\n # transformer.text_embed.text_embed.weight -> text_embed.weight\n if \"text_embed\" in key:\n text_embed_dict[key.replace(\"transformer.text_embed.\", \"\")] = dict_state[key]\n return text_embed_dict\n\n\nclass F5TTS(object):\n def __init__(\n self,\n config,\n debug_mode=True,\n stream: Optional[torch.cuda.Stream] = None,\n tllm_model_dir: Optional[str] = None,\n model_path: Optional[str] = None,\n vocab_size: Optional[int] = None,\n ):\n self.dtype = config[\"pretrained_config\"][\"dtype\"]\n\n rank = tensorrt_llm.mpi_rank()\n world_size = config[\"pretrained_config\"][\"mapping\"][\"world_size\"]\n cp_size = config[\"pretrained_config\"][\"mapping\"][\"cp_size\"]\n tp_size = config[\"pretrained_config\"][\"mapping\"][\"tp_size\"]\n pp_size = config[\"pretrained_config\"][\"mapping\"][\"pp_size\"]\n assert pp_size == 1\n self.mapping = tensorrt_llm.Mapping(\n world_size=world_size, rank=rank, cp_size=cp_size, tp_size=tp_size, pp_size=1, gpus_per_node=1\n )\n\n local_rank = rank % self.mapping.gpus_per_node\n self.device = torch.device(f\"cuda:{local_rank}\")\n\n torch.cuda.set_device(self.device)\n\n self.stream = stream\n if self.stream is None:\n self.stream = torch.cuda.Stream(self.device)\n torch.cuda.set_stream(self.stream)\n\n engine_file = os.path.join(tllm_model_dir, f\"rank{rank}.engine\")\n logger.info(f\"Loading engine from {engine_file}\")\n with open(engine_file, \"rb\") as f:\n engine_buffer = f.read()\n\n assert engine_buffer is not None\n\n self.session = Session.from_serialized_engine(engine_buffer)\n\n self.debug_mode = debug_mode\n\n self.inputs = {}\n self.outputs = {}\n self.buffer_allocated = False\n\n expected_tensor_names = [\"noise\", \"cond\", \"time\", \"rope_cos\", \"rope_sin\", \"input_lengths\", \"denoised\"]\n\n found_tensor_names = [self.session.engine.get_tensor_name(i) for i in range(self.session.engine.num_io_tensors)]\n if not self.debug_mode and set(expected_tensor_names) != set(found_tensor_names):\n logger.error(\n f\"The following expected tensors are not found: {set(expected_tensor_names).difference(set(found_tensor_names))}\"\n )\n logger.error(\n f\"Those tensors in engine are not expected: {set(found_tensor_names).difference(set(expected_tensor_names))}\"\n )\n logger.error(f\"Expected tensor names: {expected_tensor_names}\")\n logger.error(f\"Found tensor names: {found_tensor_names}\")\n raise RuntimeError(\"Tensor names in engine are not the same as expected.\")\n if self.debug_mode:\n self.debug_tensors = list(set(found_tensor_names) - set(expected_tensor_names))\n\n self.max_mel_len = 4096\n self.text_embedding = TextEmbedding(\n text_num_embeds=vocab_size, text_dim=512, conv_layers=4, precompute_max_pos=self.max_mel_len\n ).to(self.device)\n self.text_embedding.load_state_dict(load_checkpoint(model_path), strict=True)\n\n self.target_audio_sample_rate = 24000\n self.target_rms = 0.15 # target rms for audio\n self.n_fft = 1024\n self.win_length = 1024\n self.hop_length = 256\n self.n_mel_channels = 100\n # self.max_mel_len = 3000\n self.head_dim = 64\n self.base_rescale_factor = 1.0\n self.interpolation_factor = 1.0\n base = 10000.0 * self.base_rescale_factor ** (self.head_dim / (self.head_dim - 2))\n inv_freq = 1.0 / (base ** (torch.arange(0, self.head_dim, 2).float() / self.head_dim))\n freqs = torch.outer(torch.arange(self.max_mel_len, dtype=torch.float32), inv_freq) / self.interpolation_factor\n self.freqs = freqs.repeat_interleave(2, dim=-1).unsqueeze(0)\n self.rope_cos = self.freqs.cos().half()\n self.rope_sin = self.freqs.sin().half()\n self.nfe_steps = 16\n t = torch.linspace(0, 1, self.nfe_steps + 1, dtype=torch.float32)\n time_step = t + (-1.0) * (torch.cos(torch.pi * 0.5 * t) - 1 + t)\n delta_t = torch.diff(time_step)\n # WAR: hard coding 256 here\n tmp_dim = 256\n time_expand = torch.zeros((1, self.nfe_steps, tmp_dim), dtype=torch.float32)\n half_dim = tmp_dim // 2\n emb_factor = math.log(10000) / (half_dim - 1)\n emb_factor = 1000.0 * torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb_factor)\n for i in range(self.nfe_steps):\n emb = time_step[i] * emb_factor\n time_expand[:, i, :] = torch.cat((emb.sin(), emb.cos()), dim=-1)\n self.time_expand = time_expand.to(self.device)\n self.delta_t = torch.cat((delta_t, delta_t), dim=0).contiguous().to(self.device)\n\n def _tensor_dtype(self, name):\n # return torch dtype given tensor name for convenience\n dtype = trt_dtype_to_torch(self.session.engine.get_tensor_dtype(name))\n return dtype\n\n def _setup(self, batch_size, seq_len):\n for i in range(self.session.engine.num_io_tensors):\n name = self.session.engine.get_tensor_name(i)\n if self.session.engine.get_tensor_mode(name) == trt.TensorIOMode.OUTPUT:\n shape = list(self.session.engine.get_tensor_shape(name))\n shape[0] = batch_size\n shape[1] = seq_len\n self.outputs[name] = torch.empty(shape, dtype=self._tensor_dtype(name), device=self.device)\n\n self.buffer_allocated = True\n\n def cuda_stream_guard(func):\n \"\"\"Sync external stream and set current stream to the one bound to the session. Reset on exit.\"\"\"\n\n @wraps(func)\n def wrapper(self, *args, **kwargs):\n external_stream = torch.cuda.current_stream()\n if external_stream != self.stream:\n external_stream.synchronize()\n torch.cuda.set_stream(self.stream)\n ret = func(self, *args, **kwargs)\n if external_stream != self.stream:\n self.stream.synchronize()\n torch.cuda.set_stream(external_stream)\n return ret\n\n return wrapper\n\n @cuda_stream_guard\n def forward(\n self,\n noise: torch.Tensor,\n cond: torch.Tensor,\n time_expand: torch.Tensor,\n rope_cos: torch.Tensor,\n rope_sin: torch.Tensor,\n input_lengths: torch.Tensor,\n delta_t: torch.Tensor,\n use_perf: bool = False,\n ):\n if use_perf:\n torch.cuda.nvtx.range_push(\"flow matching\")\n cfg_strength = 2.0\n batch_size = noise.shape[0]\n half_batch = batch_size // 2\n noise_half = noise[:half_batch] # Store the initial half of noise\n\n input_type = str_dtype_to_torch(self.dtype)\n\n # Keep a copy of the initial tensors\n cond = cond.to(input_type)\n rope_cos = rope_cos.to(input_type)\n rope_sin = rope_sin.to(input_type)\n input_lengths = input_lengths.to(str_dtype_to_torch(\"int32\"))\n\n # Instead of iteratively updating noise within a single model context,\n # we'll do a single forward pass for each iteration with fresh context setup\n for i in range(self.nfe_steps):\n # Re-setup the buffers for clean execution\n self._setup(batch_size, noise.shape[1])\n if not self.buffer_allocated:\n raise RuntimeError(\"Buffer not allocated, please call setup first!\")\n\n # Re-create combined noises for this iteration\n current_noise = torch.cat([noise_half, noise_half], dim=0).to(input_type)\n\n # Get time step for this iteration\n current_time = time_expand[:, i].to(input_type)\n\n # Create fresh input dictionary for this iteration\n current_inputs = {\n \"noise\": current_noise,\n \"cond\": cond,\n \"time\": current_time,\n \"rope_cos\": rope_cos,\n \"rope_sin\": rope_sin,\n \"input_lengths\": input_lengths,\n }\n\n # Update inputs and set shapes\n self.inputs.clear() # Clear previous inputs\n self.inputs.update(**current_inputs)\n self.session.set_shapes(self.inputs)\n\n if use_perf:\n torch.cuda.nvtx.range_push(f\"execute {i}\")\n ok = self.session.run(self.inputs, self.outputs, self.stream.cuda_stream)\n assert ok, \"Failed to execute model\"\n # self.session.context.execute_async_v3(self.stream.cuda_stream)\n if use_perf:\n torch.cuda.nvtx.range_pop()\n # Process results\n t_scale = delta_t[i].unsqueeze(0).to(input_type)\n\n # Extract predictions\n pred_cond = self.outputs[\"denoised\"][:half_batch]\n pred_uncond = self.outputs[\"denoised\"][half_batch:]\n\n # Apply classifier-free guidance with safeguards\n guidance = pred_cond + (pred_cond - pred_uncond) * cfg_strength\n # Calculate update for noise\n noise_half = noise_half + guidance * t_scale\n if use_perf:\n torch.cuda.nvtx.range_pop()\n return noise_half\n\n def sample(\n self,\n text_pad_sequence: torch.Tensor,\n ref_mel_batch: torch.Tensor,\n ref_mel_len_batch: torch.Tensor,\n estimated_reference_target_mel_len: List[int],\n remove_input_padding: bool = False,\n use_perf: bool = False,\n ):\n if use_perf:\n torch.cuda.nvtx.range_push(\"text embedding\")\n batch = text_pad_sequence.shape[0]\n max_seq_len = ref_mel_batch.shape[1]\n\n text_pad_sequence_drop = torch.cat(\n (text_pad_sequence, torch.zeros((1, text_pad_sequence.shape[1]), dtype=torch.int32).to(self.device)), dim=0\n )\n\n text_embedding_drop_list = []\n for i in range(batch + 1):\n text_embedding_drop_list.append(self.text_embedding(text_pad_sequence_drop[i].unsqueeze(0).to(self.device)))\n text_embedding_drop_condition = torch.cat(text_embedding_drop_list, dim=0)\n\n text_embedding = text_embedding_drop_condition[:-1]\n # text_embedding_drop B,T,C batch should be the same\n text_embedding_drop = text_embedding_drop_condition[-1].unsqueeze(0).repeat(batch, 1, 1)\n\n noise = torch.randn_like(ref_mel_batch).to(self.device)\n rope_cos = self.rope_cos[:, :max_seq_len, :].float().repeat(batch, 1, 1)\n rope_sin = self.rope_sin[:, :max_seq_len, :].float().repeat(batch, 1, 1)\n\n cat_mel_text = torch.cat((ref_mel_batch, text_embedding), dim=-1)\n cat_mel_text_drop = torch.cat(\n (\n torch.zeros((batch, max_seq_len, self.n_mel_channels), dtype=torch.float32).to(self.device),\n text_embedding_drop,\n ),\n dim=-1,\n )\n\n time_expand = self.time_expand.repeat(2 * batch, 1, 1).contiguous()\n\n # Convert estimated_reference_target_mel_len to tensor\n input_lengths = torch.tensor(estimated_reference_target_mel_len, dtype=torch.int32)\n\n # combine above along the batch dimension\n inputs = {\n \"noise\": torch.cat((noise, noise), dim=0).contiguous(),\n \"cond\": torch.cat((cat_mel_text, cat_mel_text_drop), dim=0).contiguous(),\n \"time_expand\": time_expand,\n \"rope_cos\": torch.cat((rope_cos, rope_cos), dim=0).contiguous(),\n \"rope_sin\": torch.cat((rope_sin, rope_sin), dim=0).contiguous(),\n \"input_lengths\": torch.cat((input_lengths, input_lengths), dim=0).contiguous(),\n \"delta_t\": self.delta_t,\n }\n if use_perf and remove_input_padding:\n torch.cuda.nvtx.range_push(\"remove input padding\")\n if remove_input_padding:\n max_seq_len = inputs[\"cond\"].shape[1]\n inputs[\"noise\"] = remove_tensor_padding(inputs[\"noise\"], inputs[\"input_lengths\"])\n inputs[\"cond\"] = remove_tensor_padding(inputs[\"cond\"], inputs[\"input_lengths\"])\n # for time_expand, convert from B,D to B,T,D by repeat\n inputs[\"time_expand\"] = inputs[\"time_expand\"].unsqueeze(1).repeat(1, max_seq_len, 1, 1)\n inputs[\"time_expand\"] = remove_tensor_padding(inputs[\"time_expand\"], inputs[\"input_lengths\"])\n inputs[\"rope_cos\"] = remove_tensor_padding(inputs[\"rope_cos\"], inputs[\"input_lengths\"])\n inputs[\"rope_sin\"] = remove_tensor_padding(inputs[\"rope_sin\"], inputs[\"input_lengths\"])\n if use_perf and remove_input_padding:\n torch.cuda.nvtx.range_pop()\n for key in inputs:\n inputs[key] = inputs[key].to(self.device)\n if use_perf:\n torch.cuda.nvtx.range_pop()\n start_time = time.time()\n denoised = self.forward(**inputs, use_perf=use_perf)\n cost_time = time.time() - start_time\n if use_perf and remove_input_padding:\n torch.cuda.nvtx.range_push(\"remove input padding output\")\n if remove_input_padding:\n denoised_list = []\n start_idx = 0\n for i in range(batch):\n denoised_list.append(denoised[start_idx : start_idx + inputs[\"input_lengths\"][i]])\n start_idx += inputs[\"input_lengths\"][i]\n if use_perf and remove_input_padding:\n torch.cuda.nvtx.range_pop()\n return denoised_list, cost_time\n return denoised, cost_time","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.remove_tensor_padding","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.remove_tensor_padding#L17-L31","kind":"function","name":"remove_tensor_padding","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":17,"end_line":31,"context_start_line":1,"context_end_line":51,"code":"import math\nimport os\nimport time\nfrom functools import wraps\nfrom typing import List, Optional\n\nimport tensorrt as trt\nimport tensorrt_llm\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nfrom tensorrt_llm._utils import str_dtype_to_torch, trt_dtype_to_torch\nfrom tensorrt_llm.logger import logger\nfrom tensorrt_llm.runtime.session import Session\n\n\ndef remove_tensor_padding(input_tensor, input_tensor_lengths=None):\n # Audio tensor case: batch, seq_len, feature_len\n # position_ids case: batch, seq_len\n assert input_tensor_lengths is not None, \"input_tensor_lengths must be provided for 3D input_tensor\"\n\n # Initialize a list to collect valid sequences\n valid_sequences = []\n\n for i in range(input_tensor.shape[0]):\n valid_length = input_tensor_lengths[i]\n valid_sequences.append(input_tensor[i, :valid_length])\n\n # Concatenate all valid sequences along the batch dimension\n output_tensor = torch.cat(valid_sequences, dim=0).contiguous()\n return output_tensor\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, text_num_embeds, text_dim, conv_layers=0, conv_mult=2, precompute_max_pos=4096):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim) # use 0 as filler token\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(text_dim, precompute_max_pos), persistent=False)\n self.text_blocks = nn.Sequential(*[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)])\n\n def forward(self, text):\n # only keep tensors with value not -1\n text_mask = text != -1\n text_pad_cut_off_index = text_mask.sum(dim=1).max()\n\n text = text[:, :text_pad_cut_off_index]\n text = self.text_embed(text)\n text = text + self.freqs_cis[: text.shape[1], :]\n for block in self.text_blocks:\n text = block(text)\n # padding text to the original length","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.TextEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.TextEmbedding#L34-L55","kind":"class","name":"TextEmbedding","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":34,"end_line":55,"context_start_line":14,"context_end_line":75,"code":"from tensorrt_llm.runtime.session import Session\n\n\ndef remove_tensor_padding(input_tensor, input_tensor_lengths=None):\n # Audio tensor case: batch, seq_len, feature_len\n # position_ids case: batch, seq_len\n assert input_tensor_lengths is not None, \"input_tensor_lengths must be provided for 3D input_tensor\"\n\n # Initialize a list to collect valid sequences\n valid_sequences = []\n\n for i in range(input_tensor.shape[0]):\n valid_length = input_tensor_lengths[i]\n valid_sequences.append(input_tensor[i, :valid_length])\n\n # Concatenate all valid sequences along the batch dimension\n output_tensor = torch.cat(valid_sequences, dim=0).contiguous()\n return output_tensor\n\n\nclass TextEmbedding(nn.Module):\n def __init__(self, text_num_embeds, text_dim, conv_layers=0, conv_mult=2, precompute_max_pos=4096):\n super().__init__()\n self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim) # use 0 as filler token\n self.register_buffer(\"freqs_cis\", precompute_freqs_cis(text_dim, precompute_max_pos), persistent=False)\n self.text_blocks = nn.Sequential(*[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)])\n\n def forward(self, text):\n # only keep tensors with value not -1\n text_mask = text != -1\n text_pad_cut_off_index = text_mask.sum(dim=1).max()\n\n text = text[:, :text_pad_cut_off_index]\n text = self.text_embed(text)\n text = text + self.freqs_cis[: text.shape[1], :]\n for block in self.text_blocks:\n text = block(text)\n # padding text to the original length\n # text shape: B,seq_len,C\n # pad at the second dimension\n text = F.pad(text, (0, 0, 0, text_mask.shape[1] - text.shape[1], 0, 0), value=0)\n return text\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.gamma = nn.Parameter(torch.zeros(1, 1, dim))\n self.beta = nn.Parameter(torch.zeros(1, 1, dim))\n\n def forward(self, x):\n Gx = torch.norm(x, p=2, dim=1, keepdim=True)\n Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)\n return self.gamma * (x * Nx) + self.beta + x\n\n\nclass ConvNeXtV2Block(nn.Module):\n def __init__(\n self,\n dim: int,\n intermediate_dim: int,\n dilation: int = 1,","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.GRN","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.GRN#L58-L67","kind":"class","name":"GRN","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":58,"end_line":67,"context_start_line":38,"context_end_line":87,"code":" self.register_buffer(\"freqs_cis\", precompute_freqs_cis(text_dim, precompute_max_pos), persistent=False)\n self.text_blocks = nn.Sequential(*[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)])\n\n def forward(self, text):\n # only keep tensors with value not -1\n text_mask = text != -1\n text_pad_cut_off_index = text_mask.sum(dim=1).max()\n\n text = text[:, :text_pad_cut_off_index]\n text = self.text_embed(text)\n text = text + self.freqs_cis[: text.shape[1], :]\n for block in self.text_blocks:\n text = block(text)\n # padding text to the original length\n # text shape: B,seq_len,C\n # pad at the second dimension\n text = F.pad(text, (0, 0, 0, text_mask.shape[1] - text.shape[1], 0, 0), value=0)\n return text\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.gamma = nn.Parameter(torch.zeros(1, 1, dim))\n self.beta = nn.Parameter(torch.zeros(1, 1, dim))\n\n def forward(self, x):\n Gx = torch.norm(x, p=2, dim=1, keepdim=True)\n Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)\n return self.gamma * (x * Nx) + self.beta + x\n\n\nclass ConvNeXtV2Block(nn.Module):\n def __init__(\n self,\n dim: int,\n intermediate_dim: int,\n dilation: int = 1,\n ):\n super().__init__()\n padding = (dilation * (7 - 1)) // 2\n self.dwconv = nn.Conv1d(\n dim, dim, kernel_size=7, padding=padding, groups=dim, dilation=dilation\n ) # depthwise conv\n self.norm = nn.LayerNorm(dim, eps=1e-6)\n self.pwconv1 = nn.Linear(dim, intermediate_dim) # pointwise/1x1 convs, implemented with linear layers\n self.act = nn.GELU()\n self.grn = GRN(intermediate_dim)\n self.pwconv2 = nn.Linear(intermediate_dim, dim)\n","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.ConvNeXtV2Block","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.ConvNeXtV2Block#L70-L98","kind":"class","name":"ConvNeXtV2Block","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":70,"end_line":98,"context_start_line":50,"context_end_line":118,"code":" text = block(text)\n # padding text to the original length\n # text shape: B,seq_len,C\n # pad at the second dimension\n text = F.pad(text, (0, 0, 0, text_mask.shape[1] - text.shape[1], 0, 0), value=0)\n return text\n\n\nclass GRN(nn.Module):\n def __init__(self, dim):\n super().__init__()\n self.gamma = nn.Parameter(torch.zeros(1, 1, dim))\n self.beta = nn.Parameter(torch.zeros(1, 1, dim))\n\n def forward(self, x):\n Gx = torch.norm(x, p=2, dim=1, keepdim=True)\n Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)\n return self.gamma * (x * Nx) + self.beta + x\n\n\nclass ConvNeXtV2Block(nn.Module):\n def __init__(\n self,\n dim: int,\n intermediate_dim: int,\n dilation: int = 1,\n ):\n super().__init__()\n padding = (dilation * (7 - 1)) // 2\n self.dwconv = nn.Conv1d(\n dim, dim, kernel_size=7, padding=padding, groups=dim, dilation=dilation\n ) # depthwise conv\n self.norm = nn.LayerNorm(dim, eps=1e-6)\n self.pwconv1 = nn.Linear(dim, intermediate_dim) # pointwise/1x1 convs, implemented with linear layers\n self.act = nn.GELU()\n self.grn = GRN(intermediate_dim)\n self.pwconv2 = nn.Linear(intermediate_dim, dim)\n\n def forward(self, x: torch.Tensor) -> torch.Tensor:\n residual = x\n x = x.transpose(1, 2) # b n d -> b d n\n x = self.dwconv(x)\n x = x.transpose(1, 2) # b d n -> b n d\n x = self.norm(x)\n x = self.pwconv1(x)\n x = self.act(x)\n x = self.grn(x)\n x = self.pwconv2(x)\n return residual + x\n\n\ndef precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0):\n # proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning\n # has some connection to NTK literature\n # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/\n # https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py\n theta *= theta_rescale_factor ** (dim / (dim - 2))\n freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))\n t = torch.arange(end, device=freqs.device) # type: ignore\n freqs = torch.outer(t, freqs).float() # type: ignore\n freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef load_checkpoint(ckpt_path, use_ema=True):\n checkpoint = torch.load(ckpt_path, weights_only=True)\n if use_ema:\n checkpoint[\"model_state_dict\"] = {","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.precompute_freqs_cis","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.precompute_freqs_cis#L101-L112","kind":"function","name":"precompute_freqs_cis","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":101,"end_line":112,"context_start_line":81,"context_end_line":132,"code":" ) # depthwise conv\n self.norm = nn.LayerNorm(dim, eps=1e-6)\n self.pwconv1 = nn.Linear(dim, intermediate_dim) # pointwise/1x1 convs, implemented with linear layers\n self.act = nn.GELU()\n self.grn = GRN(intermediate_dim)\n self.pwconv2 = nn.Linear(intermediate_dim, dim)\n\n def forward(self, x: torch.Tensor) -> torch.Tensor:\n residual = x\n x = x.transpose(1, 2) # b n d -> b d n\n x = self.dwconv(x)\n x = x.transpose(1, 2) # b d n -> b n d\n x = self.norm(x)\n x = self.pwconv1(x)\n x = self.act(x)\n x = self.grn(x)\n x = self.pwconv2(x)\n return residual + x\n\n\ndef precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0):\n # proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning\n # has some connection to NTK literature\n # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/\n # https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py\n theta *= theta_rescale_factor ** (dim / (dim - 2))\n freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))\n t = torch.arange(end, device=freqs.device) # type: ignore\n freqs = torch.outer(t, freqs).float() # type: ignore\n freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef load_checkpoint(ckpt_path, use_ema=True):\n checkpoint = torch.load(ckpt_path, weights_only=True)\n if use_ema:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n dict_state = checkpoint[\"model_state_dict\"]\n text_embed_dict = {}\n for key in dict_state.keys():\n # transformer.text_embed.text_embed.weight -> text_embed.weight\n if \"text_embed\" in key:\n text_embed_dict[key.replace(\"transformer.text_embed.\", \"\")] = dict_state[key]\n return text_embed_dict\n\n\nclass F5TTS(object):","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.load_checkpoint","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.load_checkpoint#L115-L129","kind":"function","name":"load_checkpoint","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":115,"end_line":129,"context_start_line":95,"context_end_line":149,"code":" x = self.act(x)\n x = self.grn(x)\n x = self.pwconv2(x)\n return residual + x\n\n\ndef precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0):\n # proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning\n # has some connection to NTK literature\n # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/\n # https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py\n theta *= theta_rescale_factor ** (dim / (dim - 2))\n freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))\n t = torch.arange(end, device=freqs.device) # type: ignore\n freqs = torch.outer(t, freqs).float() # type: ignore\n freqs_cos = torch.cos(freqs) # real part\n freqs_sin = torch.sin(freqs) # imaginary part\n return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef load_checkpoint(ckpt_path, use_ema=True):\n checkpoint = torch.load(ckpt_path, weights_only=True)\n if use_ema:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n dict_state = checkpoint[\"model_state_dict\"]\n text_embed_dict = {}\n for key in dict_state.keys():\n # transformer.text_embed.text_embed.weight -> text_embed.weight\n if \"text_embed\" in key:\n text_embed_dict[key.replace(\"transformer.text_embed.\", \"\")] = dict_state[key]\n return text_embed_dict\n\n\nclass F5TTS(object):\n def __init__(\n self,\n config,\n debug_mode=True,\n stream: Optional[torch.cuda.Stream] = None,\n tllm_model_dir: Optional[str] = None,\n model_path: Optional[str] = None,\n vocab_size: Optional[int] = None,\n ):\n self.dtype = config[\"pretrained_config\"][\"dtype\"]\n\n rank = tensorrt_llm.mpi_rank()\n world_size = config[\"pretrained_config\"][\"mapping\"][\"world_size\"]\n cp_size = config[\"pretrained_config\"][\"mapping\"][\"cp_size\"]\n tp_size = config[\"pretrained_config\"][\"mapping\"][\"tp_size\"]\n pp_size = config[\"pretrained_config\"][\"mapping\"][\"pp_size\"]\n assert pp_size == 1","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.F5TTS","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.F5TTS#L132-L430","kind":"class","name":"F5TTS","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":132,"end_line":430,"context_start_line":112,"context_end_line":430,"code":" return torch.cat([freqs_cos, freqs_sin], dim=-1)\n\n\ndef load_checkpoint(ckpt_path, use_ema=True):\n checkpoint = torch.load(ckpt_path, weights_only=True)\n if use_ema:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n dict_state = checkpoint[\"model_state_dict\"]\n text_embed_dict = {}\n for key in dict_state.keys():\n # transformer.text_embed.text_embed.weight -> text_embed.weight\n if \"text_embed\" in key:\n text_embed_dict[key.replace(\"transformer.text_embed.\", \"\")] = dict_state[key]\n return text_embed_dict\n\n\nclass F5TTS(object):\n def __init__(\n self,\n config,\n debug_mode=True,\n stream: Optional[torch.cuda.Stream] = None,\n tllm_model_dir: Optional[str] = None,\n model_path: Optional[str] = None,\n vocab_size: Optional[int] = None,\n ):\n self.dtype = config[\"pretrained_config\"][\"dtype\"]\n\n rank = tensorrt_llm.mpi_rank()\n world_size = config[\"pretrained_config\"][\"mapping\"][\"world_size\"]\n cp_size = config[\"pretrained_config\"][\"mapping\"][\"cp_size\"]\n tp_size = config[\"pretrained_config\"][\"mapping\"][\"tp_size\"]\n pp_size = config[\"pretrained_config\"][\"mapping\"][\"pp_size\"]\n assert pp_size == 1\n self.mapping = tensorrt_llm.Mapping(\n world_size=world_size, rank=rank, cp_size=cp_size, tp_size=tp_size, pp_size=1, gpus_per_node=1\n )\n\n local_rank = rank % self.mapping.gpus_per_node\n self.device = torch.device(f\"cuda:{local_rank}\")\n\n torch.cuda.set_device(self.device)\n\n self.stream = stream\n if self.stream is None:\n self.stream = torch.cuda.Stream(self.device)\n torch.cuda.set_stream(self.stream)\n\n engine_file = os.path.join(tllm_model_dir, f\"rank{rank}.engine\")\n logger.info(f\"Loading engine from {engine_file}\")\n with open(engine_file, \"rb\") as f:\n engine_buffer = f.read()\n\n assert engine_buffer is not None\n\n self.session = Session.from_serialized_engine(engine_buffer)\n\n self.debug_mode = debug_mode\n\n self.inputs = {}\n self.outputs = {}\n self.buffer_allocated = False\n\n expected_tensor_names = [\"noise\", \"cond\", \"time\", \"rope_cos\", \"rope_sin\", \"input_lengths\", \"denoised\"]\n\n found_tensor_names = [self.session.engine.get_tensor_name(i) for i in range(self.session.engine.num_io_tensors)]\n if not self.debug_mode and set(expected_tensor_names) != set(found_tensor_names):\n logger.error(\n f\"The following expected tensors are not found: {set(expected_tensor_names).difference(set(found_tensor_names))}\"\n )\n logger.error(\n f\"Those tensors in engine are not expected: {set(found_tensor_names).difference(set(expected_tensor_names))}\"\n )\n logger.error(f\"Expected tensor names: {expected_tensor_names}\")\n logger.error(f\"Found tensor names: {found_tensor_names}\")\n raise RuntimeError(\"Tensor names in engine are not the same as expected.\")\n if self.debug_mode:\n self.debug_tensors = list(set(found_tensor_names) - set(expected_tensor_names))\n\n self.max_mel_len = 4096\n self.text_embedding = TextEmbedding(\n text_num_embeds=vocab_size, text_dim=512, conv_layers=4, precompute_max_pos=self.max_mel_len\n ).to(self.device)\n self.text_embedding.load_state_dict(load_checkpoint(model_path), strict=True)\n\n self.target_audio_sample_rate = 24000\n self.target_rms = 0.15 # target rms for audio\n self.n_fft = 1024\n self.win_length = 1024\n self.hop_length = 256\n self.n_mel_channels = 100\n # self.max_mel_len = 3000\n self.head_dim = 64\n self.base_rescale_factor = 1.0\n self.interpolation_factor = 1.0\n base = 10000.0 * self.base_rescale_factor ** (self.head_dim / (self.head_dim - 2))\n inv_freq = 1.0 / (base ** (torch.arange(0, self.head_dim, 2).float() / self.head_dim))\n freqs = torch.outer(torch.arange(self.max_mel_len, dtype=torch.float32), inv_freq) / self.interpolation_factor\n self.freqs = freqs.repeat_interleave(2, dim=-1).unsqueeze(0)\n self.rope_cos = self.freqs.cos().half()\n self.rope_sin = self.freqs.sin().half()\n self.nfe_steps = 16\n t = torch.linspace(0, 1, self.nfe_steps + 1, dtype=torch.float32)\n time_step = t + (-1.0) * (torch.cos(torch.pi * 0.5 * t) - 1 + t)\n delta_t = torch.diff(time_step)\n # WAR: hard coding 256 here\n tmp_dim = 256\n time_expand = torch.zeros((1, self.nfe_steps, tmp_dim), dtype=torch.float32)\n half_dim = tmp_dim // 2\n emb_factor = math.log(10000) / (half_dim - 1)\n emb_factor = 1000.0 * torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb_factor)\n for i in range(self.nfe_steps):\n emb = time_step[i] * emb_factor\n time_expand[:, i, :] = torch.cat((emb.sin(), emb.cos()), dim=-1)\n self.time_expand = time_expand.to(self.device)\n self.delta_t = torch.cat((delta_t, delta_t), dim=0).contiguous().to(self.device)\n\n def _tensor_dtype(self, name):\n # return torch dtype given tensor name for convenience\n dtype = trt_dtype_to_torch(self.session.engine.get_tensor_dtype(name))\n return dtype\n\n def _setup(self, batch_size, seq_len):\n for i in range(self.session.engine.num_io_tensors):\n name = self.session.engine.get_tensor_name(i)\n if self.session.engine.get_tensor_mode(name) == trt.TensorIOMode.OUTPUT:\n shape = list(self.session.engine.get_tensor_shape(name))\n shape[0] = batch_size\n shape[1] = seq_len\n self.outputs[name] = torch.empty(shape, dtype=self._tensor_dtype(name), device=self.device)\n\n self.buffer_allocated = True\n\n def cuda_stream_guard(func):\n \"\"\"Sync external stream and set current stream to the one bound to the session. Reset on exit.\"\"\"\n\n @wraps(func)\n def wrapper(self, *args, **kwargs):\n external_stream = torch.cuda.current_stream()\n if external_stream != self.stream:\n external_stream.synchronize()\n torch.cuda.set_stream(self.stream)\n ret = func(self, *args, **kwargs)\n if external_stream != self.stream:\n self.stream.synchronize()\n torch.cuda.set_stream(external_stream)\n return ret\n\n return wrapper\n\n @cuda_stream_guard\n def forward(\n self,\n noise: torch.Tensor,\n cond: torch.Tensor,\n time_expand: torch.Tensor,\n rope_cos: torch.Tensor,\n rope_sin: torch.Tensor,\n input_lengths: torch.Tensor,\n delta_t: torch.Tensor,\n use_perf: bool = False,\n ):\n if use_perf:\n torch.cuda.nvtx.range_push(\"flow matching\")\n cfg_strength = 2.0\n batch_size = noise.shape[0]\n half_batch = batch_size // 2\n noise_half = noise[:half_batch] # Store the initial half of noise\n\n input_type = str_dtype_to_torch(self.dtype)\n\n # Keep a copy of the initial tensors\n cond = cond.to(input_type)\n rope_cos = rope_cos.to(input_type)\n rope_sin = rope_sin.to(input_type)\n input_lengths = input_lengths.to(str_dtype_to_torch(\"int32\"))\n\n # Instead of iteratively updating noise within a single model context,\n # we'll do a single forward pass for each iteration with fresh context setup\n for i in range(self.nfe_steps):\n # Re-setup the buffers for clean execution\n self._setup(batch_size, noise.shape[1])\n if not self.buffer_allocated:\n raise RuntimeError(\"Buffer not allocated, please call setup first!\")\n\n # Re-create combined noises for this iteration\n current_noise = torch.cat([noise_half, noise_half], dim=0).to(input_type)\n\n # Get time step for this iteration\n current_time = time_expand[:, i].to(input_type)\n\n # Create fresh input dictionary for this iteration\n current_inputs = {\n \"noise\": current_noise,\n \"cond\": cond,\n \"time\": current_time,\n \"rope_cos\": rope_cos,\n \"rope_sin\": rope_sin,\n \"input_lengths\": input_lengths,\n }\n\n # Update inputs and set shapes\n self.inputs.clear() # Clear previous inputs\n self.inputs.update(**current_inputs)\n self.session.set_shapes(self.inputs)\n\n if use_perf:\n torch.cuda.nvtx.range_push(f\"execute {i}\")\n ok = self.session.run(self.inputs, self.outputs, self.stream.cuda_stream)\n assert ok, \"Failed to execute model\"\n # self.session.context.execute_async_v3(self.stream.cuda_stream)\n if use_perf:\n torch.cuda.nvtx.range_pop()\n # Process results\n t_scale = delta_t[i].unsqueeze(0).to(input_type)\n\n # Extract predictions\n pred_cond = self.outputs[\"denoised\"][:half_batch]\n pred_uncond = self.outputs[\"denoised\"][half_batch:]\n\n # Apply classifier-free guidance with safeguards\n guidance = pred_cond + (pred_cond - pred_uncond) * cfg_strength\n # Calculate update for noise\n noise_half = noise_half + guidance * t_scale\n if use_perf:\n torch.cuda.nvtx.range_pop()\n return noise_half\n\n def sample(\n self,\n text_pad_sequence: torch.Tensor,\n ref_mel_batch: torch.Tensor,\n ref_mel_len_batch: torch.Tensor,\n estimated_reference_target_mel_len: List[int],\n remove_input_padding: bool = False,\n use_perf: bool = False,\n ):\n if use_perf:\n torch.cuda.nvtx.range_push(\"text embedding\")\n batch = text_pad_sequence.shape[0]\n max_seq_len = ref_mel_batch.shape[1]\n\n text_pad_sequence_drop = torch.cat(\n (text_pad_sequence, torch.zeros((1, text_pad_sequence.shape[1]), dtype=torch.int32).to(self.device)), dim=0\n )\n\n text_embedding_drop_list = []\n for i in range(batch + 1):\n text_embedding_drop_list.append(self.text_embedding(text_pad_sequence_drop[i].unsqueeze(0).to(self.device)))\n text_embedding_drop_condition = torch.cat(text_embedding_drop_list, dim=0)\n\n text_embedding = text_embedding_drop_condition[:-1]\n # text_embedding_drop B,T,C batch should be the same\n text_embedding_drop = text_embedding_drop_condition[-1].unsqueeze(0).repeat(batch, 1, 1)\n\n noise = torch.randn_like(ref_mel_batch).to(self.device)\n rope_cos = self.rope_cos[:, :max_seq_len, :].float().repeat(batch, 1, 1)\n rope_sin = self.rope_sin[:, :max_seq_len, :].float().repeat(batch, 1, 1)\n\n cat_mel_text = torch.cat((ref_mel_batch, text_embedding), dim=-1)\n cat_mel_text_drop = torch.cat(\n (\n torch.zeros((batch, max_seq_len, self.n_mel_channels), dtype=torch.float32).to(self.device),\n text_embedding_drop,\n ),\n dim=-1,\n )\n\n time_expand = self.time_expand.repeat(2 * batch, 1, 1).contiguous()\n\n # Convert estimated_reference_target_mel_len to tensor\n input_lengths = torch.tensor(estimated_reference_target_mel_len, dtype=torch.int32)\n\n # combine above along the batch dimension\n inputs = {\n \"noise\": torch.cat((noise, noise), dim=0).contiguous(),\n \"cond\": torch.cat((cat_mel_text, cat_mel_text_drop), dim=0).contiguous(),\n \"time_expand\": time_expand,\n \"rope_cos\": torch.cat((rope_cos, rope_cos), dim=0).contiguous(),\n \"rope_sin\": torch.cat((rope_sin, rope_sin), dim=0).contiguous(),\n \"input_lengths\": torch.cat((input_lengths, input_lengths), dim=0).contiguous(),\n \"delta_t\": self.delta_t,\n }\n if use_perf and remove_input_padding:\n torch.cuda.nvtx.range_push(\"remove input padding\")\n if remove_input_padding:\n max_seq_len = inputs[\"cond\"].shape[1]\n inputs[\"noise\"] = remove_tensor_padding(inputs[\"noise\"], inputs[\"input_lengths\"])\n inputs[\"cond\"] = remove_tensor_padding(inputs[\"cond\"], inputs[\"input_lengths\"])\n # for time_expand, convert from B,D to B,T,D by repeat\n inputs[\"time_expand\"] = inputs[\"time_expand\"].unsqueeze(1).repeat(1, max_seq_len, 1, 1)\n inputs[\"time_expand\"] = remove_tensor_padding(inputs[\"time_expand\"], inputs[\"input_lengths\"])\n inputs[\"rope_cos\"] = remove_tensor_padding(inputs[\"rope_cos\"], inputs[\"input_lengths\"])\n inputs[\"rope_sin\"] = remove_tensor_padding(inputs[\"rope_sin\"], inputs[\"input_lengths\"])\n if use_perf and remove_input_padding:\n torch.cuda.nvtx.range_pop()\n for key in inputs:\n inputs[key] = inputs[key].to(self.device)\n if use_perf:\n torch.cuda.nvtx.range_pop()\n start_time = time.time()\n denoised = self.forward(**inputs, use_perf=use_perf)\n cost_time = time.time() - start_time\n if use_perf and remove_input_padding:\n torch.cuda.nvtx.range_push(\"remove input padding output\")\n if remove_input_padding:\n denoised_list = []\n start_idx = 0\n for i in range(batch):\n denoised_list.append(denoised[start_idx : start_idx + inputs[\"input_lengths\"][i]])\n start_idx += inputs[\"input_lengths\"][i]\n if use_perf and remove_input_padding:\n torch.cuda.nvtx.range_pop()\n return denoised_list, cost_time\n return denoised, cost_time","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.__init__#L133-L231","kind":"function","name":"__init__","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":133,"end_line":231,"context_start_line":113,"context_end_line":251,"code":"\n\ndef load_checkpoint(ckpt_path, use_ema=True):\n checkpoint = torch.load(ckpt_path, weights_only=True)\n if use_ema:\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n dict_state = checkpoint[\"model_state_dict\"]\n text_embed_dict = {}\n for key in dict_state.keys():\n # transformer.text_embed.text_embed.weight -> text_embed.weight\n if \"text_embed\" in key:\n text_embed_dict[key.replace(\"transformer.text_embed.\", \"\")] = dict_state[key]\n return text_embed_dict\n\n\nclass F5TTS(object):\n def __init__(\n self,\n config,\n debug_mode=True,\n stream: Optional[torch.cuda.Stream] = None,\n tllm_model_dir: Optional[str] = None,\n model_path: Optional[str] = None,\n vocab_size: Optional[int] = None,\n ):\n self.dtype = config[\"pretrained_config\"][\"dtype\"]\n\n rank = tensorrt_llm.mpi_rank()\n world_size = config[\"pretrained_config\"][\"mapping\"][\"world_size\"]\n cp_size = config[\"pretrained_config\"][\"mapping\"][\"cp_size\"]\n tp_size = config[\"pretrained_config\"][\"mapping\"][\"tp_size\"]\n pp_size = config[\"pretrained_config\"][\"mapping\"][\"pp_size\"]\n assert pp_size == 1\n self.mapping = tensorrt_llm.Mapping(\n world_size=world_size, rank=rank, cp_size=cp_size, tp_size=tp_size, pp_size=1, gpus_per_node=1\n )\n\n local_rank = rank % self.mapping.gpus_per_node\n self.device = torch.device(f\"cuda:{local_rank}\")\n\n torch.cuda.set_device(self.device)\n\n self.stream = stream\n if self.stream is None:\n self.stream = torch.cuda.Stream(self.device)\n torch.cuda.set_stream(self.stream)\n\n engine_file = os.path.join(tllm_model_dir, f\"rank{rank}.engine\")\n logger.info(f\"Loading engine from {engine_file}\")\n with open(engine_file, \"rb\") as f:\n engine_buffer = f.read()\n\n assert engine_buffer is not None\n\n self.session = Session.from_serialized_engine(engine_buffer)\n\n self.debug_mode = debug_mode\n\n self.inputs = {}\n self.outputs = {}\n self.buffer_allocated = False\n\n expected_tensor_names = [\"noise\", \"cond\", \"time\", \"rope_cos\", \"rope_sin\", \"input_lengths\", \"denoised\"]\n\n found_tensor_names = [self.session.engine.get_tensor_name(i) for i in range(self.session.engine.num_io_tensors)]\n if not self.debug_mode and set(expected_tensor_names) != set(found_tensor_names):\n logger.error(\n f\"The following expected tensors are not found: {set(expected_tensor_names).difference(set(found_tensor_names))}\"\n )\n logger.error(\n f\"Those tensors in engine are not expected: {set(found_tensor_names).difference(set(expected_tensor_names))}\"\n )\n logger.error(f\"Expected tensor names: {expected_tensor_names}\")\n logger.error(f\"Found tensor names: {found_tensor_names}\")\n raise RuntimeError(\"Tensor names in engine are not the same as expected.\")\n if self.debug_mode:\n self.debug_tensors = list(set(found_tensor_names) - set(expected_tensor_names))\n\n self.max_mel_len = 4096\n self.text_embedding = TextEmbedding(\n text_num_embeds=vocab_size, text_dim=512, conv_layers=4, precompute_max_pos=self.max_mel_len\n ).to(self.device)\n self.text_embedding.load_state_dict(load_checkpoint(model_path), strict=True)\n\n self.target_audio_sample_rate = 24000\n self.target_rms = 0.15 # target rms for audio\n self.n_fft = 1024\n self.win_length = 1024\n self.hop_length = 256\n self.n_mel_channels = 100\n # self.max_mel_len = 3000\n self.head_dim = 64\n self.base_rescale_factor = 1.0\n self.interpolation_factor = 1.0\n base = 10000.0 * self.base_rescale_factor ** (self.head_dim / (self.head_dim - 2))\n inv_freq = 1.0 / (base ** (torch.arange(0, self.head_dim, 2).float() / self.head_dim))\n freqs = torch.outer(torch.arange(self.max_mel_len, dtype=torch.float32), inv_freq) / self.interpolation_factor\n self.freqs = freqs.repeat_interleave(2, dim=-1).unsqueeze(0)\n self.rope_cos = self.freqs.cos().half()\n self.rope_sin = self.freqs.sin().half()\n self.nfe_steps = 16\n t = torch.linspace(0, 1, self.nfe_steps + 1, dtype=torch.float32)\n time_step = t + (-1.0) * (torch.cos(torch.pi * 0.5 * t) - 1 + t)\n delta_t = torch.diff(time_step)\n # WAR: hard coding 256 here\n tmp_dim = 256\n time_expand = torch.zeros((1, self.nfe_steps, tmp_dim), dtype=torch.float32)\n half_dim = tmp_dim // 2\n emb_factor = math.log(10000) / (half_dim - 1)\n emb_factor = 1000.0 * torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb_factor)\n for i in range(self.nfe_steps):\n emb = time_step[i] * emb_factor\n time_expand[:, i, :] = torch.cat((emb.sin(), emb.cos()), dim=-1)\n self.time_expand = time_expand.to(self.device)\n self.delta_t = torch.cat((delta_t, delta_t), dim=0).contiguous().to(self.device)\n\n def _tensor_dtype(self, name):\n # return torch dtype given tensor name for convenience\n dtype = trt_dtype_to_torch(self.session.engine.get_tensor_dtype(name))\n return dtype\n\n def _setup(self, batch_size, seq_len):\n for i in range(self.session.engine.num_io_tensors):\n name = self.session.engine.get_tensor_name(i)\n if self.session.engine.get_tensor_mode(name) == trt.TensorIOMode.OUTPUT:\n shape = list(self.session.engine.get_tensor_shape(name))\n shape[0] = batch_size\n shape[1] = seq_len\n self.outputs[name] = torch.empty(shape, dtype=self._tensor_dtype(name), device=self.device)\n\n self.buffer_allocated = True\n\n def cuda_stream_guard(func):\n \"\"\"Sync external stream and set current stream to the one bound to the session. Reset on exit.\"\"\"\n","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.forward","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.forward#L267-L342","kind":"function","name":"forward","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":267,"end_line":342,"context_start_line":247,"context_end_line":362,"code":" self.buffer_allocated = True\n\n def cuda_stream_guard(func):\n \"\"\"Sync external stream and set current stream to the one bound to the session. Reset on exit.\"\"\"\n\n @wraps(func)\n def wrapper(self, *args, **kwargs):\n external_stream = torch.cuda.current_stream()\n if external_stream != self.stream:\n external_stream.synchronize()\n torch.cuda.set_stream(self.stream)\n ret = func(self, *args, **kwargs)\n if external_stream != self.stream:\n self.stream.synchronize()\n torch.cuda.set_stream(external_stream)\n return ret\n\n return wrapper\n\n @cuda_stream_guard\n def forward(\n self,\n noise: torch.Tensor,\n cond: torch.Tensor,\n time_expand: torch.Tensor,\n rope_cos: torch.Tensor,\n rope_sin: torch.Tensor,\n input_lengths: torch.Tensor,\n delta_t: torch.Tensor,\n use_perf: bool = False,\n ):\n if use_perf:\n torch.cuda.nvtx.range_push(\"flow matching\")\n cfg_strength = 2.0\n batch_size = noise.shape[0]\n half_batch = batch_size // 2\n noise_half = noise[:half_batch] # Store the initial half of noise\n\n input_type = str_dtype_to_torch(self.dtype)\n\n # Keep a copy of the initial tensors\n cond = cond.to(input_type)\n rope_cos = rope_cos.to(input_type)\n rope_sin = rope_sin.to(input_type)\n input_lengths = input_lengths.to(str_dtype_to_torch(\"int32\"))\n\n # Instead of iteratively updating noise within a single model context,\n # we'll do a single forward pass for each iteration with fresh context setup\n for i in range(self.nfe_steps):\n # Re-setup the buffers for clean execution\n self._setup(batch_size, noise.shape[1])\n if not self.buffer_allocated:\n raise RuntimeError(\"Buffer not allocated, please call setup first!\")\n\n # Re-create combined noises for this iteration\n current_noise = torch.cat([noise_half, noise_half], dim=0).to(input_type)\n\n # Get time step for this iteration\n current_time = time_expand[:, i].to(input_type)\n\n # Create fresh input dictionary for this iteration\n current_inputs = {\n \"noise\": current_noise,\n \"cond\": cond,\n \"time\": current_time,\n \"rope_cos\": rope_cos,\n \"rope_sin\": rope_sin,\n \"input_lengths\": input_lengths,\n }\n\n # Update inputs and set shapes\n self.inputs.clear() # Clear previous inputs\n self.inputs.update(**current_inputs)\n self.session.set_shapes(self.inputs)\n\n if use_perf:\n torch.cuda.nvtx.range_push(f\"execute {i}\")\n ok = self.session.run(self.inputs, self.outputs, self.stream.cuda_stream)\n assert ok, \"Failed to execute model\"\n # self.session.context.execute_async_v3(self.stream.cuda_stream)\n if use_perf:\n torch.cuda.nvtx.range_pop()\n # Process results\n t_scale = delta_t[i].unsqueeze(0).to(input_type)\n\n # Extract predictions\n pred_cond = self.outputs[\"denoised\"][:half_batch]\n pred_uncond = self.outputs[\"denoised\"][half_batch:]\n\n # Apply classifier-free guidance with safeguards\n guidance = pred_cond + (pred_cond - pred_uncond) * cfg_strength\n # Calculate update for noise\n noise_half = noise_half + guidance * t_scale\n if use_perf:\n torch.cuda.nvtx.range_pop()\n return noise_half\n\n def sample(\n self,\n text_pad_sequence: torch.Tensor,\n ref_mel_batch: torch.Tensor,\n ref_mel_len_batch: torch.Tensor,\n estimated_reference_target_mel_len: List[int],\n remove_input_padding: bool = False,\n use_perf: bool = False,\n ):\n if use_perf:\n torch.cuda.nvtx.range_push(\"text embedding\")\n batch = text_pad_sequence.shape[0]\n max_seq_len = ref_mel_batch.shape[1]\n\n text_pad_sequence_drop = torch.cat(\n (text_pad_sequence, torch.zeros((1, text_pad_sequence.shape[1]), dtype=torch.int32).to(self.device)), dim=0\n )\n\n text_embedding_drop_list = []","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm._tensor_dtype","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm._tensor_dtype#L233-L236","kind":"function","name":"_tensor_dtype","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":233,"end_line":236,"context_start_line":213,"context_end_line":256,"code":" freqs = torch.outer(torch.arange(self.max_mel_len, dtype=torch.float32), inv_freq) / self.interpolation_factor\n self.freqs = freqs.repeat_interleave(2, dim=-1).unsqueeze(0)\n self.rope_cos = self.freqs.cos().half()\n self.rope_sin = self.freqs.sin().half()\n self.nfe_steps = 16\n t = torch.linspace(0, 1, self.nfe_steps + 1, dtype=torch.float32)\n time_step = t + (-1.0) * (torch.cos(torch.pi * 0.5 * t) - 1 + t)\n delta_t = torch.diff(time_step)\n # WAR: hard coding 256 here\n tmp_dim = 256\n time_expand = torch.zeros((1, self.nfe_steps, tmp_dim), dtype=torch.float32)\n half_dim = tmp_dim // 2\n emb_factor = math.log(10000) / (half_dim - 1)\n emb_factor = 1000.0 * torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb_factor)\n for i in range(self.nfe_steps):\n emb = time_step[i] * emb_factor\n time_expand[:, i, :] = torch.cat((emb.sin(), emb.cos()), dim=-1)\n self.time_expand = time_expand.to(self.device)\n self.delta_t = torch.cat((delta_t, delta_t), dim=0).contiguous().to(self.device)\n\n def _tensor_dtype(self, name):\n # return torch dtype given tensor name for convenience\n dtype = trt_dtype_to_torch(self.session.engine.get_tensor_dtype(name))\n return dtype\n\n def _setup(self, batch_size, seq_len):\n for i in range(self.session.engine.num_io_tensors):\n name = self.session.engine.get_tensor_name(i)\n if self.session.engine.get_tensor_mode(name) == trt.TensorIOMode.OUTPUT:\n shape = list(self.session.engine.get_tensor_shape(name))\n shape[0] = batch_size\n shape[1] = seq_len\n self.outputs[name] = torch.empty(shape, dtype=self._tensor_dtype(name), device=self.device)\n\n self.buffer_allocated = True\n\n def cuda_stream_guard(func):\n \"\"\"Sync external stream and set current stream to the one bound to the session. Reset on exit.\"\"\"\n\n @wraps(func)\n def wrapper(self, *args, **kwargs):\n external_stream = torch.cuda.current_stream()\n if external_stream != self.stream:\n external_stream.synchronize()","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm._setup","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm._setup#L238-L247","kind":"function","name":"_setup","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":238,"end_line":247,"context_start_line":218,"context_end_line":267,"code":" t = torch.linspace(0, 1, self.nfe_steps + 1, dtype=torch.float32)\n time_step = t + (-1.0) * (torch.cos(torch.pi * 0.5 * t) - 1 + t)\n delta_t = torch.diff(time_step)\n # WAR: hard coding 256 here\n tmp_dim = 256\n time_expand = torch.zeros((1, self.nfe_steps, tmp_dim), dtype=torch.float32)\n half_dim = tmp_dim // 2\n emb_factor = math.log(10000) / (half_dim - 1)\n emb_factor = 1000.0 * torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb_factor)\n for i in range(self.nfe_steps):\n emb = time_step[i] * emb_factor\n time_expand[:, i, :] = torch.cat((emb.sin(), emb.cos()), dim=-1)\n self.time_expand = time_expand.to(self.device)\n self.delta_t = torch.cat((delta_t, delta_t), dim=0).contiguous().to(self.device)\n\n def _tensor_dtype(self, name):\n # return torch dtype given tensor name for convenience\n dtype = trt_dtype_to_torch(self.session.engine.get_tensor_dtype(name))\n return dtype\n\n def _setup(self, batch_size, seq_len):\n for i in range(self.session.engine.num_io_tensors):\n name = self.session.engine.get_tensor_name(i)\n if self.session.engine.get_tensor_mode(name) == trt.TensorIOMode.OUTPUT:\n shape = list(self.session.engine.get_tensor_shape(name))\n shape[0] = batch_size\n shape[1] = seq_len\n self.outputs[name] = torch.empty(shape, dtype=self._tensor_dtype(name), device=self.device)\n\n self.buffer_allocated = True\n\n def cuda_stream_guard(func):\n \"\"\"Sync external stream and set current stream to the one bound to the session. Reset on exit.\"\"\"\n\n @wraps(func)\n def wrapper(self, *args, **kwargs):\n external_stream = torch.cuda.current_stream()\n if external_stream != self.stream:\n external_stream.synchronize()\n torch.cuda.set_stream(self.stream)\n ret = func(self, *args, **kwargs)\n if external_stream != self.stream:\n self.stream.synchronize()\n torch.cuda.set_stream(external_stream)\n return ret\n\n return wrapper\n\n @cuda_stream_guard\n def forward(","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.cuda_stream_guard","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.cuda_stream_guard#L249-L264","kind":"function","name":"cuda_stream_guard","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":249,"end_line":264,"context_start_line":229,"context_end_line":284,"code":" time_expand[:, i, :] = torch.cat((emb.sin(), emb.cos()), dim=-1)\n self.time_expand = time_expand.to(self.device)\n self.delta_t = torch.cat((delta_t, delta_t), dim=0).contiguous().to(self.device)\n\n def _tensor_dtype(self, name):\n # return torch dtype given tensor name for convenience\n dtype = trt_dtype_to_torch(self.session.engine.get_tensor_dtype(name))\n return dtype\n\n def _setup(self, batch_size, seq_len):\n for i in range(self.session.engine.num_io_tensors):\n name = self.session.engine.get_tensor_name(i)\n if self.session.engine.get_tensor_mode(name) == trt.TensorIOMode.OUTPUT:\n shape = list(self.session.engine.get_tensor_shape(name))\n shape[0] = batch_size\n shape[1] = seq_len\n self.outputs[name] = torch.empty(shape, dtype=self._tensor_dtype(name), device=self.device)\n\n self.buffer_allocated = True\n\n def cuda_stream_guard(func):\n \"\"\"Sync external stream and set current stream to the one bound to the session. Reset on exit.\"\"\"\n\n @wraps(func)\n def wrapper(self, *args, **kwargs):\n external_stream = torch.cuda.current_stream()\n if external_stream != self.stream:\n external_stream.synchronize()\n torch.cuda.set_stream(self.stream)\n ret = func(self, *args, **kwargs)\n if external_stream != self.stream:\n self.stream.synchronize()\n torch.cuda.set_stream(external_stream)\n return ret\n\n return wrapper\n\n @cuda_stream_guard\n def forward(\n self,\n noise: torch.Tensor,\n cond: torch.Tensor,\n time_expand: torch.Tensor,\n rope_cos: torch.Tensor,\n rope_sin: torch.Tensor,\n input_lengths: torch.Tensor,\n delta_t: torch.Tensor,\n use_perf: bool = False,\n ):\n if use_perf:\n torch.cuda.nvtx.range_push(\"flow matching\")\n cfg_strength = 2.0\n batch_size = noise.shape[0]\n half_batch = batch_size // 2\n noise_half = noise[:half_batch] # Store the initial half of noise\n","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.sample","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.sample#L344-L430","kind":"function","name":"sample","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":344,"end_line":430,"context_start_line":324,"context_end_line":430,"code":" ok = self.session.run(self.inputs, self.outputs, self.stream.cuda_stream)\n assert ok, \"Failed to execute model\"\n # self.session.context.execute_async_v3(self.stream.cuda_stream)\n if use_perf:\n torch.cuda.nvtx.range_pop()\n # Process results\n t_scale = delta_t[i].unsqueeze(0).to(input_type)\n\n # Extract predictions\n pred_cond = self.outputs[\"denoised\"][:half_batch]\n pred_uncond = self.outputs[\"denoised\"][half_batch:]\n\n # Apply classifier-free guidance with safeguards\n guidance = pred_cond + (pred_cond - pred_uncond) * cfg_strength\n # Calculate update for noise\n noise_half = noise_half + guidance * t_scale\n if use_perf:\n torch.cuda.nvtx.range_pop()\n return noise_half\n\n def sample(\n self,\n text_pad_sequence: torch.Tensor,\n ref_mel_batch: torch.Tensor,\n ref_mel_len_batch: torch.Tensor,\n estimated_reference_target_mel_len: List[int],\n remove_input_padding: bool = False,\n use_perf: bool = False,\n ):\n if use_perf:\n torch.cuda.nvtx.range_push(\"text embedding\")\n batch = text_pad_sequence.shape[0]\n max_seq_len = ref_mel_batch.shape[1]\n\n text_pad_sequence_drop = torch.cat(\n (text_pad_sequence, torch.zeros((1, text_pad_sequence.shape[1]), dtype=torch.int32).to(self.device)), dim=0\n )\n\n text_embedding_drop_list = []\n for i in range(batch + 1):\n text_embedding_drop_list.append(self.text_embedding(text_pad_sequence_drop[i].unsqueeze(0).to(self.device)))\n text_embedding_drop_condition = torch.cat(text_embedding_drop_list, dim=0)\n\n text_embedding = text_embedding_drop_condition[:-1]\n # text_embedding_drop B,T,C batch should be the same\n text_embedding_drop = text_embedding_drop_condition[-1].unsqueeze(0).repeat(batch, 1, 1)\n\n noise = torch.randn_like(ref_mel_batch).to(self.device)\n rope_cos = self.rope_cos[:, :max_seq_len, :].float().repeat(batch, 1, 1)\n rope_sin = self.rope_sin[:, :max_seq_len, :].float().repeat(batch, 1, 1)\n\n cat_mel_text = torch.cat((ref_mel_batch, text_embedding), dim=-1)\n cat_mel_text_drop = torch.cat(\n (\n torch.zeros((batch, max_seq_len, self.n_mel_channels), dtype=torch.float32).to(self.device),\n text_embedding_drop,\n ),\n dim=-1,\n )\n\n time_expand = self.time_expand.repeat(2 * batch, 1, 1).contiguous()\n\n # Convert estimated_reference_target_mel_len to tensor\n input_lengths = torch.tensor(estimated_reference_target_mel_len, dtype=torch.int32)\n\n # combine above along the batch dimension\n inputs = {\n \"noise\": torch.cat((noise, noise), dim=0).contiguous(),\n \"cond\": torch.cat((cat_mel_text, cat_mel_text_drop), dim=0).contiguous(),\n \"time_expand\": time_expand,\n \"rope_cos\": torch.cat((rope_cos, rope_cos), dim=0).contiguous(),\n \"rope_sin\": torch.cat((rope_sin, rope_sin), dim=0).contiguous(),\n \"input_lengths\": torch.cat((input_lengths, input_lengths), dim=0).contiguous(),\n \"delta_t\": self.delta_t,\n }\n if use_perf and remove_input_padding:\n torch.cuda.nvtx.range_push(\"remove input padding\")\n if remove_input_padding:\n max_seq_len = inputs[\"cond\"].shape[1]\n inputs[\"noise\"] = remove_tensor_padding(inputs[\"noise\"], inputs[\"input_lengths\"])\n inputs[\"cond\"] = remove_tensor_padding(inputs[\"cond\"], inputs[\"input_lengths\"])\n # for time_expand, convert from B,D to B,T,D by repeat\n inputs[\"time_expand\"] = inputs[\"time_expand\"].unsqueeze(1).repeat(1, max_seq_len, 1, 1)\n inputs[\"time_expand\"] = remove_tensor_padding(inputs[\"time_expand\"], inputs[\"input_lengths\"])\n inputs[\"rope_cos\"] = remove_tensor_padding(inputs[\"rope_cos\"], inputs[\"input_lengths\"])\n inputs[\"rope_sin\"] = remove_tensor_padding(inputs[\"rope_sin\"], inputs[\"input_lengths\"])\n if use_perf and remove_input_padding:\n torch.cuda.nvtx.range_pop()\n for key in inputs:\n inputs[key] = inputs[key].to(self.device)\n if use_perf:\n torch.cuda.nvtx.range_pop()\n start_time = time.time()\n denoised = self.forward(**inputs, use_perf=use_perf)\n cost_time = time.time() - start_time\n if use_perf and remove_input_padding:\n torch.cuda.nvtx.range_push(\"remove input padding output\")\n if remove_input_padding:\n denoised_list = []\n start_idx = 0\n for i in range(batch):\n denoised_list.append(denoised[start_idx : start_idx + inputs[\"input_lengths\"][i]])\n start_idx += inputs[\"input_lengths\"][i]\n if use_perf and remove_input_padding:\n torch.cuda.nvtx.range_pop()\n return denoised_list, cost_time\n return denoised, cost_time","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.wrapper","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.f5_tts_trtllm.wrapper#L253-L262","kind":"function","name":"wrapper","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":253,"end_line":262,"context_start_line":233,"context_end_line":282,"code":" def _tensor_dtype(self, name):\n # return torch dtype given tensor name for convenience\n dtype = trt_dtype_to_torch(self.session.engine.get_tensor_dtype(name))\n return dtype\n\n def _setup(self, batch_size, seq_len):\n for i in range(self.session.engine.num_io_tensors):\n name = self.session.engine.get_tensor_name(i)\n if self.session.engine.get_tensor_mode(name) == trt.TensorIOMode.OUTPUT:\n shape = list(self.session.engine.get_tensor_shape(name))\n shape[0] = batch_size\n shape[1] = seq_len\n self.outputs[name] = torch.empty(shape, dtype=self._tensor_dtype(name), device=self.device)\n\n self.buffer_allocated = True\n\n def cuda_stream_guard(func):\n \"\"\"Sync external stream and set current stream to the one bound to the session. Reset on exit.\"\"\"\n\n @wraps(func)\n def wrapper(self, *args, **kwargs):\n external_stream = torch.cuda.current_stream()\n if external_stream != self.stream:\n external_stream.synchronize()\n torch.cuda.set_stream(self.stream)\n ret = func(self, *args, **kwargs)\n if external_stream != self.stream:\n self.stream.synchronize()\n torch.cuda.set_stream(external_stream)\n return ret\n\n return wrapper\n\n @cuda_stream_guard\n def forward(\n self,\n noise: torch.Tensor,\n cond: torch.Tensor,\n time_expand: torch.Tensor,\n rope_cos: torch.Tensor,\n rope_sin: torch.Tensor,\n input_lengths: torch.Tensor,\n delta_t: torch.Tensor,\n use_perf: bool = False,\n ):\n if use_perf:\n torch.cuda.nvtx.range_push(\"flow matching\")\n cfg_strength = 2.0\n batch_size = noise.shape[0]\n half_batch = batch_size // 2","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model","uri":"program://DMOSpeech2/module/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model#L1-L278","kind":"module","name":"src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","language":"python","start_line":1,"end_line":278,"context_start_line":1,"context_end_line":278,"code":"# Copyright 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.\n#\n# Redistribution and use in source and binary forms, with or without\n# modification, are permitted provided that the following conditions\n# are met:\n# * Redistributions of source code must retain the above copyright\n# notice, this list of conditions and the following disclaimer.\n# * Redistributions in binary form must reproduce the above copyright\n# notice, this list of conditions and the following disclaimer in the\n# documentation and/or other materials provided with the distribution.\n# * Neither the name of NVIDIA CORPORATION nor the names of its\n# contributors may be used to endorse or promote products derived\n# from this software without specific prior written permission.\n#\n# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY\n# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE\n# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR\n# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR\n# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,\n# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,\n# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR\n# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY\n# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT\n# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE\n# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\nimport json\nimport os\n\nimport jieba\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nimport triton_python_backend_utils as pb_utils\nfrom f5_tts_trtllm import F5TTS\nfrom pypinyin import Style, lazy_pinyin\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torch.utils.dlpack import from_dlpack, to_dlpack\n\n\ndef get_tokenizer(vocab_file_path: str):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file\n - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n with open(vocab_file_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n return vocab_char_map, vocab_size\n\n\ndef convert_char_to_pinyin(reference_target_texts_list, polyphone=True):\n final_reference_target_texts_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n\n for text in reference_target_texts_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)\n elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)\n elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_reference_target_texts_list.append(char_list)\n\n return final_reference_target_texts_list\n\n\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n): # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n return list_idx_tensors\n\n\nclass TritonPythonModel:\n def initialize(self, args):\n self.use_perf = True\n self.device = torch.device(\"cuda\")\n self.target_audio_sample_rate = 24000\n self.target_rms = 0.15 # target rms for audio\n self.n_fft = 1024\n self.win_length = 1024\n self.hop_length = 256\n self.n_mel_channels = 100\n self.max_mel_len = 3000\n self.head_dim = 64\n\n parameters = json.loads(args[\"model_config\"])[\"parameters\"]\n for key, value in parameters.items():\n parameters[key] = value[\"string_value\"]\n\n self.vocab_char_map, self.vocab_size = get_tokenizer(parameters[\"vocab_file\"])\n self.reference_sample_rate = int(parameters[\"reference_audio_sample_rate\"])\n self.resampler = torchaudio.transforms.Resample(self.reference_sample_rate, self.target_audio_sample_rate)\n\n self.tllm_model_dir = parameters[\"tllm_model_dir\"]\n config_file = os.path.join(self.tllm_model_dir, \"config.json\")\n with open(config_file) as f:\n config = json.load(f)\n self.model = F5TTS(\n config,\n debug_mode=False,\n tllm_model_dir=self.tllm_model_dir,\n model_path=parameters[\"model_path\"],\n vocab_size=self.vocab_size,\n )\n\n self.vocoder = parameters[\"vocoder\"]\n assert self.vocoder in [\"vocos\", \"bigvgan\"]\n if self.vocoder == \"vocos\":\n self.mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=self.target_audio_sample_rate,\n n_fft=self.n_fft,\n win_length=self.win_length,\n hop_length=self.hop_length,\n n_mels=self.n_mel_channels,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(self.device)\n self.compute_mel_fn = self.get_vocos_mel_spectrogram\n elif self.vocoder == \"bigvgan\":\n self.compute_mel_fn = self.get_bigvgan_mel_spectrogram\n\n def get_vocos_mel_spectrogram(self, waveform):\n mel = self.mel_stft(waveform)\n mel = mel.clamp(min=1e-5).log()\n return mel.transpose(1, 2)\n\n def forward_vocoder(self, mel):\n mel = mel.to(torch.float32).contiguous().cpu()\n input_tensor_0 = pb_utils.Tensor.from_dlpack(\"mel\", to_dlpack(mel))\n\n inference_request = pb_utils.InferenceRequest(\n model_name=\"vocoder\", requested_output_names=[\"waveform\"], inputs=[input_tensor_0]\n )\n inference_response = inference_request.exec()\n if inference_response.has_error():\n raise pb_utils.TritonModelException(inference_response.error().message())\n else:\n waveform = pb_utils.get_output_tensor_by_name(inference_response, \"waveform\")\n waveform = torch.utils.dlpack.from_dlpack(waveform.to_dlpack()).cpu()\n\n return waveform\n\n def execute(self, requests):\n (\n reference_text_list,\n target_text_list,\n reference_target_texts_list,\n estimated_reference_target_mel_len,\n reference_mel_len,\n ) = [], [], [], [], []\n mel_features_list = []\n if self.use_perf:\n torch.cuda.nvtx.range_push(\"preprocess\")\n for request in requests:\n wav_tensor = pb_utils.get_input_tensor_by_name(request, \"reference_wav\")\n wav_lens = pb_utils.get_input_tensor_by_name(request, \"reference_wav_len\")\n\n reference_text = pb_utils.get_input_tensor_by_name(request, \"reference_text\").as_numpy()\n reference_text = reference_text[0][0].decode(\"utf-8\")\n reference_text_list.append(reference_text)\n target_text = pb_utils.get_input_tensor_by_name(request, \"target_text\").as_numpy()\n target_text = target_text[0][0].decode(\"utf-8\")\n target_text_list.append(target_text)\n\n text = reference_text + target_text\n reference_target_texts_list.append(text)\n\n wav = from_dlpack(wav_tensor.to_dlpack())\n wav_len = from_dlpack(wav_lens.to_dlpack())\n wav_len = wav_len.squeeze()\n assert wav.shape[0] == 1, \"Only support batch size 1 for now.\"\n wav = wav[:, :wav_len]\n\n ref_rms = torch.sqrt(torch.mean(torch.square(wav)))\n if ref_rms < self.target_rms:\n wav = wav * self.target_rms / ref_rms\n if self.reference_sample_rate != self.target_audio_sample_rate:\n wav = self.resampler(wav)\n wav = wav.to(self.device)\n if self.use_perf:\n torch.cuda.nvtx.range_push(\"compute_mel\")\n mel_features = self.compute_mel_fn(wav)\n if self.use_perf:\n torch.cuda.nvtx.range_pop()\n mel_features_list.append(mel_features)\n\n reference_mel_len.append(mel_features.shape[1])\n estimated_reference_target_mel_len.append(\n int(\n mel_features.shape[1] * (1 + len(target_text.encode(\"utf-8\")) / len(reference_text.encode(\"utf-8\")))\n )\n )\n\n max_seq_len = min(max(estimated_reference_target_mel_len), self.max_mel_len)\n\n batch = len(requests)\n mel_features = torch.zeros((batch, max_seq_len, self.n_mel_channels), dtype=torch.float16).to(self.device)\n for i, mel in enumerate(mel_features_list):\n mel_features[i, : mel.shape[1], :] = mel\n\n reference_mel_len_tensor = torch.LongTensor(reference_mel_len).to(self.device)\n\n pinyin_list = convert_char_to_pinyin(reference_target_texts_list, polyphone=True)\n text_pad_sequence = list_str_to_idx(pinyin_list, self.vocab_char_map)\n\n for i, item in enumerate(text_pad_sequence):\n text_pad_sequence[i] = F.pad(\n item, (0, estimated_reference_target_mel_len[i] - len(item)), mode=\"constant\", value=-1\n )\n text_pad_sequence[i] += 1 # WAR: 0 is reserved for padding token, hard coding in F5-TTS\n text_pad_sequence = pad_sequence(text_pad_sequence, padding_value=-1, batch_first=True).to(self.device)\n text_pad_sequence = F.pad(\n text_pad_sequence, (0, max_seq_len - text_pad_sequence.shape[1]), mode=\"constant\", value=-1\n )\n if self.use_perf:\n torch.cuda.nvtx.range_pop()\n\n denoised, cost_time = self.model.sample(\n text_pad_sequence,\n mel_features,\n reference_mel_len_tensor,\n estimated_reference_target_mel_len,\n remove_input_padding=False,\n use_perf=self.use_perf,\n )\n if self.use_perf:\n torch.cuda.nvtx.range_push(\"vocoder\")\n\n responses = []\n for i in range(batch):\n ref_me_len = reference_mel_len[i]\n estimated_mel_len = estimated_reference_target_mel_len[i]\n denoised_one_item = denoised[i, ref_me_len:estimated_mel_len, :].unsqueeze(0).transpose(1, 2)\n audio = self.forward_vocoder(denoised_one_item)\n rms = torch.sqrt(torch.mean(torch.square(audio)))\n if rms < self.target_rms:\n audio = audio * self.target_rms / rms\n\n audio = pb_utils.Tensor.from_dlpack(\"waveform\", to_dlpack(audio))\n inference_response = pb_utils.InferenceResponse(output_tensors=[audio])\n responses.append(inference_response)\n if self.use_perf:\n torch.cuda.nvtx.range_pop()\n return responses","source_hash":"36b4bb696585c91cc596932fded5339316d7a0af0d39c9bbdeceb42a7b8f786c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.get_tokenizer","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.get_tokenizer#L40-L55","kind":"function","name":"get_tokenizer","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","language":"python","start_line":40,"end_line":55,"context_start_line":20,"context_end_line":75,"code":"# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,\n# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR\n# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY\n# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT\n# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE\n# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\nimport json\nimport os\n\nimport jieba\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nimport triton_python_backend_utils as pb_utils\nfrom f5_tts_trtllm import F5TTS\nfrom pypinyin import Style, lazy_pinyin\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torch.utils.dlpack import from_dlpack, to_dlpack\n\n\ndef get_tokenizer(vocab_file_path: str):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file\n - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n with open(vocab_file_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n return vocab_char_map, vocab_size\n\n\ndef convert_char_to_pinyin(reference_target_texts_list, polyphone=True):\n final_reference_target_texts_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n\n for text in reference_target_texts_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)","source_hash":"36b4bb696585c91cc596932fded5339316d7a0af0d39c9bbdeceb42a7b8f786c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.convert_char_to_pinyin","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.convert_char_to_pinyin#L58-L93","kind":"function","name":"convert_char_to_pinyin","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","language":"python","start_line":58,"end_line":93,"context_start_line":38,"context_end_line":113,"code":"\n\ndef get_tokenizer(vocab_file_path: str):\n \"\"\"\n tokenizer - \"pinyin\" do g2p for only chinese characters, need .txt vocab_file\n - \"char\" for char-wise tokenizer, need .txt vocab_file\n - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n with open(vocab_file_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n return vocab_char_map, vocab_size\n\n\ndef convert_char_to_pinyin(reference_target_texts_list, polyphone=True):\n final_reference_target_texts_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n\n for text in reference_target_texts_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)\n elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)\n elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_reference_target_texts_list.append(char_list)\n\n return final_reference_target_texts_list\n\n\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n): # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n return list_idx_tensors\n\n\nclass TritonPythonModel:\n def initialize(self, args):\n self.use_perf = True\n self.device = torch.device(\"cuda\")\n self.target_audio_sample_rate = 24000\n self.target_rms = 0.15 # target rms for audio\n self.n_fft = 1024\n self.win_length = 1024\n self.hop_length = 256","source_hash":"36b4bb696585c91cc596932fded5339316d7a0af0d39c9bbdeceb42a7b8f786c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.list_str_to_idx","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.list_str_to_idx#L96-L102","kind":"function","name":"list_str_to_idx","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","language":"python","start_line":96,"end_line":102,"context_start_line":76,"context_end_line":122,"code":" elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)\n elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_reference_target_texts_list.append(char_list)\n\n return final_reference_target_texts_list\n\n\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n): # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n return list_idx_tensors\n\n\nclass TritonPythonModel:\n def initialize(self, args):\n self.use_perf = True\n self.device = torch.device(\"cuda\")\n self.target_audio_sample_rate = 24000\n self.target_rms = 0.15 # target rms for audio\n self.n_fft = 1024\n self.win_length = 1024\n self.hop_length = 256\n self.n_mel_channels = 100\n self.max_mel_len = 3000\n self.head_dim = 64\n\n parameters = json.loads(args[\"model_config\"])[\"parameters\"]\n for key, value in parameters.items():\n parameters[key] = value[\"string_value\"]\n\n self.vocab_char_map, self.vocab_size = get_tokenizer(parameters[\"vocab_file\"])","source_hash":"36b4bb696585c91cc596932fded5339316d7a0af0d39c9bbdeceb42a7b8f786c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.TritonPythonModel","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.TritonPythonModel#L105-L278","kind":"class","name":"TritonPythonModel","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","language":"python","start_line":105,"end_line":278,"context_start_line":85,"context_end_line":278,"code":" char_list.extend(c)\n elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_reference_target_texts_list.append(char_list)\n\n return final_reference_target_texts_list\n\n\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n): # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n return list_idx_tensors\n\n\nclass TritonPythonModel:\n def initialize(self, args):\n self.use_perf = True\n self.device = torch.device(\"cuda\")\n self.target_audio_sample_rate = 24000\n self.target_rms = 0.15 # target rms for audio\n self.n_fft = 1024\n self.win_length = 1024\n self.hop_length = 256\n self.n_mel_channels = 100\n self.max_mel_len = 3000\n self.head_dim = 64\n\n parameters = json.loads(args[\"model_config\"])[\"parameters\"]\n for key, value in parameters.items():\n parameters[key] = value[\"string_value\"]\n\n self.vocab_char_map, self.vocab_size = get_tokenizer(parameters[\"vocab_file\"])\n self.reference_sample_rate = int(parameters[\"reference_audio_sample_rate\"])\n self.resampler = torchaudio.transforms.Resample(self.reference_sample_rate, self.target_audio_sample_rate)\n\n self.tllm_model_dir = parameters[\"tllm_model_dir\"]\n config_file = os.path.join(self.tllm_model_dir, \"config.json\")\n with open(config_file) as f:\n config = json.load(f)\n self.model = F5TTS(\n config,\n debug_mode=False,\n tllm_model_dir=self.tllm_model_dir,\n model_path=parameters[\"model_path\"],\n vocab_size=self.vocab_size,\n )\n\n self.vocoder = parameters[\"vocoder\"]\n assert self.vocoder in [\"vocos\", \"bigvgan\"]\n if self.vocoder == \"vocos\":\n self.mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=self.target_audio_sample_rate,\n n_fft=self.n_fft,\n win_length=self.win_length,\n hop_length=self.hop_length,\n n_mels=self.n_mel_channels,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(self.device)\n self.compute_mel_fn = self.get_vocos_mel_spectrogram\n elif self.vocoder == \"bigvgan\":\n self.compute_mel_fn = self.get_bigvgan_mel_spectrogram\n\n def get_vocos_mel_spectrogram(self, waveform):\n mel = self.mel_stft(waveform)\n mel = mel.clamp(min=1e-5).log()\n return mel.transpose(1, 2)\n\n def forward_vocoder(self, mel):\n mel = mel.to(torch.float32).contiguous().cpu()\n input_tensor_0 = pb_utils.Tensor.from_dlpack(\"mel\", to_dlpack(mel))\n\n inference_request = pb_utils.InferenceRequest(\n model_name=\"vocoder\", requested_output_names=[\"waveform\"], inputs=[input_tensor_0]\n )\n inference_response = inference_request.exec()\n if inference_response.has_error():\n raise pb_utils.TritonModelException(inference_response.error().message())\n else:\n waveform = pb_utils.get_output_tensor_by_name(inference_response, \"waveform\")\n waveform = torch.utils.dlpack.from_dlpack(waveform.to_dlpack()).cpu()\n\n return waveform\n\n def execute(self, requests):\n (\n reference_text_list,\n target_text_list,\n reference_target_texts_list,\n estimated_reference_target_mel_len,\n reference_mel_len,\n ) = [], [], [], [], []\n mel_features_list = []\n if self.use_perf:\n torch.cuda.nvtx.range_push(\"preprocess\")\n for request in requests:\n wav_tensor = pb_utils.get_input_tensor_by_name(request, \"reference_wav\")\n wav_lens = pb_utils.get_input_tensor_by_name(request, \"reference_wav_len\")\n\n reference_text = pb_utils.get_input_tensor_by_name(request, \"reference_text\").as_numpy()\n reference_text = reference_text[0][0].decode(\"utf-8\")\n reference_text_list.append(reference_text)\n target_text = pb_utils.get_input_tensor_by_name(request, \"target_text\").as_numpy()\n target_text = target_text[0][0].decode(\"utf-8\")\n target_text_list.append(target_text)\n\n text = reference_text + target_text\n reference_target_texts_list.append(text)\n\n wav = from_dlpack(wav_tensor.to_dlpack())\n wav_len = from_dlpack(wav_lens.to_dlpack())\n wav_len = wav_len.squeeze()\n assert wav.shape[0] == 1, \"Only support batch size 1 for now.\"\n wav = wav[:, :wav_len]\n\n ref_rms = torch.sqrt(torch.mean(torch.square(wav)))\n if ref_rms < self.target_rms:\n wav = wav * self.target_rms / ref_rms\n if self.reference_sample_rate != self.target_audio_sample_rate:\n wav = self.resampler(wav)\n wav = wav.to(self.device)\n if self.use_perf:\n torch.cuda.nvtx.range_push(\"compute_mel\")\n mel_features = self.compute_mel_fn(wav)\n if self.use_perf:\n torch.cuda.nvtx.range_pop()\n mel_features_list.append(mel_features)\n\n reference_mel_len.append(mel_features.shape[1])\n estimated_reference_target_mel_len.append(\n int(\n mel_features.shape[1] * (1 + len(target_text.encode(\"utf-8\")) / len(reference_text.encode(\"utf-8\")))\n )\n )\n\n max_seq_len = min(max(estimated_reference_target_mel_len), self.max_mel_len)\n\n batch = len(requests)\n mel_features = torch.zeros((batch, max_seq_len, self.n_mel_channels), dtype=torch.float16).to(self.device)\n for i, mel in enumerate(mel_features_list):\n mel_features[i, : mel.shape[1], :] = mel\n\n reference_mel_len_tensor = torch.LongTensor(reference_mel_len).to(self.device)\n\n pinyin_list = convert_char_to_pinyin(reference_target_texts_list, polyphone=True)\n text_pad_sequence = list_str_to_idx(pinyin_list, self.vocab_char_map)\n\n for i, item in enumerate(text_pad_sequence):\n text_pad_sequence[i] = F.pad(\n item, (0, estimated_reference_target_mel_len[i] - len(item)), mode=\"constant\", value=-1\n )\n text_pad_sequence[i] += 1 # WAR: 0 is reserved for padding token, hard coding in F5-TTS\n text_pad_sequence = pad_sequence(text_pad_sequence, padding_value=-1, batch_first=True).to(self.device)\n text_pad_sequence = F.pad(\n text_pad_sequence, (0, max_seq_len - text_pad_sequence.shape[1]), mode=\"constant\", value=-1\n )\n if self.use_perf:\n torch.cuda.nvtx.range_pop()\n\n denoised, cost_time = self.model.sample(\n text_pad_sequence,\n mel_features,\n reference_mel_len_tensor,\n estimated_reference_target_mel_len,\n remove_input_padding=False,\n use_perf=self.use_perf,\n )\n if self.use_perf:\n torch.cuda.nvtx.range_push(\"vocoder\")\n\n responses = []\n for i in range(batch):\n ref_me_len = reference_mel_len[i]\n estimated_mel_len = estimated_reference_target_mel_len[i]\n denoised_one_item = denoised[i, ref_me_len:estimated_mel_len, :].unsqueeze(0).transpose(1, 2)\n audio = self.forward_vocoder(denoised_one_item)\n rms = torch.sqrt(torch.mean(torch.square(audio)))\n if rms < self.target_rms:\n audio = audio * self.target_rms / rms\n\n audio = pb_utils.Tensor.from_dlpack(\"waveform\", to_dlpack(audio))\n inference_response = pb_utils.InferenceResponse(output_tensors=[audio])\n responses.append(inference_response)\n if self.use_perf:\n torch.cuda.nvtx.range_pop()\n return responses","source_hash":"36b4bb696585c91cc596932fded5339316d7a0af0d39c9bbdeceb42a7b8f786c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.is_chinese","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.is_chinese#L64-L65","kind":"function","name":"is_chinese","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","language":"python","start_line":64,"end_line":65,"context_start_line":44,"context_end_line":85,"code":" - \"byte\" for utf-8 tokenizer\n - \"custom\" if you're directly passing in a path to the vocab.txt you want to use\n vocab_size - if use \"pinyin\", all available pinyin types, common alphabets (also those with accent) and symbols\n - if use \"char\", derived from unfiltered character & symbol counts of custom dataset\n - if use \"byte\", set to 256 (unicode byte range)\n \"\"\"\n with open(vocab_file_path, \"r\", encoding=\"utf-8\") as f:\n vocab_char_map = {}\n for i, char in enumerate(f):\n vocab_char_map[char[:-1]] = i\n vocab_size = len(vocab_char_map)\n return vocab_char_map, vocab_size\n\n\ndef convert_char_to_pinyin(reference_target_texts_list, polyphone=True):\n final_reference_target_texts_list = []\n custom_trans = str.maketrans(\n {\";\": \",\", \"“\": '\"', \"”\": '\"', \"‘\": \"'\", \"’\": \"'\"}\n ) # add custom trans here, to address oov\n\n def is_chinese(c):\n return \"\\u3100\" <= c <= \"\\u9fff\" # common chinese characters\n\n for text in reference_target_texts_list:\n char_list = []\n text = text.translate(custom_trans)\n for seg in jieba.cut(text):\n seg_byte_len = len(bytes(seg, \"UTF-8\"))\n if seg_byte_len == len(seg): # if pure alphabets and symbols\n if char_list and seg_byte_len > 1 and char_list[-1] not in \" :'\\\"\":\n char_list.append(\" \")\n char_list.extend(seg)\n elif polyphone and seg_byte_len == 3 * len(seg): # if pure east asian characters\n seg_ = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)\n for i, c in enumerate(seg):\n if is_chinese(c):\n char_list.append(\" \")\n char_list.append(seg_[i])\n else: # if mixed characters, alphabets and symbols\n for c in seg:\n if ord(c) < 256:\n char_list.extend(c)","source_hash":"36b4bb696585c91cc596932fded5339316d7a0af0d39c9bbdeceb42a7b8f786c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.initialize","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.initialize#L106-L154","kind":"function","name":"initialize","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","language":"python","start_line":106,"end_line":154,"context_start_line":86,"context_end_line":174,"code":" elif is_chinese(c):\n char_list.append(\" \")\n char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))\n else:\n char_list.append(c)\n final_reference_target_texts_list.append(char_list)\n\n return final_reference_target_texts_list\n\n\ndef list_str_to_idx(\n text: list[str] | list[list[str]],\n vocab_char_map: dict[str, int], # {char: idx}\n padding_value=-1,\n): # noqa: F722\n list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text] # pinyin or char style\n return list_idx_tensors\n\n\nclass TritonPythonModel:\n def initialize(self, args):\n self.use_perf = True\n self.device = torch.device(\"cuda\")\n self.target_audio_sample_rate = 24000\n self.target_rms = 0.15 # target rms for audio\n self.n_fft = 1024\n self.win_length = 1024\n self.hop_length = 256\n self.n_mel_channels = 100\n self.max_mel_len = 3000\n self.head_dim = 64\n\n parameters = json.loads(args[\"model_config\"])[\"parameters\"]\n for key, value in parameters.items():\n parameters[key] = value[\"string_value\"]\n\n self.vocab_char_map, self.vocab_size = get_tokenizer(parameters[\"vocab_file\"])\n self.reference_sample_rate = int(parameters[\"reference_audio_sample_rate\"])\n self.resampler = torchaudio.transforms.Resample(self.reference_sample_rate, self.target_audio_sample_rate)\n\n self.tllm_model_dir = parameters[\"tllm_model_dir\"]\n config_file = os.path.join(self.tllm_model_dir, \"config.json\")\n with open(config_file) as f:\n config = json.load(f)\n self.model = F5TTS(\n config,\n debug_mode=False,\n tllm_model_dir=self.tllm_model_dir,\n model_path=parameters[\"model_path\"],\n vocab_size=self.vocab_size,\n )\n\n self.vocoder = parameters[\"vocoder\"]\n assert self.vocoder in [\"vocos\", \"bigvgan\"]\n if self.vocoder == \"vocos\":\n self.mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=self.target_audio_sample_rate,\n n_fft=self.n_fft,\n win_length=self.win_length,\n hop_length=self.hop_length,\n n_mels=self.n_mel_channels,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(self.device)\n self.compute_mel_fn = self.get_vocos_mel_spectrogram\n elif self.vocoder == \"bigvgan\":\n self.compute_mel_fn = self.get_bigvgan_mel_spectrogram\n\n def get_vocos_mel_spectrogram(self, waveform):\n mel = self.mel_stft(waveform)\n mel = mel.clamp(min=1e-5).log()\n return mel.transpose(1, 2)\n\n def forward_vocoder(self, mel):\n mel = mel.to(torch.float32).contiguous().cpu()\n input_tensor_0 = pb_utils.Tensor.from_dlpack(\"mel\", to_dlpack(mel))\n\n inference_request = pb_utils.InferenceRequest(\n model_name=\"vocoder\", requested_output_names=[\"waveform\"], inputs=[input_tensor_0]\n )\n inference_response = inference_request.exec()\n if inference_response.has_error():\n raise pb_utils.TritonModelException(inference_response.error().message())\n else:\n waveform = pb_utils.get_output_tensor_by_name(inference_response, \"waveform\")\n waveform = torch.utils.dlpack.from_dlpack(waveform.to_dlpack()).cpu()\n","source_hash":"36b4bb696585c91cc596932fded5339316d7a0af0d39c9bbdeceb42a7b8f786c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.get_vocos_mel_spectrogram","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.get_vocos_mel_spectrogram#L156-L159","kind":"function","name":"get_vocos_mel_spectrogram","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","language":"python","start_line":156,"end_line":159,"context_start_line":136,"context_end_line":179,"code":" )\n\n self.vocoder = parameters[\"vocoder\"]\n assert self.vocoder in [\"vocos\", \"bigvgan\"]\n if self.vocoder == \"vocos\":\n self.mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=self.target_audio_sample_rate,\n n_fft=self.n_fft,\n win_length=self.win_length,\n hop_length=self.hop_length,\n n_mels=self.n_mel_channels,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(self.device)\n self.compute_mel_fn = self.get_vocos_mel_spectrogram\n elif self.vocoder == \"bigvgan\":\n self.compute_mel_fn = self.get_bigvgan_mel_spectrogram\n\n def get_vocos_mel_spectrogram(self, waveform):\n mel = self.mel_stft(waveform)\n mel = mel.clamp(min=1e-5).log()\n return mel.transpose(1, 2)\n\n def forward_vocoder(self, mel):\n mel = mel.to(torch.float32).contiguous().cpu()\n input_tensor_0 = pb_utils.Tensor.from_dlpack(\"mel\", to_dlpack(mel))\n\n inference_request = pb_utils.InferenceRequest(\n model_name=\"vocoder\", requested_output_names=[\"waveform\"], inputs=[input_tensor_0]\n )\n inference_response = inference_request.exec()\n if inference_response.has_error():\n raise pb_utils.TritonModelException(inference_response.error().message())\n else:\n waveform = pb_utils.get_output_tensor_by_name(inference_response, \"waveform\")\n waveform = torch.utils.dlpack.from_dlpack(waveform.to_dlpack()).cpu()\n\n return waveform\n\n def execute(self, requests):\n (\n reference_text_list,","source_hash":"36b4bb696585c91cc596932fded5339316d7a0af0d39c9bbdeceb42a7b8f786c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.forward_vocoder","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.forward_vocoder#L161-L175","kind":"function","name":"forward_vocoder","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","language":"python","start_line":161,"end_line":175,"context_start_line":141,"context_end_line":195,"code":" self.mel_stft = torchaudio.transforms.MelSpectrogram(\n sample_rate=self.target_audio_sample_rate,\n n_fft=self.n_fft,\n win_length=self.win_length,\n hop_length=self.hop_length,\n n_mels=self.n_mel_channels,\n power=1,\n center=True,\n normalized=False,\n norm=None,\n ).to(self.device)\n self.compute_mel_fn = self.get_vocos_mel_spectrogram\n elif self.vocoder == \"bigvgan\":\n self.compute_mel_fn = self.get_bigvgan_mel_spectrogram\n\n def get_vocos_mel_spectrogram(self, waveform):\n mel = self.mel_stft(waveform)\n mel = mel.clamp(min=1e-5).log()\n return mel.transpose(1, 2)\n\n def forward_vocoder(self, mel):\n mel = mel.to(torch.float32).contiguous().cpu()\n input_tensor_0 = pb_utils.Tensor.from_dlpack(\"mel\", to_dlpack(mel))\n\n inference_request = pb_utils.InferenceRequest(\n model_name=\"vocoder\", requested_output_names=[\"waveform\"], inputs=[input_tensor_0]\n )\n inference_response = inference_request.exec()\n if inference_response.has_error():\n raise pb_utils.TritonModelException(inference_response.error().message())\n else:\n waveform = pb_utils.get_output_tensor_by_name(inference_response, \"waveform\")\n waveform = torch.utils.dlpack.from_dlpack(waveform.to_dlpack()).cpu()\n\n return waveform\n\n def execute(self, requests):\n (\n reference_text_list,\n target_text_list,\n reference_target_texts_list,\n estimated_reference_target_mel_len,\n reference_mel_len,\n ) = [], [], [], [], []\n mel_features_list = []\n if self.use_perf:\n torch.cuda.nvtx.range_push(\"preprocess\")\n for request in requests:\n wav_tensor = pb_utils.get_input_tensor_by_name(request, \"reference_wav\")\n wav_lens = pb_utils.get_input_tensor_by_name(request, \"reference_wav_len\")\n\n reference_text = pb_utils.get_input_tensor_by_name(request, \"reference_text\").as_numpy()\n reference_text = reference_text[0][0].decode(\"utf-8\")\n reference_text_list.append(reference_text)\n target_text = pb_utils.get_input_tensor_by_name(request, \"target_text\").as_numpy()","source_hash":"36b4bb696585c91cc596932fded5339316d7a0af0d39c9bbdeceb42a7b8f786c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.execute","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.model_repo_f5_tts.f5_tts.1.model.execute#L177-L278","kind":"function","name":"execute","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","language":"python","start_line":177,"end_line":278,"context_start_line":157,"context_end_line":278,"code":" mel = self.mel_stft(waveform)\n mel = mel.clamp(min=1e-5).log()\n return mel.transpose(1, 2)\n\n def forward_vocoder(self, mel):\n mel = mel.to(torch.float32).contiguous().cpu()\n input_tensor_0 = pb_utils.Tensor.from_dlpack(\"mel\", to_dlpack(mel))\n\n inference_request = pb_utils.InferenceRequest(\n model_name=\"vocoder\", requested_output_names=[\"waveform\"], inputs=[input_tensor_0]\n )\n inference_response = inference_request.exec()\n if inference_response.has_error():\n raise pb_utils.TritonModelException(inference_response.error().message())\n else:\n waveform = pb_utils.get_output_tensor_by_name(inference_response, \"waveform\")\n waveform = torch.utils.dlpack.from_dlpack(waveform.to_dlpack()).cpu()\n\n return waveform\n\n def execute(self, requests):\n (\n reference_text_list,\n target_text_list,\n reference_target_texts_list,\n estimated_reference_target_mel_len,\n reference_mel_len,\n ) = [], [], [], [], []\n mel_features_list = []\n if self.use_perf:\n torch.cuda.nvtx.range_push(\"preprocess\")\n for request in requests:\n wav_tensor = pb_utils.get_input_tensor_by_name(request, \"reference_wav\")\n wav_lens = pb_utils.get_input_tensor_by_name(request, \"reference_wav_len\")\n\n reference_text = pb_utils.get_input_tensor_by_name(request, \"reference_text\").as_numpy()\n reference_text = reference_text[0][0].decode(\"utf-8\")\n reference_text_list.append(reference_text)\n target_text = pb_utils.get_input_tensor_by_name(request, \"target_text\").as_numpy()\n target_text = target_text[0][0].decode(\"utf-8\")\n target_text_list.append(target_text)\n\n text = reference_text + target_text\n reference_target_texts_list.append(text)\n\n wav = from_dlpack(wav_tensor.to_dlpack())\n wav_len = from_dlpack(wav_lens.to_dlpack())\n wav_len = wav_len.squeeze()\n assert wav.shape[0] == 1, \"Only support batch size 1 for now.\"\n wav = wav[:, :wav_len]\n\n ref_rms = torch.sqrt(torch.mean(torch.square(wav)))\n if ref_rms < self.target_rms:\n wav = wav * self.target_rms / ref_rms\n if self.reference_sample_rate != self.target_audio_sample_rate:\n wav = self.resampler(wav)\n wav = wav.to(self.device)\n if self.use_perf:\n torch.cuda.nvtx.range_push(\"compute_mel\")\n mel_features = self.compute_mel_fn(wav)\n if self.use_perf:\n torch.cuda.nvtx.range_pop()\n mel_features_list.append(mel_features)\n\n reference_mel_len.append(mel_features.shape[1])\n estimated_reference_target_mel_len.append(\n int(\n mel_features.shape[1] * (1 + len(target_text.encode(\"utf-8\")) / len(reference_text.encode(\"utf-8\")))\n )\n )\n\n max_seq_len = min(max(estimated_reference_target_mel_len), self.max_mel_len)\n\n batch = len(requests)\n mel_features = torch.zeros((batch, max_seq_len, self.n_mel_channels), dtype=torch.float16).to(self.device)\n for i, mel in enumerate(mel_features_list):\n mel_features[i, : mel.shape[1], :] = mel\n\n reference_mel_len_tensor = torch.LongTensor(reference_mel_len).to(self.device)\n\n pinyin_list = convert_char_to_pinyin(reference_target_texts_list, polyphone=True)\n text_pad_sequence = list_str_to_idx(pinyin_list, self.vocab_char_map)\n\n for i, item in enumerate(text_pad_sequence):\n text_pad_sequence[i] = F.pad(\n item, (0, estimated_reference_target_mel_len[i] - len(item)), mode=\"constant\", value=-1\n )\n text_pad_sequence[i] += 1 # WAR: 0 is reserved for padding token, hard coding in F5-TTS\n text_pad_sequence = pad_sequence(text_pad_sequence, padding_value=-1, batch_first=True).to(self.device)\n text_pad_sequence = F.pad(\n text_pad_sequence, (0, max_seq_len - text_pad_sequence.shape[1]), mode=\"constant\", value=-1\n )\n if self.use_perf:\n torch.cuda.nvtx.range_pop()\n\n denoised, cost_time = self.model.sample(\n text_pad_sequence,\n mel_features,\n reference_mel_len_tensor,\n estimated_reference_target_mel_len,\n remove_input_padding=False,\n use_perf=self.use_perf,\n )\n if self.use_perf:\n torch.cuda.nvtx.range_push(\"vocoder\")\n\n responses = []\n for i in range(batch):\n ref_me_len = reference_mel_len[i]\n estimated_mel_len = estimated_reference_target_mel_len[i]\n denoised_one_item = denoised[i, ref_me_len:estimated_mel_len, :].unsqueeze(0).transpose(1, 2)\n audio = self.forward_vocoder(denoised_one_item)\n rms = torch.sqrt(torch.mean(torch.square(audio)))\n if rms < self.target_rms:\n audio = audio * self.target_rms / rms\n\n audio = pb_utils.Tensor.from_dlpack(\"waveform\", to_dlpack(audio))\n inference_response = pb_utils.InferenceResponse(output_tensors=[audio])\n responses.append(inference_response)\n if self.use_perf:\n torch.cuda.nvtx.range_pop()\n return responses","source_hash":"36b4bb696585c91cc596932fded5339316d7a0af0d39c9bbdeceb42a7b8f786c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.model","uri":"program://DMOSpeech2/module/src.f5_tts.runtime.triton_trtllm.patch.f5tts.model#L1-L222","kind":"module","name":"src.f5_tts.runtime.triton_trtllm.patch.f5tts.model","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/model.py","language":"python","start_line":1,"end_line":222,"context_start_line":1,"context_end_line":222,"code":"from __future__ import annotations\n\nimport os\nimport sys\nfrom collections import OrderedDict\n\nimport tensorrt as trt\nfrom tensorrt_llm._common import default_net\n\nfrom ..._utils import str_dtype_to_trt\nfrom ...functional import Tensor, concat\nfrom ...layers import Linear\nfrom ...module import Module, ModuleList\nfrom ...plugin import current_all_reduce_helper\nfrom ..modeling_utils import PretrainedConfig, PretrainedModel\nfrom .modules import AdaLayerNormZero_Final, ConvPositionEmbedding, DiTBlock, TimestepEmbedding\n\n\ncurrent_file_path = os.path.abspath(__file__)\nparent_dir = os.path.dirname(current_file_path)\nsys.path.append(parent_dir)\n\n\nclass InputEmbedding(Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x, cond):\n x = self.proj(concat([x, cond], dim=-1))\n return self.conv_pos_embed(x) + x\n\n\nclass F5TTS(PretrainedModel):\n def __init__(self, config: PretrainedConfig):\n super().__init__(config)\n self.dtype = str_dtype_to_trt(config.dtype)\n\n self.time_embed = TimestepEmbedding(config.hidden_size)\n self.input_embed = InputEmbedding(config.mel_dim, config.text_dim, config.hidden_size)\n\n self.dim = config.hidden_size\n self.depth = config.num_hidden_layers\n self.transformer_blocks = ModuleList(\n [\n DiTBlock(\n dim=self.dim,\n heads=config.num_attention_heads,\n dim_head=config.dim_head,\n ff_mult=config.ff_mult,\n dropout=config.dropout,\n )\n for _ in range(self.depth)\n ]\n )\n\n self.norm_out = AdaLayerNormZero_Final(config.hidden_size) # final modulation\n self.proj_out = Linear(config.hidden_size, config.mel_dim)\n\n def forward(\n self,\n noise, # nosied input audio\n cond, # masked cond audio\n time, # time step\n rope_cos,\n rope_sin,\n input_lengths,\n scale=1.0,\n ):\n t = self.time_embed(time)\n x = self.input_embed(noise, cond)\n for block in self.transformer_blocks:\n x = block(x, t, rope_cos=rope_cos, rope_sin=rope_sin, input_lengths=input_lengths, scale=scale)\n denoise = self.proj_out(self.norm_out(x, t))\n denoise.mark_output(\"denoised\", self.dtype)\n return denoise\n\n def prepare_inputs(self, **kwargs):\n max_batch_size = kwargs[\"max_batch_size\"]\n batch_size_range = [2, 2, max_batch_size]\n mel_size = 100\n max_seq_len = 3000\n num_frames_range = [200, 2 * max_seq_len, max_seq_len * max_batch_size]\n hidden_size = 512\n concat_feature_dim = mel_size + hidden_size\n freq_embed_dim = 256\n head_dim = 64\n mapping = self.config.mapping\n if mapping.tp_size > 1:\n current_all_reduce_helper().set_workspace_tensor(mapping, 1)\n if default_net().plugin_config.remove_input_padding:\n noise = Tensor(\n name=\"noise\",\n dtype=self.dtype,\n shape=[-1, mel_size],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"n_mels\", [mel_size]),\n ]\n ),\n )\n cond = Tensor(\n name=\"cond\",\n dtype=self.dtype,\n shape=[-1, concat_feature_dim],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"embeded_length\", [concat_feature_dim]),\n ]\n ),\n )\n time = Tensor(\n name=\"time\",\n dtype=self.dtype,\n shape=[-1, freq_embed_dim],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"freq_dim\", [freq_embed_dim]),\n ]\n ),\n )\n rope_cos = Tensor(\n name=\"rope_cos\",\n dtype=self.dtype,\n shape=[-1, head_dim],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"head_dim\", [head_dim]),\n ]\n ),\n )\n rope_sin = Tensor(\n name=\"rope_sin\",\n dtype=self.dtype,\n shape=[-1, head_dim],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"head_dim\", [head_dim]),\n ]\n ),\n )\n\n else:\n noise = Tensor(\n name=\"noise\",\n dtype=self.dtype,\n shape=[-1, -1, mel_size],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"max_duratuion\", [[100, max_seq_len // 2, max_seq_len]]),\n (\"n_mels\", [mel_size]),\n ]\n ),\n )\n cond = Tensor(\n name=\"cond\",\n dtype=self.dtype,\n shape=[-1, -1, concat_feature_dim],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"max_duratuion\", [[100, max_seq_len // 2, max_seq_len]]),\n (\"embeded_length\", [concat_feature_dim]),\n ]\n ),\n )\n time = Tensor(\n name=\"time\",\n dtype=self.dtype,\n shape=[-1, freq_embed_dim],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"freq_dim\", [freq_embed_dim]),\n ]\n ),\n )\n rope_cos = Tensor(\n name=\"rope_cos\",\n dtype=self.dtype,\n shape=[-1, -1, head_dim],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"max_duratuion\", [[100, max_seq_len // 2, max_seq_len]]),\n (\"head_dim\", [head_dim]),\n ]\n ),\n )\n rope_sin = Tensor(\n name=\"rope_sin\",\n dtype=self.dtype,\n shape=[-1, -1, head_dim],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"max_duratuion\", [[100, max_seq_len // 2, max_seq_len]]),\n (\"head_dim\", [head_dim]),\n ]\n ),\n )\n input_lengths = Tensor(\n name=\"input_lengths\",\n dtype=trt.int32,\n shape=[-1],\n dim_range=OrderedDict([(\"batch_size\", [batch_size_range])]),\n )\n return {\n \"noise\": noise,\n \"cond\": cond,\n \"time\": time,\n \"rope_cos\": rope_cos,\n \"rope_sin\": rope_sin,\n \"input_lengths\": input_lengths,\n }","source_hash":"d67c74e53ee7cb463f0df30b2b0461c72b3535beee99192466313eb247dc4c9c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.model.InputEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.patch.f5tts.model.InputEmbedding#L24-L32","kind":"class","name":"InputEmbedding","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/model.py","language":"python","start_line":24,"end_line":32,"context_start_line":4,"context_end_line":52,"code":"import sys\nfrom collections import OrderedDict\n\nimport tensorrt as trt\nfrom tensorrt_llm._common import default_net\n\nfrom ..._utils import str_dtype_to_trt\nfrom ...functional import Tensor, concat\nfrom ...layers import Linear\nfrom ...module import Module, ModuleList\nfrom ...plugin import current_all_reduce_helper\nfrom ..modeling_utils import PretrainedConfig, PretrainedModel\nfrom .modules import AdaLayerNormZero_Final, ConvPositionEmbedding, DiTBlock, TimestepEmbedding\n\n\ncurrent_file_path = os.path.abspath(__file__)\nparent_dir = os.path.dirname(current_file_path)\nsys.path.append(parent_dir)\n\n\nclass InputEmbedding(Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x, cond):\n x = self.proj(concat([x, cond], dim=-1))\n return self.conv_pos_embed(x) + x\n\n\nclass F5TTS(PretrainedModel):\n def __init__(self, config: PretrainedConfig):\n super().__init__(config)\n self.dtype = str_dtype_to_trt(config.dtype)\n\n self.time_embed = TimestepEmbedding(config.hidden_size)\n self.input_embed = InputEmbedding(config.mel_dim, config.text_dim, config.hidden_size)\n\n self.dim = config.hidden_size\n self.depth = config.num_hidden_layers\n self.transformer_blocks = ModuleList(\n [\n DiTBlock(\n dim=self.dim,\n heads=config.num_attention_heads,\n dim_head=config.dim_head,\n ff_mult=config.ff_mult,\n dropout=config.dropout,","source_hash":"d67c74e53ee7cb463f0df30b2b0461c72b3535beee99192466313eb247dc4c9c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.model.F5TTS","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.patch.f5tts.model.F5TTS#L35-L222","kind":"class","name":"F5TTS","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/model.py","language":"python","start_line":35,"end_line":222,"context_start_line":15,"context_end_line":222,"code":"from ..modeling_utils import PretrainedConfig, PretrainedModel\nfrom .modules import AdaLayerNormZero_Final, ConvPositionEmbedding, DiTBlock, TimestepEmbedding\n\n\ncurrent_file_path = os.path.abspath(__file__)\nparent_dir = os.path.dirname(current_file_path)\nsys.path.append(parent_dir)\n\n\nclass InputEmbedding(Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x, cond):\n x = self.proj(concat([x, cond], dim=-1))\n return self.conv_pos_embed(x) + x\n\n\nclass F5TTS(PretrainedModel):\n def __init__(self, config: PretrainedConfig):\n super().__init__(config)\n self.dtype = str_dtype_to_trt(config.dtype)\n\n self.time_embed = TimestepEmbedding(config.hidden_size)\n self.input_embed = InputEmbedding(config.mel_dim, config.text_dim, config.hidden_size)\n\n self.dim = config.hidden_size\n self.depth = config.num_hidden_layers\n self.transformer_blocks = ModuleList(\n [\n DiTBlock(\n dim=self.dim,\n heads=config.num_attention_heads,\n dim_head=config.dim_head,\n ff_mult=config.ff_mult,\n dropout=config.dropout,\n )\n for _ in range(self.depth)\n ]\n )\n\n self.norm_out = AdaLayerNormZero_Final(config.hidden_size) # final modulation\n self.proj_out = Linear(config.hidden_size, config.mel_dim)\n\n def forward(\n self,\n noise, # nosied input audio\n cond, # masked cond audio\n time, # time step\n rope_cos,\n rope_sin,\n input_lengths,\n scale=1.0,\n ):\n t = self.time_embed(time)\n x = self.input_embed(noise, cond)\n for block in self.transformer_blocks:\n x = block(x, t, rope_cos=rope_cos, rope_sin=rope_sin, input_lengths=input_lengths, scale=scale)\n denoise = self.proj_out(self.norm_out(x, t))\n denoise.mark_output(\"denoised\", self.dtype)\n return denoise\n\n def prepare_inputs(self, **kwargs):\n max_batch_size = kwargs[\"max_batch_size\"]\n batch_size_range = [2, 2, max_batch_size]\n mel_size = 100\n max_seq_len = 3000\n num_frames_range = [200, 2 * max_seq_len, max_seq_len * max_batch_size]\n hidden_size = 512\n concat_feature_dim = mel_size + hidden_size\n freq_embed_dim = 256\n head_dim = 64\n mapping = self.config.mapping\n if mapping.tp_size > 1:\n current_all_reduce_helper().set_workspace_tensor(mapping, 1)\n if default_net().plugin_config.remove_input_padding:\n noise = Tensor(\n name=\"noise\",\n dtype=self.dtype,\n shape=[-1, mel_size],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"n_mels\", [mel_size]),\n ]\n ),\n )\n cond = Tensor(\n name=\"cond\",\n dtype=self.dtype,\n shape=[-1, concat_feature_dim],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"embeded_length\", [concat_feature_dim]),\n ]\n ),\n )\n time = Tensor(\n name=\"time\",\n dtype=self.dtype,\n shape=[-1, freq_embed_dim],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"freq_dim\", [freq_embed_dim]),\n ]\n ),\n )\n rope_cos = Tensor(\n name=\"rope_cos\",\n dtype=self.dtype,\n shape=[-1, head_dim],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"head_dim\", [head_dim]),\n ]\n ),\n )\n rope_sin = Tensor(\n name=\"rope_sin\",\n dtype=self.dtype,\n shape=[-1, head_dim],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"head_dim\", [head_dim]),\n ]\n ),\n )\n\n else:\n noise = Tensor(\n name=\"noise\",\n dtype=self.dtype,\n shape=[-1, -1, mel_size],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"max_duratuion\", [[100, max_seq_len // 2, max_seq_len]]),\n (\"n_mels\", [mel_size]),\n ]\n ),\n )\n cond = Tensor(\n name=\"cond\",\n dtype=self.dtype,\n shape=[-1, -1, concat_feature_dim],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"max_duratuion\", [[100, max_seq_len // 2, max_seq_len]]),\n (\"embeded_length\", [concat_feature_dim]),\n ]\n ),\n )\n time = Tensor(\n name=\"time\",\n dtype=self.dtype,\n shape=[-1, freq_embed_dim],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"freq_dim\", [freq_embed_dim]),\n ]\n ),\n )\n rope_cos = Tensor(\n name=\"rope_cos\",\n dtype=self.dtype,\n shape=[-1, -1, head_dim],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"max_duratuion\", [[100, max_seq_len // 2, max_seq_len]]),\n (\"head_dim\", [head_dim]),\n ]\n ),\n )\n rope_sin = Tensor(\n name=\"rope_sin\",\n dtype=self.dtype,\n shape=[-1, -1, head_dim],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"max_duratuion\", [[100, max_seq_len // 2, max_seq_len]]),\n (\"head_dim\", [head_dim]),\n ]\n ),\n )\n input_lengths = Tensor(\n name=\"input_lengths\",\n dtype=trt.int32,\n shape=[-1],\n dim_range=OrderedDict([(\"batch_size\", [batch_size_range])]),\n )\n return {\n \"noise\": noise,\n \"cond\": cond,\n \"time\": time,\n \"rope_cos\": rope_cos,\n \"rope_sin\": rope_sin,\n \"input_lengths\": input_lengths,\n }","source_hash":"d67c74e53ee7cb463f0df30b2b0461c72b3535beee99192466313eb247dc4c9c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.model.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.patch.f5tts.model.__init__#L36-L59","kind":"function","name":"__init__","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/model.py","language":"python","start_line":36,"end_line":59,"context_start_line":16,"context_end_line":79,"code":"from .modules import AdaLayerNormZero_Final, ConvPositionEmbedding, DiTBlock, TimestepEmbedding\n\n\ncurrent_file_path = os.path.abspath(__file__)\nparent_dir = os.path.dirname(current_file_path)\nsys.path.append(parent_dir)\n\n\nclass InputEmbedding(Module):\n def __init__(self, mel_dim, text_dim, out_dim):\n super().__init__()\n self.proj = Linear(mel_dim * 2 + text_dim, out_dim)\n self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)\n\n def forward(self, x, cond):\n x = self.proj(concat([x, cond], dim=-1))\n return self.conv_pos_embed(x) + x\n\n\nclass F5TTS(PretrainedModel):\n def __init__(self, config: PretrainedConfig):\n super().__init__(config)\n self.dtype = str_dtype_to_trt(config.dtype)\n\n self.time_embed = TimestepEmbedding(config.hidden_size)\n self.input_embed = InputEmbedding(config.mel_dim, config.text_dim, config.hidden_size)\n\n self.dim = config.hidden_size\n self.depth = config.num_hidden_layers\n self.transformer_blocks = ModuleList(\n [\n DiTBlock(\n dim=self.dim,\n heads=config.num_attention_heads,\n dim_head=config.dim_head,\n ff_mult=config.ff_mult,\n dropout=config.dropout,\n )\n for _ in range(self.depth)\n ]\n )\n\n self.norm_out = AdaLayerNormZero_Final(config.hidden_size) # final modulation\n self.proj_out = Linear(config.hidden_size, config.mel_dim)\n\n def forward(\n self,\n noise, # nosied input audio\n cond, # masked cond audio\n time, # time step\n rope_cos,\n rope_sin,\n input_lengths,\n scale=1.0,\n ):\n t = self.time_embed(time)\n x = self.input_embed(noise, cond)\n for block in self.transformer_blocks:\n x = block(x, t, rope_cos=rope_cos, rope_sin=rope_sin, input_lengths=input_lengths, scale=scale)\n denoise = self.proj_out(self.norm_out(x, t))\n denoise.mark_output(\"denoised\", self.dtype)\n return denoise\n\n def prepare_inputs(self, **kwargs):","source_hash":"d67c74e53ee7cb463f0df30b2b0461c72b3535beee99192466313eb247dc4c9c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.model.forward","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.patch.f5tts.model.forward#L61-L77","kind":"function","name":"forward","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/model.py","language":"python","start_line":61,"end_line":77,"context_start_line":41,"context_end_line":97,"code":" self.input_embed = InputEmbedding(config.mel_dim, config.text_dim, config.hidden_size)\n\n self.dim = config.hidden_size\n self.depth = config.num_hidden_layers\n self.transformer_blocks = ModuleList(\n [\n DiTBlock(\n dim=self.dim,\n heads=config.num_attention_heads,\n dim_head=config.dim_head,\n ff_mult=config.ff_mult,\n dropout=config.dropout,\n )\n for _ in range(self.depth)\n ]\n )\n\n self.norm_out = AdaLayerNormZero_Final(config.hidden_size) # final modulation\n self.proj_out = Linear(config.hidden_size, config.mel_dim)\n\n def forward(\n self,\n noise, # nosied input audio\n cond, # masked cond audio\n time, # time step\n rope_cos,\n rope_sin,\n input_lengths,\n scale=1.0,\n ):\n t = self.time_embed(time)\n x = self.input_embed(noise, cond)\n for block in self.transformer_blocks:\n x = block(x, t, rope_cos=rope_cos, rope_sin=rope_sin, input_lengths=input_lengths, scale=scale)\n denoise = self.proj_out(self.norm_out(x, t))\n denoise.mark_output(\"denoised\", self.dtype)\n return denoise\n\n def prepare_inputs(self, **kwargs):\n max_batch_size = kwargs[\"max_batch_size\"]\n batch_size_range = [2, 2, max_batch_size]\n mel_size = 100\n max_seq_len = 3000\n num_frames_range = [200, 2 * max_seq_len, max_seq_len * max_batch_size]\n hidden_size = 512\n concat_feature_dim = mel_size + hidden_size\n freq_embed_dim = 256\n head_dim = 64\n mapping = self.config.mapping\n if mapping.tp_size > 1:\n current_all_reduce_helper().set_workspace_tensor(mapping, 1)\n if default_net().plugin_config.remove_input_padding:\n noise = Tensor(\n name=\"noise\",\n dtype=self.dtype,\n shape=[-1, mel_size],\n dim_range=OrderedDict(","source_hash":"d67c74e53ee7cb463f0df30b2b0461c72b3535beee99192466313eb247dc4c9c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.model.prepare_inputs","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.patch.f5tts.model.prepare_inputs#L79-L222","kind":"function","name":"prepare_inputs","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/model.py","language":"python","start_line":79,"end_line":222,"context_start_line":59,"context_end_line":222,"code":" self.proj_out = Linear(config.hidden_size, config.mel_dim)\n\n def forward(\n self,\n noise, # nosied input audio\n cond, # masked cond audio\n time, # time step\n rope_cos,\n rope_sin,\n input_lengths,\n scale=1.0,\n ):\n t = self.time_embed(time)\n x = self.input_embed(noise, cond)\n for block in self.transformer_blocks:\n x = block(x, t, rope_cos=rope_cos, rope_sin=rope_sin, input_lengths=input_lengths, scale=scale)\n denoise = self.proj_out(self.norm_out(x, t))\n denoise.mark_output(\"denoised\", self.dtype)\n return denoise\n\n def prepare_inputs(self, **kwargs):\n max_batch_size = kwargs[\"max_batch_size\"]\n batch_size_range = [2, 2, max_batch_size]\n mel_size = 100\n max_seq_len = 3000\n num_frames_range = [200, 2 * max_seq_len, max_seq_len * max_batch_size]\n hidden_size = 512\n concat_feature_dim = mel_size + hidden_size\n freq_embed_dim = 256\n head_dim = 64\n mapping = self.config.mapping\n if mapping.tp_size > 1:\n current_all_reduce_helper().set_workspace_tensor(mapping, 1)\n if default_net().plugin_config.remove_input_padding:\n noise = Tensor(\n name=\"noise\",\n dtype=self.dtype,\n shape=[-1, mel_size],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"n_mels\", [mel_size]),\n ]\n ),\n )\n cond = Tensor(\n name=\"cond\",\n dtype=self.dtype,\n shape=[-1, concat_feature_dim],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"embeded_length\", [concat_feature_dim]),\n ]\n ),\n )\n time = Tensor(\n name=\"time\",\n dtype=self.dtype,\n shape=[-1, freq_embed_dim],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"freq_dim\", [freq_embed_dim]),\n ]\n ),\n )\n rope_cos = Tensor(\n name=\"rope_cos\",\n dtype=self.dtype,\n shape=[-1, head_dim],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"head_dim\", [head_dim]),\n ]\n ),\n )\n rope_sin = Tensor(\n name=\"rope_sin\",\n dtype=self.dtype,\n shape=[-1, head_dim],\n dim_range=OrderedDict(\n [\n (\"num_frames\", [num_frames_range]),\n (\"head_dim\", [head_dim]),\n ]\n ),\n )\n\n else:\n noise = Tensor(\n name=\"noise\",\n dtype=self.dtype,\n shape=[-1, -1, mel_size],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"max_duratuion\", [[100, max_seq_len // 2, max_seq_len]]),\n (\"n_mels\", [mel_size]),\n ]\n ),\n )\n cond = Tensor(\n name=\"cond\",\n dtype=self.dtype,\n shape=[-1, -1, concat_feature_dim],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"max_duratuion\", [[100, max_seq_len // 2, max_seq_len]]),\n (\"embeded_length\", [concat_feature_dim]),\n ]\n ),\n )\n time = Tensor(\n name=\"time\",\n dtype=self.dtype,\n shape=[-1, freq_embed_dim],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"freq_dim\", [freq_embed_dim]),\n ]\n ),\n )\n rope_cos = Tensor(\n name=\"rope_cos\",\n dtype=self.dtype,\n shape=[-1, -1, head_dim],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"max_duratuion\", [[100, max_seq_len // 2, max_seq_len]]),\n (\"head_dim\", [head_dim]),\n ]\n ),\n )\n rope_sin = Tensor(\n name=\"rope_sin\",\n dtype=self.dtype,\n shape=[-1, -1, head_dim],\n dim_range=OrderedDict(\n [\n (\"batch_size\", [batch_size_range]),\n (\"max_duratuion\", [[100, max_seq_len // 2, max_seq_len]]),\n (\"head_dim\", [head_dim]),\n ]\n ),\n )\n input_lengths = Tensor(\n name=\"input_lengths\",\n dtype=trt.int32,\n shape=[-1],\n dim_range=OrderedDict([(\"batch_size\", [batch_size_range])]),\n )\n return {\n \"noise\": noise,\n \"cond\": cond,\n \"time\": time,\n \"rope_cos\": rope_cos,\n \"rope_sin\": rope_sin,\n \"input_lengths\": input_lengths,\n }","source_hash":"d67c74e53ee7cb463f0df30b2b0461c72b3535beee99192466313eb247dc4c9c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules","uri":"program://DMOSpeech2/module/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules#L1-L412","kind":"module","name":"src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":1,"end_line":412,"context_start_line":1,"context_end_line":412,"code":"from __future__ import annotations\n\nimport math\nfrom typing import Optional\n\nimport numpy as np\nimport torch\nimport torch.nn.functional as F\nfrom tensorrt_llm._common import default_net\n\nfrom ..._utils import str_dtype_to_trt, trt_dtype_to_np\nfrom ...functional import (\n Tensor,\n bert_attention,\n cast,\n chunk,\n concat,\n constant,\n expand,\n expand_dims,\n expand_dims_like,\n expand_mask,\n gelu,\n matmul,\n permute,\n shape,\n silu,\n slice,\n softmax,\n squeeze,\n unsqueeze,\n view,\n)\nfrom ...layers import ColumnLinear, Conv1d, LayerNorm, Linear, Mish, RowLinear\nfrom ...module import Module\n\n\nclass FeedForward(Module):\n def __init__(self, dim, dim_out=None, mult=4, dropout=0.0):\n super().__init__()\n inner_dim = int(dim * mult)\n dim_out = dim_out if dim_out is not None else dim\n\n self.project_in = Linear(dim, inner_dim)\n self.ff = Linear(inner_dim, dim_out)\n\n def forward(self, x):\n return self.ff(gelu(self.project_in(x)))\n\n\nclass AdaLayerNormZero(Module):\n def __init__(self, dim):\n super().__init__()\n\n self.linear = Linear(dim, dim * 6)\n self.norm = LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb=None):\n emb = self.linear(silu(emb))\n shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = chunk(emb, 6, dim=1)\n x = self.norm(x)\n ones = constant(np.ones(1, dtype=np.float32)).cast(x.dtype)\n if default_net().plugin_config.remove_input_padding:\n x = x * (ones + scale_msa) + shift_msa\n else:\n x = x * (ones + unsqueeze(scale_msa, 1)) + unsqueeze(shift_msa, 1)\n return x, gate_msa, shift_mlp, scale_mlp, gate_mlp\n\n\nclass AdaLayerNormZero_Final(Module):\n def __init__(self, dim):\n super().__init__()\n\n self.linear = Linear(dim, dim * 2)\n\n self.norm = LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb):\n emb = self.linear(silu(emb))\n scale, shift = chunk(emb, 2, dim=1)\n ones = constant(np.ones(1, dtype=np.float32)).cast(x.dtype)\n if default_net().plugin_config.remove_input_padding:\n x = self.norm(x) * (ones + scale) + shift\n else:\n x = self.norm(x) * unsqueeze((ones + scale), 1)\n x = x + unsqueeze(shift, 1)\n return x\n\n\nclass ConvPositionEmbedding(Module):\n def __init__(self, dim, kernel_size=31, groups=16):\n super().__init__()\n assert kernel_size % 2 != 0\n self.conv1d1 = Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2)\n self.conv1d2 = Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2)\n self.mish = Mish()\n\n def forward(self, x, mask=None): # noqa: F722\n if default_net().plugin_config.remove_input_padding:\n x = unsqueeze(x, 0)\n x = permute(x, [0, 2, 1])\n x = self.mish(self.conv1d2(self.mish(self.conv1d1(x))))\n out = permute(x, [0, 2, 1])\n if default_net().plugin_config.remove_input_padding:\n out = squeeze(out, 0)\n return out\n\n\nclass Attention(Module):\n def __init__(\n self,\n processor: AttnProcessor,\n dim: int,\n heads: int = 16,\n dim_head: int = 64,\n dropout: float = 0.0,\n context_dim: Optional[int] = None, # if not None -> joint attention\n context_pre_only=None,\n ):\n super().__init__()\n\n if not hasattr(F, \"scaled_dot_product_attention\"):\n raise ImportError(\"Attention equires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.\")\n\n self.processor = processor\n\n self.dim = dim # hidden_size\n self.heads = heads\n self.inner_dim = dim_head * heads\n self.dropout = dropout\n self.attention_head_size = dim_head\n self.context_dim = context_dim\n self.context_pre_only = context_pre_only\n self.tp_size = 1\n self.num_attention_heads = heads // self.tp_size\n self.num_attention_kv_heads = heads // self.tp_size # 8\n self.dtype = str_dtype_to_trt(\"float32\")\n self.attention_hidden_size = self.attention_head_size * self.num_attention_heads\n self.to_q = ColumnLinear(\n dim,\n self.tp_size * self.num_attention_heads * self.attention_head_size,\n bias=True,\n dtype=self.dtype,\n tp_group=None,\n tp_size=self.tp_size,\n )\n self.to_k = ColumnLinear(\n dim,\n self.tp_size * self.num_attention_heads * self.attention_head_size,\n bias=True,\n dtype=self.dtype,\n tp_group=None,\n tp_size=self.tp_size,\n )\n self.to_v = ColumnLinear(\n dim,\n self.tp_size * self.num_attention_heads * self.attention_head_size,\n bias=True,\n dtype=self.dtype,\n tp_group=None,\n tp_size=self.tp_size,\n )\n\n if self.context_dim is not None:\n self.to_k_c = Linear(context_dim, self.inner_dim)\n self.to_v_c = Linear(context_dim, self.inner_dim)\n if self.context_pre_only is not None:\n self.to_q_c = Linear(context_dim, self.inner_dim)\n\n self.to_out = RowLinear(\n self.tp_size * self.num_attention_heads * self.attention_head_size,\n dim,\n bias=True,\n dtype=self.dtype,\n tp_group=None,\n tp_size=self.tp_size,\n )\n\n if self.context_pre_only is not None and not self.context_pre_only:\n self.to_out_c = Linear(self.inner_dim, dim)\n\n def forward(\n self,\n x, # noised input x\n rope_cos,\n rope_sin,\n input_lengths,\n c=None, # context c\n scale=1.0,\n rope=None,\n c_rope=None, # rotary position embedding for c\n ) -> torch.Tensor:\n if c is not None:\n return self.processor(self, x, c=c, input_lengths=input_lengths, scale=scale, rope=rope, c_rope=c_rope)\n else:\n return self.processor(\n self, x, rope_cos=rope_cos, rope_sin=rope_sin, input_lengths=input_lengths, scale=scale\n )\n\n\ndef rotate_every_two_3dim(tensor: Tensor) -> Tensor:\n shape_tensor = concat(\n [shape(tensor, i) / 2 if i == (tensor.ndim() - 1) else shape(tensor, i) for i in range(tensor.ndim())]\n )\n if default_net().plugin_config.remove_input_padding:\n assert tensor.ndim() == 2\n x1 = slice(tensor, [0, 0], shape_tensor, [1, 2])\n x2 = slice(tensor, [0, 1], shape_tensor, [1, 2])\n x1 = expand_dims(x1, 2)\n x2 = expand_dims(x2, 2)\n zero = constant(np.ascontiguousarray(np.zeros([1], dtype=trt_dtype_to_np(tensor.dtype))))\n x2 = zero - x2\n x = concat([x2, x1], 2)\n out = view(x, concat([shape(x, 0), shape(x, 1) * 2]))\n else:\n assert tensor.ndim() == 3\n\n x1 = slice(tensor, [0, 0, 0], shape_tensor, [1, 1, 2])\n x2 = slice(tensor, [0, 0, 1], shape_tensor, [1, 1, 2])\n x1 = expand_dims(x1, 3)\n x2 = expand_dims(x2, 3)\n zero = constant(np.ascontiguousarray(np.zeros([1], dtype=trt_dtype_to_np(tensor.dtype))))\n x2 = zero - x2\n x = concat([x2, x1], 3)\n out = view(x, concat([shape(x, 0), shape(x, 1), shape(x, 2) * 2]))\n\n return out\n\n\ndef apply_rotary_pos_emb_3dim(x, rope_cos, rope_sin):\n if default_net().plugin_config.remove_input_padding:\n rot_dim = shape(rope_cos, -1) # 64\n new_t_shape = concat([shape(x, 0), rot_dim]) # (-1, 64)\n x_ = slice(x, [0, 0], new_t_shape, [1, 1])\n end_dim = shape(x, -1) - shape(rope_cos, -1)\n new_t_unrotated_shape = concat([shape(x, 0), end_dim]) # (2, -1, 960)\n x_unrotated = slice(x, concat([0, rot_dim]), new_t_unrotated_shape, [1, 1])\n out = concat([x_ * rope_cos + rotate_every_two_3dim(x_) * rope_sin, x_unrotated], dim=-1)\n else:\n rot_dim = shape(rope_cos, 2) # 64\n new_t_shape = concat([shape(x, 0), shape(x, 1), rot_dim]) # (2, -1, 64)\n x_ = slice(x, [0, 0, 0], new_t_shape, [1, 1, 1])\n end_dim = shape(x, 2) - shape(rope_cos, 2)\n new_t_unrotated_shape = concat([shape(x, 0), shape(x, 1), end_dim]) # (2, -1, 960)\n x_unrotated = slice(x, concat([0, 0, rot_dim]), new_t_unrotated_shape, [1, 1, 1])\n out = concat([x_ * rope_cos + rotate_every_two_3dim(x_) * rope_sin, x_unrotated], dim=-1)\n return out\n\n\nclass AttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn,\n x, # noised input x\n rope_cos,\n rope_sin,\n input_lengths,\n scale=1.0,\n rope=None,\n ) -> torch.FloatTensor:\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n # k,v,q all (2,1226,1024)\n query = apply_rotary_pos_emb_3dim(query, rope_cos, rope_sin)\n key = apply_rotary_pos_emb_3dim(key, rope_cos, rope_sin)\n\n # attention\n inner_dim = key.shape[-1]\n norm_factor = math.sqrt(attn.attention_head_size)\n q_scaling = 1.0 / norm_factor\n mask = None\n if not default_net().plugin_config.remove_input_padding:\n N = shape(x, 1)\n B = shape(x, 0)\n seq_len_2d = concat([1, N])\n max_position_embeddings = 4096\n # create position ids\n position_ids_buffer = constant(np.expand_dims(np.arange(max_position_embeddings).astype(np.int32), 0))\n tmp_position_ids = slice(position_ids_buffer, starts=[0, 0], sizes=seq_len_2d)\n tmp_position_ids = expand(tmp_position_ids, concat([B, N])) # BxL\n tmp_input_lengths = unsqueeze(input_lengths, 1) # Bx1\n tmp_input_lengths = expand(tmp_input_lengths, concat([B, N])) # BxL\n mask = tmp_position_ids < tmp_input_lengths # BxL\n mask = mask.cast(\"int32\")\n\n if default_net().plugin_config.bert_attention_plugin:\n qkv = concat([query, key, value], dim=-1)\n # TRT plugin mode\n assert input_lengths is not None\n if default_net().plugin_config.remove_input_padding:\n qkv = qkv.view(concat([-1, 3 * inner_dim]))\n max_input_length = constant(\n np.zeros(\n [\n 2048,\n ],\n dtype=np.int32,\n )\n )\n else:\n max_input_length = None\n context = bert_attention(\n qkv,\n input_lengths,\n attn.num_attention_heads,\n attn.attention_head_size,\n q_scaling=q_scaling,\n max_input_length=max_input_length,\n )\n else:\n assert not default_net().plugin_config.remove_input_padding\n\n def transpose_for_scores(x):\n new_x_shape = concat([shape(x, 0), shape(x, 1), attn.num_attention_heads, attn.attention_head_size])\n\n y = x.view(new_x_shape)\n y = y.transpose(1, 2)\n return y\n\n def transpose_for_scores_k(x):\n new_x_shape = concat([shape(x, 0), shape(x, 1), attn.num_attention_heads, attn.attention_head_size])\n\n y = x.view(new_x_shape)\n y = y.permute([0, 2, 3, 1])\n return y\n\n query = transpose_for_scores(query)\n key = transpose_for_scores_k(key)\n value = transpose_for_scores(value)\n\n attention_scores = matmul(query, key, use_fp32_acc=False)\n\n if mask is not None:\n attention_mask = expand_mask(mask, shape(query, 2))\n attention_mask = cast(attention_mask, attention_scores.dtype)\n attention_scores = attention_scores + attention_mask\n\n attention_probs = softmax(attention_scores, dim=-1)\n\n context = matmul(attention_probs, value, use_fp32_acc=False).transpose(1, 2)\n context = context.view(concat([shape(context, 0), shape(context, 1), attn.attention_hidden_size]))\n context = attn.to_out(context)\n if mask is not None:\n mask = mask.view(concat([shape(mask, 0), shape(mask, 1), 1]))\n mask = expand_dims_like(mask, context)\n mask = cast(mask, context.dtype)\n context = context * mask\n return context\n\n\n# DiT Block\nclass DiTBlock(Module):\n def __init__(self, dim, heads, dim_head, ff_mult=2, dropout=0.1):\n super().__init__()\n\n self.attn_norm = AdaLayerNormZero(dim)\n self.attn = Attention(\n processor=AttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n )\n\n self.ff_norm = LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n self.ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout)\n\n def forward(\n self, x, t, rope_cos, rope_sin, input_lengths, scale=1.0, rope=ModuleNotFoundError\n ): # x: noised input, t: time embedding\n # pre-norm & modulation for attention input\n norm, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.attn_norm(x, emb=t)\n # attention\n # norm ----> (2,1226,1024)\n attn_output = self.attn(x=norm, rope_cos=rope_cos, rope_sin=rope_sin, input_lengths=input_lengths, scale=scale)\n\n # process attention output for input x\n if default_net().plugin_config.remove_input_padding:\n x = x + gate_msa * attn_output\n else:\n x = x + unsqueeze(gate_msa, 1) * attn_output\n ones = constant(np.ones(1, dtype=np.float32)).cast(x.dtype)\n if default_net().plugin_config.remove_input_padding:\n norm = self.ff_norm(x) * (ones + scale_mlp) + shift_mlp\n else:\n norm = self.ff_norm(x) * (ones + unsqueeze(scale_mlp, 1)) + unsqueeze(shift_mlp, 1)\n # norm = self.ff_norm(x) * (ones + scale_mlp) + shift_mlp\n ff_output = self.ff(norm)\n if default_net().plugin_config.remove_input_padding:\n x = x + gate_mlp * ff_output\n else:\n x = x + unsqueeze(gate_mlp, 1) * ff_output\n\n return x\n\n\nclass TimestepEmbedding(Module):\n def __init__(self, dim, freq_embed_dim=256, dtype=None):\n super().__init__()\n # self.time_embed = SinusPositionEmbedding(freq_embed_dim)\n self.mlp1 = Linear(freq_embed_dim, dim, bias=True, dtype=dtype)\n self.mlp2 = Linear(dim, dim, bias=True, dtype=dtype)\n\n def forward(self, timestep):\n t_freq = self.mlp1(timestep)\n t_freq = silu(t_freq)\n t_emb = self.mlp2(t_freq)\n return t_emb","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.FeedForward","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.FeedForward#L38-L48","kind":"class","name":"FeedForward","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":38,"end_line":48,"context_start_line":18,"context_end_line":68,"code":" constant,\n expand,\n expand_dims,\n expand_dims_like,\n expand_mask,\n gelu,\n matmul,\n permute,\n shape,\n silu,\n slice,\n softmax,\n squeeze,\n unsqueeze,\n view,\n)\nfrom ...layers import ColumnLinear, Conv1d, LayerNorm, Linear, Mish, RowLinear\nfrom ...module import Module\n\n\nclass FeedForward(Module):\n def __init__(self, dim, dim_out=None, mult=4, dropout=0.0):\n super().__init__()\n inner_dim = int(dim * mult)\n dim_out = dim_out if dim_out is not None else dim\n\n self.project_in = Linear(dim, inner_dim)\n self.ff = Linear(inner_dim, dim_out)\n\n def forward(self, x):\n return self.ff(gelu(self.project_in(x)))\n\n\nclass AdaLayerNormZero(Module):\n def __init__(self, dim):\n super().__init__()\n\n self.linear = Linear(dim, dim * 6)\n self.norm = LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb=None):\n emb = self.linear(silu(emb))\n shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = chunk(emb, 6, dim=1)\n x = self.norm(x)\n ones = constant(np.ones(1, dtype=np.float32)).cast(x.dtype)\n if default_net().plugin_config.remove_input_padding:\n x = x * (ones + scale_msa) + shift_msa\n else:\n x = x * (ones + unsqueeze(scale_msa, 1)) + unsqueeze(shift_msa, 1)\n return x, gate_msa, shift_mlp, scale_mlp, gate_mlp\n","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.AdaLayerNormZero","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.AdaLayerNormZero#L51-L67","kind":"class","name":"AdaLayerNormZero","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":51,"end_line":67,"context_start_line":31,"context_end_line":87,"code":" unsqueeze,\n view,\n)\nfrom ...layers import ColumnLinear, Conv1d, LayerNorm, Linear, Mish, RowLinear\nfrom ...module import Module\n\n\nclass FeedForward(Module):\n def __init__(self, dim, dim_out=None, mult=4, dropout=0.0):\n super().__init__()\n inner_dim = int(dim * mult)\n dim_out = dim_out if dim_out is not None else dim\n\n self.project_in = Linear(dim, inner_dim)\n self.ff = Linear(inner_dim, dim_out)\n\n def forward(self, x):\n return self.ff(gelu(self.project_in(x)))\n\n\nclass AdaLayerNormZero(Module):\n def __init__(self, dim):\n super().__init__()\n\n self.linear = Linear(dim, dim * 6)\n self.norm = LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb=None):\n emb = self.linear(silu(emb))\n shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = chunk(emb, 6, dim=1)\n x = self.norm(x)\n ones = constant(np.ones(1, dtype=np.float32)).cast(x.dtype)\n if default_net().plugin_config.remove_input_padding:\n x = x * (ones + scale_msa) + shift_msa\n else:\n x = x * (ones + unsqueeze(scale_msa, 1)) + unsqueeze(shift_msa, 1)\n return x, gate_msa, shift_mlp, scale_mlp, gate_mlp\n\n\nclass AdaLayerNormZero_Final(Module):\n def __init__(self, dim):\n super().__init__()\n\n self.linear = Linear(dim, dim * 2)\n\n self.norm = LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb):\n emb = self.linear(silu(emb))\n scale, shift = chunk(emb, 2, dim=1)\n ones = constant(np.ones(1, dtype=np.float32)).cast(x.dtype)\n if default_net().plugin_config.remove_input_padding:\n x = self.norm(x) * (ones + scale) + shift\n else:\n x = self.norm(x) * unsqueeze((ones + scale), 1)\n x = x + unsqueeze(shift, 1)\n return x","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.AdaLayerNormZero_Final","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.AdaLayerNormZero_Final#L70-L87","kind":"class","name":"AdaLayerNormZero_Final","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":70,"end_line":87,"context_start_line":50,"context_end_line":107,"code":"\nclass AdaLayerNormZero(Module):\n def __init__(self, dim):\n super().__init__()\n\n self.linear = Linear(dim, dim * 6)\n self.norm = LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb=None):\n emb = self.linear(silu(emb))\n shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = chunk(emb, 6, dim=1)\n x = self.norm(x)\n ones = constant(np.ones(1, dtype=np.float32)).cast(x.dtype)\n if default_net().plugin_config.remove_input_padding:\n x = x * (ones + scale_msa) + shift_msa\n else:\n x = x * (ones + unsqueeze(scale_msa, 1)) + unsqueeze(shift_msa, 1)\n return x, gate_msa, shift_mlp, scale_mlp, gate_mlp\n\n\nclass AdaLayerNormZero_Final(Module):\n def __init__(self, dim):\n super().__init__()\n\n self.linear = Linear(dim, dim * 2)\n\n self.norm = LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb):\n emb = self.linear(silu(emb))\n scale, shift = chunk(emb, 2, dim=1)\n ones = constant(np.ones(1, dtype=np.float32)).cast(x.dtype)\n if default_net().plugin_config.remove_input_padding:\n x = self.norm(x) * (ones + scale) + shift\n else:\n x = self.norm(x) * unsqueeze((ones + scale), 1)\n x = x + unsqueeze(shift, 1)\n return x\n\n\nclass ConvPositionEmbedding(Module):\n def __init__(self, dim, kernel_size=31, groups=16):\n super().__init__()\n assert kernel_size % 2 != 0\n self.conv1d1 = Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2)\n self.conv1d2 = Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2)\n self.mish = Mish()\n\n def forward(self, x, mask=None): # noqa: F722\n if default_net().plugin_config.remove_input_padding:\n x = unsqueeze(x, 0)\n x = permute(x, [0, 2, 1])\n x = self.mish(self.conv1d2(self.mish(self.conv1d1(x))))\n out = permute(x, [0, 2, 1])\n if default_net().plugin_config.remove_input_padding:\n out = squeeze(out, 0)\n return out\n","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.ConvPositionEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.ConvPositionEmbedding#L90-L106","kind":"class","name":"ConvPositionEmbedding","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":90,"end_line":106,"context_start_line":70,"context_end_line":126,"code":"class AdaLayerNormZero_Final(Module):\n def __init__(self, dim):\n super().__init__()\n\n self.linear = Linear(dim, dim * 2)\n\n self.norm = LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n\n def forward(self, x, emb):\n emb = self.linear(silu(emb))\n scale, shift = chunk(emb, 2, dim=1)\n ones = constant(np.ones(1, dtype=np.float32)).cast(x.dtype)\n if default_net().plugin_config.remove_input_padding:\n x = self.norm(x) * (ones + scale) + shift\n else:\n x = self.norm(x) * unsqueeze((ones + scale), 1)\n x = x + unsqueeze(shift, 1)\n return x\n\n\nclass ConvPositionEmbedding(Module):\n def __init__(self, dim, kernel_size=31, groups=16):\n super().__init__()\n assert kernel_size % 2 != 0\n self.conv1d1 = Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2)\n self.conv1d2 = Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2)\n self.mish = Mish()\n\n def forward(self, x, mask=None): # noqa: F722\n if default_net().plugin_config.remove_input_padding:\n x = unsqueeze(x, 0)\n x = permute(x, [0, 2, 1])\n x = self.mish(self.conv1d2(self.mish(self.conv1d1(x))))\n out = permute(x, [0, 2, 1])\n if default_net().plugin_config.remove_input_padding:\n out = squeeze(out, 0)\n return out\n\n\nclass Attention(Module):\n def __init__(\n self,\n processor: AttnProcessor,\n dim: int,\n heads: int = 16,\n dim_head: int = 64,\n dropout: float = 0.0,\n context_dim: Optional[int] = None, # if not None -> joint attention\n context_pre_only=None,\n ):\n super().__init__()\n\n if not hasattr(F, \"scaled_dot_product_attention\"):\n raise ImportError(\"Attention equires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.\")\n\n self.processor = processor\n","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.Attention","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.Attention#L109-L198","kind":"class","name":"Attention","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":109,"end_line":198,"context_start_line":89,"context_end_line":218,"code":"\nclass ConvPositionEmbedding(Module):\n def __init__(self, dim, kernel_size=31, groups=16):\n super().__init__()\n assert kernel_size % 2 != 0\n self.conv1d1 = Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2)\n self.conv1d2 = Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2)\n self.mish = Mish()\n\n def forward(self, x, mask=None): # noqa: F722\n if default_net().plugin_config.remove_input_padding:\n x = unsqueeze(x, 0)\n x = permute(x, [0, 2, 1])\n x = self.mish(self.conv1d2(self.mish(self.conv1d1(x))))\n out = permute(x, [0, 2, 1])\n if default_net().plugin_config.remove_input_padding:\n out = squeeze(out, 0)\n return out\n\n\nclass Attention(Module):\n def __init__(\n self,\n processor: AttnProcessor,\n dim: int,\n heads: int = 16,\n dim_head: int = 64,\n dropout: float = 0.0,\n context_dim: Optional[int] = None, # if not None -> joint attention\n context_pre_only=None,\n ):\n super().__init__()\n\n if not hasattr(F, \"scaled_dot_product_attention\"):\n raise ImportError(\"Attention equires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.\")\n\n self.processor = processor\n\n self.dim = dim # hidden_size\n self.heads = heads\n self.inner_dim = dim_head * heads\n self.dropout = dropout\n self.attention_head_size = dim_head\n self.context_dim = context_dim\n self.context_pre_only = context_pre_only\n self.tp_size = 1\n self.num_attention_heads = heads // self.tp_size\n self.num_attention_kv_heads = heads // self.tp_size # 8\n self.dtype = str_dtype_to_trt(\"float32\")\n self.attention_hidden_size = self.attention_head_size * self.num_attention_heads\n self.to_q = ColumnLinear(\n dim,\n self.tp_size * self.num_attention_heads * self.attention_head_size,\n bias=True,\n dtype=self.dtype,\n tp_group=None,\n tp_size=self.tp_size,\n )\n self.to_k = ColumnLinear(\n dim,\n self.tp_size * self.num_attention_heads * self.attention_head_size,\n bias=True,\n dtype=self.dtype,\n tp_group=None,\n tp_size=self.tp_size,\n )\n self.to_v = ColumnLinear(\n dim,\n self.tp_size * self.num_attention_heads * self.attention_head_size,\n bias=True,\n dtype=self.dtype,\n tp_group=None,\n tp_size=self.tp_size,\n )\n\n if self.context_dim is not None:\n self.to_k_c = Linear(context_dim, self.inner_dim)\n self.to_v_c = Linear(context_dim, self.inner_dim)\n if self.context_pre_only is not None:\n self.to_q_c = Linear(context_dim, self.inner_dim)\n\n self.to_out = RowLinear(\n self.tp_size * self.num_attention_heads * self.attention_head_size,\n dim,\n bias=True,\n dtype=self.dtype,\n tp_group=None,\n tp_size=self.tp_size,\n )\n\n if self.context_pre_only is not None and not self.context_pre_only:\n self.to_out_c = Linear(self.inner_dim, dim)\n\n def forward(\n self,\n x, # noised input x\n rope_cos,\n rope_sin,\n input_lengths,\n c=None, # context c\n scale=1.0,\n rope=None,\n c_rope=None, # rotary position embedding for c\n ) -> torch.Tensor:\n if c is not None:\n return self.processor(self, x, c=c, input_lengths=input_lengths, scale=scale, rope=rope, c_rope=c_rope)\n else:\n return self.processor(\n self, x, rope_cos=rope_cos, rope_sin=rope_sin, input_lengths=input_lengths, scale=scale\n )\n\n\ndef rotate_every_two_3dim(tensor: Tensor) -> Tensor:\n shape_tensor = concat(\n [shape(tensor, i) / 2 if i == (tensor.ndim() - 1) else shape(tensor, i) for i in range(tensor.ndim())]\n )\n if default_net().plugin_config.remove_input_padding:\n assert tensor.ndim() == 2\n x1 = slice(tensor, [0, 0], shape_tensor, [1, 2])\n x2 = slice(tensor, [0, 1], shape_tensor, [1, 2])\n x1 = expand_dims(x1, 2)\n x2 = expand_dims(x2, 2)\n zero = constant(np.ascontiguousarray(np.zeros([1], dtype=trt_dtype_to_np(tensor.dtype))))\n x2 = zero - x2\n x = concat([x2, x1], 2)\n out = view(x, concat([shape(x, 0), shape(x, 1) * 2]))\n else:\n assert tensor.ndim() == 3\n\n x1 = slice(tensor, [0, 0, 0], shape_tensor, [1, 1, 2])","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.rotate_every_two_3dim","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.rotate_every_two_3dim#L201-L227","kind":"function","name":"rotate_every_two_3dim","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":201,"end_line":227,"context_start_line":181,"context_end_line":247,"code":"\n def forward(\n self,\n x, # noised input x\n rope_cos,\n rope_sin,\n input_lengths,\n c=None, # context c\n scale=1.0,\n rope=None,\n c_rope=None, # rotary position embedding for c\n ) -> torch.Tensor:\n if c is not None:\n return self.processor(self, x, c=c, input_lengths=input_lengths, scale=scale, rope=rope, c_rope=c_rope)\n else:\n return self.processor(\n self, x, rope_cos=rope_cos, rope_sin=rope_sin, input_lengths=input_lengths, scale=scale\n )\n\n\ndef rotate_every_two_3dim(tensor: Tensor) -> Tensor:\n shape_tensor = concat(\n [shape(tensor, i) / 2 if i == (tensor.ndim() - 1) else shape(tensor, i) for i in range(tensor.ndim())]\n )\n if default_net().plugin_config.remove_input_padding:\n assert tensor.ndim() == 2\n x1 = slice(tensor, [0, 0], shape_tensor, [1, 2])\n x2 = slice(tensor, [0, 1], shape_tensor, [1, 2])\n x1 = expand_dims(x1, 2)\n x2 = expand_dims(x2, 2)\n zero = constant(np.ascontiguousarray(np.zeros([1], dtype=trt_dtype_to_np(tensor.dtype))))\n x2 = zero - x2\n x = concat([x2, x1], 2)\n out = view(x, concat([shape(x, 0), shape(x, 1) * 2]))\n else:\n assert tensor.ndim() == 3\n\n x1 = slice(tensor, [0, 0, 0], shape_tensor, [1, 1, 2])\n x2 = slice(tensor, [0, 0, 1], shape_tensor, [1, 1, 2])\n x1 = expand_dims(x1, 3)\n x2 = expand_dims(x2, 3)\n zero = constant(np.ascontiguousarray(np.zeros([1], dtype=trt_dtype_to_np(tensor.dtype))))\n x2 = zero - x2\n x = concat([x2, x1], 3)\n out = view(x, concat([shape(x, 0), shape(x, 1), shape(x, 2) * 2]))\n\n return out\n\n\ndef apply_rotary_pos_emb_3dim(x, rope_cos, rope_sin):\n if default_net().plugin_config.remove_input_padding:\n rot_dim = shape(rope_cos, -1) # 64\n new_t_shape = concat([shape(x, 0), rot_dim]) # (-1, 64)\n x_ = slice(x, [0, 0], new_t_shape, [1, 1])\n end_dim = shape(x, -1) - shape(rope_cos, -1)\n new_t_unrotated_shape = concat([shape(x, 0), end_dim]) # (2, -1, 960)\n x_unrotated = slice(x, concat([0, rot_dim]), new_t_unrotated_shape, [1, 1])\n out = concat([x_ * rope_cos + rotate_every_two_3dim(x_) * rope_sin, x_unrotated], dim=-1)\n else:\n rot_dim = shape(rope_cos, 2) # 64\n new_t_shape = concat([shape(x, 0), shape(x, 1), rot_dim]) # (2, -1, 64)\n x_ = slice(x, [0, 0, 0], new_t_shape, [1, 1, 1])\n end_dim = shape(x, 2) - shape(rope_cos, 2)\n new_t_unrotated_shape = concat([shape(x, 0), shape(x, 1), end_dim]) # (2, -1, 960)\n x_unrotated = slice(x, concat([0, 0, rot_dim]), new_t_unrotated_shape, [1, 1, 1])\n out = concat([x_ * rope_cos + rotate_every_two_3dim(x_) * rope_sin, x_unrotated], dim=-1)\n return out","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.apply_rotary_pos_emb_3dim","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.apply_rotary_pos_emb_3dim#L230-L247","kind":"function","name":"apply_rotary_pos_emb_3dim","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":230,"end_line":247,"context_start_line":210,"context_end_line":267,"code":" x2 = expand_dims(x2, 2)\n zero = constant(np.ascontiguousarray(np.zeros([1], dtype=trt_dtype_to_np(tensor.dtype))))\n x2 = zero - x2\n x = concat([x2, x1], 2)\n out = view(x, concat([shape(x, 0), shape(x, 1) * 2]))\n else:\n assert tensor.ndim() == 3\n\n x1 = slice(tensor, [0, 0, 0], shape_tensor, [1, 1, 2])\n x2 = slice(tensor, [0, 0, 1], shape_tensor, [1, 1, 2])\n x1 = expand_dims(x1, 3)\n x2 = expand_dims(x2, 3)\n zero = constant(np.ascontiguousarray(np.zeros([1], dtype=trt_dtype_to_np(tensor.dtype))))\n x2 = zero - x2\n x = concat([x2, x1], 3)\n out = view(x, concat([shape(x, 0), shape(x, 1), shape(x, 2) * 2]))\n\n return out\n\n\ndef apply_rotary_pos_emb_3dim(x, rope_cos, rope_sin):\n if default_net().plugin_config.remove_input_padding:\n rot_dim = shape(rope_cos, -1) # 64\n new_t_shape = concat([shape(x, 0), rot_dim]) # (-1, 64)\n x_ = slice(x, [0, 0], new_t_shape, [1, 1])\n end_dim = shape(x, -1) - shape(rope_cos, -1)\n new_t_unrotated_shape = concat([shape(x, 0), end_dim]) # (2, -1, 960)\n x_unrotated = slice(x, concat([0, rot_dim]), new_t_unrotated_shape, [1, 1])\n out = concat([x_ * rope_cos + rotate_every_two_3dim(x_) * rope_sin, x_unrotated], dim=-1)\n else:\n rot_dim = shape(rope_cos, 2) # 64\n new_t_shape = concat([shape(x, 0), shape(x, 1), rot_dim]) # (2, -1, 64)\n x_ = slice(x, [0, 0, 0], new_t_shape, [1, 1, 1])\n end_dim = shape(x, 2) - shape(rope_cos, 2)\n new_t_unrotated_shape = concat([shape(x, 0), shape(x, 1), end_dim]) # (2, -1, 960)\n x_unrotated = slice(x, concat([0, 0, rot_dim]), new_t_unrotated_shape, [1, 1, 1])\n out = concat([x_ * rope_cos + rotate_every_two_3dim(x_) * rope_sin, x_unrotated], dim=-1)\n return out\n\n\nclass AttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn,\n x, # noised input x\n rope_cos,\n rope_sin,\n input_lengths,\n scale=1.0,\n rope=None,\n ) -> torch.FloatTensor:\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n # k,v,q all (2,1226,1024)","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.AttnProcessor","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.AttnProcessor#L250-L352","kind":"class","name":"AttnProcessor","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":250,"end_line":352,"context_start_line":230,"context_end_line":372,"code":"def apply_rotary_pos_emb_3dim(x, rope_cos, rope_sin):\n if default_net().plugin_config.remove_input_padding:\n rot_dim = shape(rope_cos, -1) # 64\n new_t_shape = concat([shape(x, 0), rot_dim]) # (-1, 64)\n x_ = slice(x, [0, 0], new_t_shape, [1, 1])\n end_dim = shape(x, -1) - shape(rope_cos, -1)\n new_t_unrotated_shape = concat([shape(x, 0), end_dim]) # (2, -1, 960)\n x_unrotated = slice(x, concat([0, rot_dim]), new_t_unrotated_shape, [1, 1])\n out = concat([x_ * rope_cos + rotate_every_two_3dim(x_) * rope_sin, x_unrotated], dim=-1)\n else:\n rot_dim = shape(rope_cos, 2) # 64\n new_t_shape = concat([shape(x, 0), shape(x, 1), rot_dim]) # (2, -1, 64)\n x_ = slice(x, [0, 0, 0], new_t_shape, [1, 1, 1])\n end_dim = shape(x, 2) - shape(rope_cos, 2)\n new_t_unrotated_shape = concat([shape(x, 0), shape(x, 1), end_dim]) # (2, -1, 960)\n x_unrotated = slice(x, concat([0, 0, rot_dim]), new_t_unrotated_shape, [1, 1, 1])\n out = concat([x_ * rope_cos + rotate_every_two_3dim(x_) * rope_sin, x_unrotated], dim=-1)\n return out\n\n\nclass AttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn,\n x, # noised input x\n rope_cos,\n rope_sin,\n input_lengths,\n scale=1.0,\n rope=None,\n ) -> torch.FloatTensor:\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n # k,v,q all (2,1226,1024)\n query = apply_rotary_pos_emb_3dim(query, rope_cos, rope_sin)\n key = apply_rotary_pos_emb_3dim(key, rope_cos, rope_sin)\n\n # attention\n inner_dim = key.shape[-1]\n norm_factor = math.sqrt(attn.attention_head_size)\n q_scaling = 1.0 / norm_factor\n mask = None\n if not default_net().plugin_config.remove_input_padding:\n N = shape(x, 1)\n B = shape(x, 0)\n seq_len_2d = concat([1, N])\n max_position_embeddings = 4096\n # create position ids\n position_ids_buffer = constant(np.expand_dims(np.arange(max_position_embeddings).astype(np.int32), 0))\n tmp_position_ids = slice(position_ids_buffer, starts=[0, 0], sizes=seq_len_2d)\n tmp_position_ids = expand(tmp_position_ids, concat([B, N])) # BxL\n tmp_input_lengths = unsqueeze(input_lengths, 1) # Bx1\n tmp_input_lengths = expand(tmp_input_lengths, concat([B, N])) # BxL\n mask = tmp_position_ids < tmp_input_lengths # BxL\n mask = mask.cast(\"int32\")\n\n if default_net().plugin_config.bert_attention_plugin:\n qkv = concat([query, key, value], dim=-1)\n # TRT plugin mode\n assert input_lengths is not None\n if default_net().plugin_config.remove_input_padding:\n qkv = qkv.view(concat([-1, 3 * inner_dim]))\n max_input_length = constant(\n np.zeros(\n [\n 2048,\n ],\n dtype=np.int32,\n )\n )\n else:\n max_input_length = None\n context = bert_attention(\n qkv,\n input_lengths,\n attn.num_attention_heads,\n attn.attention_head_size,\n q_scaling=q_scaling,\n max_input_length=max_input_length,\n )\n else:\n assert not default_net().plugin_config.remove_input_padding\n\n def transpose_for_scores(x):\n new_x_shape = concat([shape(x, 0), shape(x, 1), attn.num_attention_heads, attn.attention_head_size])\n\n y = x.view(new_x_shape)\n y = y.transpose(1, 2)\n return y\n\n def transpose_for_scores_k(x):\n new_x_shape = concat([shape(x, 0), shape(x, 1), attn.num_attention_heads, attn.attention_head_size])\n\n y = x.view(new_x_shape)\n y = y.permute([0, 2, 3, 1])\n return y\n\n query = transpose_for_scores(query)\n key = transpose_for_scores_k(key)\n value = transpose_for_scores(value)\n\n attention_scores = matmul(query, key, use_fp32_acc=False)\n\n if mask is not None:\n attention_mask = expand_mask(mask, shape(query, 2))\n attention_mask = cast(attention_mask, attention_scores.dtype)\n attention_scores = attention_scores + attention_mask\n\n attention_probs = softmax(attention_scores, dim=-1)\n\n context = matmul(attention_probs, value, use_fp32_acc=False).transpose(1, 2)\n context = context.view(concat([shape(context, 0), shape(context, 1), attn.attention_hidden_size]))\n context = attn.to_out(context)\n if mask is not None:\n mask = mask.view(concat([shape(mask, 0), shape(mask, 1), 1]))\n mask = expand_dims_like(mask, context)\n mask = cast(mask, context.dtype)\n context = context * mask\n return context\n\n\n# DiT Block\nclass DiTBlock(Module):\n def __init__(self, dim, heads, dim_head, ff_mult=2, dropout=0.1):\n super().__init__()\n\n self.attn_norm = AdaLayerNormZero(dim)\n self.attn = Attention(\n processor=AttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n )\n\n self.ff_norm = LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n self.ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout)\n\n def forward(","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.DiTBlock","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.DiTBlock#L356-L398","kind":"class","name":"DiTBlock","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":356,"end_line":398,"context_start_line":336,"context_end_line":412,"code":"\n if mask is not None:\n attention_mask = expand_mask(mask, shape(query, 2))\n attention_mask = cast(attention_mask, attention_scores.dtype)\n attention_scores = attention_scores + attention_mask\n\n attention_probs = softmax(attention_scores, dim=-1)\n\n context = matmul(attention_probs, value, use_fp32_acc=False).transpose(1, 2)\n context = context.view(concat([shape(context, 0), shape(context, 1), attn.attention_hidden_size]))\n context = attn.to_out(context)\n if mask is not None:\n mask = mask.view(concat([shape(mask, 0), shape(mask, 1), 1]))\n mask = expand_dims_like(mask, context)\n mask = cast(mask, context.dtype)\n context = context * mask\n return context\n\n\n# DiT Block\nclass DiTBlock(Module):\n def __init__(self, dim, heads, dim_head, ff_mult=2, dropout=0.1):\n super().__init__()\n\n self.attn_norm = AdaLayerNormZero(dim)\n self.attn = Attention(\n processor=AttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n )\n\n self.ff_norm = LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n self.ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout)\n\n def forward(\n self, x, t, rope_cos, rope_sin, input_lengths, scale=1.0, rope=ModuleNotFoundError\n ): # x: noised input, t: time embedding\n # pre-norm & modulation for attention input\n norm, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.attn_norm(x, emb=t)\n # attention\n # norm ----> (2,1226,1024)\n attn_output = self.attn(x=norm, rope_cos=rope_cos, rope_sin=rope_sin, input_lengths=input_lengths, scale=scale)\n\n # process attention output for input x\n if default_net().plugin_config.remove_input_padding:\n x = x + gate_msa * attn_output\n else:\n x = x + unsqueeze(gate_msa, 1) * attn_output\n ones = constant(np.ones(1, dtype=np.float32)).cast(x.dtype)\n if default_net().plugin_config.remove_input_padding:\n norm = self.ff_norm(x) * (ones + scale_mlp) + shift_mlp\n else:\n norm = self.ff_norm(x) * (ones + unsqueeze(scale_mlp, 1)) + unsqueeze(shift_mlp, 1)\n # norm = self.ff_norm(x) * (ones + scale_mlp) + shift_mlp\n ff_output = self.ff(norm)\n if default_net().plugin_config.remove_input_padding:\n x = x + gate_mlp * ff_output\n else:\n x = x + unsqueeze(gate_mlp, 1) * ff_output\n\n return x\n\n\nclass TimestepEmbedding(Module):\n def __init__(self, dim, freq_embed_dim=256, dtype=None):\n super().__init__()\n # self.time_embed = SinusPositionEmbedding(freq_embed_dim)\n self.mlp1 = Linear(freq_embed_dim, dim, bias=True, dtype=dtype)\n self.mlp2 = Linear(dim, dim, bias=True, dtype=dtype)\n\n def forward(self, timestep):\n t_freq = self.mlp1(timestep)\n t_freq = silu(t_freq)\n t_emb = self.mlp2(t_freq)\n return t_emb","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.TimestepEmbedding","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.TimestepEmbedding#L401-L412","kind":"class","name":"TimestepEmbedding","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":401,"end_line":412,"context_start_line":381,"context_end_line":412,"code":" # process attention output for input x\n if default_net().plugin_config.remove_input_padding:\n x = x + gate_msa * attn_output\n else:\n x = x + unsqueeze(gate_msa, 1) * attn_output\n ones = constant(np.ones(1, dtype=np.float32)).cast(x.dtype)\n if default_net().plugin_config.remove_input_padding:\n norm = self.ff_norm(x) * (ones + scale_mlp) + shift_mlp\n else:\n norm = self.ff_norm(x) * (ones + unsqueeze(scale_mlp, 1)) + unsqueeze(shift_mlp, 1)\n # norm = self.ff_norm(x) * (ones + scale_mlp) + shift_mlp\n ff_output = self.ff(norm)\n if default_net().plugin_config.remove_input_padding:\n x = x + gate_mlp * ff_output\n else:\n x = x + unsqueeze(gate_mlp, 1) * ff_output\n\n return x\n\n\nclass TimestepEmbedding(Module):\n def __init__(self, dim, freq_embed_dim=256, dtype=None):\n super().__init__()\n # self.time_embed = SinusPositionEmbedding(freq_embed_dim)\n self.mlp1 = Linear(freq_embed_dim, dim, bias=True, dtype=dtype)\n self.mlp2 = Linear(dim, dim, bias=True, dtype=dtype)\n\n def forward(self, timestep):\n t_freq = self.mlp1(timestep)\n t_freq = silu(t_freq)\n t_emb = self.mlp2(t_freq)\n return t_emb","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.__init__#L402-L406","kind":"function","name":"__init__","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":402,"end_line":406,"context_start_line":382,"context_end_line":412,"code":" if default_net().plugin_config.remove_input_padding:\n x = x + gate_msa * attn_output\n else:\n x = x + unsqueeze(gate_msa, 1) * attn_output\n ones = constant(np.ones(1, dtype=np.float32)).cast(x.dtype)\n if default_net().plugin_config.remove_input_padding:\n norm = self.ff_norm(x) * (ones + scale_mlp) + shift_mlp\n else:\n norm = self.ff_norm(x) * (ones + unsqueeze(scale_mlp, 1)) + unsqueeze(shift_mlp, 1)\n # norm = self.ff_norm(x) * (ones + scale_mlp) + shift_mlp\n ff_output = self.ff(norm)\n if default_net().plugin_config.remove_input_padding:\n x = x + gate_mlp * ff_output\n else:\n x = x + unsqueeze(gate_mlp, 1) * ff_output\n\n return x\n\n\nclass TimestepEmbedding(Module):\n def __init__(self, dim, freq_embed_dim=256, dtype=None):\n super().__init__()\n # self.time_embed = SinusPositionEmbedding(freq_embed_dim)\n self.mlp1 = Linear(freq_embed_dim, dim, bias=True, dtype=dtype)\n self.mlp2 = Linear(dim, dim, bias=True, dtype=dtype)\n\n def forward(self, timestep):\n t_freq = self.mlp1(timestep)\n t_freq = silu(t_freq)\n t_emb = self.mlp2(t_freq)\n return t_emb","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.forward","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.forward#L408-L412","kind":"function","name":"forward","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":408,"end_line":412,"context_start_line":388,"context_end_line":412,"code":" norm = self.ff_norm(x) * (ones + scale_mlp) + shift_mlp\n else:\n norm = self.ff_norm(x) * (ones + unsqueeze(scale_mlp, 1)) + unsqueeze(shift_mlp, 1)\n # norm = self.ff_norm(x) * (ones + scale_mlp) + shift_mlp\n ff_output = self.ff(norm)\n if default_net().plugin_config.remove_input_padding:\n x = x + gate_mlp * ff_output\n else:\n x = x + unsqueeze(gate_mlp, 1) * ff_output\n\n return x\n\n\nclass TimestepEmbedding(Module):\n def __init__(self, dim, freq_embed_dim=256, dtype=None):\n super().__init__()\n # self.time_embed = SinusPositionEmbedding(freq_embed_dim)\n self.mlp1 = Linear(freq_embed_dim, dim, bias=True, dtype=dtype)\n self.mlp2 = Linear(dim, dim, bias=True, dtype=dtype)\n\n def forward(self, timestep):\n t_freq = self.mlp1(timestep)\n t_freq = silu(t_freq)\n t_emb = self.mlp2(t_freq)\n return t_emb","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.__call__","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.__call__#L254-L352","kind":"function","name":"__call__","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":254,"end_line":352,"context_start_line":234,"context_end_line":372,"code":" x_ = slice(x, [0, 0], new_t_shape, [1, 1])\n end_dim = shape(x, -1) - shape(rope_cos, -1)\n new_t_unrotated_shape = concat([shape(x, 0), end_dim]) # (2, -1, 960)\n x_unrotated = slice(x, concat([0, rot_dim]), new_t_unrotated_shape, [1, 1])\n out = concat([x_ * rope_cos + rotate_every_two_3dim(x_) * rope_sin, x_unrotated], dim=-1)\n else:\n rot_dim = shape(rope_cos, 2) # 64\n new_t_shape = concat([shape(x, 0), shape(x, 1), rot_dim]) # (2, -1, 64)\n x_ = slice(x, [0, 0, 0], new_t_shape, [1, 1, 1])\n end_dim = shape(x, 2) - shape(rope_cos, 2)\n new_t_unrotated_shape = concat([shape(x, 0), shape(x, 1), end_dim]) # (2, -1, 960)\n x_unrotated = slice(x, concat([0, 0, rot_dim]), new_t_unrotated_shape, [1, 1, 1])\n out = concat([x_ * rope_cos + rotate_every_two_3dim(x_) * rope_sin, x_unrotated], dim=-1)\n return out\n\n\nclass AttnProcessor:\n def __init__(self):\n pass\n\n def __call__(\n self,\n attn,\n x, # noised input x\n rope_cos,\n rope_sin,\n input_lengths,\n scale=1.0,\n rope=None,\n ) -> torch.FloatTensor:\n query = attn.to_q(x)\n key = attn.to_k(x)\n value = attn.to_v(x)\n # k,v,q all (2,1226,1024)\n query = apply_rotary_pos_emb_3dim(query, rope_cos, rope_sin)\n key = apply_rotary_pos_emb_3dim(key, rope_cos, rope_sin)\n\n # attention\n inner_dim = key.shape[-1]\n norm_factor = math.sqrt(attn.attention_head_size)\n q_scaling = 1.0 / norm_factor\n mask = None\n if not default_net().plugin_config.remove_input_padding:\n N = shape(x, 1)\n B = shape(x, 0)\n seq_len_2d = concat([1, N])\n max_position_embeddings = 4096\n # create position ids\n position_ids_buffer = constant(np.expand_dims(np.arange(max_position_embeddings).astype(np.int32), 0))\n tmp_position_ids = slice(position_ids_buffer, starts=[0, 0], sizes=seq_len_2d)\n tmp_position_ids = expand(tmp_position_ids, concat([B, N])) # BxL\n tmp_input_lengths = unsqueeze(input_lengths, 1) # Bx1\n tmp_input_lengths = expand(tmp_input_lengths, concat([B, N])) # BxL\n mask = tmp_position_ids < tmp_input_lengths # BxL\n mask = mask.cast(\"int32\")\n\n if default_net().plugin_config.bert_attention_plugin:\n qkv = concat([query, key, value], dim=-1)\n # TRT plugin mode\n assert input_lengths is not None\n if default_net().plugin_config.remove_input_padding:\n qkv = qkv.view(concat([-1, 3 * inner_dim]))\n max_input_length = constant(\n np.zeros(\n [\n 2048,\n ],\n dtype=np.int32,\n )\n )\n else:\n max_input_length = None\n context = bert_attention(\n qkv,\n input_lengths,\n attn.num_attention_heads,\n attn.attention_head_size,\n q_scaling=q_scaling,\n max_input_length=max_input_length,\n )\n else:\n assert not default_net().plugin_config.remove_input_padding\n\n def transpose_for_scores(x):\n new_x_shape = concat([shape(x, 0), shape(x, 1), attn.num_attention_heads, attn.attention_head_size])\n\n y = x.view(new_x_shape)\n y = y.transpose(1, 2)\n return y\n\n def transpose_for_scores_k(x):\n new_x_shape = concat([shape(x, 0), shape(x, 1), attn.num_attention_heads, attn.attention_head_size])\n\n y = x.view(new_x_shape)\n y = y.permute([0, 2, 3, 1])\n return y\n\n query = transpose_for_scores(query)\n key = transpose_for_scores_k(key)\n value = transpose_for_scores(value)\n\n attention_scores = matmul(query, key, use_fp32_acc=False)\n\n if mask is not None:\n attention_mask = expand_mask(mask, shape(query, 2))\n attention_mask = cast(attention_mask, attention_scores.dtype)\n attention_scores = attention_scores + attention_mask\n\n attention_probs = softmax(attention_scores, dim=-1)\n\n context = matmul(attention_probs, value, use_fp32_acc=False).transpose(1, 2)\n context = context.view(concat([shape(context, 0), shape(context, 1), attn.attention_hidden_size]))\n context = attn.to_out(context)\n if mask is not None:\n mask = mask.view(concat([shape(mask, 0), shape(mask, 1), 1]))\n mask = expand_dims_like(mask, context)\n mask = cast(mask, context.dtype)\n context = context * mask\n return context\n\n\n# DiT Block\nclass DiTBlock(Module):\n def __init__(self, dim, heads, dim_head, ff_mult=2, dropout=0.1):\n super().__init__()\n\n self.attn_norm = AdaLayerNormZero(dim)\n self.attn = Attention(\n processor=AttnProcessor(),\n dim=dim,\n heads=heads,\n dim_head=dim_head,\n dropout=dropout,\n )\n\n self.ff_norm = LayerNorm(dim, elementwise_affine=False, eps=1e-6)\n self.ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout)\n\n def forward(","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.transpose_for_scores","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.transpose_for_scores#L317-L322","kind":"function","name":"transpose_for_scores","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":317,"end_line":322,"context_start_line":297,"context_end_line":342,"code":" np.zeros(\n [\n 2048,\n ],\n dtype=np.int32,\n )\n )\n else:\n max_input_length = None\n context = bert_attention(\n qkv,\n input_lengths,\n attn.num_attention_heads,\n attn.attention_head_size,\n q_scaling=q_scaling,\n max_input_length=max_input_length,\n )\n else:\n assert not default_net().plugin_config.remove_input_padding\n\n def transpose_for_scores(x):\n new_x_shape = concat([shape(x, 0), shape(x, 1), attn.num_attention_heads, attn.attention_head_size])\n\n y = x.view(new_x_shape)\n y = y.transpose(1, 2)\n return y\n\n def transpose_for_scores_k(x):\n new_x_shape = concat([shape(x, 0), shape(x, 1), attn.num_attention_heads, attn.attention_head_size])\n\n y = x.view(new_x_shape)\n y = y.permute([0, 2, 3, 1])\n return y\n\n query = transpose_for_scores(query)\n key = transpose_for_scores_k(key)\n value = transpose_for_scores(value)\n\n attention_scores = matmul(query, key, use_fp32_acc=False)\n\n if mask is not None:\n attention_mask = expand_mask(mask, shape(query, 2))\n attention_mask = cast(attention_mask, attention_scores.dtype)\n attention_scores = attention_scores + attention_mask\n\n attention_probs = softmax(attention_scores, dim=-1)","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.transpose_for_scores_k","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.patch.f5tts.modules.transpose_for_scores_k#L324-L329","kind":"function","name":"transpose_for_scores_k","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":324,"end_line":329,"context_start_line":304,"context_end_line":349,"code":" else:\n max_input_length = None\n context = bert_attention(\n qkv,\n input_lengths,\n attn.num_attention_heads,\n attn.attention_head_size,\n q_scaling=q_scaling,\n max_input_length=max_input_length,\n )\n else:\n assert not default_net().plugin_config.remove_input_padding\n\n def transpose_for_scores(x):\n new_x_shape = concat([shape(x, 0), shape(x, 1), attn.num_attention_heads, attn.attention_head_size])\n\n y = x.view(new_x_shape)\n y = y.transpose(1, 2)\n return y\n\n def transpose_for_scores_k(x):\n new_x_shape = concat([shape(x, 0), shape(x, 1), attn.num_attention_heads, attn.attention_head_size])\n\n y = x.view(new_x_shape)\n y = y.permute([0, 2, 3, 1])\n return y\n\n query = transpose_for_scores(query)\n key = transpose_for_scores_k(key)\n value = transpose_for_scores(value)\n\n attention_scores = matmul(query, key, use_fp32_acc=False)\n\n if mask is not None:\n attention_mask = expand_mask(mask, shape(query, 2))\n attention_mask = cast(attention_mask, attention_scores.dtype)\n attention_scores = attention_scores + attention_mask\n\n attention_probs = softmax(attention_scores, dim=-1)\n\n context = matmul(attention_probs, value, use_fp32_acc=False).transpose(1, 2)\n context = context.view(concat([shape(context, 0), shape(context, 1), attn.attention_hidden_size]))\n context = attn.to_out(context)\n if mask is not None:\n mask = mask.view(concat([shape(mask, 0), shape(mask, 1), 1]))\n mask = expand_dims_like(mask, context)","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.conv_stft","uri":"program://DMOSpeech2/module/src.f5_tts.runtime.triton_trtllm.scripts.conv_stft#L1-L248","kind":"module","name":"src.f5_tts.runtime.triton_trtllm.scripts.conv_stft","path":"src/f5_tts/runtime/triton_trtllm/scripts/conv_stft.py","language":"python","start_line":1,"end_line":248,"context_start_line":1,"context_end_line":248,"code":"# Modified from https://github.com/echocatzh/conv-stft/blob/master/conv_stft/conv_stft.py\n\n# Copyright (c) 2024, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\n# MIT License\n\n# Copyright (c) 2020 Shimin Zhang\n\n# Permission is hereby granted, free of charge, to any person obtaining a copy\n# of this software and associated documentation files (the \"Software\"), to deal\n# in the Software without restriction, including without limitation the rights\n# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n# copies of the Software, and to permit persons to whom the Software is\n# furnished to do so, subject to the following conditions:\n\n# The above copyright notice and this permission notice shall be included in all\n# copies or substantial portions of the Software.\n\n# THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE\n# SOFTWARE.\n\nimport torch as th\nimport torch.nn.functional as F\nfrom scipy.signal import check_COLA, get_window\n\n\nsupport_clp_op = None\nif th.__version__ >= \"1.7.0\":\n from torch.fft import rfft as fft\n\n support_clp_op = True\nelse:\n from torch import rfft as fft\n\n\nclass STFT(th.nn.Module):\n def __init__(\n self,\n win_len=1024,\n win_hop=512,\n fft_len=1024,\n enframe_mode=\"continue\",\n win_type=\"hann\",\n win_sqrt=False,\n pad_center=True,\n ):\n \"\"\"\n Implement of STFT using 1D convolution and 1D transpose convolutions.\n Implement of framing the signal in 2 ways, `break` and `continue`.\n `break` method is a kaldi-like framing.\n `continue` method is a librosa-like framing.\n\n More information about `perfect reconstruction`:\n 1. https://ww2.mathworks.cn/help/signal/ref/stft.html\n 2. https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.get_window.html\n\n Args:\n win_len (int): Number of points in one frame. Defaults to 1024.\n win_hop (int): Number of framing stride. Defaults to 512.\n fft_len (int): Number of DFT points. Defaults to 1024.\n enframe_mode (str, optional): `break` and `continue`. Defaults to 'continue'.\n win_type (str, optional): The type of window to create. Defaults to 'hann'.\n win_sqrt (bool, optional): using square root window. Defaults to True.\n pad_center (bool, optional): `perfect reconstruction` opts. Defaults to True.\n \"\"\"\n super(STFT, self).__init__()\n assert enframe_mode in [\"break\", \"continue\"]\n assert fft_len >= win_len\n self.win_len = win_len\n self.win_hop = win_hop\n self.fft_len = fft_len\n self.mode = enframe_mode\n self.win_type = win_type\n self.win_sqrt = win_sqrt\n self.pad_center = pad_center\n self.pad_amount = self.fft_len // 2\n\n en_k, fft_k, ifft_k, ola_k = self.__init_kernel__()\n self.register_buffer(\"en_k\", en_k)\n self.register_buffer(\"fft_k\", fft_k)\n self.register_buffer(\"ifft_k\", ifft_k)\n self.register_buffer(\"ola_k\", ola_k)\n\n def __init_kernel__(self):\n \"\"\"\n Generate enframe_kernel, fft_kernel, ifft_kernel and overlap-add kernel.\n ** enframe_kernel: Using conv1d layer and identity matrix.\n ** fft_kernel: Using linear layer for matrix multiplication. In fact,\n enframe_kernel and fft_kernel can be combined, But for the sake of\n readability, I took the two apart.\n ** ifft_kernel, pinv of fft_kernel.\n ** overlap-add kernel, just like enframe_kernel, but transposed.\n\n Returns:\n tuple: four kernels.\n \"\"\"\n enframed_kernel = th.eye(self.fft_len)[:, None, :]\n if support_clp_op:\n tmp = fft(th.eye(self.fft_len))\n fft_kernel = th.stack([tmp.real, tmp.imag], dim=2)\n else:\n fft_kernel = fft(th.eye(self.fft_len), 1)\n if self.mode == \"break\":\n enframed_kernel = th.eye(self.win_len)[:, None, :]\n fft_kernel = fft_kernel[: self.win_len]\n fft_kernel = th.cat((fft_kernel[:, :, 0], fft_kernel[:, :, 1]), dim=1)\n ifft_kernel = th.pinverse(fft_kernel)[:, None, :]\n window = get_window(self.win_type, self.win_len)\n\n self.perfect_reconstruct = check_COLA(window, self.win_len, self.win_len - self.win_hop)\n window = th.FloatTensor(window)\n if self.mode == \"continue\":\n left_pad = (self.fft_len - self.win_len) // 2\n right_pad = left_pad + (self.fft_len - self.win_len) % 2\n window = F.pad(window, (left_pad, right_pad))\n if self.win_sqrt:\n self.padded_window = window\n window = th.sqrt(window)\n else:\n self.padded_window = window**2\n\n fft_kernel = fft_kernel.T * window\n ifft_kernel = ifft_kernel * window\n ola_kernel = th.eye(self.fft_len)[: self.win_len, None, :]\n if self.mode == \"continue\":\n ola_kernel = th.eye(self.fft_len)[:, None, : self.fft_len]\n return enframed_kernel, fft_kernel, ifft_kernel, ola_kernel\n\n def is_perfect(self):\n \"\"\"\n Whether the parameters win_len, win_hop and win_sqrt\n obey constants overlap-add(COLA)\n\n Returns:\n bool: Return true if parameters obey COLA.\n \"\"\"\n return self.perfect_reconstruct and self.pad_center\n\n def transform(self, inputs, return_type=\"complex\"):\n \"\"\"Take input data (audio) to STFT domain.\n\n Args:\n inputs (tensor): Tensor of floats, with shape (num_batch, num_samples)\n return_type (str, optional): return (mag, phase) when `magphase`,\n return (real, imag) when `realimag` and complex(real, imag) when `complex`.\n Defaults to 'complex'.\n\n Returns:\n tuple: (mag, phase) when `magphase`, return (real, imag) when\n `realimag`. Defaults to 'complex', each elements with shape\n [num_batch, num_frequencies, num_frames]\n \"\"\"\n assert return_type in [\"magphase\", \"realimag\", \"complex\"]\n if inputs.dim() == 2:\n inputs = th.unsqueeze(inputs, 1)\n self.num_samples = inputs.size(-1)\n if self.pad_center:\n inputs = F.pad(inputs, (self.pad_amount, self.pad_amount), mode=\"reflect\")\n enframe_inputs = F.conv1d(inputs, self.en_k, stride=self.win_hop)\n outputs = th.transpose(enframe_inputs, 1, 2)\n outputs = F.linear(outputs, self.fft_k)\n outputs = th.transpose(outputs, 1, 2)\n dim = self.fft_len // 2 + 1\n real = outputs[:, :dim, :]\n imag = outputs[:, dim:, :]\n if return_type == \"realimag\":\n return real, imag\n elif return_type == \"complex\":\n assert support_clp_op\n return th.complex(real, imag)\n else:\n mags = th.sqrt(real**2 + imag**2)\n phase = th.atan2(imag, real)\n return mags, phase\n\n def inverse(self, input1, input2=None, input_type=\"magphase\"):\n \"\"\"Call the inverse STFT (iSTFT), given tensors produced\n by the `transform` function.\n\n Args:\n input1 (tensors): Magnitude/Real-part of STFT with shape\n [num_batch, num_frequencies, num_frames]\n input2 (tensors): Phase/Imag-part of STFT with shape\n [num_batch, num_frequencies, num_frames]\n input_type (str, optional): Mathematical meaning of input tensor's.\n Defaults to 'magphase'.\n\n Returns:\n tensors: Reconstructed audio given magnitude and phase. Of\n shape [num_batch, num_samples]\n \"\"\"\n assert input_type in [\"magphase\", \"realimag\"]\n if input_type == \"realimag\":\n real, imag = None, None\n if support_clp_op and th.is_complex(input1):\n real, imag = input1.real, input1.imag\n else:\n real, imag = input1, input2\n else:\n real = input1 * th.cos(input2)\n imag = input1 * th.sin(input2)\n inputs = th.cat([real, imag], dim=1)\n outputs = F.conv_transpose1d(inputs, self.ifft_k, stride=self.win_hop)\n t = (self.padded_window[None, :, None]).repeat(1, 1, inputs.size(-1))\n t = t.to(inputs.device)\n coff = F.conv_transpose1d(t, self.ola_k, stride=self.win_hop)\n\n num_frames = input1.size(-1)\n num_samples = num_frames * self.win_hop\n\n rm_start, rm_end = self.pad_amount, self.pad_amount + num_samples\n\n outputs = outputs[..., rm_start:rm_end]\n coff = coff[..., rm_start:rm_end]\n coffidx = th.where(coff > 1e-8)\n outputs[coffidx] = outputs[coffidx] / (coff[coffidx])\n return outputs.squeeze(dim=1)\n\n def forward(self, inputs):\n \"\"\"Take input data (audio) to STFT domain and then back to audio.\n\n Args:\n inputs (tensor): Tensor of floats, with shape [num_batch, num_samples]\n\n Returns:\n tensor: Reconstructed audio given magnitude and phase.\n Of shape [num_batch, num_samples]\n \"\"\"\n mag, phase = self.transform(inputs)\n rec_wav = self.inverse(mag, phase)\n return rec_wav","source_hash":"9284255817c07d01ef0c1f72ccdcdc6f0fd6cb65a00ba9170b0bfa46f2c614d2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.STFT","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.STFT#L53-L248","kind":"class","name":"STFT","path":"src/f5_tts/runtime/triton_trtllm/scripts/conv_stft.py","language":"python","start_line":53,"end_line":248,"context_start_line":33,"context_end_line":248,"code":"# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE\n# SOFTWARE.\n\nimport torch as th\nimport torch.nn.functional as F\nfrom scipy.signal import check_COLA, get_window\n\n\nsupport_clp_op = None\nif th.__version__ >= \"1.7.0\":\n from torch.fft import rfft as fft\n\n support_clp_op = True\nelse:\n from torch import rfft as fft\n\n\nclass STFT(th.nn.Module):\n def __init__(\n self,\n win_len=1024,\n win_hop=512,\n fft_len=1024,\n enframe_mode=\"continue\",\n win_type=\"hann\",\n win_sqrt=False,\n pad_center=True,\n ):\n \"\"\"\n Implement of STFT using 1D convolution and 1D transpose convolutions.\n Implement of framing the signal in 2 ways, `break` and `continue`.\n `break` method is a kaldi-like framing.\n `continue` method is a librosa-like framing.\n\n More information about `perfect reconstruction`:\n 1. https://ww2.mathworks.cn/help/signal/ref/stft.html\n 2. https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.get_window.html\n\n Args:\n win_len (int): Number of points in one frame. Defaults to 1024.\n win_hop (int): Number of framing stride. Defaults to 512.\n fft_len (int): Number of DFT points. Defaults to 1024.\n enframe_mode (str, optional): `break` and `continue`. Defaults to 'continue'.\n win_type (str, optional): The type of window to create. Defaults to 'hann'.\n win_sqrt (bool, optional): using square root window. Defaults to True.\n pad_center (bool, optional): `perfect reconstruction` opts. Defaults to True.\n \"\"\"\n super(STFT, self).__init__()\n assert enframe_mode in [\"break\", \"continue\"]\n assert fft_len >= win_len\n self.win_len = win_len\n self.win_hop = win_hop\n self.fft_len = fft_len\n self.mode = enframe_mode\n self.win_type = win_type\n self.win_sqrt = win_sqrt\n self.pad_center = pad_center\n self.pad_amount = self.fft_len // 2\n\n en_k, fft_k, ifft_k, ola_k = self.__init_kernel__()\n self.register_buffer(\"en_k\", en_k)\n self.register_buffer(\"fft_k\", fft_k)\n self.register_buffer(\"ifft_k\", ifft_k)\n self.register_buffer(\"ola_k\", ola_k)\n\n def __init_kernel__(self):\n \"\"\"\n Generate enframe_kernel, fft_kernel, ifft_kernel and overlap-add kernel.\n ** enframe_kernel: Using conv1d layer and identity matrix.\n ** fft_kernel: Using linear layer for matrix multiplication. In fact,\n enframe_kernel and fft_kernel can be combined, But for the sake of\n readability, I took the two apart.\n ** ifft_kernel, pinv of fft_kernel.\n ** overlap-add kernel, just like enframe_kernel, but transposed.\n\n Returns:\n tuple: four kernels.\n \"\"\"\n enframed_kernel = th.eye(self.fft_len)[:, None, :]\n if support_clp_op:\n tmp = fft(th.eye(self.fft_len))\n fft_kernel = th.stack([tmp.real, tmp.imag], dim=2)\n else:\n fft_kernel = fft(th.eye(self.fft_len), 1)\n if self.mode == \"break\":\n enframed_kernel = th.eye(self.win_len)[:, None, :]\n fft_kernel = fft_kernel[: self.win_len]\n fft_kernel = th.cat((fft_kernel[:, :, 0], fft_kernel[:, :, 1]), dim=1)\n ifft_kernel = th.pinverse(fft_kernel)[:, None, :]\n window = get_window(self.win_type, self.win_len)\n\n self.perfect_reconstruct = check_COLA(window, self.win_len, self.win_len - self.win_hop)\n window = th.FloatTensor(window)\n if self.mode == \"continue\":\n left_pad = (self.fft_len - self.win_len) // 2\n right_pad = left_pad + (self.fft_len - self.win_len) % 2\n window = F.pad(window, (left_pad, right_pad))\n if self.win_sqrt:\n self.padded_window = window\n window = th.sqrt(window)\n else:\n self.padded_window = window**2\n\n fft_kernel = fft_kernel.T * window\n ifft_kernel = ifft_kernel * window\n ola_kernel = th.eye(self.fft_len)[: self.win_len, None, :]\n if self.mode == \"continue\":\n ola_kernel = th.eye(self.fft_len)[:, None, : self.fft_len]\n return enframed_kernel, fft_kernel, ifft_kernel, ola_kernel\n\n def is_perfect(self):\n \"\"\"\n Whether the parameters win_len, win_hop and win_sqrt\n obey constants overlap-add(COLA)\n\n Returns:\n bool: Return true if parameters obey COLA.\n \"\"\"\n return self.perfect_reconstruct and self.pad_center\n\n def transform(self, inputs, return_type=\"complex\"):\n \"\"\"Take input data (audio) to STFT domain.\n\n Args:\n inputs (tensor): Tensor of floats, with shape (num_batch, num_samples)\n return_type (str, optional): return (mag, phase) when `magphase`,\n return (real, imag) when `realimag` and complex(real, imag) when `complex`.\n Defaults to 'complex'.\n\n Returns:\n tuple: (mag, phase) when `magphase`, return (real, imag) when\n `realimag`. Defaults to 'complex', each elements with shape\n [num_batch, num_frequencies, num_frames]\n \"\"\"\n assert return_type in [\"magphase\", \"realimag\", \"complex\"]\n if inputs.dim() == 2:\n inputs = th.unsqueeze(inputs, 1)\n self.num_samples = inputs.size(-1)\n if self.pad_center:\n inputs = F.pad(inputs, (self.pad_amount, self.pad_amount), mode=\"reflect\")\n enframe_inputs = F.conv1d(inputs, self.en_k, stride=self.win_hop)\n outputs = th.transpose(enframe_inputs, 1, 2)\n outputs = F.linear(outputs, self.fft_k)\n outputs = th.transpose(outputs, 1, 2)\n dim = self.fft_len // 2 + 1\n real = outputs[:, :dim, :]\n imag = outputs[:, dim:, :]\n if return_type == \"realimag\":\n return real, imag\n elif return_type == \"complex\":\n assert support_clp_op\n return th.complex(real, imag)\n else:\n mags = th.sqrt(real**2 + imag**2)\n phase = th.atan2(imag, real)\n return mags, phase\n\n def inverse(self, input1, input2=None, input_type=\"magphase\"):\n \"\"\"Call the inverse STFT (iSTFT), given tensors produced\n by the `transform` function.\n\n Args:\n input1 (tensors): Magnitude/Real-part of STFT with shape\n [num_batch, num_frequencies, num_frames]\n input2 (tensors): Phase/Imag-part of STFT with shape\n [num_batch, num_frequencies, num_frames]\n input_type (str, optional): Mathematical meaning of input tensor's.\n Defaults to 'magphase'.\n\n Returns:\n tensors: Reconstructed audio given magnitude and phase. Of\n shape [num_batch, num_samples]\n \"\"\"\n assert input_type in [\"magphase\", \"realimag\"]\n if input_type == \"realimag\":\n real, imag = None, None\n if support_clp_op and th.is_complex(input1):\n real, imag = input1.real, input1.imag\n else:\n real, imag = input1, input2\n else:\n real = input1 * th.cos(input2)\n imag = input1 * th.sin(input2)\n inputs = th.cat([real, imag], dim=1)\n outputs = F.conv_transpose1d(inputs, self.ifft_k, stride=self.win_hop)\n t = (self.padded_window[None, :, None]).repeat(1, 1, inputs.size(-1))\n t = t.to(inputs.device)\n coff = F.conv_transpose1d(t, self.ola_k, stride=self.win_hop)\n\n num_frames = input1.size(-1)\n num_samples = num_frames * self.win_hop\n\n rm_start, rm_end = self.pad_amount, self.pad_amount + num_samples\n\n outputs = outputs[..., rm_start:rm_end]\n coff = coff[..., rm_start:rm_end]\n coffidx = th.where(coff > 1e-8)\n outputs[coffidx] = outputs[coffidx] / (coff[coffidx])\n return outputs.squeeze(dim=1)\n\n def forward(self, inputs):\n \"\"\"Take input data (audio) to STFT domain and then back to audio.\n\n Args:\n inputs (tensor): Tensor of floats, with shape [num_batch, num_samples]\n\n Returns:\n tensor: Reconstructed audio given magnitude and phase.\n Of shape [num_batch, num_samples]\n \"\"\"\n mag, phase = self.transform(inputs)\n rec_wav = self.inverse(mag, phase)\n return rec_wav","source_hash":"9284255817c07d01ef0c1f72ccdcdc6f0fd6cb65a00ba9170b0bfa46f2c614d2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.__init__#L54-L99","kind":"function","name":"__init__","path":"src/f5_tts/runtime/triton_trtllm/scripts/conv_stft.py","language":"python","start_line":54,"end_line":99,"context_start_line":34,"context_end_line":119,"code":"# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE\n# SOFTWARE.\n\nimport torch as th\nimport torch.nn.functional as F\nfrom scipy.signal import check_COLA, get_window\n\n\nsupport_clp_op = None\nif th.__version__ >= \"1.7.0\":\n from torch.fft import rfft as fft\n\n support_clp_op = True\nelse:\n from torch import rfft as fft\n\n\nclass STFT(th.nn.Module):\n def __init__(\n self,\n win_len=1024,\n win_hop=512,\n fft_len=1024,\n enframe_mode=\"continue\",\n win_type=\"hann\",\n win_sqrt=False,\n pad_center=True,\n ):\n \"\"\"\n Implement of STFT using 1D convolution and 1D transpose convolutions.\n Implement of framing the signal in 2 ways, `break` and `continue`.\n `break` method is a kaldi-like framing.\n `continue` method is a librosa-like framing.\n\n More information about `perfect reconstruction`:\n 1. https://ww2.mathworks.cn/help/signal/ref/stft.html\n 2. https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.get_window.html\n\n Args:\n win_len (int): Number of points in one frame. Defaults to 1024.\n win_hop (int): Number of framing stride. Defaults to 512.\n fft_len (int): Number of DFT points. Defaults to 1024.\n enframe_mode (str, optional): `break` and `continue`. Defaults to 'continue'.\n win_type (str, optional): The type of window to create. Defaults to 'hann'.\n win_sqrt (bool, optional): using square root window. Defaults to True.\n pad_center (bool, optional): `perfect reconstruction` opts. Defaults to True.\n \"\"\"\n super(STFT, self).__init__()\n assert enframe_mode in [\"break\", \"continue\"]\n assert fft_len >= win_len\n self.win_len = win_len\n self.win_hop = win_hop\n self.fft_len = fft_len\n self.mode = enframe_mode\n self.win_type = win_type\n self.win_sqrt = win_sqrt\n self.pad_center = pad_center\n self.pad_amount = self.fft_len // 2\n\n en_k, fft_k, ifft_k, ola_k = self.__init_kernel__()\n self.register_buffer(\"en_k\", en_k)\n self.register_buffer(\"fft_k\", fft_k)\n self.register_buffer(\"ifft_k\", ifft_k)\n self.register_buffer(\"ola_k\", ola_k)\n\n def __init_kernel__(self):\n \"\"\"\n Generate enframe_kernel, fft_kernel, ifft_kernel and overlap-add kernel.\n ** enframe_kernel: Using conv1d layer and identity matrix.\n ** fft_kernel: Using linear layer for matrix multiplication. In fact,\n enframe_kernel and fft_kernel can be combined, But for the sake of\n readability, I took the two apart.\n ** ifft_kernel, pinv of fft_kernel.\n ** overlap-add kernel, just like enframe_kernel, but transposed.\n\n Returns:\n tuple: four kernels.\n \"\"\"\n enframed_kernel = th.eye(self.fft_len)[:, None, :]\n if support_clp_op:\n tmp = fft(th.eye(self.fft_len))\n fft_kernel = th.stack([tmp.real, tmp.imag], dim=2)\n else:\n fft_kernel = fft(th.eye(self.fft_len), 1)","source_hash":"9284255817c07d01ef0c1f72ccdcdc6f0fd6cb65a00ba9170b0bfa46f2c614d2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.__init_kernel__","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.__init_kernel__#L101-L144","kind":"function","name":"__init_kernel__","path":"src/f5_tts/runtime/triton_trtllm/scripts/conv_stft.py","language":"python","start_line":101,"end_line":144,"context_start_line":81,"context_end_line":164,"code":" pad_center (bool, optional): `perfect reconstruction` opts. Defaults to True.\n \"\"\"\n super(STFT, self).__init__()\n assert enframe_mode in [\"break\", \"continue\"]\n assert fft_len >= win_len\n self.win_len = win_len\n self.win_hop = win_hop\n self.fft_len = fft_len\n self.mode = enframe_mode\n self.win_type = win_type\n self.win_sqrt = win_sqrt\n self.pad_center = pad_center\n self.pad_amount = self.fft_len // 2\n\n en_k, fft_k, ifft_k, ola_k = self.__init_kernel__()\n self.register_buffer(\"en_k\", en_k)\n self.register_buffer(\"fft_k\", fft_k)\n self.register_buffer(\"ifft_k\", ifft_k)\n self.register_buffer(\"ola_k\", ola_k)\n\n def __init_kernel__(self):\n \"\"\"\n Generate enframe_kernel, fft_kernel, ifft_kernel and overlap-add kernel.\n ** enframe_kernel: Using conv1d layer and identity matrix.\n ** fft_kernel: Using linear layer for matrix multiplication. In fact,\n enframe_kernel and fft_kernel can be combined, But for the sake of\n readability, I took the two apart.\n ** ifft_kernel, pinv of fft_kernel.\n ** overlap-add kernel, just like enframe_kernel, but transposed.\n\n Returns:\n tuple: four kernels.\n \"\"\"\n enframed_kernel = th.eye(self.fft_len)[:, None, :]\n if support_clp_op:\n tmp = fft(th.eye(self.fft_len))\n fft_kernel = th.stack([tmp.real, tmp.imag], dim=2)\n else:\n fft_kernel = fft(th.eye(self.fft_len), 1)\n if self.mode == \"break\":\n enframed_kernel = th.eye(self.win_len)[:, None, :]\n fft_kernel = fft_kernel[: self.win_len]\n fft_kernel = th.cat((fft_kernel[:, :, 0], fft_kernel[:, :, 1]), dim=1)\n ifft_kernel = th.pinverse(fft_kernel)[:, None, :]\n window = get_window(self.win_type, self.win_len)\n\n self.perfect_reconstruct = check_COLA(window, self.win_len, self.win_len - self.win_hop)\n window = th.FloatTensor(window)\n if self.mode == \"continue\":\n left_pad = (self.fft_len - self.win_len) // 2\n right_pad = left_pad + (self.fft_len - self.win_len) % 2\n window = F.pad(window, (left_pad, right_pad))\n if self.win_sqrt:\n self.padded_window = window\n window = th.sqrt(window)\n else:\n self.padded_window = window**2\n\n fft_kernel = fft_kernel.T * window\n ifft_kernel = ifft_kernel * window\n ola_kernel = th.eye(self.fft_len)[: self.win_len, None, :]\n if self.mode == \"continue\":\n ola_kernel = th.eye(self.fft_len)[:, None, : self.fft_len]\n return enframed_kernel, fft_kernel, ifft_kernel, ola_kernel\n\n def is_perfect(self):\n \"\"\"\n Whether the parameters win_len, win_hop and win_sqrt\n obey constants overlap-add(COLA)\n\n Returns:\n bool: Return true if parameters obey COLA.\n \"\"\"\n return self.perfect_reconstruct and self.pad_center\n\n def transform(self, inputs, return_type=\"complex\"):\n \"\"\"Take input data (audio) to STFT domain.\n\n Args:\n inputs (tensor): Tensor of floats, with shape (num_batch, num_samples)\n return_type (str, optional): return (mag, phase) when `magphase`,\n return (real, imag) when `realimag` and complex(real, imag) when `complex`.\n Defaults to 'complex'.\n","source_hash":"9284255817c07d01ef0c1f72ccdcdc6f0fd6cb65a00ba9170b0bfa46f2c614d2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.is_perfect","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.is_perfect#L146-L154","kind":"function","name":"is_perfect","path":"src/f5_tts/runtime/triton_trtllm/scripts/conv_stft.py","language":"python","start_line":146,"end_line":154,"context_start_line":126,"context_end_line":174,"code":"\n self.perfect_reconstruct = check_COLA(window, self.win_len, self.win_len - self.win_hop)\n window = th.FloatTensor(window)\n if self.mode == \"continue\":\n left_pad = (self.fft_len - self.win_len) // 2\n right_pad = left_pad + (self.fft_len - self.win_len) % 2\n window = F.pad(window, (left_pad, right_pad))\n if self.win_sqrt:\n self.padded_window = window\n window = th.sqrt(window)\n else:\n self.padded_window = window**2\n\n fft_kernel = fft_kernel.T * window\n ifft_kernel = ifft_kernel * window\n ola_kernel = th.eye(self.fft_len)[: self.win_len, None, :]\n if self.mode == \"continue\":\n ola_kernel = th.eye(self.fft_len)[:, None, : self.fft_len]\n return enframed_kernel, fft_kernel, ifft_kernel, ola_kernel\n\n def is_perfect(self):\n \"\"\"\n Whether the parameters win_len, win_hop and win_sqrt\n obey constants overlap-add(COLA)\n\n Returns:\n bool: Return true if parameters obey COLA.\n \"\"\"\n return self.perfect_reconstruct and self.pad_center\n\n def transform(self, inputs, return_type=\"complex\"):\n \"\"\"Take input data (audio) to STFT domain.\n\n Args:\n inputs (tensor): Tensor of floats, with shape (num_batch, num_samples)\n return_type (str, optional): return (mag, phase) when `magphase`,\n return (real, imag) when `realimag` and complex(real, imag) when `complex`.\n Defaults to 'complex'.\n\n Returns:\n tuple: (mag, phase) when `magphase`, return (real, imag) when\n `realimag`. Defaults to 'complex', each elements with shape\n [num_batch, num_frequencies, num_frames]\n \"\"\"\n assert return_type in [\"magphase\", \"realimag\", \"complex\"]\n if inputs.dim() == 2:\n inputs = th.unsqueeze(inputs, 1)\n self.num_samples = inputs.size(-1)\n if self.pad_center:","source_hash":"9284255817c07d01ef0c1f72ccdcdc6f0fd6cb65a00ba9170b0bfa46f2c614d2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.transform","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.transform#L156-L191","kind":"function","name":"transform","path":"src/f5_tts/runtime/triton_trtllm/scripts/conv_stft.py","language":"python","start_line":156,"end_line":191,"context_start_line":136,"context_end_line":211,"code":" else:\n self.padded_window = window**2\n\n fft_kernel = fft_kernel.T * window\n ifft_kernel = ifft_kernel * window\n ola_kernel = th.eye(self.fft_len)[: self.win_len, None, :]\n if self.mode == \"continue\":\n ola_kernel = th.eye(self.fft_len)[:, None, : self.fft_len]\n return enframed_kernel, fft_kernel, ifft_kernel, ola_kernel\n\n def is_perfect(self):\n \"\"\"\n Whether the parameters win_len, win_hop and win_sqrt\n obey constants overlap-add(COLA)\n\n Returns:\n bool: Return true if parameters obey COLA.\n \"\"\"\n return self.perfect_reconstruct and self.pad_center\n\n def transform(self, inputs, return_type=\"complex\"):\n \"\"\"Take input data (audio) to STFT domain.\n\n Args:\n inputs (tensor): Tensor of floats, with shape (num_batch, num_samples)\n return_type (str, optional): return (mag, phase) when `magphase`,\n return (real, imag) when `realimag` and complex(real, imag) when `complex`.\n Defaults to 'complex'.\n\n Returns:\n tuple: (mag, phase) when `magphase`, return (real, imag) when\n `realimag`. Defaults to 'complex', each elements with shape\n [num_batch, num_frequencies, num_frames]\n \"\"\"\n assert return_type in [\"magphase\", \"realimag\", \"complex\"]\n if inputs.dim() == 2:\n inputs = th.unsqueeze(inputs, 1)\n self.num_samples = inputs.size(-1)\n if self.pad_center:\n inputs = F.pad(inputs, (self.pad_amount, self.pad_amount), mode=\"reflect\")\n enframe_inputs = F.conv1d(inputs, self.en_k, stride=self.win_hop)\n outputs = th.transpose(enframe_inputs, 1, 2)\n outputs = F.linear(outputs, self.fft_k)\n outputs = th.transpose(outputs, 1, 2)\n dim = self.fft_len // 2 + 1\n real = outputs[:, :dim, :]\n imag = outputs[:, dim:, :]\n if return_type == \"realimag\":\n return real, imag\n elif return_type == \"complex\":\n assert support_clp_op\n return th.complex(real, imag)\n else:\n mags = th.sqrt(real**2 + imag**2)\n phase = th.atan2(imag, real)\n return mags, phase\n\n def inverse(self, input1, input2=None, input_type=\"magphase\"):\n \"\"\"Call the inverse STFT (iSTFT), given tensors produced\n by the `transform` function.\n\n Args:\n input1 (tensors): Magnitude/Real-part of STFT with shape\n [num_batch, num_frequencies, num_frames]\n input2 (tensors): Phase/Imag-part of STFT with shape\n [num_batch, num_frequencies, num_frames]\n input_type (str, optional): Mathematical meaning of input tensor's.\n Defaults to 'magphase'.\n\n Returns:\n tensors: Reconstructed audio given magnitude and phase. Of\n shape [num_batch, num_samples]\n \"\"\"\n assert input_type in [\"magphase\", \"realimag\"]\n if input_type == \"realimag\":\n real, imag = None, None","source_hash":"9284255817c07d01ef0c1f72ccdcdc6f0fd6cb65a00ba9170b0bfa46f2c614d2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.inverse","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.inverse#L193-L234","kind":"function","name":"inverse","path":"src/f5_tts/runtime/triton_trtllm/scripts/conv_stft.py","language":"python","start_line":193,"end_line":234,"context_start_line":173,"context_end_line":248,"code":" self.num_samples = inputs.size(-1)\n if self.pad_center:\n inputs = F.pad(inputs, (self.pad_amount, self.pad_amount), mode=\"reflect\")\n enframe_inputs = F.conv1d(inputs, self.en_k, stride=self.win_hop)\n outputs = th.transpose(enframe_inputs, 1, 2)\n outputs = F.linear(outputs, self.fft_k)\n outputs = th.transpose(outputs, 1, 2)\n dim = self.fft_len // 2 + 1\n real = outputs[:, :dim, :]\n imag = outputs[:, dim:, :]\n if return_type == \"realimag\":\n return real, imag\n elif return_type == \"complex\":\n assert support_clp_op\n return th.complex(real, imag)\n else:\n mags = th.sqrt(real**2 + imag**2)\n phase = th.atan2(imag, real)\n return mags, phase\n\n def inverse(self, input1, input2=None, input_type=\"magphase\"):\n \"\"\"Call the inverse STFT (iSTFT), given tensors produced\n by the `transform` function.\n\n Args:\n input1 (tensors): Magnitude/Real-part of STFT with shape\n [num_batch, num_frequencies, num_frames]\n input2 (tensors): Phase/Imag-part of STFT with shape\n [num_batch, num_frequencies, num_frames]\n input_type (str, optional): Mathematical meaning of input tensor's.\n Defaults to 'magphase'.\n\n Returns:\n tensors: Reconstructed audio given magnitude and phase. Of\n shape [num_batch, num_samples]\n \"\"\"\n assert input_type in [\"magphase\", \"realimag\"]\n if input_type == \"realimag\":\n real, imag = None, None\n if support_clp_op and th.is_complex(input1):\n real, imag = input1.real, input1.imag\n else:\n real, imag = input1, input2\n else:\n real = input1 * th.cos(input2)\n imag = input1 * th.sin(input2)\n inputs = th.cat([real, imag], dim=1)\n outputs = F.conv_transpose1d(inputs, self.ifft_k, stride=self.win_hop)\n t = (self.padded_window[None, :, None]).repeat(1, 1, inputs.size(-1))\n t = t.to(inputs.device)\n coff = F.conv_transpose1d(t, self.ola_k, stride=self.win_hop)\n\n num_frames = input1.size(-1)\n num_samples = num_frames * self.win_hop\n\n rm_start, rm_end = self.pad_amount, self.pad_amount + num_samples\n\n outputs = outputs[..., rm_start:rm_end]\n coff = coff[..., rm_start:rm_end]\n coffidx = th.where(coff > 1e-8)\n outputs[coffidx] = outputs[coffidx] / (coff[coffidx])\n return outputs.squeeze(dim=1)\n\n def forward(self, inputs):\n \"\"\"Take input data (audio) to STFT domain and then back to audio.\n\n Args:\n inputs (tensor): Tensor of floats, with shape [num_batch, num_samples]\n\n Returns:\n tensor: Reconstructed audio given magnitude and phase.\n Of shape [num_batch, num_samples]\n \"\"\"\n mag, phase = self.transform(inputs)\n rec_wav = self.inverse(mag, phase)\n return rec_wav","source_hash":"9284255817c07d01ef0c1f72ccdcdc6f0fd6cb65a00ba9170b0bfa46f2c614d2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.forward","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.conv_stft.forward#L236-L248","kind":"function","name":"forward","path":"src/f5_tts/runtime/triton_trtllm/scripts/conv_stft.py","language":"python","start_line":236,"end_line":248,"context_start_line":216,"context_end_line":248,"code":" else:\n real = input1 * th.cos(input2)\n imag = input1 * th.sin(input2)\n inputs = th.cat([real, imag], dim=1)\n outputs = F.conv_transpose1d(inputs, self.ifft_k, stride=self.win_hop)\n t = (self.padded_window[None, :, None]).repeat(1, 1, inputs.size(-1))\n t = t.to(inputs.device)\n coff = F.conv_transpose1d(t, self.ola_k, stride=self.win_hop)\n\n num_frames = input1.size(-1)\n num_samples = num_frames * self.win_hop\n\n rm_start, rm_end = self.pad_amount, self.pad_amount + num_samples\n\n outputs = outputs[..., rm_start:rm_end]\n coff = coff[..., rm_start:rm_end]\n coffidx = th.where(coff > 1e-8)\n outputs[coffidx] = outputs[coffidx] / (coff[coffidx])\n return outputs.squeeze(dim=1)\n\n def forward(self, inputs):\n \"\"\"Take input data (audio) to STFT domain and then back to audio.\n\n Args:\n inputs (tensor): Tensor of floats, with shape [num_batch, num_samples]\n\n Returns:\n tensor: Reconstructed audio given magnitude and phase.\n Of shape [num_batch, num_samples]\n \"\"\"\n mag, phase = self.transform(inputs)\n rec_wav = self.inverse(mag, phase)\n return rec_wav","source_hash":"9284255817c07d01ef0c1f72ccdcdc6f0fd6cb65a00ba9170b0bfa46f2c614d2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx","uri":"program://DMOSpeech2/module/src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx#L1-L138","kind":"module","name":"src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx","path":"src/f5_tts/runtime/triton_trtllm/scripts/export_vocoder_to_onnx.py","language":"python","start_line":1,"end_line":138,"context_start_line":1,"context_end_line":138,"code":"# Copyright (c) 2024, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nimport argparse\n\nimport torch\nimport torch.nn as nn\nfrom conv_stft import STFT\nfrom huggingface_hub import hf_hub_download\nfrom vocos import Vocos\n\n\nopset_version = 17\n\n\ndef get_args():\n parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)\n parser.add_argument(\n \"--vocoder\",\n type=str,\n default=\"vocos\",\n choices=[\"vocos\", \"bigvgan\"],\n help=\"Vocoder to export\",\n )\n parser.add_argument(\n \"--output-path\",\n type=str,\n default=\"./vocos_vocoder.onnx\",\n help=\"Output path\",\n )\n return parser.parse_args()\n\n\nclass ISTFTHead(nn.Module):\n def __init__(self, n_fft: int, hop_length: int):\n super().__init__()\n self.out = None\n self.stft = STFT(fft_len=n_fft, win_hop=hop_length, win_len=n_fft)\n\n def forward(self, x: torch.Tensor):\n x = self.out(x).transpose(1, 2)\n mag, p = x.chunk(2, dim=1)\n mag = torch.exp(mag)\n mag = torch.clip(mag, max=1e2)\n real = mag * torch.cos(p)\n imag = mag * torch.sin(p)\n audio = self.stft.inverse(input1=real, input2=imag, input_type=\"realimag\")\n return audio\n\n\nclass VocosVocoder(nn.Module):\n def __init__(self, vocos_vocoder):\n super(VocosVocoder, self).__init__()\n self.vocos_vocoder = vocos_vocoder\n istft_head_out = self.vocos_vocoder.head.out\n n_fft = self.vocos_vocoder.head.istft.n_fft\n hop_length = self.vocos_vocoder.head.istft.hop_length\n istft_head_for_export = ISTFTHead(n_fft, hop_length)\n istft_head_for_export.out = istft_head_out\n self.vocos_vocoder.head = istft_head_for_export\n\n def forward(self, mel):\n waveform = self.vocos_vocoder.decode(mel)\n return waveform\n\n\ndef export_VocosVocoder(vocos_vocoder, output_path, verbose):\n vocos_vocoder = VocosVocoder(vocos_vocoder).cuda()\n vocos_vocoder.eval()\n\n dummy_batch_size = 8\n dummy_input_length = 500\n\n dummy_mel = torch.randn(dummy_batch_size, 100, dummy_input_length).cuda()\n\n with torch.no_grad():\n dummy_waveform = vocos_vocoder(mel=dummy_mel)\n print(dummy_waveform.shape)\n\n dummy_input = dummy_mel\n\n torch.onnx.export(\n vocos_vocoder,\n dummy_input,\n output_path,\n opset_version=opset_version,\n do_constant_folding=True,\n input_names=[\"mel\"],\n output_names=[\"waveform\"],\n dynamic_axes={\n \"mel\": {0: \"batch_size\", 2: \"input_length\"},\n \"waveform\": {0: \"batch_size\", 1: \"output_length\"},\n },\n verbose=verbose,\n )\n\n print(\"Exported to {}\".format(output_path))\n\n\ndef load_vocoder(vocoder_name=\"vocos\", is_local=False, local_path=\"\", device=\"cpu\", hf_cache_dir=None):\n if vocoder_name == \"vocos\":\n # vocoder = Vocos.from_pretrained(\"charactr/vocos-mel-24khz\").to(device)\n if is_local:\n print(f\"Load vocos from local path {local_path}\")\n config_path = f\"{local_path}/config.yaml\"\n model_path = f\"{local_path}/pytorch_model.bin\"\n else:\n print(\"Download Vocos from huggingface charactr/vocos-mel-24khz\")\n repo_id = \"charactr/vocos-mel-24khz\"\n config_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"config.yaml\")\n model_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"pytorch_model.bin\")\n vocoder = Vocos.from_hparams(config_path)\n state_dict = torch.load(model_path, map_location=\"cpu\", weights_only=True)\n vocoder.load_state_dict(state_dict)\n vocoder = vocoder.eval().to(device)\n elif vocoder_name == \"bigvgan\":\n raise NotImplementedError(\"BigVGAN is not supported yet\")\n vocoder.remove_weight_norm()\n vocoder = vocoder.eval().to(device)\n return vocoder\n\n\nif __name__ == \"__main__\":\n args = get_args()\n vocoder = load_vocoder(vocoder_name=args.vocoder, device=\"cpu\", hf_cache_dir=None)\n if args.vocoder == \"vocos\":\n export_VocosVocoder(vocoder, args.output_path, verbose=False)","source_hash":"72c9953957b8be407255e0642519f0817158c4fdb5550dc8ccc57b124828bd98","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.get_args","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.get_args#L27-L42","kind":"function","name":"get_args","path":"src/f5_tts/runtime/triton_trtllm/scripts/export_vocoder_to_onnx.py","language":"python","start_line":27,"end_line":42,"context_start_line":7,"context_end_line":62,"code":"# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nimport argparse\n\nimport torch\nimport torch.nn as nn\nfrom conv_stft import STFT\nfrom huggingface_hub import hf_hub_download\nfrom vocos import Vocos\n\n\nopset_version = 17\n\n\ndef get_args():\n parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)\n parser.add_argument(\n \"--vocoder\",\n type=str,\n default=\"vocos\",\n choices=[\"vocos\", \"bigvgan\"],\n help=\"Vocoder to export\",\n )\n parser.add_argument(\n \"--output-path\",\n type=str,\n default=\"./vocos_vocoder.onnx\",\n help=\"Output path\",\n )\n return parser.parse_args()\n\n\nclass ISTFTHead(nn.Module):\n def __init__(self, n_fft: int, hop_length: int):\n super().__init__()\n self.out = None\n self.stft = STFT(fft_len=n_fft, win_hop=hop_length, win_len=n_fft)\n\n def forward(self, x: torch.Tensor):\n x = self.out(x).transpose(1, 2)\n mag, p = x.chunk(2, dim=1)\n mag = torch.exp(mag)\n mag = torch.clip(mag, max=1e2)\n real = mag * torch.cos(p)\n imag = mag * torch.sin(p)\n audio = self.stft.inverse(input1=real, input2=imag, input_type=\"realimag\")\n return audio\n\n\nclass VocosVocoder(nn.Module):","source_hash":"72c9953957b8be407255e0642519f0817158c4fdb5550dc8ccc57b124828bd98","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.ISTFTHead","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.ISTFTHead#L45-L59","kind":"class","name":"ISTFTHead","path":"src/f5_tts/runtime/triton_trtllm/scripts/export_vocoder_to_onnx.py","language":"python","start_line":45,"end_line":59,"context_start_line":25,"context_end_line":79,"code":"\n\ndef get_args():\n parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)\n parser.add_argument(\n \"--vocoder\",\n type=str,\n default=\"vocos\",\n choices=[\"vocos\", \"bigvgan\"],\n help=\"Vocoder to export\",\n )\n parser.add_argument(\n \"--output-path\",\n type=str,\n default=\"./vocos_vocoder.onnx\",\n help=\"Output path\",\n )\n return parser.parse_args()\n\n\nclass ISTFTHead(nn.Module):\n def __init__(self, n_fft: int, hop_length: int):\n super().__init__()\n self.out = None\n self.stft = STFT(fft_len=n_fft, win_hop=hop_length, win_len=n_fft)\n\n def forward(self, x: torch.Tensor):\n x = self.out(x).transpose(1, 2)\n mag, p = x.chunk(2, dim=1)\n mag = torch.exp(mag)\n mag = torch.clip(mag, max=1e2)\n real = mag * torch.cos(p)\n imag = mag * torch.sin(p)\n audio = self.stft.inverse(input1=real, input2=imag, input_type=\"realimag\")\n return audio\n\n\nclass VocosVocoder(nn.Module):\n def __init__(self, vocos_vocoder):\n super(VocosVocoder, self).__init__()\n self.vocos_vocoder = vocos_vocoder\n istft_head_out = self.vocos_vocoder.head.out\n n_fft = self.vocos_vocoder.head.istft.n_fft\n hop_length = self.vocos_vocoder.head.istft.hop_length\n istft_head_for_export = ISTFTHead(n_fft, hop_length)\n istft_head_for_export.out = istft_head_out\n self.vocos_vocoder.head = istft_head_for_export\n\n def forward(self, mel):\n waveform = self.vocos_vocoder.decode(mel)\n return waveform\n\n\ndef export_VocosVocoder(vocos_vocoder, output_path, verbose):\n vocos_vocoder = VocosVocoder(vocos_vocoder).cuda()","source_hash":"72c9953957b8be407255e0642519f0817158c4fdb5550dc8ccc57b124828bd98","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.VocosVocoder","uri":"program://DMOSpeech2/class/src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.VocosVocoder#L62-L75","kind":"class","name":"VocosVocoder","path":"src/f5_tts/runtime/triton_trtllm/scripts/export_vocoder_to_onnx.py","language":"python","start_line":62,"end_line":75,"context_start_line":42,"context_end_line":95,"code":" return parser.parse_args()\n\n\nclass ISTFTHead(nn.Module):\n def __init__(self, n_fft: int, hop_length: int):\n super().__init__()\n self.out = None\n self.stft = STFT(fft_len=n_fft, win_hop=hop_length, win_len=n_fft)\n\n def forward(self, x: torch.Tensor):\n x = self.out(x).transpose(1, 2)\n mag, p = x.chunk(2, dim=1)\n mag = torch.exp(mag)\n mag = torch.clip(mag, max=1e2)\n real = mag * torch.cos(p)\n imag = mag * torch.sin(p)\n audio = self.stft.inverse(input1=real, input2=imag, input_type=\"realimag\")\n return audio\n\n\nclass VocosVocoder(nn.Module):\n def __init__(self, vocos_vocoder):\n super(VocosVocoder, self).__init__()\n self.vocos_vocoder = vocos_vocoder\n istft_head_out = self.vocos_vocoder.head.out\n n_fft = self.vocos_vocoder.head.istft.n_fft\n hop_length = self.vocos_vocoder.head.istft.hop_length\n istft_head_for_export = ISTFTHead(n_fft, hop_length)\n istft_head_for_export.out = istft_head_out\n self.vocos_vocoder.head = istft_head_for_export\n\n def forward(self, mel):\n waveform = self.vocos_vocoder.decode(mel)\n return waveform\n\n\ndef export_VocosVocoder(vocos_vocoder, output_path, verbose):\n vocos_vocoder = VocosVocoder(vocos_vocoder).cuda()\n vocos_vocoder.eval()\n\n dummy_batch_size = 8\n dummy_input_length = 500\n\n dummy_mel = torch.randn(dummy_batch_size, 100, dummy_input_length).cuda()\n\n with torch.no_grad():\n dummy_waveform = vocos_vocoder(mel=dummy_mel)\n print(dummy_waveform.shape)\n\n dummy_input = dummy_mel\n\n torch.onnx.export(\n vocos_vocoder,\n dummy_input,","source_hash":"72c9953957b8be407255e0642519f0817158c4fdb5550dc8ccc57b124828bd98","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.export_VocosVocoder","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.export_VocosVocoder#L78-L108","kind":"function","name":"export_VocosVocoder","path":"src/f5_tts/runtime/triton_trtllm/scripts/export_vocoder_to_onnx.py","language":"python","start_line":78,"end_line":108,"context_start_line":58,"context_end_line":128,"code":" audio = self.stft.inverse(input1=real, input2=imag, input_type=\"realimag\")\n return audio\n\n\nclass VocosVocoder(nn.Module):\n def __init__(self, vocos_vocoder):\n super(VocosVocoder, self).__init__()\n self.vocos_vocoder = vocos_vocoder\n istft_head_out = self.vocos_vocoder.head.out\n n_fft = self.vocos_vocoder.head.istft.n_fft\n hop_length = self.vocos_vocoder.head.istft.hop_length\n istft_head_for_export = ISTFTHead(n_fft, hop_length)\n istft_head_for_export.out = istft_head_out\n self.vocos_vocoder.head = istft_head_for_export\n\n def forward(self, mel):\n waveform = self.vocos_vocoder.decode(mel)\n return waveform\n\n\ndef export_VocosVocoder(vocos_vocoder, output_path, verbose):\n vocos_vocoder = VocosVocoder(vocos_vocoder).cuda()\n vocos_vocoder.eval()\n\n dummy_batch_size = 8\n dummy_input_length = 500\n\n dummy_mel = torch.randn(dummy_batch_size, 100, dummy_input_length).cuda()\n\n with torch.no_grad():\n dummy_waveform = vocos_vocoder(mel=dummy_mel)\n print(dummy_waveform.shape)\n\n dummy_input = dummy_mel\n\n torch.onnx.export(\n vocos_vocoder,\n dummy_input,\n output_path,\n opset_version=opset_version,\n do_constant_folding=True,\n input_names=[\"mel\"],\n output_names=[\"waveform\"],\n dynamic_axes={\n \"mel\": {0: \"batch_size\", 2: \"input_length\"},\n \"waveform\": {0: \"batch_size\", 1: \"output_length\"},\n },\n verbose=verbose,\n )\n\n print(\"Exported to {}\".format(output_path))\n\n\ndef load_vocoder(vocoder_name=\"vocos\", is_local=False, local_path=\"\", device=\"cpu\", hf_cache_dir=None):\n if vocoder_name == \"vocos\":\n # vocoder = Vocos.from_pretrained(\"charactr/vocos-mel-24khz\").to(device)\n if is_local:\n print(f\"Load vocos from local path {local_path}\")\n config_path = f\"{local_path}/config.yaml\"\n model_path = f\"{local_path}/pytorch_model.bin\"\n else:\n print(\"Download Vocos from huggingface charactr/vocos-mel-24khz\")\n repo_id = \"charactr/vocos-mel-24khz\"\n config_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"config.yaml\")\n model_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"pytorch_model.bin\")\n vocoder = Vocos.from_hparams(config_path)\n state_dict = torch.load(model_path, map_location=\"cpu\", weights_only=True)\n vocoder.load_state_dict(state_dict)\n vocoder = vocoder.eval().to(device)\n elif vocoder_name == \"bigvgan\":\n raise NotImplementedError(\"BigVGAN is not supported yet\")","source_hash":"72c9953957b8be407255e0642519f0817158c4fdb5550dc8ccc57b124828bd98","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.load_vocoder","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.load_vocoder#L111-L131","kind":"function","name":"load_vocoder","path":"src/f5_tts/runtime/triton_trtllm/scripts/export_vocoder_to_onnx.py","language":"python","start_line":111,"end_line":131,"context_start_line":91,"context_end_line":138,"code":" dummy_input = dummy_mel\n\n torch.onnx.export(\n vocos_vocoder,\n dummy_input,\n output_path,\n opset_version=opset_version,\n do_constant_folding=True,\n input_names=[\"mel\"],\n output_names=[\"waveform\"],\n dynamic_axes={\n \"mel\": {0: \"batch_size\", 2: \"input_length\"},\n \"waveform\": {0: \"batch_size\", 1: \"output_length\"},\n },\n verbose=verbose,\n )\n\n print(\"Exported to {}\".format(output_path))\n\n\ndef load_vocoder(vocoder_name=\"vocos\", is_local=False, local_path=\"\", device=\"cpu\", hf_cache_dir=None):\n if vocoder_name == \"vocos\":\n # vocoder = Vocos.from_pretrained(\"charactr/vocos-mel-24khz\").to(device)\n if is_local:\n print(f\"Load vocos from local path {local_path}\")\n config_path = f\"{local_path}/config.yaml\"\n model_path = f\"{local_path}/pytorch_model.bin\"\n else:\n print(\"Download Vocos from huggingface charactr/vocos-mel-24khz\")\n repo_id = \"charactr/vocos-mel-24khz\"\n config_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"config.yaml\")\n model_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"pytorch_model.bin\")\n vocoder = Vocos.from_hparams(config_path)\n state_dict = torch.load(model_path, map_location=\"cpu\", weights_only=True)\n vocoder.load_state_dict(state_dict)\n vocoder = vocoder.eval().to(device)\n elif vocoder_name == \"bigvgan\":\n raise NotImplementedError(\"BigVGAN is not supported yet\")\n vocoder.remove_weight_norm()\n vocoder = vocoder.eval().to(device)\n return vocoder\n\n\nif __name__ == \"__main__\":\n args = get_args()\n vocoder = load_vocoder(vocoder_name=args.vocoder, device=\"cpu\", hf_cache_dir=None)\n if args.vocoder == \"vocos\":\n export_VocosVocoder(vocoder, args.output_path, verbose=False)","source_hash":"72c9953957b8be407255e0642519f0817158c4fdb5550dc8ccc57b124828bd98","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.__init__","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.__init__#L63-L71","kind":"function","name":"__init__","path":"src/f5_tts/runtime/triton_trtllm/scripts/export_vocoder_to_onnx.py","language":"python","start_line":63,"end_line":71,"context_start_line":43,"context_end_line":91,"code":"\n\nclass ISTFTHead(nn.Module):\n def __init__(self, n_fft: int, hop_length: int):\n super().__init__()\n self.out = None\n self.stft = STFT(fft_len=n_fft, win_hop=hop_length, win_len=n_fft)\n\n def forward(self, x: torch.Tensor):\n x = self.out(x).transpose(1, 2)\n mag, p = x.chunk(2, dim=1)\n mag = torch.exp(mag)\n mag = torch.clip(mag, max=1e2)\n real = mag * torch.cos(p)\n imag = mag * torch.sin(p)\n audio = self.stft.inverse(input1=real, input2=imag, input_type=\"realimag\")\n return audio\n\n\nclass VocosVocoder(nn.Module):\n def __init__(self, vocos_vocoder):\n super(VocosVocoder, self).__init__()\n self.vocos_vocoder = vocos_vocoder\n istft_head_out = self.vocos_vocoder.head.out\n n_fft = self.vocos_vocoder.head.istft.n_fft\n hop_length = self.vocos_vocoder.head.istft.hop_length\n istft_head_for_export = ISTFTHead(n_fft, hop_length)\n istft_head_for_export.out = istft_head_out\n self.vocos_vocoder.head = istft_head_for_export\n\n def forward(self, mel):\n waveform = self.vocos_vocoder.decode(mel)\n return waveform\n\n\ndef export_VocosVocoder(vocos_vocoder, output_path, verbose):\n vocos_vocoder = VocosVocoder(vocos_vocoder).cuda()\n vocos_vocoder.eval()\n\n dummy_batch_size = 8\n dummy_input_length = 500\n\n dummy_mel = torch.randn(dummy_batch_size, 100, dummy_input_length).cuda()\n\n with torch.no_grad():\n dummy_waveform = vocos_vocoder(mel=dummy_mel)\n print(dummy_waveform.shape)\n\n dummy_input = dummy_mel","source_hash":"72c9953957b8be407255e0642519f0817158c4fdb5550dc8ccc57b124828bd98","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.forward","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.export_vocoder_to_onnx.forward#L73-L75","kind":"function","name":"forward","path":"src/f5_tts/runtime/triton_trtllm/scripts/export_vocoder_to_onnx.py","language":"python","start_line":73,"end_line":75,"context_start_line":53,"context_end_line":95,"code":" mag, p = x.chunk(2, dim=1)\n mag = torch.exp(mag)\n mag = torch.clip(mag, max=1e2)\n real = mag * torch.cos(p)\n imag = mag * torch.sin(p)\n audio = self.stft.inverse(input1=real, input2=imag, input_type=\"realimag\")\n return audio\n\n\nclass VocosVocoder(nn.Module):\n def __init__(self, vocos_vocoder):\n super(VocosVocoder, self).__init__()\n self.vocos_vocoder = vocos_vocoder\n istft_head_out = self.vocos_vocoder.head.out\n n_fft = self.vocos_vocoder.head.istft.n_fft\n hop_length = self.vocos_vocoder.head.istft.hop_length\n istft_head_for_export = ISTFTHead(n_fft, hop_length)\n istft_head_for_export.out = istft_head_out\n self.vocos_vocoder.head = istft_head_for_export\n\n def forward(self, mel):\n waveform = self.vocos_vocoder.decode(mel)\n return waveform\n\n\ndef export_VocosVocoder(vocos_vocoder, output_path, verbose):\n vocos_vocoder = VocosVocoder(vocos_vocoder).cuda()\n vocos_vocoder.eval()\n\n dummy_batch_size = 8\n dummy_input_length = 500\n\n dummy_mel = torch.randn(dummy_batch_size, 100, dummy_input_length).cuda()\n\n with torch.no_grad():\n dummy_waveform = vocos_vocoder(mel=dummy_mel)\n print(dummy_waveform.shape)\n\n dummy_input = dummy_mel\n\n torch.onnx.export(\n vocos_vocoder,\n dummy_input,","source_hash":"72c9953957b8be407255e0642519f0817158c4fdb5550dc8ccc57b124828bd98","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint","uri":"program://DMOSpeech2/module/src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint#L1-L358","kind":"module","name":"src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint","path":"src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","language":"python","start_line":1,"end_line":358,"context_start_line":1,"context_end_line":358,"code":"import argparse\nimport json\nimport os\nimport re\nimport time\nimport traceback\nfrom concurrent.futures import ThreadPoolExecutor, as_completed\n\nimport safetensors.torch\nimport torch\nfrom tensorrt_llm import str_dtype_to_torch\nfrom tensorrt_llm.mapping import Mapping\nfrom tensorrt_llm.models.convert_utils import split, split_matrix_tp\n\n\ndef split_q_tp(v, n_head, n_hidden, tensor_parallel, rank):\n split_v = split(v, tensor_parallel, rank, dim=1)\n return split_v.contiguous()\n\n\ndef split_q_bias_tp(v, n_head, n_hidden, tensor_parallel, rank):\n split_v = split(v, tensor_parallel, rank, dim=0)\n return split_v.contiguous()\n\n\nFACEBOOK_DIT_NAME_MAPPING = {\n \"^time_embed.time_mlp.0.weight$\": \"time_embed.mlp1.weight\",\n \"^time_embed.time_mlp.0.bias$\": \"time_embed.mlp1.bias\",\n \"^time_embed.time_mlp.2.weight$\": \"time_embed.mlp2.weight\",\n \"^time_embed.time_mlp.2.bias$\": \"time_embed.mlp2.bias\",\n \"^input_embed.conv_pos_embed.conv1d.0.weight$\": \"input_embed.conv_pos_embed.conv1d1.weight\",\n \"^input_embed.conv_pos_embed.conv1d.0.bias$\": \"input_embed.conv_pos_embed.conv1d1.bias\",\n \"^input_embed.conv_pos_embed.conv1d.2.weight$\": \"input_embed.conv_pos_embed.conv1d2.weight\",\n \"^input_embed.conv_pos_embed.conv1d.2.bias$\": \"input_embed.conv_pos_embed.conv1d2.bias\",\n \"^transformer_blocks.0.attn.to_out.0.weight$\": \"transformer_blocks.0.attn.to_out.weight\",\n \"^transformer_blocks.0.attn.to_out.0.bias$\": \"transformer_blocks.0.attn.to_out.bias\",\n \"^transformer_blocks.1.attn.to_out.0.weight$\": \"transformer_blocks.1.attn.to_out.weight\",\n \"^transformer_blocks.1.attn.to_out.0.bias$\": \"transformer_blocks.1.attn.to_out.bias\",\n \"^transformer_blocks.2.attn.to_out.0.weight$\": \"transformer_blocks.2.attn.to_out.weight\",\n \"^transformer_blocks.2.attn.to_out.0.bias$\": \"transformer_blocks.2.attn.to_out.bias\",\n \"^transformer_blocks.3.attn.to_out.0.weight$\": \"transformer_blocks.3.attn.to_out.weight\",\n \"^transformer_blocks.3.attn.to_out.0.bias$\": \"transformer_blocks.3.attn.to_out.bias\",\n \"^transformer_blocks.4.attn.to_out.0.weight$\": \"transformer_blocks.4.attn.to_out.weight\",\n \"^transformer_blocks.4.attn.to_out.0.bias$\": \"transformer_blocks.4.attn.to_out.bias\",\n \"^transformer_blocks.5.attn.to_out.0.weight$\": \"transformer_blocks.5.attn.to_out.weight\",\n \"^transformer_blocks.5.attn.to_out.0.bias$\": \"transformer_blocks.5.attn.to_out.bias\",\n \"^transformer_blocks.6.attn.to_out.0.weight$\": \"transformer_blocks.6.attn.to_out.weight\",\n \"^transformer_blocks.6.attn.to_out.0.bias$\": \"transformer_blocks.6.attn.to_out.bias\",\n \"^transformer_blocks.7.attn.to_out.0.weight$\": \"transformer_blocks.7.attn.to_out.weight\",\n \"^transformer_blocks.7.attn.to_out.0.bias$\": \"transformer_blocks.7.attn.to_out.bias\",\n \"^transformer_blocks.8.attn.to_out.0.weight$\": \"transformer_blocks.8.attn.to_out.weight\",\n \"^transformer_blocks.8.attn.to_out.0.bias$\": \"transformer_blocks.8.attn.to_out.bias\",\n \"^transformer_blocks.9.attn.to_out.0.weight$\": \"transformer_blocks.9.attn.to_out.weight\",\n \"^transformer_blocks.9.attn.to_out.0.bias$\": \"transformer_blocks.9.attn.to_out.bias\",\n \"^transformer_blocks.10.attn.to_out.0.weight$\": \"transformer_blocks.10.attn.to_out.weight\",\n \"^transformer_blocks.10.attn.to_out.0.bias$\": \"transformer_blocks.10.attn.to_out.bias\",\n \"^transformer_blocks.11.attn.to_out.0.weight$\": \"transformer_blocks.11.attn.to_out.weight\",\n \"^transformer_blocks.11.attn.to_out.0.bias$\": \"transformer_blocks.11.attn.to_out.bias\",\n \"^transformer_blocks.12.attn.to_out.0.weight$\": \"transformer_blocks.12.attn.to_out.weight\",\n \"^transformer_blocks.12.attn.to_out.0.bias$\": \"transformer_blocks.12.attn.to_out.bias\",\n \"^transformer_blocks.13.attn.to_out.0.weight$\": \"transformer_blocks.13.attn.to_out.weight\",\n \"^transformer_blocks.13.attn.to_out.0.bias$\": \"transformer_blocks.13.attn.to_out.bias\",\n \"^transformer_blocks.14.attn.to_out.0.weight$\": \"transformer_blocks.14.attn.to_out.weight\",\n \"^transformer_blocks.14.attn.to_out.0.bias$\": \"transformer_blocks.14.attn.to_out.bias\",\n \"^transformer_blocks.15.attn.to_out.0.weight$\": \"transformer_blocks.15.attn.to_out.weight\",\n \"^transformer_blocks.15.attn.to_out.0.bias$\": \"transformer_blocks.15.attn.to_out.bias\",\n \"^transformer_blocks.16.attn.to_out.0.weight$\": \"transformer_blocks.16.attn.to_out.weight\",\n \"^transformer_blocks.16.attn.to_out.0.bias$\": \"transformer_blocks.16.attn.to_out.bias\",\n \"^transformer_blocks.17.attn.to_out.0.weight$\": \"transformer_blocks.17.attn.to_out.weight\",\n \"^transformer_blocks.17.attn.to_out.0.bias$\": \"transformer_blocks.17.attn.to_out.bias\",\n \"^transformer_blocks.18.attn.to_out.0.weight$\": \"transformer_blocks.18.attn.to_out.weight\",\n \"^transformer_blocks.18.attn.to_out.0.bias$\": \"transformer_blocks.18.attn.to_out.bias\",\n \"^transformer_blocks.19.attn.to_out.0.weight$\": \"transformer_blocks.19.attn.to_out.weight\",\n \"^transformer_blocks.19.attn.to_out.0.bias$\": \"transformer_blocks.19.attn.to_out.bias\",\n \"^transformer_blocks.20.attn.to_out.0.weight$\": \"transformer_blocks.20.attn.to_out.weight\",\n \"^transformer_blocks.20.attn.to_out.0.bias$\": \"transformer_blocks.20.attn.to_out.bias\",\n \"^transformer_blocks.21.attn.to_out.0.weight$\": \"transformer_blocks.21.attn.to_out.weight\",\n \"^transformer_blocks.21.attn.to_out.0.bias$\": \"transformer_blocks.21.attn.to_out.bias\",\n \"^transformer_blocks.0.ff.ff.0.0.weight$\": \"transformer_blocks.0.ff.project_in.weight\",\n \"^transformer_blocks.0.ff.ff.0.0.bias$\": \"transformer_blocks.0.ff.project_in.bias\",\n \"^transformer_blocks.0.ff.ff.2.weight$\": \"transformer_blocks.0.ff.ff.weight\",\n \"^transformer_blocks.0.ff.ff.2.bias$\": \"transformer_blocks.0.ff.ff.bias\",\n \"^transformer_blocks.1.ff.ff.0.0.weight$\": \"transformer_blocks.1.ff.project_in.weight\",\n \"^transformer_blocks.1.ff.ff.0.0.bias$\": \"transformer_blocks.1.ff.project_in.bias\",\n \"^transformer_blocks.1.ff.ff.2.weight$\": \"transformer_blocks.1.ff.ff.weight\",\n \"^transformer_blocks.1.ff.ff.2.bias$\": \"transformer_blocks.1.ff.ff.bias\",\n \"^transformer_blocks.2.ff.ff.0.0.weight$\": \"transformer_blocks.2.ff.project_in.weight\",\n \"^transformer_blocks.2.ff.ff.0.0.bias$\": \"transformer_blocks.2.ff.project_in.bias\",\n \"^transformer_blocks.2.ff.ff.2.weight$\": \"transformer_blocks.2.ff.ff.weight\",\n \"^transformer_blocks.2.ff.ff.2.bias$\": \"transformer_blocks.2.ff.ff.bias\",\n \"^transformer_blocks.3.ff.ff.0.0.weight$\": \"transformer_blocks.3.ff.project_in.weight\",\n \"^transformer_blocks.3.ff.ff.0.0.bias$\": \"transformer_blocks.3.ff.project_in.bias\",\n \"^transformer_blocks.3.ff.ff.2.weight$\": \"transformer_blocks.3.ff.ff.weight\",\n \"^transformer_blocks.3.ff.ff.2.bias$\": \"transformer_blocks.3.ff.ff.bias\",\n \"^transformer_blocks.4.ff.ff.0.0.weight$\": \"transformer_blocks.4.ff.project_in.weight\",\n \"^transformer_blocks.4.ff.ff.0.0.bias$\": \"transformer_blocks.4.ff.project_in.bias\",\n \"^transformer_blocks.4.ff.ff.2.weight$\": \"transformer_blocks.4.ff.ff.weight\",\n \"^transformer_blocks.4.ff.ff.2.bias$\": \"transformer_blocks.4.ff.ff.bias\",\n \"^transformer_blocks.5.ff.ff.0.0.weight$\": \"transformer_blocks.5.ff.project_in.weight\",\n \"^transformer_blocks.5.ff.ff.0.0.bias$\": \"transformer_blocks.5.ff.project_in.bias\",\n \"^transformer_blocks.5.ff.ff.2.weight$\": \"transformer_blocks.5.ff.ff.weight\",\n \"^transformer_blocks.5.ff.ff.2.bias$\": \"transformer_blocks.5.ff.ff.bias\",\n \"^transformer_blocks.6.ff.ff.0.0.weight$\": \"transformer_blocks.6.ff.project_in.weight\",\n \"^transformer_blocks.6.ff.ff.0.0.bias$\": \"transformer_blocks.6.ff.project_in.bias\",\n \"^transformer_blocks.6.ff.ff.2.weight$\": \"transformer_blocks.6.ff.ff.weight\",\n \"^transformer_blocks.6.ff.ff.2.bias$\": \"transformer_blocks.6.ff.ff.bias\",\n \"^transformer_blocks.7.ff.ff.0.0.weight$\": \"transformer_blocks.7.ff.project_in.weight\",\n \"^transformer_blocks.7.ff.ff.0.0.bias$\": \"transformer_blocks.7.ff.project_in.bias\",\n \"^transformer_blocks.7.ff.ff.2.weight$\": \"transformer_blocks.7.ff.ff.weight\",\n \"^transformer_blocks.7.ff.ff.2.bias$\": \"transformer_blocks.7.ff.ff.bias\",\n \"^transformer_blocks.8.ff.ff.0.0.weight$\": \"transformer_blocks.8.ff.project_in.weight\",\n \"^transformer_blocks.8.ff.ff.0.0.bias$\": \"transformer_blocks.8.ff.project_in.bias\",\n \"^transformer_blocks.8.ff.ff.2.weight$\": \"transformer_blocks.8.ff.ff.weight\",\n \"^transformer_blocks.8.ff.ff.2.bias$\": \"transformer_blocks.8.ff.ff.bias\",\n \"^transformer_blocks.9.ff.ff.0.0.weight$\": \"transformer_blocks.9.ff.project_in.weight\",\n \"^transformer_blocks.9.ff.ff.0.0.bias$\": \"transformer_blocks.9.ff.project_in.bias\",\n \"^transformer_blocks.9.ff.ff.2.weight$\": \"transformer_blocks.9.ff.ff.weight\",\n \"^transformer_blocks.9.ff.ff.2.bias$\": \"transformer_blocks.9.ff.ff.bias\",\n \"^transformer_blocks.10.ff.ff.0.0.weight$\": \"transformer_blocks.10.ff.project_in.weight\",\n \"^transformer_blocks.10.ff.ff.0.0.bias$\": \"transformer_blocks.10.ff.project_in.bias\",\n \"^transformer_blocks.10.ff.ff.2.weight$\": \"transformer_blocks.10.ff.ff.weight\",\n \"^transformer_blocks.10.ff.ff.2.bias$\": \"transformer_blocks.10.ff.ff.bias\",\n \"^transformer_blocks.11.ff.ff.0.0.weight$\": \"transformer_blocks.11.ff.project_in.weight\",\n \"^transformer_blocks.11.ff.ff.0.0.bias$\": \"transformer_blocks.11.ff.project_in.bias\",\n \"^transformer_blocks.11.ff.ff.2.weight$\": \"transformer_blocks.11.ff.ff.weight\",\n \"^transformer_blocks.11.ff.ff.2.bias$\": \"transformer_blocks.11.ff.ff.bias\",\n \"^transformer_blocks.12.ff.ff.0.0.weight$\": \"transformer_blocks.12.ff.project_in.weight\",\n \"^transformer_blocks.12.ff.ff.0.0.bias$\": \"transformer_blocks.12.ff.project_in.bias\",\n \"^transformer_blocks.12.ff.ff.2.weight$\": \"transformer_blocks.12.ff.ff.weight\",\n \"^transformer_blocks.12.ff.ff.2.bias$\": \"transformer_blocks.12.ff.ff.bias\",\n \"^transformer_blocks.13.ff.ff.0.0.weight$\": \"transformer_blocks.13.ff.project_in.weight\",\n \"^transformer_blocks.13.ff.ff.0.0.bias$\": \"transformer_blocks.13.ff.project_in.bias\",\n \"^transformer_blocks.13.ff.ff.2.weight$\": \"transformer_blocks.13.ff.ff.weight\",\n \"^transformer_blocks.13.ff.ff.2.bias$\": \"transformer_blocks.13.ff.ff.bias\",\n \"^transformer_blocks.14.ff.ff.0.0.weight$\": \"transformer_blocks.14.ff.project_in.weight\",\n \"^transformer_blocks.14.ff.ff.0.0.bias$\": \"transformer_blocks.14.ff.project_in.bias\",\n \"^transformer_blocks.14.ff.ff.2.weight$\": \"transformer_blocks.14.ff.ff.weight\",\n \"^transformer_blocks.14.ff.ff.2.bias$\": \"transformer_blocks.14.ff.ff.bias\",\n \"^transformer_blocks.15.ff.ff.0.0.weight$\": \"transformer_blocks.15.ff.project_in.weight\",\n \"^transformer_blocks.15.ff.ff.0.0.bias$\": \"transformer_blocks.15.ff.project_in.bias\",\n \"^transformer_blocks.15.ff.ff.2.weight$\": \"transformer_blocks.15.ff.ff.weight\",\n \"^transformer_blocks.15.ff.ff.2.bias$\": \"transformer_blocks.15.ff.ff.bias\",\n \"^transformer_blocks.16.ff.ff.0.0.weight$\": \"transformer_blocks.16.ff.project_in.weight\",\n \"^transformer_blocks.16.ff.ff.0.0.bias$\": \"transformer_blocks.16.ff.project_in.bias\",\n \"^transformer_blocks.16.ff.ff.2.weight$\": \"transformer_blocks.16.ff.ff.weight\",\n \"^transformer_blocks.16.ff.ff.2.bias$\": \"transformer_blocks.16.ff.ff.bias\",\n \"^transformer_blocks.17.ff.ff.0.0.weight$\": \"transformer_blocks.17.ff.project_in.weight\",\n \"^transformer_blocks.17.ff.ff.0.0.bias$\": \"transformer_blocks.17.ff.project_in.bias\",\n \"^transformer_blocks.17.ff.ff.2.weight$\": \"transformer_blocks.17.ff.ff.weight\",\n \"^transformer_blocks.17.ff.ff.2.bias$\": \"transformer_blocks.17.ff.ff.bias\",\n \"^transformer_blocks.18.ff.ff.0.0.weight$\": \"transformer_blocks.18.ff.project_in.weight\",\n \"^transformer_blocks.18.ff.ff.0.0.bias$\": \"transformer_blocks.18.ff.project_in.bias\",\n \"^transformer_blocks.18.ff.ff.2.weight$\": \"transformer_blocks.18.ff.ff.weight\",\n \"^transformer_blocks.18.ff.ff.2.bias$\": \"transformer_blocks.18.ff.ff.bias\",\n \"^transformer_blocks.19.ff.ff.0.0.weight$\": \"transformer_blocks.19.ff.project_in.weight\",\n \"^transformer_blocks.19.ff.ff.0.0.bias$\": \"transformer_blocks.19.ff.project_in.bias\",\n \"^transformer_blocks.19.ff.ff.2.weight$\": \"transformer_blocks.19.ff.ff.weight\",\n \"^transformer_blocks.19.ff.ff.2.bias$\": \"transformer_blocks.19.ff.ff.bias\",\n \"^transformer_blocks.20.ff.ff.0.0.weight$\": \"transformer_blocks.20.ff.project_in.weight\",\n \"^transformer_blocks.20.ff.ff.0.0.bias$\": \"transformer_blocks.20.ff.project_in.bias\",\n \"^transformer_blocks.20.ff.ff.2.weight$\": \"transformer_blocks.20.ff.ff.weight\",\n \"^transformer_blocks.20.ff.ff.2.bias$\": \"transformer_blocks.20.ff.ff.bias\",\n \"^transformer_blocks.21.ff.ff.0.0.weight$\": \"transformer_blocks.21.ff.project_in.weight\",\n \"^transformer_blocks.21.ff.ff.0.0.bias$\": \"transformer_blocks.21.ff.project_in.bias\",\n \"^transformer_blocks.21.ff.ff.2.weight$\": \"transformer_blocks.21.ff.ff.weight\",\n \"^transformer_blocks.21.ff.ff.2.bias$\": \"transformer_blocks.21.ff.ff.bias\",\n}\n\n\ndef parse_arguments():\n parser = argparse.ArgumentParser()\n parser.add_argument(\n \"--model_name\",\n type=str,\n default=\"F5TTS_Base\",\n choices=[\n \"F5TTS_Base\",\n ],\n ) # TODO: support F5TTS_v1_Base\n parser.add_argument(\"--timm_ckpt\", type=str, default=\"./ckpts/model_1200000.pt\")\n parser.add_argument(\n \"--output_dir\", type=str, default=\"./tllm_checkpoint\", help=\"The path to save the TensorRT-LLM checkpoint\"\n )\n parser.add_argument(\"--hidden_size\", type=int, default=1024, help=\"The hidden size of DiT\")\n parser.add_argument(\"--depth\", type=int, default=22, help=\"The number of DiTBlock layers\")\n parser.add_argument(\"--num_heads\", type=int, default=16, help=\"The number of heads of attention module\")\n parser.add_argument(\"--cfg_scale\", type=float, default=4.0)\n parser.add_argument(\"--tp_size\", type=int, default=1, help=\"N-way tensor parallelism size\")\n parser.add_argument(\"--cp_size\", type=int, default=1, help=\"Context parallelism size\")\n parser.add_argument(\"--pp_size\", type=int, default=1, help=\"N-way pipeline parallelism size\")\n parser.add_argument(\"--dtype\", type=str, default=\"float16\", choices=[\"float32\", \"bfloat16\", \"float16\"])\n parser.add_argument(\"--fp8_linear\", action=\"store_true\", help=\"Whether use FP8 for linear layers\")\n parser.add_argument(\n \"--workers\", type=int, default=1, help=\"The number of workers for converting checkpoint in parallel\"\n )\n args = parser.parse_args()\n return args\n\n\ndef convert_timm_dit(args, mapping, dtype=\"float32\"):\n weights = {}\n tik = time.time()\n torch_dtype = str_dtype_to_torch(dtype)\n tensor_parallel = mapping.tp_size\n\n model_params = dict(torch.load(args.timm_ckpt))\n model_params = {\n k: v for k, v in model_params[\"ema_model_state_dict\"].items() if k.startswith(\"ema_model.transformer\")\n }\n prefix = \"ema_model.transformer.\"\n model_params = {key[len(prefix) :] if key.startswith(prefix) else key: value for key, value in model_params.items()}\n\n timm_to_trtllm_name = FACEBOOK_DIT_NAME_MAPPING\n\n def get_trtllm_name(timm_name):\n for k, v in timm_to_trtllm_name.items():\n m = re.match(k, timm_name)\n if m is not None:\n if \"*\" in v:\n v = v.replace(\"*\", m.groups()[0])\n return v\n return timm_name\n\n weights = dict()\n for name, param in model_params.items():\n if name == \"input_embed.conv_pos_embed.conv1d.0.weight\" or name == \"input_embed.conv_pos_embed.conv1d.2.weight\":\n weights[get_trtllm_name(name)] = param.contiguous().to(torch_dtype).unsqueeze(-1)\n else:\n weights[get_trtllm_name(name)] = param.contiguous().to(torch_dtype)\n\n assert len(weights) == len(model_params)\n\n # new_prefix = 'f5_transformer.'\n new_prefix = \"\"\n weights = {new_prefix + key: value for key, value in weights.items()}\n import math\n\n scale_factor = math.pow(64, -0.25)\n for k, v in weights.items():\n if re.match(\"^transformer_blocks.*.attn.to_k.weight$\", k):\n weights[k] *= scale_factor\n weights[k] = split_q_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n\n elif re.match(\"^transformer_blocks.*.attn.to_k.bias$\", k):\n weights[k] *= scale_factor\n weights[k] = split_q_bias_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n\n elif re.match(\"^transformer_blocks.*.attn.to_q.weight$\", k):\n weights[k] = split_q_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n weights[k] *= scale_factor\n\n elif re.match(\"^transformer_blocks.*.attn.to_q.bias$\", k):\n weights[k] = split_q_bias_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n weights[k] *= scale_factor\n\n elif re.match(\"^transformer_blocks.*.attn.to_v.weight$\", k):\n weights[k] = split_q_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n\n elif re.match(\"^transformer_blocks.*.attn.to_v.bias$\", k):\n weights[k] = split_q_bias_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n\n elif re.match(\"^transformer_blocks.*.attn.to_out.weight$\", k):\n weights[k] = split_matrix_tp(v, tensor_parallel, mapping.tp_rank, dim=1)\n\n tok = time.time()\n t = time.strftime(\"%H:%M:%S\", time.gmtime(tok - tik))\n print(f\"Weights loaded. Total time: {t}\")\n return weights\n\n\ndef save_config(args):\n if not os.path.exists(args.output_dir):\n os.makedirs(args.output_dir)\n config = {\n \"architecture\": \"F5TTS\",\n \"dtype\": args.dtype,\n \"hidden_size\": 1024,\n \"num_hidden_layers\": 22,\n \"num_attention_heads\": 16,\n \"dim_head\": 64,\n \"dropout\": 0.1,\n \"ff_mult\": 2,\n \"mel_dim\": 100,\n \"text_num_embeds\": 256,\n \"text_dim\": 512,\n \"conv_layers\": 4,\n \"long_skip_connection\": False,\n \"mapping\": {\n \"world_size\": args.cp_size * args.tp_size * args.pp_size,\n \"cp_size\": args.cp_size,\n \"tp_size\": args.tp_size,\n \"pp_size\": args.pp_size,\n },\n }\n if args.fp8_linear:\n config[\"quantization\"] = {\n \"quant_algo\": \"FP8\",\n # TODO: add support for exclude modules.\n # 'exclude_modules': \"*final_layer*\",\n }\n\n with open(os.path.join(args.output_dir, \"config.json\"), \"w\") as f:\n json.dump(config, f, indent=4)\n\n\ndef covert_and_save(args, rank):\n if rank == 0:\n save_config(args)\n\n mapping = Mapping(\n world_size=args.cp_size * args.tp_size * args.pp_size,\n rank=rank,\n cp_size=args.cp_size,\n tp_size=args.tp_size,\n pp_size=args.pp_size,\n )\n\n weights = convert_timm_dit(args, mapping, dtype=args.dtype)\n\n safetensors.torch.save_file(weights, os.path.join(args.output_dir, f\"rank{rank}.safetensors\"))\n\n\ndef execute(workers, func, args):\n if workers == 1:\n for rank, f in enumerate(func):\n f(args, rank)\n else:\n with ThreadPoolExecutor(max_workers=workers) as p:\n futures = [p.submit(f, args, rank) for rank, f in enumerate(func)]\n exceptions = []\n for future in as_completed(futures):\n try:\n future.result()\n except Exception as e:\n traceback.print_exc()\n exceptions.append(e)\n assert len(exceptions) == 0, \"Checkpoint conversion failed, please check error log.\"\n\n\ndef main():\n args = parse_arguments()\n world_size = args.cp_size * args.tp_size * args.pp_size\n\n assert args.pp_size == 1, \"PP is not supported yet.\"\n\n tik = time.time()\n if args.timm_ckpt is None:\n return\n print(\"start execute\")\n execute(args.workers, [covert_and_save] * world_size, args)\n\n tok = time.time()\n t = time.strfti\n# ... truncated ...","source_hash":"453fc4ae05851022942ced5d7df412f8d66f1d0fe85850978205a2c35becd32b","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.split_q_tp","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.split_q_tp#L16-L18","kind":"function","name":"split_q_tp","path":"src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","language":"python","start_line":16,"end_line":18,"context_start_line":1,"context_end_line":38,"code":"import argparse\nimport json\nimport os\nimport re\nimport time\nimport traceback\nfrom concurrent.futures import ThreadPoolExecutor, as_completed\n\nimport safetensors.torch\nimport torch\nfrom tensorrt_llm import str_dtype_to_torch\nfrom tensorrt_llm.mapping import Mapping\nfrom tensorrt_llm.models.convert_utils import split, split_matrix_tp\n\n\ndef split_q_tp(v, n_head, n_hidden, tensor_parallel, rank):\n split_v = split(v, tensor_parallel, rank, dim=1)\n return split_v.contiguous()\n\n\ndef split_q_bias_tp(v, n_head, n_hidden, tensor_parallel, rank):\n split_v = split(v, tensor_parallel, rank, dim=0)\n return split_v.contiguous()\n\n\nFACEBOOK_DIT_NAME_MAPPING = {\n \"^time_embed.time_mlp.0.weight$\": \"time_embed.mlp1.weight\",\n \"^time_embed.time_mlp.0.bias$\": \"time_embed.mlp1.bias\",\n \"^time_embed.time_mlp.2.weight$\": \"time_embed.mlp2.weight\",\n \"^time_embed.time_mlp.2.bias$\": \"time_embed.mlp2.bias\",\n \"^input_embed.conv_pos_embed.conv1d.0.weight$\": \"input_embed.conv_pos_embed.conv1d1.weight\",\n \"^input_embed.conv_pos_embed.conv1d.0.bias$\": \"input_embed.conv_pos_embed.conv1d1.bias\",\n \"^input_embed.conv_pos_embed.conv1d.2.weight$\": \"input_embed.conv_pos_embed.conv1d2.weight\",\n \"^input_embed.conv_pos_embed.conv1d.2.bias$\": \"input_embed.conv_pos_embed.conv1d2.bias\",\n \"^transformer_blocks.0.attn.to_out.0.weight$\": \"transformer_blocks.0.attn.to_out.weight\",\n \"^transformer_blocks.0.attn.to_out.0.bias$\": \"transformer_blocks.0.attn.to_out.bias\",\n \"^transformer_blocks.1.attn.to_out.0.weight$\": \"transformer_blocks.1.attn.to_out.weight\",\n \"^transformer_blocks.1.attn.to_out.0.bias$\": \"transformer_blocks.1.attn.to_out.bias\",","source_hash":"453fc4ae05851022942ced5d7df412f8d66f1d0fe85850978205a2c35becd32b","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.split_q_bias_tp","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.split_q_bias_tp#L21-L23","kind":"function","name":"split_q_bias_tp","path":"src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","language":"python","start_line":21,"end_line":23,"context_start_line":1,"context_end_line":43,"code":"import argparse\nimport json\nimport os\nimport re\nimport time\nimport traceback\nfrom concurrent.futures import ThreadPoolExecutor, as_completed\n\nimport safetensors.torch\nimport torch\nfrom tensorrt_llm import str_dtype_to_torch\nfrom tensorrt_llm.mapping import Mapping\nfrom tensorrt_llm.models.convert_utils import split, split_matrix_tp\n\n\ndef split_q_tp(v, n_head, n_hidden, tensor_parallel, rank):\n split_v = split(v, tensor_parallel, rank, dim=1)\n return split_v.contiguous()\n\n\ndef split_q_bias_tp(v, n_head, n_hidden, tensor_parallel, rank):\n split_v = split(v, tensor_parallel, rank, dim=0)\n return split_v.contiguous()\n\n\nFACEBOOK_DIT_NAME_MAPPING = {\n \"^time_embed.time_mlp.0.weight$\": \"time_embed.mlp1.weight\",\n \"^time_embed.time_mlp.0.bias$\": \"time_embed.mlp1.bias\",\n \"^time_embed.time_mlp.2.weight$\": \"time_embed.mlp2.weight\",\n \"^time_embed.time_mlp.2.bias$\": \"time_embed.mlp2.bias\",\n \"^input_embed.conv_pos_embed.conv1d.0.weight$\": \"input_embed.conv_pos_embed.conv1d1.weight\",\n \"^input_embed.conv_pos_embed.conv1d.0.bias$\": \"input_embed.conv_pos_embed.conv1d1.bias\",\n \"^input_embed.conv_pos_embed.conv1d.2.weight$\": \"input_embed.conv_pos_embed.conv1d2.weight\",\n \"^input_embed.conv_pos_embed.conv1d.2.bias$\": \"input_embed.conv_pos_embed.conv1d2.bias\",\n \"^transformer_blocks.0.attn.to_out.0.weight$\": \"transformer_blocks.0.attn.to_out.weight\",\n \"^transformer_blocks.0.attn.to_out.0.bias$\": \"transformer_blocks.0.attn.to_out.bias\",\n \"^transformer_blocks.1.attn.to_out.0.weight$\": \"transformer_blocks.1.attn.to_out.weight\",\n \"^transformer_blocks.1.attn.to_out.0.bias$\": \"transformer_blocks.1.attn.to_out.bias\",\n \"^transformer_blocks.2.attn.to_out.0.weight$\": \"transformer_blocks.2.attn.to_out.weight\",\n \"^transformer_blocks.2.attn.to_out.0.bias$\": \"transformer_blocks.2.attn.to_out.bias\",\n \"^transformer_blocks.3.attn.to_out.0.weight$\": \"transformer_blocks.3.attn.to_out.weight\",\n \"^transformer_blocks.3.attn.to_out.0.bias$\": \"transformer_blocks.3.attn.to_out.bias\",\n \"^transformer_blocks.4.attn.to_out.0.weight$\": \"transformer_blocks.4.attn.to_out.weight\",","source_hash":"453fc4ae05851022942ced5d7df412f8d66f1d0fe85850978205a2c35becd32b","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.parse_arguments","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.parse_arguments#L170-L197","kind":"function","name":"parse_arguments","path":"src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","language":"python","start_line":170,"end_line":197,"context_start_line":150,"context_end_line":217,"code":" \"^transformer_blocks.17.ff.ff.2.bias$\": \"transformer_blocks.17.ff.ff.bias\",\n \"^transformer_blocks.18.ff.ff.0.0.weight$\": \"transformer_blocks.18.ff.project_in.weight\",\n \"^transformer_blocks.18.ff.ff.0.0.bias$\": \"transformer_blocks.18.ff.project_in.bias\",\n \"^transformer_blocks.18.ff.ff.2.weight$\": \"transformer_blocks.18.ff.ff.weight\",\n \"^transformer_blocks.18.ff.ff.2.bias$\": \"transformer_blocks.18.ff.ff.bias\",\n \"^transformer_blocks.19.ff.ff.0.0.weight$\": \"transformer_blocks.19.ff.project_in.weight\",\n \"^transformer_blocks.19.ff.ff.0.0.bias$\": \"transformer_blocks.19.ff.project_in.bias\",\n \"^transformer_blocks.19.ff.ff.2.weight$\": \"transformer_blocks.19.ff.ff.weight\",\n \"^transformer_blocks.19.ff.ff.2.bias$\": \"transformer_blocks.19.ff.ff.bias\",\n \"^transformer_blocks.20.ff.ff.0.0.weight$\": \"transformer_blocks.20.ff.project_in.weight\",\n \"^transformer_blocks.20.ff.ff.0.0.bias$\": \"transformer_blocks.20.ff.project_in.bias\",\n \"^transformer_blocks.20.ff.ff.2.weight$\": \"transformer_blocks.20.ff.ff.weight\",\n \"^transformer_blocks.20.ff.ff.2.bias$\": \"transformer_blocks.20.ff.ff.bias\",\n \"^transformer_blocks.21.ff.ff.0.0.weight$\": \"transformer_blocks.21.ff.project_in.weight\",\n \"^transformer_blocks.21.ff.ff.0.0.bias$\": \"transformer_blocks.21.ff.project_in.bias\",\n \"^transformer_blocks.21.ff.ff.2.weight$\": \"transformer_blocks.21.ff.ff.weight\",\n \"^transformer_blocks.21.ff.ff.2.bias$\": \"transformer_blocks.21.ff.ff.bias\",\n}\n\n\ndef parse_arguments():\n parser = argparse.ArgumentParser()\n parser.add_argument(\n \"--model_name\",\n type=str,\n default=\"F5TTS_Base\",\n choices=[\n \"F5TTS_Base\",\n ],\n ) # TODO: support F5TTS_v1_Base\n parser.add_argument(\"--timm_ckpt\", type=str, default=\"./ckpts/model_1200000.pt\")\n parser.add_argument(\n \"--output_dir\", type=str, default=\"./tllm_checkpoint\", help=\"The path to save the TensorRT-LLM checkpoint\"\n )\n parser.add_argument(\"--hidden_size\", type=int, default=1024, help=\"The hidden size of DiT\")\n parser.add_argument(\"--depth\", type=int, default=22, help=\"The number of DiTBlock layers\")\n parser.add_argument(\"--num_heads\", type=int, default=16, help=\"The number of heads of attention module\")\n parser.add_argument(\"--cfg_scale\", type=float, default=4.0)\n parser.add_argument(\"--tp_size\", type=int, default=1, help=\"N-way tensor parallelism size\")\n parser.add_argument(\"--cp_size\", type=int, default=1, help=\"Context parallelism size\")\n parser.add_argument(\"--pp_size\", type=int, default=1, help=\"N-way pipeline parallelism size\")\n parser.add_argument(\"--dtype\", type=str, default=\"float16\", choices=[\"float32\", \"bfloat16\", \"float16\"])\n parser.add_argument(\"--fp8_linear\", action=\"store_true\", help=\"Whether use FP8 for linear layers\")\n parser.add_argument(\n \"--workers\", type=int, default=1, help=\"The number of workers for converting checkpoint in parallel\"\n )\n args = parser.parse_args()\n return args\n\n\ndef convert_timm_dit(args, mapping, dtype=\"float32\"):\n weights = {}\n tik = time.time()\n torch_dtype = str_dtype_to_torch(dtype)\n tensor_parallel = mapping.tp_size\n\n model_params = dict(torch.load(args.timm_ckpt))\n model_params = {\n k: v for k, v in model_params[\"ema_model_state_dict\"].items() if k.startswith(\"ema_model.transformer\")\n }\n prefix = \"ema_model.transformer.\"\n model_params = {key[len(prefix) :] if key.startswith(prefix) else key: value for key, value in model_params.items()}\n\n timm_to_trtllm_name = FACEBOOK_DIT_NAME_MAPPING\n\n def get_trtllm_name(timm_name):\n for k, v in timm_to_trtllm_name.items():\n m = re.match(k, timm_name)","source_hash":"453fc4ae05851022942ced5d7df412f8d66f1d0fe85850978205a2c35becd32b","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.convert_timm_dit","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.convert_timm_dit#L200-L268","kind":"function","name":"convert_timm_dit","path":"src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","language":"python","start_line":200,"end_line":268,"context_start_line":180,"context_end_line":288,"code":" parser.add_argument(\"--timm_ckpt\", type=str, default=\"./ckpts/model_1200000.pt\")\n parser.add_argument(\n \"--output_dir\", type=str, default=\"./tllm_checkpoint\", help=\"The path to save the TensorRT-LLM checkpoint\"\n )\n parser.add_argument(\"--hidden_size\", type=int, default=1024, help=\"The hidden size of DiT\")\n parser.add_argument(\"--depth\", type=int, default=22, help=\"The number of DiTBlock layers\")\n parser.add_argument(\"--num_heads\", type=int, default=16, help=\"The number of heads of attention module\")\n parser.add_argument(\"--cfg_scale\", type=float, default=4.0)\n parser.add_argument(\"--tp_size\", type=int, default=1, help=\"N-way tensor parallelism size\")\n parser.add_argument(\"--cp_size\", type=int, default=1, help=\"Context parallelism size\")\n parser.add_argument(\"--pp_size\", type=int, default=1, help=\"N-way pipeline parallelism size\")\n parser.add_argument(\"--dtype\", type=str, default=\"float16\", choices=[\"float32\", \"bfloat16\", \"float16\"])\n parser.add_argument(\"--fp8_linear\", action=\"store_true\", help=\"Whether use FP8 for linear layers\")\n parser.add_argument(\n \"--workers\", type=int, default=1, help=\"The number of workers for converting checkpoint in parallel\"\n )\n args = parser.parse_args()\n return args\n\n\ndef convert_timm_dit(args, mapping, dtype=\"float32\"):\n weights = {}\n tik = time.time()\n torch_dtype = str_dtype_to_torch(dtype)\n tensor_parallel = mapping.tp_size\n\n model_params = dict(torch.load(args.timm_ckpt))\n model_params = {\n k: v for k, v in model_params[\"ema_model_state_dict\"].items() if k.startswith(\"ema_model.transformer\")\n }\n prefix = \"ema_model.transformer.\"\n model_params = {key[len(prefix) :] if key.startswith(prefix) else key: value for key, value in model_params.items()}\n\n timm_to_trtllm_name = FACEBOOK_DIT_NAME_MAPPING\n\n def get_trtllm_name(timm_name):\n for k, v in timm_to_trtllm_name.items():\n m = re.match(k, timm_name)\n if m is not None:\n if \"*\" in v:\n v = v.replace(\"*\", m.groups()[0])\n return v\n return timm_name\n\n weights = dict()\n for name, param in model_params.items():\n if name == \"input_embed.conv_pos_embed.conv1d.0.weight\" or name == \"input_embed.conv_pos_embed.conv1d.2.weight\":\n weights[get_trtllm_name(name)] = param.contiguous().to(torch_dtype).unsqueeze(-1)\n else:\n weights[get_trtllm_name(name)] = param.contiguous().to(torch_dtype)\n\n assert len(weights) == len(model_params)\n\n # new_prefix = 'f5_transformer.'\n new_prefix = \"\"\n weights = {new_prefix + key: value for key, value in weights.items()}\n import math\n\n scale_factor = math.pow(64, -0.25)\n for k, v in weights.items():\n if re.match(\"^transformer_blocks.*.attn.to_k.weight$\", k):\n weights[k] *= scale_factor\n weights[k] = split_q_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n\n elif re.match(\"^transformer_blocks.*.attn.to_k.bias$\", k):\n weights[k] *= scale_factor\n weights[k] = split_q_bias_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n\n elif re.match(\"^transformer_blocks.*.attn.to_q.weight$\", k):\n weights[k] = split_q_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n weights[k] *= scale_factor\n\n elif re.match(\"^transformer_blocks.*.attn.to_q.bias$\", k):\n weights[k] = split_q_bias_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n weights[k] *= scale_factor\n\n elif re.match(\"^transformer_blocks.*.attn.to_v.weight$\", k):\n weights[k] = split_q_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n\n elif re.match(\"^transformer_blocks.*.attn.to_v.bias$\", k):\n weights[k] = split_q_bias_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n\n elif re.match(\"^transformer_blocks.*.attn.to_out.weight$\", k):\n weights[k] = split_matrix_tp(v, tensor_parallel, mapping.tp_rank, dim=1)\n\n tok = time.time()\n t = time.strftime(\"%H:%M:%S\", time.gmtime(tok - tik))\n print(f\"Weights loaded. Total time: {t}\")\n return weights\n\n\ndef save_config(args):\n if not os.path.exists(args.output_dir):\n os.makedirs(args.output_dir)\n config = {\n \"architecture\": \"F5TTS\",\n \"dtype\": args.dtype,\n \"hidden_size\": 1024,\n \"num_hidden_layers\": 22,\n \"num_attention_heads\": 16,\n \"dim_head\": 64,\n \"dropout\": 0.1,\n \"ff_mult\": 2,\n \"mel_dim\": 100,\n \"text_num_embeds\": 256,\n \"text_dim\": 512,\n \"conv_layers\": 4,\n \"long_skip_connection\": False,\n \"mapping\": {","source_hash":"453fc4ae05851022942ced5d7df412f8d66f1d0fe85850978205a2c35becd32b","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.save_config","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.save_config#L271-L303","kind":"function","name":"save_config","path":"src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","language":"python","start_line":271,"end_line":303,"context_start_line":251,"context_end_line":323,"code":"\n elif re.match(\"^transformer_blocks.*.attn.to_q.bias$\", k):\n weights[k] = split_q_bias_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n weights[k] *= scale_factor\n\n elif re.match(\"^transformer_blocks.*.attn.to_v.weight$\", k):\n weights[k] = split_q_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n\n elif re.match(\"^transformer_blocks.*.attn.to_v.bias$\", k):\n weights[k] = split_q_bias_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)\n\n elif re.match(\"^transformer_blocks.*.attn.to_out.weight$\", k):\n weights[k] = split_matrix_tp(v, tensor_parallel, mapping.tp_rank, dim=1)\n\n tok = time.time()\n t = time.strftime(\"%H:%M:%S\", time.gmtime(tok - tik))\n print(f\"Weights loaded. Total time: {t}\")\n return weights\n\n\ndef save_config(args):\n if not os.path.exists(args.output_dir):\n os.makedirs(args.output_dir)\n config = {\n \"architecture\": \"F5TTS\",\n \"dtype\": args.dtype,\n \"hidden_size\": 1024,\n \"num_hidden_layers\": 22,\n \"num_attention_heads\": 16,\n \"dim_head\": 64,\n \"dropout\": 0.1,\n \"ff_mult\": 2,\n \"mel_dim\": 100,\n \"text_num_embeds\": 256,\n \"text_dim\": 512,\n \"conv_layers\": 4,\n \"long_skip_connection\": False,\n \"mapping\": {\n \"world_size\": args.cp_size * args.tp_size * args.pp_size,\n \"cp_size\": args.cp_size,\n \"tp_size\": args.tp_size,\n \"pp_size\": args.pp_size,\n },\n }\n if args.fp8_linear:\n config[\"quantization\"] = {\n \"quant_algo\": \"FP8\",\n # TODO: add support for exclude modules.\n # 'exclude_modules': \"*final_layer*\",\n }\n\n with open(os.path.join(args.output_dir, \"config.json\"), \"w\") as f:\n json.dump(config, f, indent=4)\n\n\ndef covert_and_save(args, rank):\n if rank == 0:\n save_config(args)\n\n mapping = Mapping(\n world_size=args.cp_size * args.tp_size * args.pp_size,\n rank=rank,\n cp_size=args.cp_size,\n tp_size=args.tp_size,\n pp_size=args.pp_size,\n )\n\n weights = convert_timm_dit(args, mapping, dtype=args.dtype)\n\n safetensors.torch.save_file(weights, os.path.join(args.output_dir, f\"rank{rank}.safetensors\"))\n\n\ndef execute(workers, func, args):","source_hash":"453fc4ae05851022942ced5d7df412f8d66f1d0fe85850978205a2c35becd32b","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.covert_and_save","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.covert_and_save#L306-L320","kind":"function","name":"covert_and_save","path":"src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","language":"python","start_line":306,"end_line":320,"context_start_line":286,"context_end_line":340,"code":" \"conv_layers\": 4,\n \"long_skip_connection\": False,\n \"mapping\": {\n \"world_size\": args.cp_size * args.tp_size * args.pp_size,\n \"cp_size\": args.cp_size,\n \"tp_size\": args.tp_size,\n \"pp_size\": args.pp_size,\n },\n }\n if args.fp8_linear:\n config[\"quantization\"] = {\n \"quant_algo\": \"FP8\",\n # TODO: add support for exclude modules.\n # 'exclude_modules': \"*final_layer*\",\n }\n\n with open(os.path.join(args.output_dir, \"config.json\"), \"w\") as f:\n json.dump(config, f, indent=4)\n\n\ndef covert_and_save(args, rank):\n if rank == 0:\n save_config(args)\n\n mapping = Mapping(\n world_size=args.cp_size * args.tp_size * args.pp_size,\n rank=rank,\n cp_size=args.cp_size,\n tp_size=args.tp_size,\n pp_size=args.pp_size,\n )\n\n weights = convert_timm_dit(args, mapping, dtype=args.dtype)\n\n safetensors.torch.save_file(weights, os.path.join(args.output_dir, f\"rank{rank}.safetensors\"))\n\n\ndef execute(workers, func, args):\n if workers == 1:\n for rank, f in enumerate(func):\n f(args, rank)\n else:\n with ThreadPoolExecutor(max_workers=workers) as p:\n futures = [p.submit(f, args, rank) for rank, f in enumerate(func)]\n exceptions = []\n for future in as_completed(futures):\n try:\n future.result()\n except Exception as e:\n traceback.print_exc()\n exceptions.append(e)\n assert len(exceptions) == 0, \"Checkpoint conversion failed, please check error log.\"\n\n\ndef main():","source_hash":"453fc4ae05851022942ced5d7df412f8d66f1d0fe85850978205a2c35becd32b","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.execute","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.execute#L323-L337","kind":"function","name":"execute","path":"src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","language":"python","start_line":323,"end_line":337,"context_start_line":303,"context_end_line":357,"code":" json.dump(config, f, indent=4)\n\n\ndef covert_and_save(args, rank):\n if rank == 0:\n save_config(args)\n\n mapping = Mapping(\n world_size=args.cp_size * args.tp_size * args.pp_size,\n rank=rank,\n cp_size=args.cp_size,\n tp_size=args.tp_size,\n pp_size=args.pp_size,\n )\n\n weights = convert_timm_dit(args, mapping, dtype=args.dtype)\n\n safetensors.torch.save_file(weights, os.path.join(args.output_dir, f\"rank{rank}.safetensors\"))\n\n\ndef execute(workers, func, args):\n if workers == 1:\n for rank, f in enumerate(func):\n f(args, rank)\n else:\n with ThreadPoolExecutor(max_workers=workers) as p:\n futures = [p.submit(f, args, rank) for rank, f in enumerate(func)]\n exceptions = []\n for future in as_completed(futures):\n try:\n future.result()\n except Exception as e:\n traceback.print_exc()\n exceptions.append(e)\n assert len(exceptions) == 0, \"Checkpoint conversion failed, please check error log.\"\n\n\ndef main():\n args = parse_arguments()\n world_size = args.cp_size * args.tp_size * args.pp_size\n\n assert args.pp_size == 1, \"PP is not supported yet.\"\n\n tik = time.time()\n if args.timm_ckpt is None:\n return\n print(\"start execute\")\n execute(args.workers, [covert_and_save] * world_size, args)\n\n tok = time.time()\n t = time.strftime(\"%H:%M:%S\", time.gmtime(tok - tik))\n print(f\"Total time of converting checkpoints: {t}\")\n\n\nif __name__ == \"__main__\":","source_hash":"453fc4ae05851022942ced5d7df412f8d66f1d0fe85850978205a2c35becd32b","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.main","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.main#L340-L354","kind":"function","name":"main","path":"src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","language":"python","start_line":340,"end_line":354,"context_start_line":320,"context_end_line":358,"code":" safetensors.torch.save_file(weights, os.path.join(args.output_dir, f\"rank{rank}.safetensors\"))\n\n\ndef execute(workers, func, args):\n if workers == 1:\n for rank, f in enumerate(func):\n f(args, rank)\n else:\n with ThreadPoolExecutor(max_workers=workers) as p:\n futures = [p.submit(f, args, rank) for rank, f in enumerate(func)]\n exceptions = []\n for future in as_completed(futures):\n try:\n future.result()\n except Exception as e:\n traceback.print_exc()\n exceptions.append(e)\n assert len(exceptions) == 0, \"Checkpoint conversion failed, please check error log.\"\n\n\ndef main():\n args = parse_arguments()\n world_size = args.cp_size * args.tp_size * args.pp_size\n\n assert args.pp_size == 1, \"PP is not supported yet.\"\n\n tik = time.time()\n if args.timm_ckpt is None:\n return\n print(\"start execute\")\n execute(args.workers, [covert_and_save] * world_size, args)\n\n tok = time.time()\n t = time.strftime(\"%H:%M:%S\", time.gmtime(tok - tik))\n print(f\"Total time of converting checkpoints: {t}\")\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"453fc4ae05851022942ced5d7df412f8d66f1d0fe85850978205a2c35becd32b","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.get_trtllm_name","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.convert_checkpoint.get_trtllm_name#L215-L222","kind":"function","name":"get_trtllm_name","path":"src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","language":"python","start_line":215,"end_line":222,"context_start_line":195,"context_end_line":242,"code":" )\n args = parser.parse_args()\n return args\n\n\ndef convert_timm_dit(args, mapping, dtype=\"float32\"):\n weights = {}\n tik = time.time()\n torch_dtype = str_dtype_to_torch(dtype)\n tensor_parallel = mapping.tp_size\n\n model_params = dict(torch.load(args.timm_ckpt))\n model_params = {\n k: v for k, v in model_params[\"ema_model_state_dict\"].items() if k.startswith(\"ema_model.transformer\")\n }\n prefix = \"ema_model.transformer.\"\n model_params = {key[len(prefix) :] if key.startswith(prefix) else key: value for key, value in model_params.items()}\n\n timm_to_trtllm_name = FACEBOOK_DIT_NAME_MAPPING\n\n def get_trtllm_name(timm_name):\n for k, v in timm_to_trtllm_name.items():\n m = re.match(k, timm_name)\n if m is not None:\n if \"*\" in v:\n v = v.replace(\"*\", m.groups()[0])\n return v\n return timm_name\n\n weights = dict()\n for name, param in model_params.items():\n if name == \"input_embed.conv_pos_embed.conv1d.0.weight\" or name == \"input_embed.conv_pos_embed.conv1d.2.weight\":\n weights[get_trtllm_name(name)] = param.contiguous().to(torch_dtype).unsqueeze(-1)\n else:\n weights[get_trtllm_name(name)] = param.contiguous().to(torch_dtype)\n\n assert len(weights) == len(model_params)\n\n # new_prefix = 'f5_transformer.'\n new_prefix = \"\"\n weights = {new_prefix + key: value for key, value in weights.items()}\n import math\n\n scale_factor = math.pow(64, -0.25)\n for k, v in weights.items():\n if re.match(\"^transformer_blocks.*.attn.to_k.weight$\", k):\n weights[k] *= scale_factor\n weights[k] = split_q_tp(v, args.num_heads, args.hidden_size, tensor_parallel, mapping.tp_rank)","source_hash":"453fc4ae05851022942ced5d7df412f8d66f1d0fe85850978205a2c35becd32b","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.fill_template","uri":"program://DMOSpeech2/module/src.f5_tts.runtime.triton_trtllm.scripts.fill_template#L1-L36","kind":"module","name":"src.f5_tts.runtime.triton_trtllm.scripts.fill_template","path":"src/f5_tts/runtime/triton_trtllm/scripts/fill_template.py","language":"python","start_line":1,"end_line":36,"context_start_line":1,"context_end_line":36,"code":"#! /usr/bin/env python3\nfrom argparse import ArgumentParser\nfrom string import Template\n\n\ndef main(file_path, substitutions, in_place, participant_ids):\n with open(file_path) as f:\n pbtxt = Template(f.read())\n\n sub_dict = {\"max_queue_size\": 0}\n sub_dict[\"participant_ids\"] = participant_ids\n for sub in substitutions.split(\",\"):\n key, value = sub.split(\":\")\n sub_dict[key] = value\n\n pbtxt = pbtxt.safe_substitute(sub_dict)\n\n if in_place:\n with open(file_path, \"w\") as f:\n f.write(pbtxt)\n else:\n print(pbtxt)\n\n\nif __name__ == \"__main__\":\n parser = ArgumentParser()\n parser.add_argument(\"file_path\", help=\"path of the .pbtxt to modify\")\n parser.add_argument(\n \"substitutions\",\n help=\"substitutions to perform, in the format variable_name_1:value_1,variable_name_2:value_2...\",\n )\n parser.add_argument(\"--in_place\", \"-i\", action=\"store_true\", help=\"do the operation in-place\")\n parser.add_argument(\"--participant_ids\", help=\"Participant IDs for the model\", default=\"\")\n args = parser.parse_args()\n\n main(**vars(args))","source_hash":"2341448e22389ebd542a52552ba5c093ebd6096c2cc7d5e00c9c598e7b062654","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.runtime.triton_trtllm.scripts.fill_template.main","uri":"program://DMOSpeech2/function/src.f5_tts.runtime.triton_trtllm.scripts.fill_template.main#L6-L22","kind":"function","name":"main","path":"src/f5_tts/runtime/triton_trtllm/scripts/fill_template.py","language":"python","start_line":6,"end_line":22,"context_start_line":1,"context_end_line":36,"code":"#! /usr/bin/env python3\nfrom argparse import ArgumentParser\nfrom string import Template\n\n\ndef main(file_path, substitutions, in_place, participant_ids):\n with open(file_path) as f:\n pbtxt = Template(f.read())\n\n sub_dict = {\"max_queue_size\": 0}\n sub_dict[\"participant_ids\"] = participant_ids\n for sub in substitutions.split(\",\"):\n key, value = sub.split(\":\")\n sub_dict[key] = value\n\n pbtxt = pbtxt.safe_substitute(sub_dict)\n\n if in_place:\n with open(file_path, \"w\") as f:\n f.write(pbtxt)\n else:\n print(pbtxt)\n\n\nif __name__ == \"__main__\":\n parser = ArgumentParser()\n parser.add_argument(\"file_path\", help=\"path of the .pbtxt to modify\")\n parser.add_argument(\n \"substitutions\",\n help=\"substitutions to perform, in the format variable_name_1:value_1,variable_name_2:value_2...\",\n )\n parser.add_argument(\"--in_place\", \"-i\", action=\"store_true\", help=\"do the operation in-place\")\n parser.add_argument(\"--participant_ids\", help=\"Participant IDs for the model\", default=\"\")\n args = parser.parse_args()\n\n main(**vars(args))","source_hash":"2341448e22389ebd542a52552ba5c093ebd6096c2cc7d5e00c9c598e7b062654","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.scripts.count_max_epoch","uri":"program://DMOSpeech2/module/src.f5_tts.scripts.count_max_epoch#L1-L33","kind":"module","name":"src.f5_tts.scripts.count_max_epoch","path":"src/f5_tts/scripts/count_max_epoch.py","language":"python","start_line":1,"end_line":33,"context_start_line":1,"context_end_line":33,"code":"\"\"\"ADAPTIVE BATCH SIZE\"\"\"\n\nprint(\"Adaptive batch size: using grouping batch sampler, frames_per_gpu fixed fed in\")\nprint(\" -> least padding, gather wavs with accumulated frames in a batch\\n\")\n\n# data\ntotal_hours = 95282\nmel_hop_length = 256\nmel_sampling_rate = 24000\n\n# target\nwanted_max_updates = 1200000\n\n# train params\ngpus = 8\nframes_per_gpu = 38400 # 8 * 38400 = 307200\ngrad_accum = 1\n\n# intermediate\nmini_batch_frames = frames_per_gpu * grad_accum * gpus\nmini_batch_hours = mini_batch_frames * mel_hop_length / mel_sampling_rate / 3600\nupdates_per_epoch = total_hours / mini_batch_hours\n# steps_per_epoch = updates_per_epoch * grad_accum\n\n# result\nepochs = wanted_max_updates / updates_per_epoch\nprint(f\"epochs should be set to: {epochs:.0f} ({epochs / grad_accum:.1f} x gd_acum {grad_accum})\")\nprint(f\"progress_bar should show approx. 0/{updates_per_epoch:.0f} updates\")\n# print(f\" or approx. 0/{steps_per_epoch:.0f} steps\")\n\n# others\nprint(f\"total {total_hours:.0f} hours\")\nprint(f\"mini-batch of {mini_batch_frames:.0f} frames, {mini_batch_hours:.2f} hours per mini-batch\")","source_hash":"43c0b6aa0a0a924397c53b71cb0b00eac662736f7daa94790e413147a8bf05de","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.scripts.count_params_gflops","uri":"program://DMOSpeech2/module/src.f5_tts.scripts.count_params_gflops#L1-L40","kind":"module","name":"src.f5_tts.scripts.count_params_gflops","path":"src/f5_tts/scripts/count_params_gflops.py","language":"python","start_line":1,"end_line":40,"context_start_line":1,"context_end_line":40,"code":"import os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport thop\nimport torch\n\nfrom f5_tts.model import CFM, DiT\n\n\n\"\"\" ~155M \"\"\"\n# transformer = UNetT(dim = 768, depth = 20, heads = 12, ff_mult = 4)\n# transformer = UNetT(dim = 768, depth = 20, heads = 12, ff_mult = 4, text_dim = 512, conv_layers = 4)\n# transformer = DiT(dim = 768, depth = 18, heads = 12, ff_mult = 2)\n# transformer = DiT(dim = 768, depth = 18, heads = 12, ff_mult = 2, text_dim = 512, conv_layers = 4)\n# transformer = DiT(dim = 768, depth = 18, heads = 12, ff_mult = 2, text_dim = 512, conv_layers = 4, long_skip_connection = True)\n# transformer = MMDiT(dim = 512, depth = 16, heads = 16, ff_mult = 2)\n\n\"\"\" ~335M \"\"\"\n# FLOPs: 622.1 G, Params: 333.2 M\n# transformer = UNetT(dim = 1024, depth = 24, heads = 16, ff_mult = 4)\n# FLOPs: 363.4 G, Params: 335.8 M\ntransformer = DiT(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)\n\n\nmodel = CFM(transformer=transformer)\ntarget_sample_rate = 24000\nn_mel_channels = 100\nhop_length = 256\nduration = 20\nframe_length = int(duration * target_sample_rate / hop_length)\ntext_length = 150\n\nflops, params = thop.profile(\n model, inputs=(torch.randn(1, frame_length, n_mel_channels), torch.zeros(1, text_length, dtype=torch.long))\n)\nprint(f\"FLOPs: {flops / 1e9} G\")\nprint(f\"Params: {params / 1e6} M\")","source_hash":"ea250747154d3900b617f90c24c9447c9c992397a3b6e08f774bb84a21846af9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer","uri":"program://DMOSpeech2/module/src.f5_tts.infer.utils_infer#L1-L605","kind":"module","name":"src.f5_tts.infer.utils_infer","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":1,"end_line":605,"context_start_line":1,"context_end_line":605,"code":"# A unified script for inference process\n# Make adjustments inside functions, and consider both gradio and cli scripts if need to change func output format\nimport os\nimport sys\nfrom concurrent.futures import ThreadPoolExecutor\n\n\nos.environ[\"PYTORCH_ENABLE_MPS_FALLBACK\"] = \"1\" # for MPS device compatibility\nsys.path.append(f\"{os.path.dirname(os.path.abspath(__file__))}/../../third_party/BigVGAN/\")\n\nimport hashlib\nimport re\nimport tempfile\nfrom importlib.resources import files\n\nimport matplotlib\n\n\nmatplotlib.use(\"Agg\")\n\nimport matplotlib.pylab as plt\nimport numpy as np\nimport torch\nimport torchaudio\nimport tqdm\nfrom huggingface_hub import hf_hub_download\nfrom pydub import AudioSegment, silence\nfrom transformers import pipeline\nfrom vocos import Vocos\n\nfrom f5_tts.model import CFM\nfrom f5_tts.model.utils import convert_char_to_pinyin, get_tokenizer\n\n\n_ref_audio_cache = {}\n_ref_text_cache = {}\n\ndevice = (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n)\n\ntempfile_kwargs = {\"delete_on_close\": False} if sys.version_info >= (3, 12) else {\"delete\": False}\n\n# -----------------------------------------\n\ntarget_sample_rate = 24000\nn_mel_channels = 100\nhop_length = 256\nwin_length = 1024\nn_fft = 1024\nmel_spec_type = \"vocos\"\ntarget_rms = 0.1\ncross_fade_duration = 0.15\node_method = \"euler\"\nnfe_step = 32 # 16, 32\ncfg_strength = 2.0\nsway_sampling_coef = -1.0\nspeed = 1.0\nfix_duration = None\n\n# -----------------------------------------\n\n\n# chunk text into smaller pieces\n\n\ndef chunk_text(text, max_chars=135):\n \"\"\"\n Splits the input text into chunks, each with a maximum number of characters.\n\n Args:\n text (str): The text to be split.\n max_chars (int): The maximum number of characters per chunk.\n\n Returns:\n List[str]: A list of text chunks.\n \"\"\"\n chunks = []\n current_chunk = \"\"\n # Split the text into sentences based on punctuation followed by whitespace\n sentences = re.split(r\"(?<=[;:,.!?])\\s+|(?<=[；：，。！？])\", text)\n\n for sentence in sentences:\n if len(current_chunk.encode(\"utf-8\")) + len(sentence.encode(\"utf-8\")) <= max_chars:\n current_chunk += sentence + \" \" if sentence and len(sentence[-1].encode(\"utf-8\")) == 1 else sentence\n else:\n if current_chunk:\n chunks.append(current_chunk.strip())\n current_chunk = sentence + \" \" if sentence and len(sentence[-1].encode(\"utf-8\")) == 1 else sentence\n\n if current_chunk:\n chunks.append(current_chunk.strip())\n\n return chunks\n\n\n# load vocoder\ndef load_vocoder(vocoder_name=\"vocos\", is_local=False, local_path=\"\", device=device, hf_cache_dir=None):\n if vocoder_name == \"vocos\":\n # vocoder = Vocos.from_pretrained(\"charactr/vocos-mel-24khz\").to(device)\n if is_local:\n print(f\"Load vocos from local path {local_path}\")\n config_path = f\"{local_path}/config.yaml\"\n model_path = f\"{local_path}/pytorch_model.bin\"\n else:\n print(\"Download Vocos from huggingface charactr/vocos-mel-24khz\")\n repo_id = \"charactr/vocos-mel-24khz\"\n config_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"config.yaml\")\n model_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"pytorch_model.bin\")\n vocoder = Vocos.from_hparams(config_path)\n state_dict = torch.load(model_path, map_location=\"cpu\", weights_only=True)\n from vocos.feature_extractors import EncodecFeatures\n\n if isinstance(vocoder.feature_extractor, EncodecFeatures):\n encodec_parameters = {\n \"feature_extractor.encodec.\" + key: value\n for key, value in vocoder.feature_extractor.encodec.state_dict().items()\n }\n state_dict.update(encodec_parameters)\n vocoder.load_state_dict(state_dict)\n vocoder = vocoder.eval().to(device)\n elif vocoder_name == \"bigvgan\":\n try:\n from third_party.BigVGAN import bigvgan\n except ImportError:\n print(\"You need to follow the README to init submodule and change the BigVGAN source code.\")\n if is_local:\n # download generator from https://huggingface.co/nvidia/bigvgan_v2_24khz_100band_256x/tree/main\n vocoder = bigvgan.BigVGAN.from_pretrained(local_path, use_cuda_kernel=False)\n else:\n vocoder = bigvgan.BigVGAN.from_pretrained(\n \"nvidia/bigvgan_v2_24khz_100band_256x\", use_cuda_kernel=False, cache_dir=hf_cache_dir\n )\n\n vocoder.remove_weight_norm()\n vocoder = vocoder.eval().to(device)\n return vocoder\n\n\n# load asr pipeline\n\nasr_pipe = None\n\n\ndef initialize_asr_pipeline(device: str = device, dtype=None):\n if dtype is None:\n dtype = (\n torch.float16\n if \"cuda\" in device\n and torch.cuda.get_device_properties(device).major >= 7\n and not torch.cuda.get_device_name().endswith(\"[ZLUDA]\")\n else torch.float32\n )\n global asr_pipe\n asr_pipe = pipeline(\n \"automatic-speech-recognition\",\n model=\"openai/whisper-large-v3-turbo\",\n torch_dtype=dtype,\n device=device,\n )\n\n\n# transcribe\n\n\ndef transcribe(ref_audio, language=None):\n global asr_pipe\n if asr_pipe is None:\n initialize_asr_pipeline(device=device)\n return asr_pipe(\n ref_audio,\n chunk_length_s=30,\n batch_size=128,\n generate_kwargs={\"task\": \"transcribe\", \"language\": language} if language else {\"task\": \"transcribe\"},\n return_timestamps=False,\n )[\"text\"].strip()\n\n\n# load model checkpoint for inference\n\n\ndef load_checkpoint(model, ckpt_path, device: str, dtype=None, use_ema=True):\n if dtype is None:\n dtype = (\n torch.float16\n if \"cuda\" in device\n and torch.cuda.get_device_properties(device).major >= 7\n and not torch.cuda.get_device_name().endswith(\"[ZLUDA]\")\n else torch.float32\n )\n model = model.to(dtype)\n\n ckpt_type = ckpt_path.split(\".\")[-1]\n if ckpt_type == \"safetensors\":\n from safetensors.torch import load_file\n\n checkpoint = load_file(ckpt_path, device=device)\n else:\n checkpoint = torch.load(ckpt_path, map_location=device, weights_only=True)\n\n if use_ema:\n if ckpt_type == \"safetensors\":\n checkpoint = {\"ema_model_state_dict\": checkpoint}\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n model.load_state_dict(checkpoint[\"model_state_dict\"])\n else:\n if ckpt_type == \"safetensors\":\n checkpoint = {\"model_state_dict\": checkpoint}\n model.load_state_dict(checkpoint[\"model_state_dict\"])\n\n del checkpoint\n torch.cuda.empty_cache()\n\n return model.to(device)\n\n\n# load model for inference\n\n\ndef load_model(\n model_cls,\n model_cfg,\n ckpt_path,\n mel_spec_type=mel_spec_type,\n vocab_file=\"\",\n ode_method=ode_method,\n use_ema=True,\n device=device,\n):\n if vocab_file == \"\":\n vocab_file = str(files(\"f5_tts\").joinpath(\"infer/examples/vocab.txt\"))\n tokenizer = \"custom\"\n\n print(\"\\nvocab : \", vocab_file)\n print(\"token : \", tokenizer)\n print(\"model : \", ckpt_path, \"\\n\")\n\n vocab_char_map, vocab_size = get_tokenizer(vocab_file, tokenizer)\n model = CFM(\n transformer=model_cls(**model_cfg, text_num_embeds=vocab_size, mel_dim=n_mel_channels),\n mel_spec_kwargs=dict(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n ),\n odeint_kwargs=dict(\n method=ode_method,\n ),\n vocab_char_map=vocab_char_map,\n ).to(device)\n\n dtype = torch.float32 if mel_spec_type == \"bigvgan\" else None\n model = load_checkpoint(model, ckpt_path, device, dtype=dtype, use_ema=use_ema)\n\n return model\n\n\ndef remove_silence_edges(audio, silence_threshold=-42):\n # Remove silence from the start\n non_silent_start_idx = silence.detect_leading_silence(audio, silence_threshold=silence_threshold)\n audio = audio[non_silent_start_idx:]\n\n # Remove silence from the end\n non_silent_end_duration = audio.duration_seconds\n for ms in reversed(audio):\n if ms.dBFS > silence_threshold:\n break\n non_silent_end_duration -= 0.001\n trimmed_audio = audio[: int(non_silent_end_duration * 1000)]\n\n return trimmed_audio\n\n\n# preprocess reference audio and text\n\n\ndef preprocess_ref_audio_text(ref_audio_orig, ref_text, show_info=print):\n show_info(\"Converting audio...\")\n\n # Compute a hash of the reference audio file\n with open(ref_audio_orig, \"rb\") as audio_file:\n audio_data = audio_file.read()\n audio_hash = hashlib.md5(audio_data).hexdigest()\n\n global _ref_audio_cache\n\n if audio_hash in _ref_audio_cache:\n show_info(\"Using cached preprocessed reference audio...\")\n ref_audio = _ref_audio_cache[audio_hash]\n\n else: # first pass, do preprocess\n with tempfile.NamedTemporaryFile(suffix=\".wav\", **tempfile_kwargs) as f:\n temp_path = f.name\n\n aseg = AudioSegment.from_file(ref_audio_orig)\n\n # 1. try to find long silence for clipping\n non_silent_segs = silence.split_on_silence(\n aseg, min_silence_len=1000, silence_thresh=-50, keep_silence=1000, seek_step=10\n )\n non_silent_wave = AudioSegment.silent(duration=0)\n for non_silent_seg in non_silent_segs:\n if len(non_silent_wave) > 6000 and len(non_silent_wave + non_silent_seg) > 12000:\n show_info(\"Audio is over 12s, clipping short. (1)\")\n break\n non_silent_wave += non_silent_seg\n\n # 2. try to find short silence for clipping if 1. failed\n if len(non_silent_wave) > 12000:\n non_silent_segs = silence.split_on_silence(\n aseg, min_silence_len=100, silence_thresh=-40, keep_silence=1000, seek_step=10\n )\n non_silent_wave = AudioSegment.silent(duration=0)\n for non_silent_seg in non_silent_segs:\n if len(non_silent_wave) > 6000 and len(non_silent_wave + non_silent_seg) > 12000:\n show_info(\"Audio is over 12s, clipping short. (2)\")\n break\n non_silent_wave += non_silent_seg\n\n aseg = non_silent_wave\n\n # 3. if no proper silence found for clipping\n if len(aseg) > 12000:\n aseg = aseg[:12000]\n show_info(\"Audio is over 12s, clipping short. (3)\")\n\n aseg = remove_silence_edges(aseg) + AudioSegment.silent(duration=50)\n aseg.export(temp_path, format=\"wav\")\n ref_audio = temp_path\n\n # Cache the processed reference audio\n _ref_audio_cache[audio_hash] = ref_audio\n\n if not ref_text.strip():\n global _ref_text_cache\n if audio_hash in _ref_text_cache:\n # Use cached asr transcription\n show_info(\"Using cached reference text...\")\n ref_text = _ref_text_cache[audio_hash]\n else:\n show_info(\"No reference text provided, transcribing reference audio...\")\n ref_text = transcribe(ref_audio)\n # Cache the transcribed text (not caching custom ref_text, enabling users to do manual tweak)\n _ref_text_cache[audio_hash] = ref_text\n else:\n show_info(\"Using custom reference text...\")\n\n # Ensure ref_text ends with a proper sentence-ending punctuation\n if not ref_text.endswith(\". \") and not ref_text.endswith(\"。\"):\n if ref_text.endswith(\".\"):\n ref_text += \" \"\n else:\n ref_text += \". \"\n\n print(\"\\nref_text \", ref_text)\n\n return ref_audio, ref_text\n\n\n# infer process: chunk text -> infer batches [i.e. infer_batch_process()]\n\n\ndef infer_process(\n ref_audio,\n ref_text,\n gen_text,\n model_obj,\n vocoder,\n mel_spec_type=mel_spec_type,\n show_info=print,\n progress=tqdm,\n target_rms=target_rms,\n cross_fade_duration=cross_fade_duration,\n nfe_step=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n speed=speed,\n fix_duration=fix_duration,\n device=device,\n):\n # Split the input text into batches\n audio, sr = torchaudio.load(ref_audio)\n max_chars = int(len(ref_text.encode(\"utf-8\")) / (audio.shape[-1] / sr) * (22 - audio.shape[-1] / sr) * speed)\n gen_text_batches = chunk_text(gen_text, max_chars=max_chars)\n for i, gen_text in enumerate(gen_text_batches):\n print(f\"gen_text {i}\", gen_text)\n print(\"\\n\")\n\n show_info(f\"Generating audio in {len(gen_text_batches)} batches...\")\n return next(\n infer_batch_process(\n (audio, sr),\n ref_text,\n gen_text_batches,\n model_obj,\n vocoder,\n mel_spec_type=mel_spec_type,\n progress=progress,\n target_rms=target_rms,\n cross_fade_duration=cross_fade_duration,\n nfe_step=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n speed=speed,\n fix_duration=fix_duration,\n device=device,\n )\n )\n\n\n# infer batches\n\n\ndef infer_batch_process(\n ref_audio,\n ref_text,\n gen_text_batches,\n model_obj,\n vocoder,\n mel_spec_type=\"vocos\",\n progress=tqdm,\n target_rms=0.1,\n cross_fade_duration=0.15,\n nfe_step=32,\n cfg_strength=2.0,\n sway_sampling_coef=-1,\n speed=1,\n fix_duration=None,\n device=None,\n streaming=False,\n chunk_size=2048,\n):\n audio, sr = ref_audio\n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n\n rms = torch.sqrt(torch.mean(torch.square(audio)))\n if rms < target_rms:\n audio = audio * target_rms / rms\n if sr != target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, target_sample_rate)\n audio = resampler(audio)\n audio = audio.to(device)\n\n generated_waves = []\n spectrograms = []\n\n if len(ref_text[-1].encode(\"utf-8\")) == 1:\n ref_text = ref_text + \" \"\n\n def process_batch(gen_text):\n local_speed = speed\n if len(gen_text.encode(\"utf-8\")) < 10:\n local_speed = 0.3\n\n # Prepare the text\n text_list = [ref_text + gen_text]\n final_text_list = convert_char_to_pinyin(text_list)\n\n ref_audio_len = audio.shape[-1] // hop_length\n if fix_duration is not None:\n duration = int(fix_duration * target_sample_rate / hop_length)\n else:\n # Calculate duration\n ref_text_len = len(ref_text.encode(\"utf-8\"))\n gen_text_len = len(gen_text.encode(\"utf-8\"))\n duration = ref_audio_len + int(ref_audio_len / ref_text_len * gen_text_len / local_speed)\n\n # inference\n with torch.inference_mode():\n generated, _ = model_obj.sample(\n cond=audio,\n text=final_text_list,\n duration=duration,\n steps=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n )\n del _\n\n generated = generated.to(torch.float32) # generated mel spectrogram\n generated = generated[:, ref_audio_len:, :]\n generated = generated.permute(0, 2, 1)\n if mel_spec_type == \"vocos\":\n generated_wave = vocoder.decode(generated)\n elif mel_spec_type == \"bigvgan\":\n generated_wave = vocoder(generated)\n if rms < target_rms:\n generated_wave = generated_wave * rms / target_rms\n\n # wav -> numpy\n generated_wave = generated_wave.squeeze().cpu().numpy()\n\n if streaming:\n for j in range(0, len(generated_wave), chunk_size):\n yield generated_wave[j : j + chunk_size], target_sample_rate\n else:\n generated_cpu = generated[0].cpu().numpy()\n del generated\n yield generated_wave, generated_cpu\n\n if streaming:\n for gen_text in progress.tqdm(gen_text_batches) if progress is not None else gen_text_batches:\n for chunk in process_batch(gen_text):\n yield chunk\n else:\n with ThreadPoolExecutor() as executor:\n futures = [executor.submit(process_batch, gen_text) for gen_text in gen_text_batches]\n for future in progress.tqdm(futures) if progress is not None else futures:\n result = future.result()\n if result:\n generated_wave, generated_mel_spec = next(result)\n generated_waves.append(generated_wave)\n spectrograms.append(generated_mel_spec)\n\n if generated_waves:\n if cross_fade_duration <= 0:\n # Simply concatenate\n final_wave = np.concatenate(generated_waves)\n else:\n # Combine all generated waves with cross-fading\n final_wave = generated_waves[0]\n for i in range(1, len(generated_waves)):\n prev_wave = final_wave\n next_wave = generated_waves[i]\n\n # Calculate cross-fade samples, ensuring it does not exceed wave lengths\n cross_fade_samples = int(cross_fade_duration * target_sample_rate)\n cross_fade_samples = min(cross_fade_samples, len(prev_wave), len(next_wave))\n\n if cross_fade_samples <= 0:\n # No overlap possible, concatenate\n final_wave = np.concatenate([prev_wave, next_wave])\n continue\n\n # Overlapping parts\n prev_overlap = prev_wave[-cross_fade_samples:]\n next_overlap = next_wave[:cross_fade_samples]\n\n # Fade out and fade in\n fade_out = np.linspace(1, 0, cross_fade_samples)\n fade_in = np.linspace(0, 1, cross_fade_samples)\n\n # Cross-faded overlap\n cross_faded_overlap = prev_overlap * fade_out + next_overlap * fade_in\n\n # Combine\n new_wave = np.concatenate(\n [prev_wave[:-cross_fade_samples], cross_faded_overlap, next_wave[cross_fade_samples:]]\n )\n\n final_wave = new_wave\n\n # Create a combined spectrogram\n combined_spectrogram = np.concatenate(spectrograms, axis=1)\n\n yield final_wave, target_sample_rate, combined_spectrogram\n\n else:\n yield None, target_sample_rate, None\n\n\n# remove silence from generated wav\n\n\ndef remove_silence_for_generated_wav(filename):\n aseg = AudioSegment.from_file(filename)\n non_silent_segs = silence.split_on_silence(\n aseg, min_silence_len=1000, silence_thresh=-50, keep_silence=500, seek_step=10\n )\n non_silent_wave = AudioSegment.silent(duration=0)\n for non_silent_seg in non_silent_segs:\n non_silent_wave += non_silent_seg\n aseg = non_silent_wave\n aseg.export(filename, format=\"wav\")\n\n\n# save spectrogram\n\n\ndef save_spectrogram(spectrogram, path):\n plt.figure(figsize=(12, 4))\n plt.imshow(spectrogram, origin=\"lower\", asp\n# ... truncated ...","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer.chunk_text","uri":"program://DMOSpeech2/function/src.f5_tts.infer.utils_infer.chunk_text#L73-L100","kind":"function","name":"chunk_text","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":73,"end_line":100,"context_start_line":53,"context_end_line":120,"code":"n_mel_channels = 100\nhop_length = 256\nwin_length = 1024\nn_fft = 1024\nmel_spec_type = \"vocos\"\ntarget_rms = 0.1\ncross_fade_duration = 0.15\node_method = \"euler\"\nnfe_step = 32 # 16, 32\ncfg_strength = 2.0\nsway_sampling_coef = -1.0\nspeed = 1.0\nfix_duration = None\n\n# -----------------------------------------\n\n\n# chunk text into smaller pieces\n\n\ndef chunk_text(text, max_chars=135):\n \"\"\"\n Splits the input text into chunks, each with a maximum number of characters.\n\n Args:\n text (str): The text to be split.\n max_chars (int): The maximum number of characters per chunk.\n\n Returns:\n List[str]: A list of text chunks.\n \"\"\"\n chunks = []\n current_chunk = \"\"\n # Split the text into sentences based on punctuation followed by whitespace\n sentences = re.split(r\"(?<=[;:,.!?])\\s+|(?<=[；：，。！？])\", text)\n\n for sentence in sentences:\n if len(current_chunk.encode(\"utf-8\")) + len(sentence.encode(\"utf-8\")) <= max_chars:\n current_chunk += sentence + \" \" if sentence and len(sentence[-1].encode(\"utf-8\")) == 1 else sentence\n else:\n if current_chunk:\n chunks.append(current_chunk.strip())\n current_chunk = sentence + \" \" if sentence and len(sentence[-1].encode(\"utf-8\")) == 1 else sentence\n\n if current_chunk:\n chunks.append(current_chunk.strip())\n\n return chunks\n\n\n# load vocoder\ndef load_vocoder(vocoder_name=\"vocos\", is_local=False, local_path=\"\", device=device, hf_cache_dir=None):\n if vocoder_name == \"vocos\":\n # vocoder = Vocos.from_pretrained(\"charactr/vocos-mel-24khz\").to(device)\n if is_local:\n print(f\"Load vocos from local path {local_path}\")\n config_path = f\"{local_path}/config.yaml\"\n model_path = f\"{local_path}/pytorch_model.bin\"\n else:\n print(\"Download Vocos from huggingface charactr/vocos-mel-24khz\")\n repo_id = \"charactr/vocos-mel-24khz\"\n config_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"config.yaml\")\n model_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"pytorch_model.bin\")\n vocoder = Vocos.from_hparams(config_path)\n state_dict = torch.load(model_path, map_location=\"cpu\", weights_only=True)\n from vocos.feature_extractors import EncodecFeatures\n\n if isinstance(vocoder.feature_extractor, EncodecFeatures):","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer.load_vocoder","uri":"program://DMOSpeech2/function/src.f5_tts.infer.utils_infer.load_vocoder#L104-L143","kind":"function","name":"load_vocoder","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":104,"end_line":143,"context_start_line":84,"context_end_line":163,"code":" chunks = []\n current_chunk = \"\"\n # Split the text into sentences based on punctuation followed by whitespace\n sentences = re.split(r\"(?<=[;:,.!?])\\s+|(?<=[；：，。！？])\", text)\n\n for sentence in sentences:\n if len(current_chunk.encode(\"utf-8\")) + len(sentence.encode(\"utf-8\")) <= max_chars:\n current_chunk += sentence + \" \" if sentence and len(sentence[-1].encode(\"utf-8\")) == 1 else sentence\n else:\n if current_chunk:\n chunks.append(current_chunk.strip())\n current_chunk = sentence + \" \" if sentence and len(sentence[-1].encode(\"utf-8\")) == 1 else sentence\n\n if current_chunk:\n chunks.append(current_chunk.strip())\n\n return chunks\n\n\n# load vocoder\ndef load_vocoder(vocoder_name=\"vocos\", is_local=False, local_path=\"\", device=device, hf_cache_dir=None):\n if vocoder_name == \"vocos\":\n # vocoder = Vocos.from_pretrained(\"charactr/vocos-mel-24khz\").to(device)\n if is_local:\n print(f\"Load vocos from local path {local_path}\")\n config_path = f\"{local_path}/config.yaml\"\n model_path = f\"{local_path}/pytorch_model.bin\"\n else:\n print(\"Download Vocos from huggingface charactr/vocos-mel-24khz\")\n repo_id = \"charactr/vocos-mel-24khz\"\n config_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"config.yaml\")\n model_path = hf_hub_download(repo_id=repo_id, cache_dir=hf_cache_dir, filename=\"pytorch_model.bin\")\n vocoder = Vocos.from_hparams(config_path)\n state_dict = torch.load(model_path, map_location=\"cpu\", weights_only=True)\n from vocos.feature_extractors import EncodecFeatures\n\n if isinstance(vocoder.feature_extractor, EncodecFeatures):\n encodec_parameters = {\n \"feature_extractor.encodec.\" + key: value\n for key, value in vocoder.feature_extractor.encodec.state_dict().items()\n }\n state_dict.update(encodec_parameters)\n vocoder.load_state_dict(state_dict)\n vocoder = vocoder.eval().to(device)\n elif vocoder_name == \"bigvgan\":\n try:\n from third_party.BigVGAN import bigvgan\n except ImportError:\n print(\"You need to follow the README to init submodule and change the BigVGAN source code.\")\n if is_local:\n # download generator from https://huggingface.co/nvidia/bigvgan_v2_24khz_100band_256x/tree/main\n vocoder = bigvgan.BigVGAN.from_pretrained(local_path, use_cuda_kernel=False)\n else:\n vocoder = bigvgan.BigVGAN.from_pretrained(\n \"nvidia/bigvgan_v2_24khz_100band_256x\", use_cuda_kernel=False, cache_dir=hf_cache_dir\n )\n\n vocoder.remove_weight_norm()\n vocoder = vocoder.eval().to(device)\n return vocoder\n\n\n# load asr pipeline\n\nasr_pipe = None\n\n\ndef initialize_asr_pipeline(device: str = device, dtype=None):\n if dtype is None:\n dtype = (\n torch.float16\n if \"cuda\" in device\n and torch.cuda.get_device_properties(device).major >= 7\n and not torch.cuda.get_device_name().endswith(\"[ZLUDA]\")\n else torch.float32\n )\n global asr_pipe\n asr_pipe = pipeline(\n \"automatic-speech-recognition\",\n model=\"openai/whisper-large-v3-turbo\",","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer.initialize_asr_pipeline","uri":"program://DMOSpeech2/function/src.f5_tts.infer.utils_infer.initialize_asr_pipeline#L151-L166","kind":"function","name":"initialize_asr_pipeline","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":151,"end_line":166,"context_start_line":131,"context_end_line":186,"code":" except ImportError:\n print(\"You need to follow the README to init submodule and change the BigVGAN source code.\")\n if is_local:\n # download generator from https://huggingface.co/nvidia/bigvgan_v2_24khz_100band_256x/tree/main\n vocoder = bigvgan.BigVGAN.from_pretrained(local_path, use_cuda_kernel=False)\n else:\n vocoder = bigvgan.BigVGAN.from_pretrained(\n \"nvidia/bigvgan_v2_24khz_100band_256x\", use_cuda_kernel=False, cache_dir=hf_cache_dir\n )\n\n vocoder.remove_weight_norm()\n vocoder = vocoder.eval().to(device)\n return vocoder\n\n\n# load asr pipeline\n\nasr_pipe = None\n\n\ndef initialize_asr_pipeline(device: str = device, dtype=None):\n if dtype is None:\n dtype = (\n torch.float16\n if \"cuda\" in device\n and torch.cuda.get_device_properties(device).major >= 7\n and not torch.cuda.get_device_name().endswith(\"[ZLUDA]\")\n else torch.float32\n )\n global asr_pipe\n asr_pipe = pipeline(\n \"automatic-speech-recognition\",\n model=\"openai/whisper-large-v3-turbo\",\n torch_dtype=dtype,\n device=device,\n )\n\n\n# transcribe\n\n\ndef transcribe(ref_audio, language=None):\n global asr_pipe\n if asr_pipe is None:\n initialize_asr_pipeline(device=device)\n return asr_pipe(\n ref_audio,\n chunk_length_s=30,\n batch_size=128,\n generate_kwargs={\"task\": \"transcribe\", \"language\": language} if language else {\"task\": \"transcribe\"},\n return_timestamps=False,\n )[\"text\"].strip()\n\n\n# load model checkpoint for inference\n","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer.transcribe","uri":"program://DMOSpeech2/function/src.f5_tts.infer.utils_infer.transcribe#L172-L182","kind":"function","name":"transcribe","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":172,"end_line":182,"context_start_line":152,"context_end_line":202,"code":" if dtype is None:\n dtype = (\n torch.float16\n if \"cuda\" in device\n and torch.cuda.get_device_properties(device).major >= 7\n and not torch.cuda.get_device_name().endswith(\"[ZLUDA]\")\n else torch.float32\n )\n global asr_pipe\n asr_pipe = pipeline(\n \"automatic-speech-recognition\",\n model=\"openai/whisper-large-v3-turbo\",\n torch_dtype=dtype,\n device=device,\n )\n\n\n# transcribe\n\n\ndef transcribe(ref_audio, language=None):\n global asr_pipe\n if asr_pipe is None:\n initialize_asr_pipeline(device=device)\n return asr_pipe(\n ref_audio,\n chunk_length_s=30,\n batch_size=128,\n generate_kwargs={\"task\": \"transcribe\", \"language\": language} if language else {\"task\": \"transcribe\"},\n return_timestamps=False,\n )[\"text\"].strip()\n\n\n# load model checkpoint for inference\n\n\ndef load_checkpoint(model, ckpt_path, device: str, dtype=None, use_ema=True):\n if dtype is None:\n dtype = (\n torch.float16\n if \"cuda\" in device\n and torch.cuda.get_device_properties(device).major >= 7\n and not torch.cuda.get_device_name().endswith(\"[ZLUDA]\")\n else torch.float32\n )\n model = model.to(dtype)\n\n ckpt_type = ckpt_path.split(\".\")[-1]\n if ckpt_type == \"safetensors\":\n from safetensors.torch import load_file\n","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer.load_checkpoint","uri":"program://DMOSpeech2/function/src.f5_tts.infer.utils_infer.load_checkpoint#L188-L230","kind":"function","name":"load_checkpoint","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":188,"end_line":230,"context_start_line":168,"context_end_line":250,"code":"\n# transcribe\n\n\ndef transcribe(ref_audio, language=None):\n global asr_pipe\n if asr_pipe is None:\n initialize_asr_pipeline(device=device)\n return asr_pipe(\n ref_audio,\n chunk_length_s=30,\n batch_size=128,\n generate_kwargs={\"task\": \"transcribe\", \"language\": language} if language else {\"task\": \"transcribe\"},\n return_timestamps=False,\n )[\"text\"].strip()\n\n\n# load model checkpoint for inference\n\n\ndef load_checkpoint(model, ckpt_path, device: str, dtype=None, use_ema=True):\n if dtype is None:\n dtype = (\n torch.float16\n if \"cuda\" in device\n and torch.cuda.get_device_properties(device).major >= 7\n and not torch.cuda.get_device_name().endswith(\"[ZLUDA]\")\n else torch.float32\n )\n model = model.to(dtype)\n\n ckpt_type = ckpt_path.split(\".\")[-1]\n if ckpt_type == \"safetensors\":\n from safetensors.torch import load_file\n\n checkpoint = load_file(ckpt_path, device=device)\n else:\n checkpoint = torch.load(ckpt_path, map_location=device, weights_only=True)\n\n if use_ema:\n if ckpt_type == \"safetensors\":\n checkpoint = {\"ema_model_state_dict\": checkpoint}\n checkpoint[\"model_state_dict\"] = {\n k.replace(\"ema_model.\", \"\"): v\n for k, v in checkpoint[\"ema_model_state_dict\"].items()\n if k not in [\"initted\", \"step\"]\n }\n\n # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n model.load_state_dict(checkpoint[\"model_state_dict\"])\n else:\n if ckpt_type == \"safetensors\":\n checkpoint = {\"model_state_dict\": checkpoint}\n model.load_state_dict(checkpoint[\"model_state_dict\"])\n\n del checkpoint\n torch.cuda.empty_cache()\n\n return model.to(device)\n\n\n# load model for inference\n\n\ndef load_model(\n model_cls,\n model_cfg,\n ckpt_path,\n mel_spec_type=mel_spec_type,\n vocab_file=\"\",\n ode_method=ode_method,\n use_ema=True,\n device=device,\n):\n if vocab_file == \"\":\n vocab_file = str(files(\"f5_tts\").joinpath(\"infer/examples/vocab.txt\"))\n tokenizer = \"custom\"\n\n print(\"\\nvocab : \", vocab_file)","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer.load_model","uri":"program://DMOSpeech2/function/src.f5_tts.infer.utils_infer.load_model#L236-L274","kind":"function","name":"load_model","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":236,"end_line":274,"context_start_line":216,"context_end_line":294,"code":" # patch for backward compatibility, 305e3ea\n for key in [\"mel_spec.mel_stft.mel_scale.fb\", \"mel_spec.mel_stft.spectrogram.window\"]:\n if key in checkpoint[\"model_state_dict\"]:\n del checkpoint[\"model_state_dict\"][key]\n\n model.load_state_dict(checkpoint[\"model_state_dict\"])\n else:\n if ckpt_type == \"safetensors\":\n checkpoint = {\"model_state_dict\": checkpoint}\n model.load_state_dict(checkpoint[\"model_state_dict\"])\n\n del checkpoint\n torch.cuda.empty_cache()\n\n return model.to(device)\n\n\n# load model for inference\n\n\ndef load_model(\n model_cls,\n model_cfg,\n ckpt_path,\n mel_spec_type=mel_spec_type,\n vocab_file=\"\",\n ode_method=ode_method,\n use_ema=True,\n device=device,\n):\n if vocab_file == \"\":\n vocab_file = str(files(\"f5_tts\").joinpath(\"infer/examples/vocab.txt\"))\n tokenizer = \"custom\"\n\n print(\"\\nvocab : \", vocab_file)\n print(\"token : \", tokenizer)\n print(\"model : \", ckpt_path, \"\\n\")\n\n vocab_char_map, vocab_size = get_tokenizer(vocab_file, tokenizer)\n model = CFM(\n transformer=model_cls(**model_cfg, text_num_embeds=vocab_size, mel_dim=n_mel_channels),\n mel_spec_kwargs=dict(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n ),\n odeint_kwargs=dict(\n method=ode_method,\n ),\n vocab_char_map=vocab_char_map,\n ).to(device)\n\n dtype = torch.float32 if mel_spec_type == \"bigvgan\" else None\n model = load_checkpoint(model, ckpt_path, device, dtype=dtype, use_ema=use_ema)\n\n return model\n\n\ndef remove_silence_edges(audio, silence_threshold=-42):\n # Remove silence from the start\n non_silent_start_idx = silence.detect_leading_silence(audio, silence_threshold=silence_threshold)\n audio = audio[non_silent_start_idx:]\n\n # Remove silence from the end\n non_silent_end_duration = audio.duration_seconds\n for ms in reversed(audio):\n if ms.dBFS > silence_threshold:\n break\n non_silent_end_duration -= 0.001\n trimmed_audio = audio[: int(non_silent_end_duration * 1000)]\n\n return trimmed_audio\n\n\n# preprocess reference audio and text\n","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer.remove_silence_edges","uri":"program://DMOSpeech2/function/src.f5_tts.infer.utils_infer.remove_silence_edges#L277-L290","kind":"function","name":"remove_silence_edges","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":277,"end_line":290,"context_start_line":257,"context_end_line":310,"code":" mel_spec_kwargs=dict(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n ),\n odeint_kwargs=dict(\n method=ode_method,\n ),\n vocab_char_map=vocab_char_map,\n ).to(device)\n\n dtype = torch.float32 if mel_spec_type == \"bigvgan\" else None\n model = load_checkpoint(model, ckpt_path, device, dtype=dtype, use_ema=use_ema)\n\n return model\n\n\ndef remove_silence_edges(audio, silence_threshold=-42):\n # Remove silence from the start\n non_silent_start_idx = silence.detect_leading_silence(audio, silence_threshold=silence_threshold)\n audio = audio[non_silent_start_idx:]\n\n # Remove silence from the end\n non_silent_end_duration = audio.duration_seconds\n for ms in reversed(audio):\n if ms.dBFS > silence_threshold:\n break\n non_silent_end_duration -= 0.001\n trimmed_audio = audio[: int(non_silent_end_duration * 1000)]\n\n return trimmed_audio\n\n\n# preprocess reference audio and text\n\n\ndef preprocess_ref_audio_text(ref_audio_orig, ref_text, show_info=print):\n show_info(\"Converting audio...\")\n\n # Compute a hash of the reference audio file\n with open(ref_audio_orig, \"rb\") as audio_file:\n audio_data = audio_file.read()\n audio_hash = hashlib.md5(audio_data).hexdigest()\n\n global _ref_audio_cache\n\n if audio_hash in _ref_audio_cache:\n show_info(\"Using cached preprocessed reference audio...\")\n ref_audio = _ref_audio_cache[audio_hash]\n\n else: # first pass, do preprocess","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer.preprocess_ref_audio_text","uri":"program://DMOSpeech2/function/src.f5_tts.infer.utils_infer.preprocess_ref_audio_text#L296-L376","kind":"function","name":"preprocess_ref_audio_text","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":296,"end_line":376,"context_start_line":276,"context_end_line":396,"code":"\ndef remove_silence_edges(audio, silence_threshold=-42):\n # Remove silence from the start\n non_silent_start_idx = silence.detect_leading_silence(audio, silence_threshold=silence_threshold)\n audio = audio[non_silent_start_idx:]\n\n # Remove silence from the end\n non_silent_end_duration = audio.duration_seconds\n for ms in reversed(audio):\n if ms.dBFS > silence_threshold:\n break\n non_silent_end_duration -= 0.001\n trimmed_audio = audio[: int(non_silent_end_duration * 1000)]\n\n return trimmed_audio\n\n\n# preprocess reference audio and text\n\n\ndef preprocess_ref_audio_text(ref_audio_orig, ref_text, show_info=print):\n show_info(\"Converting audio...\")\n\n # Compute a hash of the reference audio file\n with open(ref_audio_orig, \"rb\") as audio_file:\n audio_data = audio_file.read()\n audio_hash = hashlib.md5(audio_data).hexdigest()\n\n global _ref_audio_cache\n\n if audio_hash in _ref_audio_cache:\n show_info(\"Using cached preprocessed reference audio...\")\n ref_audio = _ref_audio_cache[audio_hash]\n\n else: # first pass, do preprocess\n with tempfile.NamedTemporaryFile(suffix=\".wav\", **tempfile_kwargs) as f:\n temp_path = f.name\n\n aseg = AudioSegment.from_file(ref_audio_orig)\n\n # 1. try to find long silence for clipping\n non_silent_segs = silence.split_on_silence(\n aseg, min_silence_len=1000, silence_thresh=-50, keep_silence=1000, seek_step=10\n )\n non_silent_wave = AudioSegment.silent(duration=0)\n for non_silent_seg in non_silent_segs:\n if len(non_silent_wave) > 6000 and len(non_silent_wave + non_silent_seg) > 12000:\n show_info(\"Audio is over 12s, clipping short. (1)\")\n break\n non_silent_wave += non_silent_seg\n\n # 2. try to find short silence for clipping if 1. failed\n if len(non_silent_wave) > 12000:\n non_silent_segs = silence.split_on_silence(\n aseg, min_silence_len=100, silence_thresh=-40, keep_silence=1000, seek_step=10\n )\n non_silent_wave = AudioSegment.silent(duration=0)\n for non_silent_seg in non_silent_segs:\n if len(non_silent_wave) > 6000 and len(non_silent_wave + non_silent_seg) > 12000:\n show_info(\"Audio is over 12s, clipping short. (2)\")\n break\n non_silent_wave += non_silent_seg\n\n aseg = non_silent_wave\n\n # 3. if no proper silence found for clipping\n if len(aseg) > 12000:\n aseg = aseg[:12000]\n show_info(\"Audio is over 12s, clipping short. (3)\")\n\n aseg = remove_silence_edges(aseg) + AudioSegment.silent(duration=50)\n aseg.export(temp_path, format=\"wav\")\n ref_audio = temp_path\n\n # Cache the processed reference audio\n _ref_audio_cache[audio_hash] = ref_audio\n\n if not ref_text.strip():\n global _ref_text_cache\n if audio_hash in _ref_text_cache:\n # Use cached asr transcription\n show_info(\"Using cached reference text...\")\n ref_text = _ref_text_cache[audio_hash]\n else:\n show_info(\"No reference text provided, transcribing reference audio...\")\n ref_text = transcribe(ref_audio)\n # Cache the transcribed text (not caching custom ref_text, enabling users to do manual tweak)\n _ref_text_cache[audio_hash] = ref_text\n else:\n show_info(\"Using custom reference text...\")\n\n # Ensure ref_text ends with a proper sentence-ending punctuation\n if not ref_text.endswith(\". \") and not ref_text.endswith(\"。\"):\n if ref_text.endswith(\".\"):\n ref_text += \" \"\n else:\n ref_text += \". \"\n\n print(\"\\nref_text \", ref_text)\n\n return ref_audio, ref_text\n\n\n# infer process: chunk text -> infer batches [i.e. infer_batch_process()]\n\n\ndef infer_process(\n ref_audio,\n ref_text,\n gen_text,\n model_obj,\n vocoder,\n mel_spec_type=mel_spec_type,\n show_info=print,\n progress=tqdm,\n target_rms=target_rms,\n cross_fade_duration=cross_fade_duration,\n nfe_step=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n speed=speed,","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer.infer_process","uri":"program://DMOSpeech2/function/src.f5_tts.infer.utils_infer.infer_process#L382-L427","kind":"function","name":"infer_process","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":382,"end_line":427,"context_start_line":362,"context_end_line":447,"code":" # Cache the transcribed text (not caching custom ref_text, enabling users to do manual tweak)\n _ref_text_cache[audio_hash] = ref_text\n else:\n show_info(\"Using custom reference text...\")\n\n # Ensure ref_text ends with a proper sentence-ending punctuation\n if not ref_text.endswith(\". \") and not ref_text.endswith(\"。\"):\n if ref_text.endswith(\".\"):\n ref_text += \" \"\n else:\n ref_text += \". \"\n\n print(\"\\nref_text \", ref_text)\n\n return ref_audio, ref_text\n\n\n# infer process: chunk text -> infer batches [i.e. infer_batch_process()]\n\n\ndef infer_process(\n ref_audio,\n ref_text,\n gen_text,\n model_obj,\n vocoder,\n mel_spec_type=mel_spec_type,\n show_info=print,\n progress=tqdm,\n target_rms=target_rms,\n cross_fade_duration=cross_fade_duration,\n nfe_step=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n speed=speed,\n fix_duration=fix_duration,\n device=device,\n):\n # Split the input text into batches\n audio, sr = torchaudio.load(ref_audio)\n max_chars = int(len(ref_text.encode(\"utf-8\")) / (audio.shape[-1] / sr) * (22 - audio.shape[-1] / sr) * speed)\n gen_text_batches = chunk_text(gen_text, max_chars=max_chars)\n for i, gen_text in enumerate(gen_text_batches):\n print(f\"gen_text {i}\", gen_text)\n print(\"\\n\")\n\n show_info(f\"Generating audio in {len(gen_text_batches)} batches...\")\n return next(\n infer_batch_process(\n (audio, sr),\n ref_text,\n gen_text_batches,\n model_obj,\n vocoder,\n mel_spec_type=mel_spec_type,\n progress=progress,\n target_rms=target_rms,\n cross_fade_duration=cross_fade_duration,\n nfe_step=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n speed=speed,\n fix_duration=fix_duration,\n device=device,\n )\n )\n\n\n# infer batches\n\n\ndef infer_batch_process(\n ref_audio,\n ref_text,\n gen_text_batches,\n model_obj,\n vocoder,\n mel_spec_type=\"vocos\",\n progress=tqdm,\n target_rms=0.1,\n cross_fade_duration=0.15,\n nfe_step=32,\n cfg_strength=2.0,\n sway_sampling_coef=-1,\n speed=1,\n fix_duration=None,","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer.infer_batch_process","uri":"program://DMOSpeech2/function/src.f5_tts.infer.utils_infer.infer_batch_process#L433-L579","kind":"function","name":"infer_batch_process","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":433,"end_line":579,"context_start_line":413,"context_end_line":599,"code":" gen_text_batches,\n model_obj,\n vocoder,\n mel_spec_type=mel_spec_type,\n progress=progress,\n target_rms=target_rms,\n cross_fade_duration=cross_fade_duration,\n nfe_step=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n speed=speed,\n fix_duration=fix_duration,\n device=device,\n )\n )\n\n\n# infer batches\n\n\ndef infer_batch_process(\n ref_audio,\n ref_text,\n gen_text_batches,\n model_obj,\n vocoder,\n mel_spec_type=\"vocos\",\n progress=tqdm,\n target_rms=0.1,\n cross_fade_duration=0.15,\n nfe_step=32,\n cfg_strength=2.0,\n sway_sampling_coef=-1,\n speed=1,\n fix_duration=None,\n device=None,\n streaming=False,\n chunk_size=2048,\n):\n audio, sr = ref_audio\n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n\n rms = torch.sqrt(torch.mean(torch.square(audio)))\n if rms < target_rms:\n audio = audio * target_rms / rms\n if sr != target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, target_sample_rate)\n audio = resampler(audio)\n audio = audio.to(device)\n\n generated_waves = []\n spectrograms = []\n\n if len(ref_text[-1].encode(\"utf-8\")) == 1:\n ref_text = ref_text + \" \"\n\n def process_batch(gen_text):\n local_speed = speed\n if len(gen_text.encode(\"utf-8\")) < 10:\n local_speed = 0.3\n\n # Prepare the text\n text_list = [ref_text + gen_text]\n final_text_list = convert_char_to_pinyin(text_list)\n\n ref_audio_len = audio.shape[-1] // hop_length\n if fix_duration is not None:\n duration = int(fix_duration * target_sample_rate / hop_length)\n else:\n # Calculate duration\n ref_text_len = len(ref_text.encode(\"utf-8\"))\n gen_text_len = len(gen_text.encode(\"utf-8\"))\n duration = ref_audio_len + int(ref_audio_len / ref_text_len * gen_text_len / local_speed)\n\n # inference\n with torch.inference_mode():\n generated, _ = model_obj.sample(\n cond=audio,\n text=final_text_list,\n duration=duration,\n steps=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n )\n del _\n\n generated = generated.to(torch.float32) # generated mel spectrogram\n generated = generated[:, ref_audio_len:, :]\n generated = generated.permute(0, 2, 1)\n if mel_spec_type == \"vocos\":\n generated_wave = vocoder.decode(generated)\n elif mel_spec_type == \"bigvgan\":\n generated_wave = vocoder(generated)\n if rms < target_rms:\n generated_wave = generated_wave * rms / target_rms\n\n # wav -> numpy\n generated_wave = generated_wave.squeeze().cpu().numpy()\n\n if streaming:\n for j in range(0, len(generated_wave), chunk_size):\n yield generated_wave[j : j + chunk_size], target_sample_rate\n else:\n generated_cpu = generated[0].cpu().numpy()\n del generated\n yield generated_wave, generated_cpu\n\n if streaming:\n for gen_text in progress.tqdm(gen_text_batches) if progress is not None else gen_text_batches:\n for chunk in process_batch(gen_text):\n yield chunk\n else:\n with ThreadPoolExecutor() as executor:\n futures = [executor.submit(process_batch, gen_text) for gen_text in gen_text_batches]\n for future in progress.tqdm(futures) if progress is not None else futures:\n result = future.result()\n if result:\n generated_wave, generated_mel_spec = next(result)\n generated_waves.append(generated_wave)\n spectrograms.append(generated_mel_spec)\n\n if generated_waves:\n if cross_fade_duration <= 0:\n # Simply concatenate\n final_wave = np.concatenate(generated_waves)\n else:\n # Combine all generated waves with cross-fading\n final_wave = generated_waves[0]\n for i in range(1, len(generated_waves)):\n prev_wave = final_wave\n next_wave = generated_waves[i]\n\n # Calculate cross-fade samples, ensuring it does not exceed wave lengths\n cross_fade_samples = int(cross_fade_duration * target_sample_rate)\n cross_fade_samples = min(cross_fade_samples, len(prev_wave), len(next_wave))\n\n if cross_fade_samples <= 0:\n # No overlap possible, concatenate\n final_wave = np.concatenate([prev_wave, next_wave])\n continue\n\n # Overlapping parts\n prev_overlap = prev_wave[-cross_fade_samples:]\n next_overlap = next_wave[:cross_fade_samples]\n\n # Fade out and fade in\n fade_out = np.linspace(1, 0, cross_fade_samples)\n fade_in = np.linspace(0, 1, cross_fade_samples)\n\n # Cross-faded overlap\n cross_faded_overlap = prev_overlap * fade_out + next_overlap * fade_in\n\n # Combine\n new_wave = np.concatenate(\n [prev_wave[:-cross_fade_samples], cross_faded_overlap, next_wave[cross_fade_samples:]]\n )\n\n final_wave = new_wave\n\n # Create a combined spectrogram\n combined_spectrogram = np.concatenate(spectrograms, axis=1)\n\n yield final_wave, target_sample_rate, combined_spectrogram\n\n else:\n yield None, target_sample_rate, None\n\n\n# remove silence from generated wav\n\n\ndef remove_silence_for_generated_wav(filename):\n aseg = AudioSegment.from_file(filename)\n non_silent_segs = silence.split_on_silence(\n aseg, min_silence_len=1000, silence_thresh=-50, keep_silence=500, seek_step=10\n )\n non_silent_wave = AudioSegment.silent(duration=0)\n for non_silent_seg in non_silent_segs:\n non_silent_wave += non_silent_seg\n aseg = non_silent_wave\n aseg.export(filename, format=\"wav\")\n\n\n# save spectrogram\n\n","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer.remove_silence_for_generated_wav","uri":"program://DMOSpeech2/function/src.f5_tts.infer.utils_infer.remove_silence_for_generated_wav#L585-L594","kind":"function","name":"remove_silence_for_generated_wav","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":585,"end_line":594,"context_start_line":565,"context_end_line":605,"code":"\n # Combine\n new_wave = np.concatenate(\n [prev_wave[:-cross_fade_samples], cross_faded_overlap, next_wave[cross_fade_samples:]]\n )\n\n final_wave = new_wave\n\n # Create a combined spectrogram\n combined_spectrogram = np.concatenate(spectrograms, axis=1)\n\n yield final_wave, target_sample_rate, combined_spectrogram\n\n else:\n yield None, target_sample_rate, None\n\n\n# remove silence from generated wav\n\n\ndef remove_silence_for_generated_wav(filename):\n aseg = AudioSegment.from_file(filename)\n non_silent_segs = silence.split_on_silence(\n aseg, min_silence_len=1000, silence_thresh=-50, keep_silence=500, seek_step=10\n )\n non_silent_wave = AudioSegment.silent(duration=0)\n for non_silent_seg in non_silent_segs:\n non_silent_wave += non_silent_seg\n aseg = non_silent_wave\n aseg.export(filename, format=\"wav\")\n\n\n# save spectrogram\n\n\ndef save_spectrogram(spectrogram, path):\n plt.figure(figsize=(12, 4))\n plt.imshow(spectrogram, origin=\"lower\", aspect=\"auto\")\n plt.colorbar()\n plt.savefig(path)\n plt.close()","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer.save_spectrogram","uri":"program://DMOSpeech2/function/src.f5_tts.infer.utils_infer.save_spectrogram#L600-L605","kind":"function","name":"save_spectrogram","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":600,"end_line":605,"context_start_line":580,"context_end_line":605,"code":"\n\n# remove silence from generated wav\n\n\ndef remove_silence_for_generated_wav(filename):\n aseg = AudioSegment.from_file(filename)\n non_silent_segs = silence.split_on_silence(\n aseg, min_silence_len=1000, silence_thresh=-50, keep_silence=500, seek_step=10\n )\n non_silent_wave = AudioSegment.silent(duration=0)\n for non_silent_seg in non_silent_segs:\n non_silent_wave += non_silent_seg\n aseg = non_silent_wave\n aseg.export(filename, format=\"wav\")\n\n\n# save spectrogram\n\n\ndef save_spectrogram(spectrogram, path):\n plt.figure(figsize=(12, 4))\n plt.imshow(spectrogram, origin=\"lower\", aspect=\"auto\")\n plt.colorbar()\n plt.savefig(path)\n plt.close()","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.utils_infer.process_batch","uri":"program://DMOSpeech2/function/src.f5_tts.infer.utils_infer.process_batch#L470-L519","kind":"function","name":"process_batch","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":470,"end_line":519,"context_start_line":450,"context_end_line":539,"code":" chunk_size=2048,\n):\n audio, sr = ref_audio\n if audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\n\n rms = torch.sqrt(torch.mean(torch.square(audio)))\n if rms < target_rms:\n audio = audio * target_rms / rms\n if sr != target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, target_sample_rate)\n audio = resampler(audio)\n audio = audio.to(device)\n\n generated_waves = []\n spectrograms = []\n\n if len(ref_text[-1].encode(\"utf-8\")) == 1:\n ref_text = ref_text + \" \"\n\n def process_batch(gen_text):\n local_speed = speed\n if len(gen_text.encode(\"utf-8\")) < 10:\n local_speed = 0.3\n\n # Prepare the text\n text_list = [ref_text + gen_text]\n final_text_list = convert_char_to_pinyin(text_list)\n\n ref_audio_len = audio.shape[-1] // hop_length\n if fix_duration is not None:\n duration = int(fix_duration * target_sample_rate / hop_length)\n else:\n # Calculate duration\n ref_text_len = len(ref_text.encode(\"utf-8\"))\n gen_text_len = len(gen_text.encode(\"utf-8\"))\n duration = ref_audio_len + int(ref_audio_len / ref_text_len * gen_text_len / local_speed)\n\n # inference\n with torch.inference_mode():\n generated, _ = model_obj.sample(\n cond=audio,\n text=final_text_list,\n duration=duration,\n steps=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n )\n del _\n\n generated = generated.to(torch.float32) # generated mel spectrogram\n generated = generated[:, ref_audio_len:, :]\n generated = generated.permute(0, 2, 1)\n if mel_spec_type == \"vocos\":\n generated_wave = vocoder.decode(generated)\n elif mel_spec_type == \"bigvgan\":\n generated_wave = vocoder(generated)\n if rms < target_rms:\n generated_wave = generated_wave * rms / target_rms\n\n # wav -> numpy\n generated_wave = generated_wave.squeeze().cpu().numpy()\n\n if streaming:\n for j in range(0, len(generated_wave), chunk_size):\n yield generated_wave[j : j + chunk_size], target_sample_rate\n else:\n generated_cpu = generated[0].cpu().numpy()\n del generated\n yield generated_wave, generated_cpu\n\n if streaming:\n for gen_text in progress.tqdm(gen_text_batches) if progress is not None else gen_text_batches:\n for chunk in process_batch(gen_text):\n yield chunk\n else:\n with ThreadPoolExecutor() as executor:\n futures = [executor.submit(process_batch, gen_text) for gen_text in gen_text_batches]\n for future in progress.tqdm(futures) if progress is not None else futures:\n result = future.result()\n if result:\n generated_wave, generated_mel_spec = next(result)\n generated_waves.append(generated_wave)\n spectrograms.append(generated_mel_spec)\n\n if generated_waves:\n if cross_fade_duration <= 0:\n # Simply concatenate\n final_wave = np.concatenate(generated_waves)\n else:","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.speech_edit","uri":"program://DMOSpeech2/module/src.f5_tts.infer.speech_edit#L1-L205","kind":"module","name":"src.f5_tts.infer.speech_edit","path":"src/f5_tts/infer/speech_edit.py","language":"python","start_line":1,"end_line":205,"context_start_line":1,"context_end_line":205,"code":"import os\n\n\nos.environ[\"PYTORCH_ENABLE_MPS_FALLBACK\"] = \"1\" # for MPS device compatibility\n\nfrom importlib.resources import files\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom cached_path import cached_path\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\n\nfrom f5_tts.infer.utils_infer import load_checkpoint, load_vocoder, save_spectrogram\nfrom f5_tts.model import CFM\nfrom f5_tts.model.utils import convert_char_to_pinyin, get_tokenizer\n\n\ndevice = (\n \"cuda\"\n if torch.cuda.is_available()\n else \"xpu\"\n if torch.xpu.is_available()\n else \"mps\"\n if torch.backends.mps.is_available()\n else \"cpu\"\n)\n\n\n# ---------------------- infer setting ---------------------- #\n\nseed = None # int | None\n\nexp_name = \"F5TTS_v1_Base\" # F5TTS_v1_Base | E2TTS_Base\nckpt_step = 1250000\n\nnfe_step = 32 # 16, 32\ncfg_strength = 2.0\node_method = \"euler\" # euler | midpoint\nsway_sampling_coef = -1.0\nspeed = 1.0\ntarget_rms = 0.1\n\n\nmodel_cfg = OmegaConf.load(str(files(\"f5_tts\").joinpath(f\"configs/{exp_name}.yaml\")))\nmodel_cls = get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\nmodel_arc = model_cfg.model.arch\n\ndataset_name = model_cfg.datasets.name\ntokenizer = model_cfg.model.tokenizer\n\nmel_spec_type = model_cfg.model.mel_spec.mel_spec_type\ntarget_sample_rate = model_cfg.model.mel_spec.target_sample_rate\nn_mel_channels = model_cfg.model.mel_spec.n_mel_channels\nhop_length = model_cfg.model.mel_spec.hop_length\nwin_length = model_cfg.model.mel_spec.win_length\nn_fft = model_cfg.model.mel_spec.n_fft\n\n\n# ckpt_path = str(files(\"f5_tts\").joinpath(\"../../\")) + f\"/ckpts/{exp_name}/model_{ckpt_step}.safetensors\"\nckpt_path = str(cached_path(f\"hf://SWivid/F5-TTS/{exp_name}/model_{ckpt_step}.safetensors\"))\noutput_dir = \"tests\"\n\n\n# [leverage https://github.com/MahmoudAshraf97/ctc-forced-aligner to get char level alignment]\n# pip install git+https://github.com/MahmoudAshraf97/ctc-forced-aligner.git\n# [write the origin_text into a file, e.g. tests/test_edit.txt]\n# ctc-forced-aligner --audio_path \"src/f5_tts/infer/examples/basic/basic_ref_en.wav\" --text_path \"tests/test_edit.txt\" --language \"zho\" --romanize --split_size \"char\"\n# [result will be saved at same path of audio file]\n# [--language \"zho\" for Chinese, \"eng\" for English]\n# [if local ckpt, set --alignment_model \"../checkpoints/mms-300m-1130-forced-aligner\"]\n\naudio_to_edit = str(files(\"f5_tts\").joinpath(\"infer/examples/basic/basic_ref_en.wav\"))\norigin_text = \"Some call me nature, others call me mother nature.\"\ntarget_text = \"Some call me optimist, others call me realist.\"\nparts_to_edit = [\n [1.42, 2.44],\n [4.04, 4.9],\n] # stard_ends of \"nature\" & \"mother nature\", in seconds\nfix_duration = [\n 1.2,\n 1,\n] # fix duration for \"optimist\" & \"realist\", in seconds\n\n# audio_to_edit = \"src/f5_tts/infer/examples/basic/basic_ref_zh.wav\"\n# origin_text = \"对，这就是我，万人敬仰的太乙真人。\"\n# target_text = \"对，那就是你，万人敬仰的太白金星。\"\n# parts_to_edit = [[0.84, 1.4], [1.92, 2.4], [4.26, 6.26], ]\n# fix_duration = None # use origin text duration\n\n\n# -------------------------------------------------#\n\nuse_ema = True\n\nif not os.path.exists(output_dir):\n os.makedirs(output_dir)\n\n# Vocoder model\nlocal = False\nif mel_spec_type == \"vocos\":\n vocoder_local_path = \"../checkpoints/charactr/vocos-mel-24khz\"\nelif mel_spec_type == \"bigvgan\":\n vocoder_local_path = \"../checkpoints/bigvgan_v2_24khz_100band_256x\"\nvocoder = load_vocoder(vocoder_name=mel_spec_type, is_local=local, local_path=vocoder_local_path)\n\n# Tokenizer\nvocab_char_map, vocab_size = get_tokenizer(dataset_name, tokenizer)\n\n# Model\nmodel = CFM(\n transformer=model_cls(**model_arc, text_num_embeds=vocab_size, mel_dim=n_mel_channels),\n mel_spec_kwargs=dict(\n n_fft=n_fft,\n hop_length=hop_length,\n win_length=win_length,\n n_mel_channels=n_mel_channels,\n target_sample_rate=target_sample_rate,\n mel_spec_type=mel_spec_type,\n ),\n odeint_kwargs=dict(\n method=ode_method,\n ),\n vocab_char_map=vocab_char_map,\n).to(device)\n\ndtype = torch.float32 if mel_spec_type == \"bigvgan\" else None\nmodel = load_checkpoint(model, ckpt_path, device, dtype=dtype, use_ema=use_ema)\n\n# Audio\naudio, sr = torchaudio.load(audio_to_edit)\nif audio.shape[0] > 1:\n audio = torch.mean(audio, dim=0, keepdim=True)\nrms = torch.sqrt(torch.mean(torch.square(audio)))\nif rms < target_rms:\n audio = audio * target_rms / rms\nif sr != target_sample_rate:\n resampler = torchaudio.transforms.Resample(sr, target_sample_rate)\n audio = resampler(audio)\noffset = 0\naudio_ = torch.zeros(1, 0)\nedit_mask = torch.zeros(1, 0, dtype=torch.bool)\nfor part in parts_to_edit:\n start, end = part\n part_dur = end - start if fix_duration is None else fix_duration.pop(0)\n part_dur = part_dur * target_sample_rate\n start = start * target_sample_rate\n audio_ = torch.cat((audio_, audio[:, round(offset) : round(start)], torch.zeros(1, round(part_dur))), dim=-1)\n edit_mask = torch.cat(\n (\n edit_mask,\n torch.ones(1, round((start - offset) / hop_length), dtype=torch.bool),\n torch.zeros(1, round(part_dur / hop_length), dtype=torch.bool),\n ),\n dim=-1,\n )\n offset = end * target_sample_rate\naudio = torch.cat((audio_, audio[:, round(offset) :]), dim=-1)\nedit_mask = F.pad(edit_mask, (0, audio.shape[-1] // hop_length - edit_mask.shape[-1] + 1), value=True)\naudio = audio.to(device)\nedit_mask = edit_mask.to(device)\n\n# Text\ntext_list = [target_text]\nif tokenizer == \"pinyin\":\n final_text_list = convert_char_to_pinyin(text_list)\nelse:\n final_text_list = [text_list]\nprint(f\"text : {text_list}\")\nprint(f\"pinyin: {final_text_list}\")\n\n# Duration\nref_audio_len = 0\nduration = audio.shape[-1] // hop_length\n\n# Inference\nwith torch.inference_mode():\n generated, trajectory = model.sample(\n cond=audio,\n text=final_text_list,\n duration=duration,\n steps=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n seed=seed,\n edit_mask=edit_mask,\n )\n print(f\"Generated mel: {generated.shape}\")\n\n # Final result\n generated = generated.to(torch.float32)\n generated = generated[:, ref_audio_len:, :]\n gen_mel_spec = generated.permute(0, 2, 1)\n if mel_spec_type == \"vocos\":\n generated_wave = vocoder.decode(gen_mel_spec).cpu()\n elif mel_spec_type == \"bigvgan\":\n generated_wave = vocoder(gen_mel_spec).squeeze(0).cpu()\n\n if rms < target_rms:\n generated_wave = generated_wave * rms / target_rms\n\n save_spectrogram(gen_mel_spec[0].cpu().numpy(), f\"{output_dir}/speech_edit_out.png\")\n torchaudio.save(f\"{output_dir}/speech_edit_out.wav\", generated_wave, target_sample_rate)\n print(f\"Generated wav: {generated_wave.shape}\")","source_hash":"b91b2f03f44460701eea899837e71da92b4d92a6bf44f4e56e5f1eacd86e6540","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_cli","uri":"program://DMOSpeech2/module/src.f5_tts.infer.infer_cli#L1-L383","kind":"module","name":"src.f5_tts.infer.infer_cli","path":"src/f5_tts/infer/infer_cli.py","language":"python","start_line":1,"end_line":383,"context_start_line":1,"context_end_line":383,"code":"import argparse\nimport codecs\nimport os\nimport re\nfrom datetime import datetime\nfrom importlib.resources import files\nfrom pathlib import Path\n\nimport numpy as np\nimport soundfile as sf\nimport tomli\nfrom cached_path import cached_path\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\nfrom unidecode import unidecode\n\nfrom f5_tts.infer.utils_infer import (\n cfg_strength,\n cross_fade_duration,\n device,\n fix_duration,\n infer_process,\n load_model,\n load_vocoder,\n mel_spec_type,\n nfe_step,\n preprocess_ref_audio_text,\n remove_silence_for_generated_wav,\n speed,\n sway_sampling_coef,\n target_rms,\n)\n\n\nparser = argparse.ArgumentParser(\n prog=\"python3 infer-cli.py\",\n description=\"Commandline interface for E2/F5 TTS with Advanced Batch Processing.\",\n epilog=\"Specify options above to override one or more settings from config.\",\n)\nparser.add_argument(\n \"-c\",\n \"--config\",\n type=str,\n default=os.path.join(files(\"f5_tts\").joinpath(\"infer/examples/basic\"), \"basic.toml\"),\n help=\"The configuration file, default see infer/examples/basic/basic.toml\",\n)\n\n\n# Note. Not to provide default value here in order to read default from config file\n\nparser.add_argument(\n \"-m\",\n \"--model\",\n type=str,\n help=\"The model name: F5TTS_v1_Base | F5TTS_Base | E2TTS_Base | etc.\",\n)\nparser.add_argument(\n \"-mc\",\n \"--model_cfg\",\n type=str,\n help=\"The path to F5-TTS model config file .yaml\",\n)\nparser.add_argument(\n \"-p\",\n \"--ckpt_file\",\n type=str,\n help=\"The path to model checkpoint .pt, leave blank to use default\",\n)\nparser.add_argument(\n \"-v\",\n \"--vocab_file\",\n type=str,\n help=\"The path to vocab file .txt, leave blank to use default\",\n)\nparser.add_argument(\n \"-r\",\n \"--ref_audio\",\n type=str,\n help=\"The reference audio file.\",\n)\nparser.add_argument(\n \"-s\",\n \"--ref_text\",\n type=str,\n help=\"The transcript/subtitle for the reference audio\",\n)\nparser.add_argument(\n \"-t\",\n \"--gen_text\",\n type=str,\n help=\"The text to make model synthesize a speech\",\n)\nparser.add_argument(\n \"-f\",\n \"--gen_file\",\n type=str,\n help=\"The file with text to generate, will ignore --gen_text\",\n)\nparser.add_argument(\n \"-o\",\n \"--output_dir\",\n type=str,\n help=\"The path to output folder\",\n)\nparser.add_argument(\n \"-w\",\n \"--output_file\",\n type=str,\n help=\"The name of output file\",\n)\nparser.add_argument(\n \"--save_chunk\",\n action=\"store_true\",\n help=\"To save each audio chunks during inference\",\n)\nparser.add_argument(\n \"--no_legacy_text\",\n action=\"store_false\",\n help=\"Not to use lossy ASCII transliterations of unicode text in saved file names.\",\n)\nparser.add_argument(\n \"--remove_silence\",\n action=\"store_true\",\n help=\"To remove long silence found in ouput\",\n)\nparser.add_argument(\n \"--load_vocoder_from_local\",\n action=\"store_true\",\n help=\"To load vocoder from local dir, default to ../checkpoints/vocos-mel-24khz\",\n)\nparser.add_argument(\n \"--vocoder_name\",\n type=str,\n choices=[\"vocos\", \"bigvgan\"],\n help=f\"Used vocoder name: vocos | bigvgan, default {mel_spec_type}\",\n)\nparser.add_argument(\n \"--target_rms\",\n type=float,\n help=f\"Target output speech loudness normalization value, default {target_rms}\",\n)\nparser.add_argument(\n \"--cross_fade_duration\",\n type=float,\n help=f\"Duration of cross-fade between audio segments in seconds, default {cross_fade_duration}\",\n)\nparser.add_argument(\n \"--nfe_step\",\n type=int,\n help=f\"The number of function evaluation (denoising steps), default {nfe_step}\",\n)\nparser.add_argument(\n \"--cfg_strength\",\n type=float,\n help=f\"Classifier-free guidance strength, default {cfg_strength}\",\n)\nparser.add_argument(\n \"--sway_sampling_coef\",\n type=float,\n help=f\"Sway Sampling coefficient, default {sway_sampling_coef}\",\n)\nparser.add_argument(\n \"--speed\",\n type=float,\n help=f\"The speed of the generated audio, default {speed}\",\n)\nparser.add_argument(\n \"--fix_duration\",\n type=float,\n help=f\"Fix the total duration (ref and gen audios) in seconds, default {fix_duration}\",\n)\nparser.add_argument(\n \"--device\",\n type=str,\n help=\"Specify the device to run on\",\n)\nargs = parser.parse_args()\n\n\n# config file\n\nconfig = tomli.load(open(args.config, \"rb\"))\n\n\n# command-line interface parameters\n\nmodel = args.model or config.get(\"model\", \"F5TTS_v1_Base\")\nckpt_file = args.ckpt_file or config.get(\"ckpt_file\", \"\")\nvocab_file = args.vocab_file or config.get(\"vocab_file\", \"\")\n\nref_audio = args.ref_audio or config.get(\"ref_audio\", \"infer/examples/basic/basic_ref_en.wav\")\nref_text = (\n args.ref_text\n if args.ref_text is not None\n else config.get(\"ref_text\", \"Some call me nature, others call me mother nature.\")\n)\ngen_text = args.gen_text or config.get(\"gen_text\", \"Here we generate something just for test.\")\ngen_file = args.gen_file or config.get(\"gen_file\", \"\")\n\noutput_dir = args.output_dir or config.get(\"output_dir\", \"tests\")\noutput_file = args.output_file or config.get(\n \"output_file\", f\"infer_cli_{datetime.now().strftime(r'%Y%m%d_%H%M%S')}.wav\"\n)\n\nsave_chunk = args.save_chunk or config.get(\"save_chunk\", False)\nuse_legacy_text = args.no_legacy_text or config.get(\"no_legacy_text\", False) # no_legacy_text is a store_false arg\nif save_chunk and use_legacy_text:\n print(\n \"\\nWarning to --save_chunk: lossy ASCII transliterations of unicode text for legacy (.wav) file names, --no_legacy_text to disable.\\n\"\n )\n\nremove_silence = args.remove_silence or config.get(\"remove_silence\", False)\nload_vocoder_from_local = args.load_vocoder_from_local or config.get(\"load_vocoder_from_local\", False)\n\nvocoder_name = args.vocoder_name or config.get(\"vocoder_name\", mel_spec_type)\ntarget_rms = args.target_rms or config.get(\"target_rms\", target_rms)\ncross_fade_duration = args.cross_fade_duration or config.get(\"cross_fade_duration\", cross_fade_duration)\nnfe_step = args.nfe_step or config.get(\"nfe_step\", nfe_step)\ncfg_strength = args.cfg_strength or config.get(\"cfg_strength\", cfg_strength)\nsway_sampling_coef = args.sway_sampling_coef or config.get(\"sway_sampling_coef\", sway_sampling_coef)\nspeed = args.speed or config.get(\"speed\", speed)\nfix_duration = args.fix_duration or config.get(\"fix_duration\", fix_duration)\ndevice = args.device or config.get(\"device\", device)\n\n\n# patches for pip pkg user\nif \"infer/examples/\" in ref_audio:\n ref_audio = str(files(\"f5_tts\").joinpath(f\"{ref_audio}\"))\nif \"infer/examples/\" in gen_file:\n gen_file = str(files(\"f5_tts\").joinpath(f\"{gen_file}\"))\nif \"voices\" in config:\n for voice in config[\"voices\"]:\n voice_ref_audio = config[\"voices\"][voice][\"ref_audio\"]\n if \"infer/examples/\" in voice_ref_audio:\n config[\"voices\"][voice][\"ref_audio\"] = str(files(\"f5_tts\").joinpath(f\"{voice_ref_audio}\"))\n\n\n# ignore gen_text if gen_file provided\n\nif gen_file:\n gen_text = codecs.open(gen_file, \"r\", \"utf-8\").read()\n\n\n# output path\n\nwave_path = Path(output_dir) / output_file\n# spectrogram_path = Path(output_dir) / \"infer_cli_out.png\"\nif save_chunk:\n output_chunk_dir = os.path.join(output_dir, f\"{Path(output_file).stem}_chunks\")\n if not os.path.exists(output_chunk_dir):\n os.makedirs(output_chunk_dir)\n\n\n# load vocoder\n\nif vocoder_name == \"vocos\":\n vocoder_local_path = \"../checkpoints/vocos-mel-24khz\"\nelif vocoder_name == \"bigvgan\":\n vocoder_local_path = \"../checkpoints/bigvgan_v2_24khz_100band_256x\"\n\nvocoder = load_vocoder(\n vocoder_name=vocoder_name, is_local=load_vocoder_from_local, local_path=vocoder_local_path, device=device\n)\n\n\n# load TTS model\n\nmodel_cfg = OmegaConf.load(\n args.model_cfg or config.get(\"model_cfg\", str(files(\"f5_tts\").joinpath(f\"configs/{model}.yaml\")))\n)\nmodel_cls = get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\nmodel_arc = model_cfg.model.arch\n\nrepo_name, ckpt_step, ckpt_type = \"F5-TTS\", 1250000, \"safetensors\"\n\nif model != \"F5TTS_Base\":\n assert vocoder_name == model_cfg.model.mel_spec.mel_spec_type\n\n# override for previous models\nif model == \"F5TTS_Base\":\n if vocoder_name == \"vocos\":\n ckpt_step = 1200000\n elif vocoder_name == \"bigvgan\":\n model = \"F5TTS_Base_bigvgan\"\n ckpt_type = \"pt\"\nelif model == \"E2TTS_Base\":\n repo_name = \"E2-TTS\"\n ckpt_step = 1200000\n\nif not ckpt_file:\n ckpt_file = str(cached_path(f\"hf://SWivid/{repo_name}/{model}/model_{ckpt_step}.{ckpt_type}\"))\n\nprint(f\"Using {model}...\")\nema_model = load_model(\n model_cls, model_arc, ckpt_file, mel_spec_type=vocoder_name, vocab_file=vocab_file, device=device\n)\n\n\n# inference process\n\n\ndef main():\n main_voice = {\"ref_audio\": ref_audio, \"ref_text\": ref_text}\n if \"voices\" not in config:\n voices = {\"main\": main_voice}\n else:\n voices = config[\"voices\"]\n voices[\"main\"] = main_voice\n for voice in voices:\n print(\"Voice:\", voice)\n print(\"ref_audio \", voices[voice][\"ref_audio\"])\n voices[voice][\"ref_audio\"], voices[voice][\"ref_text\"] = preprocess_ref_audio_text(\n voices[voice][\"ref_audio\"], voices[voice][\"ref_text\"]\n )\n print(\"ref_audio_\", voices[voice][\"ref_audio\"], \"\\n\\n\")\n\n generated_audio_segments = []\n reg1 = r\"(?=\\[\\w+\\])\"\n chunks = re.split(reg1, gen_text)\n reg2 = r\"\\[(\\w+)\\]\"\n for text in chunks:\n if not text.strip():\n continue\n match = re.match(reg2, text)\n if match:\n voice = match[1]\n else:\n print(\"No voice tag found, using main.\")\n voice = \"main\"\n if voice not in voices:\n print(f\"Voice {voice} not found, using main.\")\n voice = \"main\"\n text = re.sub(reg2, \"\", text)\n ref_audio_ = voices[voice][\"ref_audio\"]\n ref_text_ = voices[voice][\"ref_text\"]\n local_speed = voices[voice].get(\"speed\", speed)\n gen_text_ = text.strip()\n print(f\"Voice: {voice}\")\n audio_segment, final_sample_rate, spectrogram = infer_process(\n ref_audio_,\n ref_text_,\n gen_text_,\n ema_model,\n vocoder,\n mel_spec_type=vocoder_name,\n target_rms=target_rms,\n cross_fade_duration=cross_fade_duration,\n nfe_step=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n speed=local_speed,\n fix_duration=fix_duration,\n device=device,\n )\n generated_audio_segments.append(audio_segment)\n\n if save_chunk:\n if len(gen_text_) > 200:\n gen_text_ = gen_text_[:200] + \" ... \"\n if use_legacy_text:\n gen_text_ = unidecode(gen_text_)\n sf.write(\n os.path.join(output_chunk_dir, f\"{len(generated_audio_segments) - 1}_{gen_text_}.wav\"),\n audio_segment,\n final_sample_rate,\n )\n\n if generated_audio_segments:\n final_wave = np.concatenate(generated_audio_segments)\n\n if not os.path.exists(output_dir):\n os.makedirs(output_dir)\n\n with open(wave_path, \"wb\") as f:\n sf.write(f.name, final_wave, final_sample_rate)\n # Remove silence\n if remove_silence:\n remove_silence_for_generated_wav(f.name)\n print(f.name)\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"5d9b324c18a7021852710940ce4e98cece7c097b2a96d66a9417339fa7dda25b","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_cli.main","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_cli.main#L302-L379","kind":"function","name":"main","path":"src/f5_tts/infer/infer_cli.py","language":"python","start_line":302,"end_line":379,"context_start_line":282,"context_end_line":383,"code":" ckpt_step = 1200000\n elif vocoder_name == \"bigvgan\":\n model = \"F5TTS_Base_bigvgan\"\n ckpt_type = \"pt\"\nelif model == \"E2TTS_Base\":\n repo_name = \"E2-TTS\"\n ckpt_step = 1200000\n\nif not ckpt_file:\n ckpt_file = str(cached_path(f\"hf://SWivid/{repo_name}/{model}/model_{ckpt_step}.{ckpt_type}\"))\n\nprint(f\"Using {model}...\")\nema_model = load_model(\n model_cls, model_arc, ckpt_file, mel_spec_type=vocoder_name, vocab_file=vocab_file, device=device\n)\n\n\n# inference process\n\n\ndef main():\n main_voice = {\"ref_audio\": ref_audio, \"ref_text\": ref_text}\n if \"voices\" not in config:\n voices = {\"main\": main_voice}\n else:\n voices = config[\"voices\"]\n voices[\"main\"] = main_voice\n for voice in voices:\n print(\"Voice:\", voice)\n print(\"ref_audio \", voices[voice][\"ref_audio\"])\n voices[voice][\"ref_audio\"], voices[voice][\"ref_text\"] = preprocess_ref_audio_text(\n voices[voice][\"ref_audio\"], voices[voice][\"ref_text\"]\n )\n print(\"ref_audio_\", voices[voice][\"ref_audio\"], \"\\n\\n\")\n\n generated_audio_segments = []\n reg1 = r\"(?=\\[\\w+\\])\"\n chunks = re.split(reg1, gen_text)\n reg2 = r\"\\[(\\w+)\\]\"\n for text in chunks:\n if not text.strip():\n continue\n match = re.match(reg2, text)\n if match:\n voice = match[1]\n else:\n print(\"No voice tag found, using main.\")\n voice = \"main\"\n if voice not in voices:\n print(f\"Voice {voice} not found, using main.\")\n voice = \"main\"\n text = re.sub(reg2, \"\", text)\n ref_audio_ = voices[voice][\"ref_audio\"]\n ref_text_ = voices[voice][\"ref_text\"]\n local_speed = voices[voice].get(\"speed\", speed)\n gen_text_ = text.strip()\n print(f\"Voice: {voice}\")\n audio_segment, final_sample_rate, spectrogram = infer_process(\n ref_audio_,\n ref_text_,\n gen_text_,\n ema_model,\n vocoder,\n mel_spec_type=vocoder_name,\n target_rms=target_rms,\n cross_fade_duration=cross_fade_duration,\n nfe_step=nfe_step,\n cfg_strength=cfg_strength,\n sway_sampling_coef=sway_sampling_coef,\n speed=local_speed,\n fix_duration=fix_duration,\n device=device,\n )\n generated_audio_segments.append(audio_segment)\n\n if save_chunk:\n if len(gen_text_) > 200:\n gen_text_ = gen_text_[:200] + \" ... \"\n if use_legacy_text:\n gen_text_ = unidecode(gen_text_)\n sf.write(\n os.path.join(output_chunk_dir, f\"{len(generated_audio_segments) - 1}_{gen_text_}.wav\"),\n audio_segment,\n final_sample_rate,\n )\n\n if generated_audio_segments:\n final_wave = np.concatenate(generated_audio_segments)\n\n if not os.path.exists(output_dir):\n os.makedirs(output_dir)\n\n with open(wave_path, \"wb\") as f:\n sf.write(f.name, final_wave, final_sample_rate)\n # Remove silence\n if remove_silence:\n remove_silence_for_generated_wav(f.name)\n print(f.name)\n\n\nif __name__ == \"__main__\":\n main()","source_hash":"5d9b324c18a7021852710940ce4e98cece7c097b2a96d66a9417339fa7dda25b","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio","uri":"program://DMOSpeech2/module/src.f5_tts.infer.infer_gradio#L1-L1121","kind":"module","name":"src.f5_tts.infer.infer_gradio","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":1,"end_line":1121,"context_start_line":1,"context_end_line":1121,"code":"# ruff: noqa: E402\n# Above allows ruff to ignore E402: module level import not at top of file\n\nimport gc\nimport json\nimport os\nimport re\nimport tempfile\nfrom collections import OrderedDict\nfrom functools import lru_cache\nfrom importlib.resources import files\n\nimport click\nimport gradio as gr\nimport numpy as np\nimport soundfile as sf\nimport torch\nimport torchaudio\nfrom cached_path import cached_path\nfrom transformers import AutoModelForCausalLM, AutoTokenizer\n\n\ntry:\n import spaces\n\n USING_SPACES = True\nexcept ImportError:\n USING_SPACES = False\n\n\ndef gpu_decorator(func):\n if USING_SPACES:\n return spaces.GPU(func)\n else:\n return func\n\n\nfrom f5_tts.infer.utils_infer import (\n infer_process,\n load_model,\n load_vocoder,\n preprocess_ref_audio_text,\n remove_silence_for_generated_wav,\n save_spectrogram,\n tempfile_kwargs,\n)\nfrom f5_tts.model import DiT, UNetT\n\n\nDEFAULT_TTS_MODEL = \"F5-TTS_v1\"\ntts_model_choice = DEFAULT_TTS_MODEL\n\nDEFAULT_TTS_MODEL_CFG = [\n \"hf://SWivid/F5-TTS/F5TTS_v1_Base/model_1250000.safetensors\",\n \"hf://SWivid/F5-TTS/F5TTS_v1_Base/vocab.txt\",\n json.dumps(dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)),\n]\n\n\n# load models\n\nvocoder = load_vocoder()\n\n\ndef load_f5tts():\n ckpt_path = str(cached_path(DEFAULT_TTS_MODEL_CFG[0]))\n F5TTS_model_cfg = json.loads(DEFAULT_TTS_MODEL_CFG[2])\n return load_model(DiT, F5TTS_model_cfg, ckpt_path)\n\n\ndef load_e2tts():\n ckpt_path = str(cached_path(\"hf://SWivid/E2-TTS/E2TTS_Base/model_1200000.safetensors\"))\n E2TTS_model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4, text_mask_padding=False, pe_attn_head=1)\n return load_model(UNetT, E2TTS_model_cfg, ckpt_path)\n\n\ndef load_custom(ckpt_path: str, vocab_path=\"\", model_cfg=None):\n ckpt_path, vocab_path = ckpt_path.strip(), vocab_path.strip()\n if ckpt_path.startswith(\"hf://\"):\n ckpt_path = str(cached_path(ckpt_path))\n if vocab_path.startswith(\"hf://\"):\n vocab_path = str(cached_path(vocab_path))\n if model_cfg is None:\n model_cfg = json.loads(DEFAULT_TTS_MODEL_CFG[2])\n elif isinstance(model_cfg, str):\n model_cfg = json.loads(model_cfg)\n return load_model(DiT, model_cfg, ckpt_path, vocab_file=vocab_path)\n\n\nF5TTS_ema_model = load_f5tts()\nE2TTS_ema_model = load_e2tts() if USING_SPACES else None\ncustom_ema_model, pre_custom_path = None, \"\"\n\nchat_model_state = None\nchat_tokenizer_state = None\n\n\n@gpu_decorator\ndef chat_model_inference(messages, model, tokenizer):\n \"\"\"Generate response using Qwen\"\"\"\n text = tokenizer.apply_chat_template(\n messages,\n tokenize=False,\n add_generation_prompt=True,\n )\n\n model_inputs = tokenizer([text], return_tensors=\"pt\").to(model.device)\n generated_ids = model.generate(\n **model_inputs,\n max_new_tokens=512,\n temperature=0.7,\n top_p=0.95,\n )\n\n generated_ids = [\n output_ids[len(input_ids) :] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)\n ]\n return tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]\n\n\n@gpu_decorator\ndef load_text_from_file(file):\n if file:\n with open(file, \"r\", encoding=\"utf-8\") as f:\n text = f.read().strip()\n else:\n text = \"\"\n return gr.update(value=text)\n\n\n@lru_cache(maxsize=1000) # NOTE. need to ensure params of infer() hashable\n@gpu_decorator\ndef infer(\n ref_audio_orig,\n ref_text,\n gen_text,\n model,\n remove_silence,\n seed,\n cross_fade_duration=0.15,\n nfe_step=32,\n speed=1,\n show_info=gr.Info,\n):\n if not ref_audio_orig:\n gr.Warning(\"Please provide reference audio.\")\n return gr.update(), gr.update(), ref_text\n\n # Set inference seed\n if seed < 0 or seed > 2**31 - 1:\n gr.Warning(\"Seed must in range 0 ~ 2147483647. Using random seed instead.\")\n seed = np.random.randint(0, 2**31 - 1)\n torch.manual_seed(seed)\n used_seed = seed\n\n if not gen_text.strip():\n gr.Warning(\"Please enter text to generate or upload a text file.\")\n return gr.update(), gr.update(), ref_text\n\n ref_audio, ref_text = preprocess_ref_audio_text(ref_audio_orig, ref_text, show_info=show_info)\n\n if model == DEFAULT_TTS_MODEL:\n ema_model = F5TTS_ema_model\n elif model == \"E2-TTS\":\n global E2TTS_ema_model\n if E2TTS_ema_model is None:\n show_info(\"Loading E2-TTS model...\")\n E2TTS_ema_model = load_e2tts()\n ema_model = E2TTS_ema_model\n elif isinstance(model, tuple) and model[0] == \"Custom\":\n assert not USING_SPACES, \"Only official checkpoints allowed in Spaces.\"\n global custom_ema_model, pre_custom_path\n if pre_custom_path != model[1]:\n show_info(\"Loading Custom TTS model...\")\n custom_ema_model = load_custom(model[1], vocab_path=model[2], model_cfg=model[3])\n pre_custom_path = model[1]\n ema_model = custom_ema_model\n\n final_wave, final_sample_rate, combined_spectrogram = infer_process(\n ref_audio,\n ref_text,\n gen_text,\n ema_model,\n vocoder,\n cross_fade_duration=cross_fade_duration,\n nfe_step=nfe_step,\n speed=speed,\n show_info=show_info,\n progress=gr.Progress(),\n )\n\n # Remove silence\n if remove_silence:\n with tempfile.NamedTemporaryFile(suffix=\".wav\", **tempfile_kwargs) as f:\n temp_path = f.name\n try:\n sf.write(temp_path, final_wave, final_sample_rate)\n remove_silence_for_generated_wav(f.name)\n final_wave, _ = torchaudio.load(f.name)\n finally:\n os.unlink(temp_path)\n final_wave = final_wave.squeeze().cpu().numpy()\n\n # Save the spectrogram\n with tempfile.NamedTemporaryFile(suffix=\".png\", **tempfile_kwargs) as tmp_spectrogram:\n spectrogram_path = tmp_spectrogram.name\n save_spectrogram(combined_spectrogram, spectrogram_path)\n\n return (final_sample_rate, final_wave), spectrogram_path, ref_text, used_seed\n\n\nwith gr.Blocks() as app_tts:\n gr.Markdown(\"# Batched TTS\")\n ref_audio_input = gr.Audio(label=\"Reference Audio\", type=\"filepath\")\n with gr.Row():\n gen_text_input = gr.Textbox(\n label=\"Text to Generate\",\n lines=10,\n max_lines=40,\n scale=4,\n )\n gen_text_file = gr.File(label=\"Load Text to Generate from File (.txt)\", file_types=[\".txt\"], scale=1)\n generate_btn = gr.Button(\"Synthesize\", variant=\"primary\")\n with gr.Accordion(\"Advanced Settings\", open=False):\n with gr.Row():\n ref_text_input = gr.Textbox(\n label=\"Reference Text\",\n info=\"Leave blank to automatically transcribe the reference audio. If you enter text or upload a file, it will override automatic transcription.\",\n lines=2,\n scale=4,\n )\n ref_text_file = gr.File(label=\"Load Reference Text from File (.txt)\", file_types=[\".txt\"], scale=1)\n with gr.Row():\n randomize_seed = gr.Checkbox(\n label=\"Randomize Seed\",\n info=\"Check to use a random seed for each generation. Uncheck to use the seed specified.\",\n value=True,\n scale=3,\n )\n seed_input = gr.Number(show_label=False, value=0, precision=0, scale=1)\n with gr.Column(scale=4):\n remove_silence = gr.Checkbox(\n label=\"Remove Silences\",\n info=\"If undesired long silence(s) produced, turn on to automatically detect and crop.\",\n value=False,\n )\n speed_slider = gr.Slider(\n label=\"Speed\",\n minimum=0.3,\n maximum=2.0,\n value=1.0,\n step=0.1,\n info=\"Adjust the speed of the audio.\",\n )\n nfe_slider = gr.Slider(\n label=\"NFE Steps\",\n minimum=4,\n maximum=64,\n value=32,\n step=2,\n info=\"Set the number of denoising steps.\",\n )\n cross_fade_duration_slider = gr.Slider(\n label=\"Cross-Fade Duration (s)\",\n minimum=0.0,\n maximum=1.0,\n value=0.15,\n step=0.01,\n info=\"Set the duration of the cross-fade between audio clips.\",\n )\n\n audio_output = gr.Audio(label=\"Synthesized Audio\")\n spectrogram_output = gr.Image(label=\"Spectrogram\")\n\n @gpu_decorator\n def basic_tts(\n ref_audio_input,\n ref_text_input,\n gen_text_input,\n remove_silence,\n randomize_seed,\n seed_input,\n cross_fade_duration_slider,\n nfe_slider,\n speed_slider,\n ):\n if randomize_seed:\n seed_input = np.random.randint(0, 2**31 - 1)\n\n audio_out, spectrogram_path, ref_text_out, used_seed = infer(\n ref_audio_input,\n ref_text_input,\n gen_text_input,\n tts_model_choice,\n remove_silence,\n seed=seed_input,\n cross_fade_duration=cross_fade_duration_slider,\n nfe_step=nfe_slider,\n speed=speed_slider,\n )\n return audio_out, spectrogram_path, ref_text_out, used_seed\n\n gen_text_file.upload(\n load_text_from_file,\n inputs=[gen_text_file],\n outputs=[gen_text_input],\n )\n\n ref_text_file.upload(\n load_text_from_file,\n inputs=[ref_text_file],\n outputs=[ref_text_input],\n )\n\n ref_audio_input.clear(\n lambda: [None, None],\n None,\n [ref_text_input, ref_text_file],\n )\n\n generate_btn.click(\n basic_tts,\n inputs=[\n ref_audio_input,\n ref_text_input,\n gen_text_input,\n remove_silence,\n randomize_seed,\n seed_input,\n cross_fade_duration_slider,\n nfe_slider,\n speed_slider,\n ],\n outputs=[audio_output, spectrogram_output, ref_text_input, seed_input],\n )\n\n\ndef parse_speechtypes_text(gen_text):\n # Pattern to find {str} or {\"name\": str, \"seed\": int, \"speed\": float}\n pattern = r\"(\\{.*?\\})\"\n\n # Split the text by the pattern\n tokens = re.split(pattern, gen_text)\n\n segments = []\n\n current_type_dict = {\n \"name\": \"Regular\",\n \"seed\": -1,\n \"speed\": 1.0,\n }\n\n for i in range(len(tokens)):\n if i % 2 == 0:\n # This is text\n text = tokens[i].strip()\n if text:\n current_type_dict[\"text\"] = text\n segments.append(current_type_dict)\n else:\n # This is type\n type_str = tokens[i].strip()\n try: # if type dict\n current_type_dict = json.loads(type_str)\n except json.decoder.JSONDecodeError:\n type_str = type_str[1:-1] # remove brace {}\n current_type_dict = {\"name\": type_str, \"seed\": -1, \"speed\": 1.0}\n\n return segments\n\n\nwith gr.Blocks() as app_multistyle:\n # New section for multistyle generation\n gr.Markdown(\n \"\"\"\n # Multiple Speech-Type Generation\n\n This section allows you to generate multiple speech types or multiple people's voices. Enter your text in the format shown below, or upload a .txt file with the same format. The system will generate speech using the appropriate type. If unspecified, the model will use the regular speech type. The current speech type will be used until the next speech type is specified.\n \"\"\"\n )\n\n with gr.Row():\n gr.Markdown(\n \"\"\"\n **Example Input:**
\n {Regular} Hello, I'd like to order a sandwich please.
\n {Surprised} What do you mean you're out of bread?
\n {Sad} I really wanted a sandwich though...
\n {Angry} You know what, darn you and your little shop!
\n {Whisper} I'll just go back home and cry now.
\n {Shouting} Why me?!\n \"\"\"\n )\n\n gr.Markdown(\n \"\"\"\n **Example Input 2:**
\n {\"name\": \"Speaker1_Happy\", \"seed\": -1, \"speed\": 1} Hello, I'd like to order a sandwich please.
\n {\"name\": \"Speaker2_Regular\", \"seed\": -1, \"speed\": 1} Sorry, we're out of bread.
\n {\"name\": \"Speaker1_Sad\", \"seed\": -1, \"speed\": 1} I really wanted a sandwich though...
\n {\"name\": \"Speaker2_Whisper\", \"seed\": -1, \"speed\": 1} I'll give you the last one I was hiding.\n \"\"\"\n )\n\n gr.Markdown(\n 'Upload different audio clips for each speech type. The first speech type is mandatory. You can add additional speech types by clicking the \"Add Speech Type\" button.'\n )\n\n # Regular speech type (mandatory)\n with gr.Row(variant=\"compact\") as regular_row:\n with gr.Column(scale=1, min_width=160):\n regular_name = gr.Textbox(value=\"Regular\", label=\"Speech Type Name\")\n regular_insert = gr.Button(\"Insert Label\", variant=\"secondary\")\n with gr.Column(scale=3):\n regular_audio = gr.Audio(label=\"Regular Reference Audio\", type=\"filepath\")\n with gr.Column(scale=3):\n regular_ref_text = gr.Textbox(label=\"Reference Text (Regular)\", lines=4)\n with gr.Row():\n regular_seed_slider = gr.Slider(\n show_label=False, minimum=-1, maximum=999, value=-1, step=1, info=\"Seed, -1 for random\"\n )\n regular_speed_slider = gr.Slider(\n show_label=False, minimum=0.3, maximum=2.0, value=1.0, step=0.1, info=\"Adjust the speed\"\n )\n with gr.Column(scale=1, min_width=160):\n regular_ref_text_file = gr.File(label=\"Load Reference Text from File (.txt)\", file_types=[\".txt\"])\n\n # Regular speech type (max 100)\n max_speech_types = 100\n speech_type_rows = [regular_row]\n speech_type_names = [regular_name]\n speech_type_audios = [regular_audio]\n speech_type_ref_texts = [regular_ref_text]\n speech_type_ref_text_files = [regular_ref_text_file]\n speech_type_seeds = [regular_seed_slider]\n speech_type_speeds = [regular_speed_slider]\n speech_type_delete_btns = [None]\n speech_type_insert_btns = [regular_insert]\n\n # Additional speech types (99 more)\n for i in range(max_speech_types - 1):\n with gr.Row(variant=\"compact\", visible=False) as row:\n with gr.Column(scale=1, min_width=160):\n name_input = gr.Textbox(label=\"Speech Type Name\")\n insert_btn = gr.Button(\"Insert Label\", variant=\"secondary\")\n delete_btn = gr.Button(\"Delete Type\", variant=\"stop\")\n with gr.Column(scale=3):\n audio_input = gr.Audio(label=\"Reference Audio\", type=\"filepath\")\n with gr.Column(scale=3):\n ref_text_input = gr.Textbox(label=\"Reference Text\", lines=4)\n with gr.Row():\n seed_input = gr.Slider(\n show_label=False, minimum=-1, maximum=999, value=-1, step=1, info=\"Seed. -1 for random\"\n )\n speed_input = gr.Slider(\n show_label=False, minimum=0.3, maximum=2.0, value=1.0, step=0.1, info=\"Adjust the speed\"\n )\n with gr.Column(scale=1, min_width=160):\n ref_text_file_input = gr.File(label=\"Load Reference Text from File (.txt)\", file_types=[\".txt\"])\n speech_type_rows.append(row)\n speech_type_names.append(name_input)\n speech_type_audios.append(audio_input)\n speech_type_ref_texts.append(ref_text_input)\n speech_type_ref_text_files.append(ref_text_file_input)\n speech_type_seeds.append(seed_input)\n speech_type_speeds.append(speed_input)\n speech_type_delete_btns.append(delete_btn)\n speech_type_insert_btns.append(insert_btn)\n\n # Global logic for all speech types\n for i in range(max_speech_types):\n speech_type_audios[i].clear(\n lambda: [None, None],\n None,\n [speech_type_ref_texts[i], speech_type_ref_text_files[i]],\n )\n speech_type_ref_text_files[i].upload(\n load_text_from_file,\n inputs=[speech_type_ref_text_files[i]],\n outputs=[speech_type_ref_texts[i]],\n )\n\n # Button to add speech type\n add_speech_type_btn = gr.Button(\"Add Speech Type\")\n\n # Keep track of autoincrement of speech types, no roll back\n speech_type_count = 1\n\n # Function to add a speech type\n def add_speech_type_fn():\n row_updates = [gr.update() for _ in range(max_speech_types)]\n global speech_type_count\n if speech_type_count < max_speech_types:\n row_updates[speech_type_count] = gr.update(visible=True)\n speech_type_count += 1\n else:\n gr.Warning(\"Exhausted maximum number of speech types. Consider restart the app.\")\n return row_updates\n\n add_speech_type_btn.click(add_speech_type_fn, outputs=speech_type_rows)\n\n # Function to delete a speech type\n def delete_speech_type_fn():\n return gr.update(visible=False), None, None, None, None\n\n # Update delete button clicks and ref text file changes\n for i in range(1, len(speech_type_delete_btns)):\n speech_type_delete_btns[i].click(\n delete_speech_type_fn,\n outputs=[\n speech_type_rows[i],\n speech_type_names[i],\n speech_type_audios[i],\n speech_type_ref_texts[i],\n speech_type_ref_text_files[i],\n ],\n )\n\n # Text input for the prompt\n with gr.Row():\n gen_text_input_multistyle = gr.Textbox(\n label=\"Text to Generate\",\n lines=10,\n max_lines=40,\n scale=4,\n placeholder=\"Enter the script with speaker names (or emotion types) at the start of each block, e.g.:\\n\\n{Regular} Hello, I'd like to order a sandwich please.\\n{Surprised} What do you mean you're out of bread?\\n{Sad} I really wanted a sandwich though...\\n{Angry} You know what, darn you and your little shop!\\n{Whisper} I'll just go back home and cry now.\\n{Shouting} Why me?!\",\n )\n gen_text_file_multistyle = gr.File(label=\"Load Text to Generate from File (.txt)\", file_types=[\".txt\"], scale=1)\n\n def make_insert_speech_type_fn(index):\n def insert_speech_type_fn(current_text, speech_type_name, speech_type_seed, speech_type_speed):\n current_text = current_text or \"\"\n if not speech_type_name:\n gr.Warning(\"Please enter speech type name before insert.\")\n return current_text\n speech_type_dict = {\n \"name\": speech_type_name,\n \"seed\": speech_type_seed,\n \"speed\": speech_type_speed,\n }\n updated_text = current_text + json.dumps(speech_type_dict) + \" \"\n return updated_text\n\n return insert_speech_type_fn\n\n for i, insert_btn in enumerate(speech_type_insert_btns):\n insert_fn = make_insert_speech_type_fn(i)\n insert_btn.click(\n insert_fn,\n inputs=[gen_text_input_multistyle, speech_type_names[i], speech_type_seeds[i], speech_type_speeds[i]],\n outputs=gen_text_input_multistyle,\n )\n\n with gr.Accordion(\"Advanced Settings\", open=True):\n with gr.Row():\n with gr.Column():\n show_cherrypick_multistyle = gr.Checkbox(\n label=\"Show Cherry-pick Interface\",\n info=\"Turn on to show interface, picking seeds from previous generations.\",\n value=False,\n )\n with gr.Column():\n remove_silence_multistyle = gr.Checkbox(\n label=\"Remove Silences\",\n info=\"Turn on to automatically d\n# ... truncated ...","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":true} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.gpu_decorator","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.gpu_decorator#L31-L35","kind":"function","name":"gpu_decorator","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":31,"end_line":35,"context_start_line":11,"context_end_line":55,"code":"from importlib.resources import files\n\nimport click\nimport gradio as gr\nimport numpy as np\nimport soundfile as sf\nimport torch\nimport torchaudio\nfrom cached_path import cached_path\nfrom transformers import AutoModelForCausalLM, AutoTokenizer\n\n\ntry:\n import spaces\n\n USING_SPACES = True\nexcept ImportError:\n USING_SPACES = False\n\n\ndef gpu_decorator(func):\n if USING_SPACES:\n return spaces.GPU(func)\n else:\n return func\n\n\nfrom f5_tts.infer.utils_infer import (\n infer_process,\n load_model,\n load_vocoder,\n preprocess_ref_audio_text,\n remove_silence_for_generated_wav,\n save_spectrogram,\n tempfile_kwargs,\n)\nfrom f5_tts.model import DiT, UNetT\n\n\nDEFAULT_TTS_MODEL = \"F5-TTS_v1\"\ntts_model_choice = DEFAULT_TTS_MODEL\n\nDEFAULT_TTS_MODEL_CFG = [\n \"hf://SWivid/F5-TTS/F5TTS_v1_Base/model_1250000.safetensors\",\n \"hf://SWivid/F5-TTS/F5TTS_v1_Base/vocab.txt\",","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.load_f5tts","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.load_f5tts#L65-L68","kind":"function","name":"load_f5tts","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":65,"end_line":68,"context_start_line":45,"context_end_line":88,"code":" tempfile_kwargs,\n)\nfrom f5_tts.model import DiT, UNetT\n\n\nDEFAULT_TTS_MODEL = \"F5-TTS_v1\"\ntts_model_choice = DEFAULT_TTS_MODEL\n\nDEFAULT_TTS_MODEL_CFG = [\n \"hf://SWivid/F5-TTS/F5TTS_v1_Base/model_1250000.safetensors\",\n \"hf://SWivid/F5-TTS/F5TTS_v1_Base/vocab.txt\",\n json.dumps(dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)),\n]\n\n\n# load models\n\nvocoder = load_vocoder()\n\n\ndef load_f5tts():\n ckpt_path = str(cached_path(DEFAULT_TTS_MODEL_CFG[0]))\n F5TTS_model_cfg = json.loads(DEFAULT_TTS_MODEL_CFG[2])\n return load_model(DiT, F5TTS_model_cfg, ckpt_path)\n\n\ndef load_e2tts():\n ckpt_path = str(cached_path(\"hf://SWivid/E2-TTS/E2TTS_Base/model_1200000.safetensors\"))\n E2TTS_model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4, text_mask_padding=False, pe_attn_head=1)\n return load_model(UNetT, E2TTS_model_cfg, ckpt_path)\n\n\ndef load_custom(ckpt_path: str, vocab_path=\"\", model_cfg=None):\n ckpt_path, vocab_path = ckpt_path.strip(), vocab_path.strip()\n if ckpt_path.startswith(\"hf://\"):\n ckpt_path = str(cached_path(ckpt_path))\n if vocab_path.startswith(\"hf://\"):\n vocab_path = str(cached_path(vocab_path))\n if model_cfg is None:\n model_cfg = json.loads(DEFAULT_TTS_MODEL_CFG[2])\n elif isinstance(model_cfg, str):\n model_cfg = json.loads(model_cfg)\n return load_model(DiT, model_cfg, ckpt_path, vocab_file=vocab_path)\n","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.load_e2tts","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.load_e2tts#L71-L74","kind":"function","name":"load_e2tts","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":71,"end_line":74,"context_start_line":51,"context_end_line":94,"code":"tts_model_choice = DEFAULT_TTS_MODEL\n\nDEFAULT_TTS_MODEL_CFG = [\n \"hf://SWivid/F5-TTS/F5TTS_v1_Base/model_1250000.safetensors\",\n \"hf://SWivid/F5-TTS/F5TTS_v1_Base/vocab.txt\",\n json.dumps(dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)),\n]\n\n\n# load models\n\nvocoder = load_vocoder()\n\n\ndef load_f5tts():\n ckpt_path = str(cached_path(DEFAULT_TTS_MODEL_CFG[0]))\n F5TTS_model_cfg = json.loads(DEFAULT_TTS_MODEL_CFG[2])\n return load_model(DiT, F5TTS_model_cfg, ckpt_path)\n\n\ndef load_e2tts():\n ckpt_path = str(cached_path(\"hf://SWivid/E2-TTS/E2TTS_Base/model_1200000.safetensors\"))\n E2TTS_model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4, text_mask_padding=False, pe_attn_head=1)\n return load_model(UNetT, E2TTS_model_cfg, ckpt_path)\n\n\ndef load_custom(ckpt_path: str, vocab_path=\"\", model_cfg=None):\n ckpt_path, vocab_path = ckpt_path.strip(), vocab_path.strip()\n if ckpt_path.startswith(\"hf://\"):\n ckpt_path = str(cached_path(ckpt_path))\n if vocab_path.startswith(\"hf://\"):\n vocab_path = str(cached_path(vocab_path))\n if model_cfg is None:\n model_cfg = json.loads(DEFAULT_TTS_MODEL_CFG[2])\n elif isinstance(model_cfg, str):\n model_cfg = json.loads(model_cfg)\n return load_model(DiT, model_cfg, ckpt_path, vocab_file=vocab_path)\n\n\nF5TTS_ema_model = load_f5tts()\nE2TTS_ema_model = load_e2tts() if USING_SPACES else None\ncustom_ema_model, pre_custom_path = None, \"\"\n\nchat_model_state = None","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.load_custom","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.load_custom#L77-L87","kind":"function","name":"load_custom","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":77,"end_line":87,"context_start_line":57,"context_end_line":107,"code":"]\n\n\n# load models\n\nvocoder = load_vocoder()\n\n\ndef load_f5tts():\n ckpt_path = str(cached_path(DEFAULT_TTS_MODEL_CFG[0]))\n F5TTS_model_cfg = json.loads(DEFAULT_TTS_MODEL_CFG[2])\n return load_model(DiT, F5TTS_model_cfg, ckpt_path)\n\n\ndef load_e2tts():\n ckpt_path = str(cached_path(\"hf://SWivid/E2-TTS/E2TTS_Base/model_1200000.safetensors\"))\n E2TTS_model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4, text_mask_padding=False, pe_attn_head=1)\n return load_model(UNetT, E2TTS_model_cfg, ckpt_path)\n\n\ndef load_custom(ckpt_path: str, vocab_path=\"\", model_cfg=None):\n ckpt_path, vocab_path = ckpt_path.strip(), vocab_path.strip()\n if ckpt_path.startswith(\"hf://\"):\n ckpt_path = str(cached_path(ckpt_path))\n if vocab_path.startswith(\"hf://\"):\n vocab_path = str(cached_path(vocab_path))\n if model_cfg is None:\n model_cfg = json.loads(DEFAULT_TTS_MODEL_CFG[2])\n elif isinstance(model_cfg, str):\n model_cfg = json.loads(model_cfg)\n return load_model(DiT, model_cfg, ckpt_path, vocab_file=vocab_path)\n\n\nF5TTS_ema_model = load_f5tts()\nE2TTS_ema_model = load_e2tts() if USING_SPACES else None\ncustom_ema_model, pre_custom_path = None, \"\"\n\nchat_model_state = None\nchat_tokenizer_state = None\n\n\n@gpu_decorator\ndef chat_model_inference(messages, model, tokenizer):\n \"\"\"Generate response using Qwen\"\"\"\n text = tokenizer.apply_chat_template(\n messages,\n tokenize=False,\n add_generation_prompt=True,\n )\n\n model_inputs = tokenizer([text], return_tensors=\"pt\").to(model.device)","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.chat_model_inference","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.chat_model_inference#L99-L118","kind":"function","name":"chat_model_inference","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":99,"end_line":118,"context_start_line":79,"context_end_line":138,"code":" if ckpt_path.startswith(\"hf://\"):\n ckpt_path = str(cached_path(ckpt_path))\n if vocab_path.startswith(\"hf://\"):\n vocab_path = str(cached_path(vocab_path))\n if model_cfg is None:\n model_cfg = json.loads(DEFAULT_TTS_MODEL_CFG[2])\n elif isinstance(model_cfg, str):\n model_cfg = json.loads(model_cfg)\n return load_model(DiT, model_cfg, ckpt_path, vocab_file=vocab_path)\n\n\nF5TTS_ema_model = load_f5tts()\nE2TTS_ema_model = load_e2tts() if USING_SPACES else None\ncustom_ema_model, pre_custom_path = None, \"\"\n\nchat_model_state = None\nchat_tokenizer_state = None\n\n\n@gpu_decorator\ndef chat_model_inference(messages, model, tokenizer):\n \"\"\"Generate response using Qwen\"\"\"\n text = tokenizer.apply_chat_template(\n messages,\n tokenize=False,\n add_generation_prompt=True,\n )\n\n model_inputs = tokenizer([text], return_tensors=\"pt\").to(model.device)\n generated_ids = model.generate(\n **model_inputs,\n max_new_tokens=512,\n temperature=0.7,\n top_p=0.95,\n )\n\n generated_ids = [\n output_ids[len(input_ids) :] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)\n ]\n return tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]\n\n\n@gpu_decorator\ndef load_text_from_file(file):\n if file:\n with open(file, \"r\", encoding=\"utf-8\") as f:\n text = f.read().strip()\n else:\n text = \"\"\n return gr.update(value=text)\n\n\n@lru_cache(maxsize=1000) # NOTE. need to ensure params of infer() hashable\n@gpu_decorator\ndef infer(\n ref_audio_orig,\n ref_text,\n gen_text,\n model,\n remove_silence,","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.load_text_from_file","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.load_text_from_file#L122-L128","kind":"function","name":"load_text_from_file","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":122,"end_line":128,"context_start_line":102,"context_end_line":148,"code":" messages,\n tokenize=False,\n add_generation_prompt=True,\n )\n\n model_inputs = tokenizer([text], return_tensors=\"pt\").to(model.device)\n generated_ids = model.generate(\n **model_inputs,\n max_new_tokens=512,\n temperature=0.7,\n top_p=0.95,\n )\n\n generated_ids = [\n output_ids[len(input_ids) :] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)\n ]\n return tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]\n\n\n@gpu_decorator\ndef load_text_from_file(file):\n if file:\n with open(file, \"r\", encoding=\"utf-8\") as f:\n text = f.read().strip()\n else:\n text = \"\"\n return gr.update(value=text)\n\n\n@lru_cache(maxsize=1000) # NOTE. need to ensure params of infer() hashable\n@gpu_decorator\ndef infer(\n ref_audio_orig,\n ref_text,\n gen_text,\n model,\n remove_silence,\n seed,\n cross_fade_duration=0.15,\n nfe_step=32,\n speed=1,\n show_info=gr.Info,\n):\n if not ref_audio_orig:\n gr.Warning(\"Please provide reference audio.\")\n return gr.update(), gr.update(), ref_text\n","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.infer","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.infer#L133-L209","kind":"function","name":"infer","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":133,"end_line":209,"context_start_line":113,"context_end_line":229,"code":" )\n\n generated_ids = [\n output_ids[len(input_ids) :] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)\n ]\n return tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]\n\n\n@gpu_decorator\ndef load_text_from_file(file):\n if file:\n with open(file, \"r\", encoding=\"utf-8\") as f:\n text = f.read().strip()\n else:\n text = \"\"\n return gr.update(value=text)\n\n\n@lru_cache(maxsize=1000) # NOTE. need to ensure params of infer() hashable\n@gpu_decorator\ndef infer(\n ref_audio_orig,\n ref_text,\n gen_text,\n model,\n remove_silence,\n seed,\n cross_fade_duration=0.15,\n nfe_step=32,\n speed=1,\n show_info=gr.Info,\n):\n if not ref_audio_orig:\n gr.Warning(\"Please provide reference audio.\")\n return gr.update(), gr.update(), ref_text\n\n # Set inference seed\n if seed < 0 or seed > 2**31 - 1:\n gr.Warning(\"Seed must in range 0 ~ 2147483647. Using random seed instead.\")\n seed = np.random.randint(0, 2**31 - 1)\n torch.manual_seed(seed)\n used_seed = seed\n\n if not gen_text.strip():\n gr.Warning(\"Please enter text to generate or upload a text file.\")\n return gr.update(), gr.update(), ref_text\n\n ref_audio, ref_text = preprocess_ref_audio_text(ref_audio_orig, ref_text, show_info=show_info)\n\n if model == DEFAULT_TTS_MODEL:\n ema_model = F5TTS_ema_model\n elif model == \"E2-TTS\":\n global E2TTS_ema_model\n if E2TTS_ema_model is None:\n show_info(\"Loading E2-TTS model...\")\n E2TTS_ema_model = load_e2tts()\n ema_model = E2TTS_ema_model\n elif isinstance(model, tuple) and model[0] == \"Custom\":\n assert not USING_SPACES, \"Only official checkpoints allowed in Spaces.\"\n global custom_ema_model, pre_custom_path\n if pre_custom_path != model[1]:\n show_info(\"Loading Custom TTS model...\")\n custom_ema_model = load_custom(model[1], vocab_path=model[2], model_cfg=model[3])\n pre_custom_path = model[1]\n ema_model = custom_ema_model\n\n final_wave, final_sample_rate, combined_spectrogram = infer_process(\n ref_audio,\n ref_text,\n gen_text,\n ema_model,\n vocoder,\n cross_fade_duration=cross_fade_duration,\n nfe_step=nfe_step,\n speed=speed,\n show_info=show_info,\n progress=gr.Progress(),\n )\n\n # Remove silence\n if remove_silence:\n with tempfile.NamedTemporaryFile(suffix=\".wav\", **tempfile_kwargs) as f:\n temp_path = f.name\n try:\n sf.write(temp_path, final_wave, final_sample_rate)\n remove_silence_for_generated_wav(f.name)\n final_wave, _ = torchaudio.load(f.name)\n finally:\n os.unlink(temp_path)\n final_wave = final_wave.squeeze().cpu().numpy()\n\n # Save the spectrogram\n with tempfile.NamedTemporaryFile(suffix=\".png\", **tempfile_kwargs) as tmp_spectrogram:\n spectrogram_path = tmp_spectrogram.name\n save_spectrogram(combined_spectrogram, spectrogram_path)\n\n return (final_sample_rate, final_wave), spectrogram_path, ref_text, used_seed\n\n\nwith gr.Blocks() as app_tts:\n gr.Markdown(\"# Batched TTS\")\n ref_audio_input = gr.Audio(label=\"Reference Audio\", type=\"filepath\")\n with gr.Row():\n gen_text_input = gr.Textbox(\n label=\"Text to Generate\",\n lines=10,\n max_lines=40,\n scale=4,\n )\n gen_text_file = gr.File(label=\"Load Text to Generate from File (.txt)\", file_types=[\".txt\"], scale=1)\n generate_btn = gr.Button(\"Synthesize\", variant=\"primary\")\n with gr.Accordion(\"Advanced Settings\", open=False):\n with gr.Row():\n ref_text_input = gr.Textbox(\n label=\"Reference Text\",\n info=\"Leave blank to automatically transcribe the reference audio. If you enter text or upload a file, it will override automatic transcription.\",\n lines=2,","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.parse_speechtypes_text","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.parse_speechtypes_text#L338-L369","kind":"function","name":"parse_speechtypes_text","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":338,"end_line":369,"context_start_line":318,"context_end_line":389,"code":" [ref_text_input, ref_text_file],\n )\n\n generate_btn.click(\n basic_tts,\n inputs=[\n ref_audio_input,\n ref_text_input,\n gen_text_input,\n remove_silence,\n randomize_seed,\n seed_input,\n cross_fade_duration_slider,\n nfe_slider,\n speed_slider,\n ],\n outputs=[audio_output, spectrogram_output, ref_text_input, seed_input],\n )\n\n\ndef parse_speechtypes_text(gen_text):\n # Pattern to find {str} or {\"name\": str, \"seed\": int, \"speed\": float}\n pattern = r\"(\\{.*?\\})\"\n\n # Split the text by the pattern\n tokens = re.split(pattern, gen_text)\n\n segments = []\n\n current_type_dict = {\n \"name\": \"Regular\",\n \"seed\": -1,\n \"speed\": 1.0,\n }\n\n for i in range(len(tokens)):\n if i % 2 == 0:\n # This is text\n text = tokens[i].strip()\n if text:\n current_type_dict[\"text\"] = text\n segments.append(current_type_dict)\n else:\n # This is type\n type_str = tokens[i].strip()\n try: # if type dict\n current_type_dict = json.loads(type_str)\n except json.decoder.JSONDecodeError:\n type_str = type_str[1:-1] # remove brace {}\n current_type_dict = {\"name\": type_str, \"seed\": -1, \"speed\": 1.0}\n\n return segments\n\n\nwith gr.Blocks() as app_multistyle:\n # New section for multistyle generation\n gr.Markdown(\n \"\"\"\n # Multiple Speech-Type Generation\n\n This section allows you to generate multiple speech types or multiple people's voices. Enter your text in the format shown below, or upload a .txt file with the same format. The system will generate speech using the appropriate type. If unspecified, the model will use the regular speech type. The current speech type will be used until the next speech type is specified.\n \"\"\"\n )\n\n with gr.Row():\n gr.Markdown(\n \"\"\"\n **Example Input:**
\n {Regular} Hello, I'd like to order a sandwich please.
\n {Surprised} What do you mean you're out of bread?
\n {Sad} I really wanted a sandwich though...
\n {Angry} You know what, darn you and your little shop!
","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.main","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.main#L1104-L1114","kind":"function","name":"main","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":1104,"end_line":1114,"context_start_line":1084,"context_end_line":1121,"code":" \"-s\",\n default=False,\n is_flag=True,\n help=\"Share the app via Gradio share link\",\n)\n@click.option(\"--api\", \"-a\", default=True, is_flag=True, help=\"Allow API access\")\n@click.option(\n \"--root_path\",\n \"-r\",\n default=None,\n type=str,\n help='The root path (or \"mount point\") of the application, if it\\'s not served from the root (\"/\") of the domain. Often used when the application is behind a reverse proxy that forwards requests to the application, e.g. set \"/myapp\" or full URL for application served at \"https://example.com/myapp\".',\n)\n@click.option(\n \"--inbrowser\",\n \"-i\",\n is_flag=True,\n default=False,\n help=\"Automatically launch the interface in the default web browser\",\n)\ndef main(port, host, share, api, root_path, inbrowser):\n global app\n print(\"Starting app...\")\n app.queue(api_open=api).launch(\n server_name=host,\n server_port=port,\n share=share,\n show_api=api,\n root_path=root_path,\n inbrowser=inbrowser,\n )\n\n\nif __name__ == \"__main__\":\n if not USING_SPACES:\n main()\n else:\n app.queue().launch()","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.basic_tts","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.basic_tts#L276-L301","kind":"function","name":"basic_tts","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":276,"end_line":301,"context_start_line":256,"context_end_line":321,"code":" label=\"NFE Steps\",\n minimum=4,\n maximum=64,\n value=32,\n step=2,\n info=\"Set the number of denoising steps.\",\n )\n cross_fade_duration_slider = gr.Slider(\n label=\"Cross-Fade Duration (s)\",\n minimum=0.0,\n maximum=1.0,\n value=0.15,\n step=0.01,\n info=\"Set the duration of the cross-fade between audio clips.\",\n )\n\n audio_output = gr.Audio(label=\"Synthesized Audio\")\n spectrogram_output = gr.Image(label=\"Spectrogram\")\n\n @gpu_decorator\n def basic_tts(\n ref_audio_input,\n ref_text_input,\n gen_text_input,\n remove_silence,\n randomize_seed,\n seed_input,\n cross_fade_duration_slider,\n nfe_slider,\n speed_slider,\n ):\n if randomize_seed:\n seed_input = np.random.randint(0, 2**31 - 1)\n\n audio_out, spectrogram_path, ref_text_out, used_seed = infer(\n ref_audio_input,\n ref_text_input,\n gen_text_input,\n tts_model_choice,\n remove_silence,\n seed=seed_input,\n cross_fade_duration=cross_fade_duration_slider,\n nfe_step=nfe_slider,\n speed=speed_slider,\n )\n return audio_out, spectrogram_path, ref_text_out, used_seed\n\n gen_text_file.upload(\n load_text_from_file,\n inputs=[gen_text_file],\n outputs=[gen_text_input],\n )\n\n ref_text_file.upload(\n load_text_from_file,\n inputs=[ref_text_file],\n outputs=[ref_text_input],\n )\n\n ref_audio_input.clear(\n lambda: [None, None],\n None,\n [ref_text_input, ref_text_file],\n )\n\n generate_btn.click(","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.add_speech_type_fn","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.add_speech_type_fn#L490-L498","kind":"function","name":"add_speech_type_fn","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":490,"end_line":498,"context_start_line":470,"context_end_line":518,"code":" # Global logic for all speech types\n for i in range(max_speech_types):\n speech_type_audios[i].clear(\n lambda: [None, None],\n None,\n [speech_type_ref_texts[i], speech_type_ref_text_files[i]],\n )\n speech_type_ref_text_files[i].upload(\n load_text_from_file,\n inputs=[speech_type_ref_text_files[i]],\n outputs=[speech_type_ref_texts[i]],\n )\n\n # Button to add speech type\n add_speech_type_btn = gr.Button(\"Add Speech Type\")\n\n # Keep track of autoincrement of speech types, no roll back\n speech_type_count = 1\n\n # Function to add a speech type\n def add_speech_type_fn():\n row_updates = [gr.update() for _ in range(max_speech_types)]\n global speech_type_count\n if speech_type_count < max_speech_types:\n row_updates[speech_type_count] = gr.update(visible=True)\n speech_type_count += 1\n else:\n gr.Warning(\"Exhausted maximum number of speech types. Consider restart the app.\")\n return row_updates\n\n add_speech_type_btn.click(add_speech_type_fn, outputs=speech_type_rows)\n\n # Function to delete a speech type\n def delete_speech_type_fn():\n return gr.update(visible=False), None, None, None, None\n\n # Update delete button clicks and ref text file changes\n for i in range(1, len(speech_type_delete_btns)):\n speech_type_delete_btns[i].click(\n delete_speech_type_fn,\n outputs=[\n speech_type_rows[i],\n speech_type_names[i],\n speech_type_audios[i],\n speech_type_ref_texts[i],\n speech_type_ref_text_files[i],\n ],\n )\n","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.delete_speech_type_fn","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.delete_speech_type_fn#L503-L504","kind":"function","name":"delete_speech_type_fn","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":503,"end_line":504,"context_start_line":483,"context_end_line":524,"code":" # Button to add speech type\n add_speech_type_btn = gr.Button(\"Add Speech Type\")\n\n # Keep track of autoincrement of speech types, no roll back\n speech_type_count = 1\n\n # Function to add a speech type\n def add_speech_type_fn():\n row_updates = [gr.update() for _ in range(max_speech_types)]\n global speech_type_count\n if speech_type_count < max_speech_types:\n row_updates[speech_type_count] = gr.update(visible=True)\n speech_type_count += 1\n else:\n gr.Warning(\"Exhausted maximum number of speech types. Consider restart the app.\")\n return row_updates\n\n add_speech_type_btn.click(add_speech_type_fn, outputs=speech_type_rows)\n\n # Function to delete a speech type\n def delete_speech_type_fn():\n return gr.update(visible=False), None, None, None, None\n\n # Update delete button clicks and ref text file changes\n for i in range(1, len(speech_type_delete_btns)):\n speech_type_delete_btns[i].click(\n delete_speech_type_fn,\n outputs=[\n speech_type_rows[i],\n speech_type_names[i],\n speech_type_audios[i],\n speech_type_ref_texts[i],\n speech_type_ref_text_files[i],\n ],\n )\n\n # Text input for the prompt\n with gr.Row():\n gen_text_input_multistyle = gr.Textbox(\n label=\"Text to Generate\",\n lines=10,\n max_lines=40,","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.make_insert_speech_type_fn","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.make_insert_speech_type_fn#L530-L544","kind":"function","name":"make_insert_speech_type_fn","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":530,"end_line":544,"context_start_line":510,"context_end_line":564,"code":" outputs=[\n speech_type_rows[i],\n speech_type_names[i],\n speech_type_audios[i],\n speech_type_ref_texts[i],\n speech_type_ref_text_files[i],\n ],\n )\n\n # Text input for the prompt\n with gr.Row():\n gen_text_input_multistyle = gr.Textbox(\n label=\"Text to Generate\",\n lines=10,\n max_lines=40,\n scale=4,\n placeholder=\"Enter the script with speaker names (or emotion types) at the start of each block, e.g.:\\n\\n{Regular} Hello, I'd like to order a sandwich please.\\n{Surprised} What do you mean you're out of bread?\\n{Sad} I really wanted a sandwich though...\\n{Angry} You know what, darn you and your little shop!\\n{Whisper} I'll just go back home and cry now.\\n{Shouting} Why me?!\",\n )\n gen_text_file_multistyle = gr.File(label=\"Load Text to Generate from File (.txt)\", file_types=[\".txt\"], scale=1)\n\n def make_insert_speech_type_fn(index):\n def insert_speech_type_fn(current_text, speech_type_name, speech_type_seed, speech_type_speed):\n current_text = current_text or \"\"\n if not speech_type_name:\n gr.Warning(\"Please enter speech type name before insert.\")\n return current_text\n speech_type_dict = {\n \"name\": speech_type_name,\n \"seed\": speech_type_seed,\n \"speed\": speech_type_speed,\n }\n updated_text = current_text + json.dumps(speech_type_dict) + \" \"\n return updated_text\n\n return insert_speech_type_fn\n\n for i, insert_btn in enumerate(speech_type_insert_btns):\n insert_fn = make_insert_speech_type_fn(i)\n insert_btn.click(\n insert_fn,\n inputs=[gen_text_input_multistyle, speech_type_names[i], speech_type_seeds[i], speech_type_speeds[i]],\n outputs=gen_text_input_multistyle,\n )\n\n with gr.Accordion(\"Advanced Settings\", open=True):\n with gr.Row():\n with gr.Column():\n show_cherrypick_multistyle = gr.Checkbox(\n label=\"Show Cherry-pick Interface\",\n info=\"Turn on to show interface, picking seeds from previous generations.\",\n value=False,\n )\n with gr.Column():\n remove_silence_multistyle = gr.Checkbox(\n label=\"Remove Silences\",","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.generate_multistyle_speech","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.generate_multistyle_speech#L600-L679","kind":"function","name":"generate_multistyle_speech","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":600,"end_line":679,"context_start_line":580,"context_end_line":699,"code":" show_copy_button=True,\n interactive=False,\n visible=False,\n )\n\n # Logic control to show/hide the cherrypick interface\n show_cherrypick_multistyle.change(\n lambda is_visible: gr.update(visible=is_visible),\n show_cherrypick_multistyle,\n cherrypick_interface_multistyle,\n )\n\n # Function to load text to generate from file\n gen_text_file_multistyle.upload(\n load_text_from_file,\n inputs=[gen_text_file_multistyle],\n outputs=[gen_text_input_multistyle],\n )\n\n @gpu_decorator\n def generate_multistyle_speech(\n gen_text,\n *args,\n ):\n speech_type_names_list = args[:max_speech_types]\n speech_type_audios_list = args[max_speech_types : 2 * max_speech_types]\n speech_type_ref_texts_list = args[2 * max_speech_types : 3 * max_speech_types]\n remove_silence = args[3 * max_speech_types]\n # Collect the speech types and their audios into a dict\n speech_types = OrderedDict()\n\n ref_text_idx = 0\n for name_input, audio_input, ref_text_input in zip(\n speech_type_names_list, speech_type_audios_list, speech_type_ref_texts_list\n ):\n if name_input and audio_input:\n speech_types[name_input] = {\"audio\": audio_input, \"ref_text\": ref_text_input}\n else:\n speech_types[f\"@{ref_text_idx}@\"] = {\"audio\": \"\", \"ref_text\": \"\"}\n ref_text_idx += 1\n\n # Parse the gen_text into segments\n segments = parse_speechtypes_text(gen_text)\n\n # For each segment, generate speech\n generated_audio_segments = []\n current_type_name = \"Regular\"\n inference_meta_data = \"\"\n\n for segment in segments:\n name = segment[\"name\"]\n seed_input = segment[\"seed\"]\n speed = segment[\"speed\"]\n text = segment[\"text\"]\n\n if name in speech_types:\n current_type_name = name\n else:\n gr.Warning(f\"Type {name} is not available, will use Regular as default.\")\n current_type_name = \"Regular\"\n\n try:\n ref_audio = speech_types[current_type_name][\"audio\"]\n except KeyError:\n gr.Warning(f\"Please provide reference audio for type {current_type_name}.\")\n return [None] + [speech_types[name][\"ref_text\"] for name in speech_types] + [None]\n ref_text = speech_types[current_type_name].get(\"ref_text\", \"\")\n\n if seed_input == -1:\n seed_input = np.random.randint(0, 2**31 - 1)\n\n # Generate or retrieve speech for this segment\n audio_out, _, ref_text_out, used_seed = infer(\n ref_audio,\n ref_text,\n text,\n tts_model_choice,\n remove_silence,\n seed=seed_input,\n cross_fade_duration=0,\n speed=speed,\n show_info=print, # no pull to top when generating\n )\n sr, audio_data = audio_out\n\n generated_audio_segments.append(audio_data)\n speech_types[current_type_name][\"ref_text\"] = ref_text_out\n inference_meta_data += json.dumps(dict(name=name, seed=used_seed, speed=speed)) + f\" {text}\\n\"\n\n # Concatenate all audio segments\n if generated_audio_segments:\n final_audio_data = np.concatenate(generated_audio_segments)\n return (\n [(sr, final_audio_data)]\n + [speech_types[name][\"ref_text\"] for name in speech_types]\n + [inference_meta_data]\n )\n else:\n gr.Warning(\"No audio generated.\")\n return [None] + [speech_types[name][\"ref_text\"] for name in speech_types] + [None]\n\n generate_multistyle_btn.click(\n generate_multistyle_speech,\n inputs=[\n gen_text_input_multistyle,\n ]\n + speech_type_names\n + speech_type_audios\n + speech_type_ref_texts\n + [\n remove_silence_multistyle,\n ],\n outputs=[audio_output_multistyle] + speech_type_ref_texts + [cherrypick_interface_multistyle],\n )\n\n # Validation function to disable Generate button if speech types are missing\n def validate_speech_types(gen_text, regular_name, *args):\n speech_type_names_list = args\n\n # Collect the speech types names","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.validate_speech_types","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.validate_speech_types#L696-L719","kind":"function","name":"validate_speech_types","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":696,"end_line":719,"context_start_line":676,"context_end_line":739,"code":" )\n else:\n gr.Warning(\"No audio generated.\")\n return [None] + [speech_types[name][\"ref_text\"] for name in speech_types] + [None]\n\n generate_multistyle_btn.click(\n generate_multistyle_speech,\n inputs=[\n gen_text_input_multistyle,\n ]\n + speech_type_names\n + speech_type_audios\n + speech_type_ref_texts\n + [\n remove_silence_multistyle,\n ],\n outputs=[audio_output_multistyle] + speech_type_ref_texts + [cherrypick_interface_multistyle],\n )\n\n # Validation function to disable Generate button if speech types are missing\n def validate_speech_types(gen_text, regular_name, *args):\n speech_type_names_list = args\n\n # Collect the speech types names\n speech_types_available = set()\n if regular_name:\n speech_types_available.add(regular_name)\n for name_input in speech_type_names_list:\n if name_input:\n speech_types_available.add(name_input)\n\n # Parse the gen_text to get the speech types used\n segments = parse_speechtypes_text(gen_text)\n speech_types_in_text = set(segment[\"name\"] for segment in segments)\n\n # Check if all speech types in text are available\n missing_speech_types = speech_types_in_text - speech_types_available\n\n if missing_speech_types:\n # Disable the generate button\n return gr.update(interactive=False)\n else:\n # Enable the generate button\n return gr.update(interactive=True)\n\n gen_text_input_multistyle.change(\n validate_speech_types,\n inputs=[gen_text_input_multistyle, regular_name] + speech_type_names,\n outputs=generate_multistyle_btn,\n )\n\n\nwith gr.Blocks() as app_chat:\n gr.Markdown(\n \"\"\"\n# Voice Chat\nHave a conversation with an AI using your reference voice!\n1. Upload a reference audio clip and optionally its transcript (via text or .txt file).\n2. Load the chat model.\n3. Record your message through your microphone or type it.\n4. The AI will respond using the reference voice.\n\"\"\"\n )\n","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.load_chat_model","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.load_chat_model#L746-L760","kind":"function","name":"load_chat_model","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":746,"end_line":760,"context_start_line":726,"context_end_line":780,"code":"\n\nwith gr.Blocks() as app_chat:\n gr.Markdown(\n \"\"\"\n# Voice Chat\nHave a conversation with an AI using your reference voice!\n1. Upload a reference audio clip and optionally its transcript (via text or .txt file).\n2. Load the chat model.\n3. Record your message through your microphone or type it.\n4. The AI will respond using the reference voice.\n\"\"\"\n )\n\n chat_model_name_list = [\n \"Qwen/Qwen2.5-3B-Instruct\",\n \"microsoft/Phi-4-mini-instruct\",\n ]\n\n @gpu_decorator\n def load_chat_model(chat_model_name):\n show_info = gr.Info\n global chat_model_state, chat_tokenizer_state\n if chat_model_state is not None:\n chat_model_state = None\n chat_tokenizer_state = None\n gc.collect()\n torch.cuda.empty_cache()\n\n show_info(f\"Loading chat model: {chat_model_name}\")\n chat_model_state = AutoModelForCausalLM.from_pretrained(chat_model_name, torch_dtype=\"auto\", device_map=\"auto\")\n chat_tokenizer_state = AutoTokenizer.from_pretrained(chat_model_name)\n show_info(f\"Chat model {chat_model_name} loaded successfully!\")\n\n return gr.update(visible=False), gr.update(visible=True)\n\n if USING_SPACES:\n load_chat_model(chat_model_name_list[0])\n\n chat_model_name_input = gr.Dropdown(\n choices=chat_model_name_list,\n value=chat_model_name_list[0],\n label=\"Chat Model Name\",\n info=\"Enter the name of a HuggingFace chat model\",\n allow_custom_value=not USING_SPACES,\n )\n load_chat_model_btn = gr.Button(\"Load Chat Model\", variant=\"primary\", visible=not USING_SPACES)\n chat_interface_container = gr.Column(visible=USING_SPACES)\n\n chat_model_name_input.change(\n lambda: gr.update(visible=True),\n None,\n load_chat_model_btn,\n show_progress=\"hidden\",\n )","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.load_last_used_custom","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.load_last_used_custom#L963-L972","kind":"function","name":"load_last_used_custom","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":963,"end_line":972,"context_start_line":943,"context_end_line":992,"code":"with gr.Blocks() as app:\n gr.Markdown(\n f\"\"\"\n# E2/F5 TTS\n\nThis is {\"a local web UI for [F5 TTS](https://github.com/SWivid/F5-TTS)\" if not USING_SPACES else \"an online demo for [F5-TTS](https://github.com/SWivid/F5-TTS)\"} with advanced batch processing support. This app supports the following TTS models:\n\n* [F5-TTS](https://arxiv.org/abs/2410.06885) (A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching)\n* [E2 TTS](https://arxiv.org/abs/2406.18009) (Embarrassingly Easy Fully Non-Autoregressive Zero-Shot TTS)\n\nThe checkpoints currently support English and Chinese.\n\nIf you're having issues, try converting your reference audio to WAV or MP3, clipping it to 12s with ✂ in the bottom right corner (otherwise might have non-optimal auto-trimmed result).\n\n**NOTE: Reference text will be automatically transcribed with Whisper if not provided. For best results, keep your reference clips short (<12s). Ensure the audio is fully uploaded before generating.**\n\"\"\"\n )\n\n last_used_custom = files(\"f5_tts\").joinpath(\"infer/.cache/last_used_custom_model_info_v1.txt\")\n\n def load_last_used_custom():\n try:\n custom = []\n with open(last_used_custom, \"r\", encoding=\"utf-8\") as f:\n for line in f:\n custom.append(line.strip())\n return custom\n except FileNotFoundError:\n last_used_custom.parent.mkdir(parents=True, exist_ok=True)\n return DEFAULT_TTS_MODEL_CFG\n\n def switch_tts_model(new_choice):\n global tts_model_choice\n if new_choice == \"Custom\": # override in case webpage is refreshed\n custom_ckpt_path, custom_vocab_path, custom_model_cfg = load_last_used_custom()\n tts_model_choice = (\"Custom\", custom_ckpt_path, custom_vocab_path, custom_model_cfg)\n return (\n gr.update(visible=True, value=custom_ckpt_path),\n gr.update(visible=True, value=custom_vocab_path),\n gr.update(visible=True, value=custom_model_cfg),\n )\n else:\n tts_model_choice = new_choice\n return gr.update(visible=False), gr.update(visible=False), gr.update(visible=False)\n\n def set_custom_model(custom_ckpt_path, custom_vocab_path, custom_model_cfg):\n global tts_model_choice\n tts_model_choice = (\"Custom\", custom_ckpt_path, custom_vocab_path, custom_model_cfg)\n with open(last_used_custom, \"w\", encoding=\"utf-8\") as f:\n f.write(custom_ckpt_path + \"\\n\" + custom_vocab_path + \"\\n\" + custom_model_cfg + \"\\n\")","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.switch_tts_model","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.switch_tts_model#L974-L986","kind":"function","name":"switch_tts_model","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":974,"end_line":986,"context_start_line":954,"context_end_line":1006,"code":"\nIf you're having issues, try converting your reference audio to WAV or MP3, clipping it to 12s with ✂ in the bottom right corner (otherwise might have non-optimal auto-trimmed result).\n\n**NOTE: Reference text will be automatically transcribed with Whisper if not provided. For best results, keep your reference clips short (<12s). Ensure the audio is fully uploaded before generating.**\n\"\"\"\n )\n\n last_used_custom = files(\"f5_tts\").joinpath(\"infer/.cache/last_used_custom_model_info_v1.txt\")\n\n def load_last_used_custom():\n try:\n custom = []\n with open(last_used_custom, \"r\", encoding=\"utf-8\") as f:\n for line in f:\n custom.append(line.strip())\n return custom\n except FileNotFoundError:\n last_used_custom.parent.mkdir(parents=True, exist_ok=True)\n return DEFAULT_TTS_MODEL_CFG\n\n def switch_tts_model(new_choice):\n global tts_model_choice\n if new_choice == \"Custom\": # override in case webpage is refreshed\n custom_ckpt_path, custom_vocab_path, custom_model_cfg = load_last_used_custom()\n tts_model_choice = (\"Custom\", custom_ckpt_path, custom_vocab_path, custom_model_cfg)\n return (\n gr.update(visible=True, value=custom_ckpt_path),\n gr.update(visible=True, value=custom_vocab_path),\n gr.update(visible=True, value=custom_model_cfg),\n )\n else:\n tts_model_choice = new_choice\n return gr.update(visible=False), gr.update(visible=False), gr.update(visible=False)\n\n def set_custom_model(custom_ckpt_path, custom_vocab_path, custom_model_cfg):\n global tts_model_choice\n tts_model_choice = (\"Custom\", custom_ckpt_path, custom_vocab_path, custom_model_cfg)\n with open(last_used_custom, \"w\", encoding=\"utf-8\") as f:\n f.write(custom_ckpt_path + \"\\n\" + custom_vocab_path + \"\\n\" + custom_model_cfg + \"\\n\")\n\n with gr.Row():\n if not USING_SPACES:\n choose_tts_model = gr.Radio(\n choices=[DEFAULT_TTS_MODEL, \"E2-TTS\", \"Custom\"], label=\"Choose TTS Model\", value=DEFAULT_TTS_MODEL\n )\n else:\n choose_tts_model = gr.Radio(\n choices=[DEFAULT_TTS_MODEL, \"E2-TTS\"], label=\"Choose TTS Model\", value=DEFAULT_TTS_MODEL\n )\n custom_ckpt_path = gr.Dropdown(\n choices=[DEFAULT_TTS_MODEL_CFG[0]],\n value=load_last_used_custom()[0],\n allow_custom_value=True,","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.set_custom_model","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.set_custom_model#L988-L992","kind":"function","name":"set_custom_model","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":988,"end_line":992,"context_start_line":968,"context_end_line":1012,"code":" custom.append(line.strip())\n return custom\n except FileNotFoundError:\n last_used_custom.parent.mkdir(parents=True, exist_ok=True)\n return DEFAULT_TTS_MODEL_CFG\n\n def switch_tts_model(new_choice):\n global tts_model_choice\n if new_choice == \"Custom\": # override in case webpage is refreshed\n custom_ckpt_path, custom_vocab_path, custom_model_cfg = load_last_used_custom()\n tts_model_choice = (\"Custom\", custom_ckpt_path, custom_vocab_path, custom_model_cfg)\n return (\n gr.update(visible=True, value=custom_ckpt_path),\n gr.update(visible=True, value=custom_vocab_path),\n gr.update(visible=True, value=custom_model_cfg),\n )\n else:\n tts_model_choice = new_choice\n return gr.update(visible=False), gr.update(visible=False), gr.update(visible=False)\n\n def set_custom_model(custom_ckpt_path, custom_vocab_path, custom_model_cfg):\n global tts_model_choice\n tts_model_choice = (\"Custom\", custom_ckpt_path, custom_vocab_path, custom_model_cfg)\n with open(last_used_custom, \"w\", encoding=\"utf-8\") as f:\n f.write(custom_ckpt_path + \"\\n\" + custom_vocab_path + \"\\n\" + custom_model_cfg + \"\\n\")\n\n with gr.Row():\n if not USING_SPACES:\n choose_tts_model = gr.Radio(\n choices=[DEFAULT_TTS_MODEL, \"E2-TTS\", \"Custom\"], label=\"Choose TTS Model\", value=DEFAULT_TTS_MODEL\n )\n else:\n choose_tts_model = gr.Radio(\n choices=[DEFAULT_TTS_MODEL, \"E2-TTS\"], label=\"Choose TTS Model\", value=DEFAULT_TTS_MODEL\n )\n custom_ckpt_path = gr.Dropdown(\n choices=[DEFAULT_TTS_MODEL_CFG[0]],\n value=load_last_used_custom()[0],\n allow_custom_value=True,\n label=\"Model: local_path | hf://user_id/repo_id/model_ckpt\",\n visible=False,\n )\n custom_vocab_path = gr.Dropdown(\n choices=[DEFAULT_TTS_MODEL_CFG[1]],\n value=load_last_used_custom()[1],","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.insert_speech_type_fn","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.insert_speech_type_fn#L531-L542","kind":"function","name":"insert_speech_type_fn","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":531,"end_line":542,"context_start_line":511,"context_end_line":562,"code":" speech_type_rows[i],\n speech_type_names[i],\n speech_type_audios[i],\n speech_type_ref_texts[i],\n speech_type_ref_text_files[i],\n ],\n )\n\n # Text input for the prompt\n with gr.Row():\n gen_text_input_multistyle = gr.Textbox(\n label=\"Text to Generate\",\n lines=10,\n max_lines=40,\n scale=4,\n placeholder=\"Enter the script with speaker names (or emotion types) at the start of each block, e.g.:\\n\\n{Regular} Hello, I'd like to order a sandwich please.\\n{Surprised} What do you mean you're out of bread?\\n{Sad} I really wanted a sandwich though...\\n{Angry} You know what, darn you and your little shop!\\n{Whisper} I'll just go back home and cry now.\\n{Shouting} Why me?!\",\n )\n gen_text_file_multistyle = gr.File(label=\"Load Text to Generate from File (.txt)\", file_types=[\".txt\"], scale=1)\n\n def make_insert_speech_type_fn(index):\n def insert_speech_type_fn(current_text, speech_type_name, speech_type_seed, speech_type_speed):\n current_text = current_text or \"\"\n if not speech_type_name:\n gr.Warning(\"Please enter speech type name before insert.\")\n return current_text\n speech_type_dict = {\n \"name\": speech_type_name,\n \"seed\": speech_type_seed,\n \"speed\": speech_type_speed,\n }\n updated_text = current_text + json.dumps(speech_type_dict) + \" \"\n return updated_text\n\n return insert_speech_type_fn\n\n for i, insert_btn in enumerate(speech_type_insert_btns):\n insert_fn = make_insert_speech_type_fn(i)\n insert_btn.click(\n insert_fn,\n inputs=[gen_text_input_multistyle, speech_type_names[i], speech_type_seeds[i], speech_type_speeds[i]],\n outputs=gen_text_input_multistyle,\n )\n\n with gr.Accordion(\"Advanced Settings\", open=True):\n with gr.Row():\n with gr.Column():\n show_cherrypick_multistyle = gr.Checkbox(\n label=\"Show Cherry-pick Interface\",\n info=\"Turn on to show interface, picking seeds from previous generations.\",\n value=False,\n )\n with gr.Column():","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.process_audio_input","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.process_audio_input#L838-L850","kind":"function","name":"process_audio_input","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":838,"end_line":850,"context_start_line":818,"context_end_line":870,"code":"\n chatbot_interface = gr.Chatbot(label=\"Conversation\", type=\"messages\")\n\n with gr.Row():\n with gr.Column():\n audio_input_chat = gr.Microphone(\n label=\"Speak your message\",\n type=\"filepath\",\n )\n audio_output_chat = gr.Audio(autoplay=True)\n with gr.Column():\n text_input_chat = gr.Textbox(\n label=\"Type your message\",\n lines=1,\n )\n send_btn_chat = gr.Button(\"Send Message\")\n clear_btn_chat = gr.Button(\"Clear Conversation\")\n\n # Modify process_audio_input to generate user input\n @gpu_decorator\n def process_audio_input(conv_state, audio_path, text):\n \"\"\"Handle audio or text input from user\"\"\"\n\n if not audio_path and not text.strip():\n return conv_state\n\n if audio_path:\n text = preprocess_ref_audio_text(audio_path, text)[1]\n if not text.strip():\n return conv_state\n\n conv_state.append({\"role\": \"user\", \"content\": text})\n return conv_state\n\n # Use model and tokenizer from state to get text response\n @gpu_decorator\n def generate_text_response(conv_state, system_prompt):\n \"\"\"Generate text response from AI\"\"\"\n\n system_prompt_state = [{\"role\": \"system\", \"content\": system_prompt}]\n response = chat_model_inference(system_prompt_state + conv_state, chat_model_state, chat_tokenizer_state)\n\n conv_state.append({\"role\": \"assistant\", \"content\": response})\n return conv_state\n\n @gpu_decorator\n def generate_audio_response(conv_state, ref_audio, ref_text, remove_silence, randomize_seed, seed_input):\n \"\"\"Generate TTS audio for AI response\"\"\"\n if not conv_state or not ref_audio:\n return None, ref_text, seed_input\n\n last_ai_response = conv_state[-1][\"content\"]\n if not last_ai_response or conv_state[-1][\"role\"] != \"assistant\":","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.generate_text_response","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.generate_text_response#L854-L861","kind":"function","name":"generate_text_response","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":854,"end_line":861,"context_start_line":834,"context_end_line":881,"code":" clear_btn_chat = gr.Button(\"Clear Conversation\")\n\n # Modify process_audio_input to generate user input\n @gpu_decorator\n def process_audio_input(conv_state, audio_path, text):\n \"\"\"Handle audio or text input from user\"\"\"\n\n if not audio_path and not text.strip():\n return conv_state\n\n if audio_path:\n text = preprocess_ref_audio_text(audio_path, text)[1]\n if not text.strip():\n return conv_state\n\n conv_state.append({\"role\": \"user\", \"content\": text})\n return conv_state\n\n # Use model and tokenizer from state to get text response\n @gpu_decorator\n def generate_text_response(conv_state, system_prompt):\n \"\"\"Generate text response from AI\"\"\"\n\n system_prompt_state = [{\"role\": \"system\", \"content\": system_prompt}]\n response = chat_model_inference(system_prompt_state + conv_state, chat_model_state, chat_tokenizer_state)\n\n conv_state.append({\"role\": \"assistant\", \"content\": response})\n return conv_state\n\n @gpu_decorator\n def generate_audio_response(conv_state, ref_audio, ref_text, remove_silence, randomize_seed, seed_input):\n \"\"\"Generate TTS audio for AI response\"\"\"\n if not conv_state or not ref_audio:\n return None, ref_text, seed_input\n\n last_ai_response = conv_state[-1][\"content\"]\n if not last_ai_response or conv_state[-1][\"role\"] != \"assistant\":\n return None, ref_text, seed_input\n\n if randomize_seed:\n seed_input = np.random.randint(0, 2**31 - 1)\n\n audio_result, _, ref_text_out, used_seed = infer(\n ref_audio,\n ref_text,\n last_ai_response,\n tts_model_choice,\n remove_silence,","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.generate_audio_response","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.generate_audio_response#L864-L887","kind":"function","name":"generate_audio_response","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":864,"end_line":887,"context_start_line":844,"context_end_line":907,"code":" if audio_path:\n text = preprocess_ref_audio_text(audio_path, text)[1]\n if not text.strip():\n return conv_state\n\n conv_state.append({\"role\": \"user\", \"content\": text})\n return conv_state\n\n # Use model and tokenizer from state to get text response\n @gpu_decorator\n def generate_text_response(conv_state, system_prompt):\n \"\"\"Generate text response from AI\"\"\"\n\n system_prompt_state = [{\"role\": \"system\", \"content\": system_prompt}]\n response = chat_model_inference(system_prompt_state + conv_state, chat_model_state, chat_tokenizer_state)\n\n conv_state.append({\"role\": \"assistant\", \"content\": response})\n return conv_state\n\n @gpu_decorator\n def generate_audio_response(conv_state, ref_audio, ref_text, remove_silence, randomize_seed, seed_input):\n \"\"\"Generate TTS audio for AI response\"\"\"\n if not conv_state or not ref_audio:\n return None, ref_text, seed_input\n\n last_ai_response = conv_state[-1][\"content\"]\n if not last_ai_response or conv_state[-1][\"role\"] != \"assistant\":\n return None, ref_text, seed_input\n\n if randomize_seed:\n seed_input = np.random.randint(0, 2**31 - 1)\n\n audio_result, _, ref_text_out, used_seed = infer(\n ref_audio,\n ref_text,\n last_ai_response,\n tts_model_choice,\n remove_silence,\n seed=seed_input,\n cross_fade_duration=0.15,\n speed=1.0,\n show_info=print, # show_info=print no pull to top when generating\n )\n return audio_result, ref_text_out, used_seed\n\n def clear_conversation():\n \"\"\"Reset the conversation\"\"\"\n return [], None\n\n ref_text_file_chat.upload(\n load_text_from_file,\n inputs=[ref_text_file_chat],\n outputs=[ref_text_chat],\n )\n\n for user_operation in [audio_input_chat.stop_recording, text_input_chat.submit, send_btn_chat.click]:\n user_operation(\n process_audio_input,\n inputs=[chatbot_interface, audio_input_chat, text_input_chat],\n outputs=[chatbot_interface],\n ).then(\n generate_text_response,\n inputs=[chatbot_interface, system_prompt_chat],\n outputs=[chatbot_interface],","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"py:src.f5_tts.infer.infer_gradio.clear_conversation","uri":"program://DMOSpeech2/function/src.f5_tts.infer.infer_gradio.clear_conversation#L889-L891","kind":"function","name":"clear_conversation","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":889,"end_line":891,"context_start_line":869,"context_end_line":911,"code":" last_ai_response = conv_state[-1][\"content\"]\n if not last_ai_response or conv_state[-1][\"role\"] != \"assistant\":\n return None, ref_text, seed_input\n\n if randomize_seed:\n seed_input = np.random.randint(0, 2**31 - 1)\n\n audio_result, _, ref_text_out, used_seed = infer(\n ref_audio,\n ref_text,\n last_ai_response,\n tts_model_choice,\n remove_silence,\n seed=seed_input,\n cross_fade_duration=0.15,\n speed=1.0,\n show_info=print, # show_info=print no pull to top when generating\n )\n return audio_result, ref_text_out, used_seed\n\n def clear_conversation():\n \"\"\"Reset the conversation\"\"\"\n return [], None\n\n ref_text_file_chat.upload(\n load_text_from_file,\n inputs=[ref_text_file_chat],\n outputs=[ref_text_chat],\n )\n\n for user_operation in [audio_input_chat.stop_recording, text_input_chat.submit, send_btn_chat.click]:\n user_operation(\n process_audio_input,\n inputs=[chatbot_interface, audio_input_chat, text_input_chat],\n outputs=[chatbot_interface],\n ).then(\n generate_text_response,\n inputs=[chatbot_interface, system_prompt_chat],\n outputs=[chatbot_interface],\n ).then(\n generate_audio_response,\n inputs=[\n chatbot_interface,","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/duration_trainer_with_prompt.py","uri":"program://DMOSpeech2/file/src/duration_trainer_with_prompt.py","kind":"file","name":"src/duration_trainer_with_prompt.py","path":"src/duration_trainer_with_prompt.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"from __future__ import annotations\n\nimport gc\nimport os\n\nimport math\n\nimport torch\nimport torchaudio\nimport wandb\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nfrom ema_pytorch import EMA\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler, Subset # <-- Added Subset import\nfrom tqdm import tqdm\n\nimport torch.nn.functional as F\n\nfrom f5_tts.model import CFM","source_hash":"20ff79d536aa888b20ce19adba2d43f1dcea96063d23f16c210e5b9fdd404874","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/dmd_trainer.py","uri":"program://DMOSpeech2/file/src/dmd_trainer.py","kind":"file","name":"src/dmd_trainer.py","path":"src/dmd_trainer.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"from __future__ import annotations\n\nimport os\nimport gc\nfrom tqdm import tqdm\nimport wandb\n\nimport torch\nimport torch.nn as nn\nfrom torch.optim import AdamW\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\n\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\n\nfrom unimodel import UniModel\nfrom f5_tts.model import CFM\nfrom f5_tts.model.utils import exists, default\nfrom f5_tts.model.dataset import DynamicBatchSampler, collate_fn\n","source_hash":"2f2618612fc8a72a725d379657533ec1e177fd919989a0850d8a287ee23d5b40","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/guidance_model.py","uri":"program://DMOSpeech2/file/src/guidance_model.py","kind":"file","name":"src/guidance_model.py","path":"src/guidance_model.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\nfrom typing import Callable\nfrom random import random\nimport numpy as np\n\nimport torch\nfrom torch import nn\nimport torch.nn.functional as F\n\nfrom f5_tts.model import DiT\n\nfrom f5_tts.model.utils import (","source_hash":"fcb03759ceb8eeb0740d3a2b732facf2b88a59ecc401fa56805fe19b89d8ec0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/duration_trainer.py","uri":"program://DMOSpeech2/file/src/duration_trainer.py","kind":"file","name":"src/duration_trainer.py","path":"src/duration_trainer.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"from __future__ import annotations\n\nimport gc\nimport os\n\nimport math\n\nimport torch\nimport torchaudio\nimport wandb\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nfrom ema_pytorch import EMA\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler, Subset # <-- Added Subset import\nfrom tqdm import tqdm\n\nimport torch.nn.functional as F\n\nfrom f5_tts.model import CFM","source_hash":"2d47b5df9155cbe6db309e3617bab7f7fbf0c03498c4a19ddf3e3229f1ea24c1","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/ecapa_tdnn.py","uri":"program://DMOSpeech2/file/src/ecapa_tdnn.py","kind":"file","name":"src/ecapa_tdnn.py","path":"src/ecapa_tdnn.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# part of the code is borrowed from https://github.com/lawlict/ECAPA-TDNN\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torchaudio.transforms as trans\nfrom ctcmodel import ConformerCTC\n# from ctcmodel_nopool import ConformerCTC as ConformerCTCNoPool\nfrom pathlib import Path\n\n''' Res2Conv1d + BatchNorm1d + ReLU\n'''\n\n\nclass Res2Conv1dReluBn(nn.Module):\n '''\n in_channels == out_channels == channels\n '''\n\n def __init__(self, channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True, scale=4):\n super().__init__()","source_hash":"2856f9a59c60f45ed4d23d46faef7d769bec4babf5bc0157bd1d32de30046a4f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/unimodel.py","uri":"program://DMOSpeech2/file/src/unimodel.py","kind":"file","name":"src/unimodel.py","path":"src/unimodel.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"from __future__ import annotations\nfrom typing import Callable\nfrom random import random\n\nimport contextlib\n\nfrom torch import nn\nimport torch \nimport copy\nimport os\n\nfrom f5_tts.model import DiT, UNetT\nfrom pathlib import Path\nfrom guidance_model import Guidance\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,","source_hash":"d0a2447ed4e6999217aa5d1177a1def4c9688256e5ff0c12fe4a7cd040b19fbd","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/duration_predictor.py","uri":"program://DMOSpeech2/file/src/duration_predictor.py","kind":"file","name":"src/duration_predictor.py","path":"src/duration_predictor.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import torch\nimport torch.nn as nn\n\n# from tts_encode import tts_encode\n\ndef calculate_remaining_lengths(mel_lengths):\n B = mel_lengths.shape[0]\n max_L = mel_lengths.max().item() # Get the maximum length in the batch\n\n # Create a range tensor: shape (max_L,), [0, 1, 2, ..., max_L-1]\n range_tensor = torch.arange(max_L, device=mel_lengths.device).expand(B, max_L)\n\n # Compute targets using broadcasting: (L-1) - range_tensor\n remain_lengths = (mel_lengths[:, None] - 1 - range_tensor).clamp(min=0)\n\n return remain_lengths\n\n\nclass PositionalEncoding(nn.Module):\n def __init__(self, hidden_dim, max_len=4096):\n super().__init__()","source_hash":"9db72e598a7cc564ed6b6fb1eb634d42d4853ed6797b38a5461c6550d0491a8e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/grpo_duration_trainer.py","uri":"program://DMOSpeech2/file/src/grpo_duration_trainer.py","kind":"file","name":"src/grpo_duration_trainer.py","path":"src/grpo_duration_trainer.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import os\nimport gc\nimport json\nimport random\nimport time\nimport io\nimport copy\nfrom typing import List, Dict, Any, Optional, Callable, Tuple, Union\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler, Subset\nfrom tqdm import tqdm\n\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nimport wandb\n","source_hash":"0c0b89d9ba699d670ccf447ede603f8e9bdb05c7f4cf7b6a5e2ef268a90e4005","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/infer.py","uri":"program://DMOSpeech2/file/src/infer.py","kind":"file","name":"src/infer.py","path":"src/infer.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import os\nimport torch\nimport torchaudio\nimport torch.nn.functional as F\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torchdiffeq import odeint\nfrom safetensors.torch import load_file\nimport IPython.display as ipd\n\n# Import F5-TTS modules\nfrom f5_tts.model import CFM, UNetT, DiT\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import (\n default, exists, list_str_to_idx, list_str_to_tensor,\n lens_to_mask, mask_from_frac_lengths, get_tokenizer\n)\nfrom f5_tts.infer.utils_infer import (\n load_vocoder, preprocess_ref_audio_text, chunk_text,\n convert_char_to_pinyin, transcribe, target_rms,\n target_sample_rate, hop_length, speed\n)","source_hash":"fa86ae8ac7f5ba4f96dd6a347bdeb679e1e66e5f65ce00e5379633478fcfead2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/discriminator_conformer.py","uri":"program://DMOSpeech2/file/src/discriminator_conformer.py","kind":"file","name":"src/discriminator_conformer.py","path":"src/discriminator_conformer.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# part of the code is borrowed from https://github.com/lawlict/ECAPA-TDNN\n\nfrom __future__ import annotations\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torchaudio.transforms as trans\nfrom pathlib import Path\nfrom torchaudio.models import Conformer\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,\n lens_to_mask,\n mask_from_frac_lengths,\n)\n\nclass ResBlock(nn.Module):","source_hash":"25cc145b2a3e168fd92617837b289ebc02a86aa041d3ffbd1b4e303892986baf","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/ctcmodel.py","uri":"program://DMOSpeech2/file/src/ctcmodel.py","kind":"file","name":"src/ctcmodel.py","path":"src/ctcmodel.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"from torch import nn\nimport torch \nimport copy\n\nfrom pathlib import Path\nfrom torchaudio.models import Conformer\n\n\nfrom f5_tts.model.utils import default\nfrom f5_tts.model.utils import exists\nfrom f5_tts.model.utils import list_str_to_idx\nfrom f5_tts.model.utils import list_str_to_tensor\nfrom f5_tts.model.utils import lens_to_mask\nfrom f5_tts.model.utils import mask_from_frac_lengths\n\n\nfrom f5_tts.model.utils import (\n default,\n exists,\n list_str_to_idx,\n list_str_to_tensor,","source_hash":"61bd6ed4c15b5affe57e2e7cf82b89734754e5cb0c0f639947fb1b307278ba06","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/api.py","uri":"program://DMOSpeech2/file/src/f5_tts/api.py","kind":"file","name":"src/f5_tts/api.py","path":"src/f5_tts/api.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import random\nimport sys\nfrom importlib.resources import files\n\nimport soundfile as sf\nimport tqdm\nfrom cached_path import cached_path\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\n\nfrom f5_tts.infer.utils_infer import (\n infer_process,\n load_model,\n load_vocoder,\n preprocess_ref_audio_text,\n remove_silence_for_generated_wav,\n save_spectrogram,\n transcribe,\n)\nfrom f5_tts.model.utils import seed_everything\n","source_hash":"520e3eafa052475f30e038f2a37addeaa4aafe3b17cf9da172d67f1037b47a53","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/socket_client.py","uri":"program://DMOSpeech2/file/src/f5_tts/socket_client.py","kind":"file","name":"src/f5_tts/socket_client.py","path":"src/f5_tts/socket_client.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import asyncio\nimport logging\nimport socket\nimport time\n\nimport numpy as np\nimport pyaudio\n\n\nlogging.basicConfig(level=logging.INFO)\nlogger = logging.getLogger(__name__)\n\n\nasync def listen_to_F5TTS(text, server_ip=\"localhost\", server_port=9998):\n client_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)\n await asyncio.get_event_loop().run_in_executor(None, client_socket.connect, (server_ip, int(server_port)))\n\n start_time = time.time()\n first_chunk_time = None\n\n async def play_audio_stream():","source_hash":"301265e66a6fca7849462861c53751be848df709d1b9b755d098798e81cf25e6","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/socket_server.py","uri":"program://DMOSpeech2/file/src/f5_tts/socket_server.py","kind":"file","name":"src/f5_tts/socket_server.py","path":"src/f5_tts/socket_server.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import argparse\nimport gc\nimport logging\nimport queue\nimport socket\nimport struct\nimport threading\nimport traceback\nimport wave\nfrom importlib.resources import files\n\nimport numpy as np\nimport torch\nimport torchaudio\nfrom huggingface_hub import hf_hub_download\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\n\nfrom f5_tts.infer.utils_infer import (\n chunk_text,\n infer_batch_process,","source_hash":"08677037c902eae913256957c4d967e2a90a6a6e7d9a44cdb8bdf972fe958250","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/train/train.py","uri":"program://DMOSpeech2/file/src/f5_tts/train/train.py","kind":"file","name":"src/f5_tts/train/train.py","path":"src/f5_tts/train/train.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# training script.\n\nimport os\nfrom importlib.resources import files\n\nimport hydra\nfrom omegaconf import OmegaConf\n\nfrom f5_tts.model import CFM, Trainer\nfrom f5_tts.model.dataset import load_dataset\nfrom f5_tts.model.utils import get_tokenizer\n\n\nos.chdir(str(files(\"f5_tts\").joinpath(\"../..\"))) # change working directory to root of project (local editable)\n\n\n@hydra.main(version_base=\"1.3\", config_path=str(files(\"f5_tts\").joinpath(\"configs\")), config_name=None)\ndef main(model_cfg):\n model_cls = hydra.utils.get_class(f\"f5_tts.model.{model_cfg.model.backbone}\")\n model_arc = model_cfg.model.arch\n tokenizer = model_cfg.model.tokenizer","source_hash":"0b23913aed7f806c7697e3bf8d390c96a5b868c8c05158129999b98c70a50ca4","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/train/finetune_cli.py","uri":"program://DMOSpeech2/file/src/f5_tts/train/finetune_cli.py","kind":"file","name":"src/f5_tts/train/finetune_cli.py","path":"src/f5_tts/train/finetune_cli.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import argparse\nimport os\nimport shutil\nfrom importlib.resources import files\n\nfrom cached_path import cached_path\n\nfrom f5_tts.model import CFM, DiT, Trainer, UNetT\nfrom f5_tts.model.dataset import load_dataset\nfrom f5_tts.model.utils import get_tokenizer\n\n\n# -------------------------- Dataset Settings --------------------------- #\ntarget_sample_rate = 24000\nn_mel_channels = 100\nhop_length = 256\nwin_length = 1024\nn_fft = 1024\nmel_spec_type = \"vocos\" # 'vocos' or 'bigvgan'\n\n","source_hash":"c04786bf9631406fbea26b1b51f469fa6e7ddb9a3607090baa217ef45c23c4ae","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/train/finetune_gradio.py","uri":"program://DMOSpeech2/file/src/f5_tts/train/finetune_gradio.py","kind":"file","name":"src/f5_tts/train/finetune_gradio.py","path":"src/f5_tts/train/finetune_gradio.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import gc\nimport json\nimport os\nimport platform\nimport queue\nimport random\nimport re\nimport shutil\nimport signal\nimport subprocess\nimport sys\nimport tempfile\nimport threading\nimport time\nfrom glob import glob\nfrom importlib.resources import files\n\nimport click\nimport gradio as gr\nimport librosa\nimport numpy as np","source_hash":"1efcecaa8389adf15a807025dcb7297b3acd435c1de17db71163e70d655c78dc","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/train/datasets/prepare_libritts.py","uri":"program://DMOSpeech2/file/src/f5_tts/train/datasets/prepare_libritts.py","kind":"file","name":"src/f5_tts/train/datasets/prepare_libritts.py","path":"src/f5_tts/train/datasets/prepare_libritts.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport json\nfrom concurrent.futures import ProcessPoolExecutor\nfrom importlib.resources import files\nfrom pathlib import Path\n\nimport soundfile as sf\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm\n\n\ndef deal_with_audio_dir(audio_dir):\n sub_result, durations = [], []\n vocab_set = set()\n audio_lists = list(audio_dir.rglob(\"*.wav\"))\n","source_hash":"b3d6cc06b84e9c295a75bc558141f87c1328b6d50fac57419a39d4658f382a20","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/train/datasets/prepare_ljspeech.py","uri":"program://DMOSpeech2/file/src/f5_tts/train/datasets/prepare_ljspeech.py","kind":"file","name":"src/f5_tts/train/datasets/prepare_ljspeech.py","path":"src/f5_tts/train/datasets/prepare_ljspeech.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport json\nfrom importlib.resources import files\nfrom pathlib import Path\n\nimport soundfile as sf\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm\n\n\ndef main():\n result = []\n duration_list = []\n text_vocab_set = set()\n\n with open(meta_info, \"r\") as f:","source_hash":"7f03a0819517ade8a7722e4ae5f0378b3b790ea4878e226ab032d9910c608750","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/train/datasets/prepare_emilia_v2.py","uri":"program://DMOSpeech2/file/src/f5_tts/train/datasets/prepare_emilia_v2.py","kind":"file","name":"src/f5_tts/train/datasets/prepare_emilia_v2.py","path":"src/f5_tts/train/datasets/prepare_emilia_v2.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# put in src/f5_tts/train/datasets/prepare_emilia_v2.py\n# prepares Emilia dataset with the new format w/ Emilia-YODAS\n\nimport json\nimport os\nfrom concurrent.futures import ProcessPoolExecutor\nfrom importlib.resources import files\nfrom pathlib import Path\n\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm\n\nfrom f5_tts.model.utils import repetition_found\n\n\n# Define filters for exclusion\nout_en = set()\nen_filters = [\"ا\", \"い\", \"て\"]\n\n\ndef process_audio_directory(audio_dir):","source_hash":"3757de14aff828d561c375a976c6c9a52ed08492d1e9c99d77df9c5673f50174","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/train/datasets/prepare_wenetspeech4tts.py","uri":"program://DMOSpeech2/file/src/f5_tts/train/datasets/prepare_wenetspeech4tts.py","kind":"file","name":"src/f5_tts/train/datasets/prepare_wenetspeech4tts.py","path":"src/f5_tts/train/datasets/prepare_wenetspeech4tts.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# generate audio text map for WenetSpeech4TTS\n# evaluate for vocab size\n\nimport os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport json\nfrom concurrent.futures import ProcessPoolExecutor\nfrom importlib.resources import files\n\nimport torchaudio\nfrom datasets import Dataset\nfrom tqdm import tqdm\n\nfrom f5_tts.model.utils import convert_char_to_pinyin\n\n\ndef deal_with_sub_path_files(dataset_path, sub_path):","source_hash":"75f27f2a8632386e881a40a58c865011761c8d5f2cdf7d9fe0d2e97e2b896cb2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/train/datasets/prepare_emilia.py","uri":"program://DMOSpeech2/file/src/f5_tts/train/datasets/prepare_emilia.py","kind":"file","name":"src/f5_tts/train/datasets/prepare_emilia.py","path":"src/f5_tts/train/datasets/prepare_emilia.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# Emilia Dataset: https://huggingface.co/datasets/amphion/Emilia-Dataset/tree/fc71e07\n# if use updated new version, i.e. WebDataset, feel free to modify / draft your own script\n\n# generate audio text map for Emilia ZH & EN\n# evaluate for vocab size\n\nimport os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport json\nfrom concurrent.futures import ProcessPoolExecutor\nfrom importlib.resources import files\nfrom pathlib import Path\n\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm\n\nfrom f5_tts.model.utils import convert_char_to_pinyin, repetition_found","source_hash":"418b1ecd9af0922bb02a7bc11617b00dda6b918620148bd8a47df64df3c9ec4e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/train/datasets/prepare_csv_wavs.py","uri":"program://DMOSpeech2/file/src/f5_tts/train/datasets/prepare_csv_wavs.py","kind":"file","name":"src/f5_tts/train/datasets/prepare_csv_wavs.py","path":"src/f5_tts/train/datasets/prepare_csv_wavs.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import concurrent.futures\nimport multiprocessing\nimport os\nimport shutil\nimport signal\nimport subprocess # For invoking ffprobe\nimport sys\nfrom contextlib import contextmanager\n\n\nsys.path.append(os.getcwd())\n\nimport argparse\nimport csv\nimport json\nfrom importlib.resources import files\nfrom pathlib import Path\n\nimport torchaudio\nfrom datasets.arrow_writer import ArrowWriter\nfrom tqdm import tqdm","source_hash":"a3a4a9fe211097c49fdbdd87c343f81bb992e127d473b27a46a970336eaff9e0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model/trainer.py","uri":"program://DMOSpeech2/file/src/f5_tts/model/trainer.py","kind":"file","name":"src/f5_tts/model/trainer.py","path":"src/f5_tts/model/trainer.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"from __future__ import annotations\n\nimport gc\nimport os\n\nimport torch\nimport torchaudio\nimport wandb\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nfrom ema_pytorch import EMA\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model import CFM\nfrom f5_tts.model.dataset import DynamicBatchSampler, collate_fn\nfrom f5_tts.model.utils import default, exists\n\n# trainer","source_hash":"2b9b27e8647e8f0a32d4b60a3a0bb41365c787bea1168dc1099632ac1d210ea7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model/cfm.py","uri":"program://DMOSpeech2/file/src/f5_tts/model/cfm.py","kind":"file","name":"src/f5_tts/model/cfm.py","path":"src/f5_tts/model/cfm.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nfrom random import random\nfrom typing import Callable\n\nimport torch\nimport torch.nn.functional as F\nfrom torch import nn\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torchdiffeq import odeint\n\nfrom f5_tts.model.modules import MelSpec","source_hash":"2e833f1d5f202c84edd5752b8412699b4cd2c92ca78790de2b19a8f31515d079","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model/dataset.py","uri":"program://DMOSpeech2/file/src/f5_tts/model/dataset.py","kind":"file","name":"src/f5_tts/model/dataset.py","path":"src/f5_tts/model/dataset.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import re\nimport json\nimport random\nfrom importlib.resources import files\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom datasets import Dataset as Dataset_\nfrom datasets import load_from_disk\nfrom torch import nn\nfrom torch.utils.data import Dataset, Sampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import default\n\n\n\n\ndef get_speaker_id(path):","source_hash":"2ed7497956c0ce04b71926e3f330898dec956dea8a40d270c1682a03f8899787","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model/__init__.py","uri":"program://DMOSpeech2/file/src/f5_tts/model/__init__.py","kind":"file","name":"src/f5_tts/model/__init__.py","path":"src/f5_tts/model/__init__.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":10,"code":"from f5_tts.model.cfm import CFM\n\nfrom f5_tts.model.backbones.unett import UNetT\nfrom f5_tts.model.backbones.dit import DiT\nfrom f5_tts.model.backbones.mmdit import MMDiT\n\nfrom f5_tts.model.trainer import Trainer\n\n\n__all__ = [\"CFM\", \"UNetT\", \"DiT\", \"MMDiT\", \"Trainer\"]","source_hash":"b1fd6734ba5c90507acbb47f4b335dd7979cc84a9831ea2a1c2996caa26f834f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model/utils.py","uri":"program://DMOSpeech2/file/src/f5_tts/model/utils.py","kind":"file","name":"src/f5_tts/model/utils.py","path":"src/f5_tts/model/utils.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"from __future__ import annotations\n\nimport os\nimport random\nfrom collections import defaultdict\nfrom importlib.resources import files\n\nimport torch\nfrom torch.nn.utils.rnn import pad_sequence\n\nimport jieba\nfrom pypinyin import lazy_pinyin, Style\n\n\n# seed everything\n\n\ndef seed_everything(seed=0):\n random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n torch.manual_seed(seed)","source_hash":"01ede9247f1240f707cc31040a79d9f3ce65e91a46f7db6a69d24714bc11e0b2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model/modules.py","uri":"program://DMOSpeech2/file/src/f5_tts/model/modules.py","kind":"file","name":"src/f5_tts/model/modules.py","path":"src/f5_tts/model/modules.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nimport math\nfrom typing import Optional\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom librosa.filters import mel as librosa_mel_fn\nfrom torch import nn\nfrom x_transformers.x_transformers import apply_rotary_pos_emb\n","source_hash":"c5661f3dbf3cc9a9da5892786e1ca84814b40733e0c9fe86795c7ea4bad7f729","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model/backbones/dit.py","uri":"program://DMOSpeech2/file/src/f5_tts/model/backbones/dit.py","kind":"file","name":"src/f5_tts/model/backbones/dit.py","path":"src/f5_tts/model/backbones/dit.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nimport torch\nfrom torch import nn\nimport torch.nn.functional as F\n\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model.modules import (\n TimestepEmbedding,\n ConvNeXtV2Block,\n ConvPositionEmbedding,","source_hash":"4647e46533416eff5d28067e6dbaf5a86f7fec6b3a8466774fbaeb77ed0f432d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model/backbones/mmdit.py","uri":"program://DMOSpeech2/file/src/f5_tts/model/backbones/mmdit.py","kind":"file","name":"src/f5_tts/model/backbones/mmdit.py","path":"src/f5_tts/model/backbones/mmdit.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nimport torch\nfrom torch import nn\nimport torch.nn.functional as F\n\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model.modules import (\n TimestepEmbedding,\n ConvPositionEmbedding,\n MMDiTBlock,","source_hash":"127234f75c330a1063a9e8032ed491029c301ed85c79db764881b250600ebdb5","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model/backbones/unett.py","uri":"program://DMOSpeech2/file/src/f5_tts/model/backbones/unett.py","kind":"file","name":"src/f5_tts/model/backbones/unett.py","path":"src/f5_tts/model/backbones/unett.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\nfrom typing import Literal\n\nimport torch\nfrom torch import nn\nimport torch.nn.functional as F\n\nfrom x_transformers import RMSNorm\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model.modules import (\n TimestepEmbedding,","source_hash":"852e63fcd735376abb5572df733fb3d7ce64af8d3377c4930a0a4d02090be47a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model_new/trainer.py","uri":"program://DMOSpeech2/file/src/f5_tts/model_new/trainer.py","kind":"file","name":"src/f5_tts/model_new/trainer.py","path":"src/f5_tts/model_new/trainer.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"from __future__ import annotations\n\nimport gc\nimport math\nimport os\n\nimport torch\nimport torchaudio\nimport wandb\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nfrom ema_pytorch import EMA\nfrom torch.optim import AdamW\nfrom torch.optim.lr_scheduler import LinearLR, SequentialLR\nfrom torch.utils.data import DataLoader, Dataset, SequentialSampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model import CFM\nfrom f5_tts.model.dataset import DynamicBatchSampler, collate_fn\nfrom f5_tts.model.utils import default, exists\n","source_hash":"f920112a2433c8a5c92100766e845ef7990d488705ac80de568096e6900fb11c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model_new/cfm.py","uri":"program://DMOSpeech2/file/src/f5_tts/model_new/cfm.py","kind":"file","name":"src/f5_tts/model_new/cfm.py","path":"src/f5_tts/model_new/cfm.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nfrom random import random\nfrom typing import Callable\n\nimport torch\nimport torch.nn.functional as F\nfrom torch import nn\nfrom torch.nn.utils.rnn import pad_sequence\nfrom torchdiffeq import odeint\n\nfrom f5_tts.model_new.modules import MelSpec","source_hash":"99fd339581ab0096169d1183c45c099bd69d572e9343efc3d9f19614c3f8629c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model_new/dataset.py","uri":"program://DMOSpeech2/file/src/f5_tts/model_new/dataset.py","kind":"file","name":"src/f5_tts/model_new/dataset.py","path":"src/f5_tts/model_new/dataset.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import json\nfrom importlib.resources import files\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom datasets import Dataset as Dataset_\nfrom datasets import load_from_disk\nfrom torch import nn\nfrom torch.utils.data import Dataset, Sampler\nfrom tqdm import tqdm\n\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import default\n\n\nclass HFDataset(Dataset):\n def __init__(\n self,\n hf_dataset: Dataset,\n target_sample_rate=24_000,","source_hash":"5ddfadf6e712d32b7c55a455c51078896ebcd239a8e60bf2d70ac7033f14f697","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model_new/__init__.py","uri":"program://DMOSpeech2/file/src/f5_tts/model_new/__init__.py","kind":"file","name":"src/f5_tts/model_new/__init__.py","path":"src/f5_tts/model_new/__init__.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":8,"code":"from f5_tts.model_new.backbones.dit import DiT\nfrom f5_tts.model_new.backbones.mmdit import MMDiT\nfrom f5_tts.model_new.backbones.unett import UNetT\nfrom f5_tts.model_new.cfm import CFM\nfrom f5_tts.model_new.trainer import Trainer\n\n\n__all__ = [\"CFM\", \"UNetT\", \"DiT\", \"MMDiT\", \"Trainer\"]","source_hash":"5b6c184ca0e8937125a67e42ce35584aaa0c665d86ef54965cd283d78c3831f6","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model_new/utils.py","uri":"program://DMOSpeech2/file/src/f5_tts/model_new/utils.py","kind":"file","name":"src/f5_tts/model_new/utils.py","path":"src/f5_tts/model_new/utils.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"from __future__ import annotations\n\nimport os\nimport random\nfrom collections import defaultdict\nfrom importlib.resources import files\n\nimport jieba\nimport torch\nfrom pypinyin import Style, lazy_pinyin\nfrom torch.nn.utils.rnn import pad_sequence\n\n\n# seed everything\n\n\ndef seed_everything(seed=0):\n random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n torch.manual_seed(seed)\n torch.cuda.manual_seed(seed)","source_hash":"202c6959e7fdc709bf6028a073d53992879cc01a979aa25388e82f09e099bef7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model_new/modules.py","uri":"program://DMOSpeech2/file/src/f5_tts/model_new/modules.py","kind":"file","name":"src/f5_tts/model_new/modules.py","path":"src/f5_tts/model_new/modules.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n# flake8: noqa\n\nfrom __future__ import annotations\n\nimport math\nfrom typing import Optional\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom librosa.filters import mel as librosa_mel_fn\nfrom torch import nn\nfrom x_transformers.x_transformers import apply_rotary_pos_emb","source_hash":"345a04e47434925dc115c7aaeb14adccaa62c38073172ceba1d971889c6408f9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model_new/backbones/dit.py","uri":"program://DMOSpeech2/file/src/f5_tts/model_new/backbones/dit.py","kind":"file","name":"src/f5_tts/model_new/backbones/dit.py","path":"src/f5_tts/model_new/backbones/dit.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nimport torch\nimport torch.nn.functional as F\nfrom torch import nn\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model_new.modules import (\n AdaLayerNorm_Final,\n ConvNeXtV2Block,\n ConvPositionEmbedding,\n DiTBlock,","source_hash":"f6bfffce2c45a7042aa1d60f709b909db7dafdd95ac4425bf7fc2ee4eb75840d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model_new/backbones/mmdit.py","uri":"program://DMOSpeech2/file/src/f5_tts/model_new/backbones/mmdit.py","kind":"file","name":"src/f5_tts/model_new/backbones/mmdit.py","path":"src/f5_tts/model_new/backbones/mmdit.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nimport torch\nfrom torch import nn\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model_new.modules import (\n AdaLayerNorm_Final,\n ConvPositionEmbedding,\n MMDiTBlock,\n TimestepEmbedding,\n get_pos_embed_indices,","source_hash":"95a35a246150d327f79a341648e833a6dbfecf112134ffbaaff6cefa97322e94","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/model_new/backbones/unett.py","uri":"program://DMOSpeech2/file/src/f5_tts/model_new/backbones/unett.py","kind":"file","name":"src/f5_tts/model_new/backbones/unett.py","path":"src/f5_tts/model_new/backbones/unett.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"\"\"\"\nein notation:\nb - batch\nn - sequence\nnt - text sequence\nnw - raw wave length\nd - dimension\n\"\"\"\n\nfrom __future__ import annotations\n\nfrom typing import Literal\n\nimport torch\nimport torch.nn.functional as F\nfrom torch import nn\nfrom x_transformers import RMSNorm\nfrom x_transformers.x_transformers import RotaryEmbedding\n\nfrom f5_tts.model_new.modules import (\n Attention,","source_hash":"108eedf2ce95cd9d8d8bdce990e98db27e089a47f9e147fb7ee3c620ae41ea50","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/eval/eval_infer_batch.sh","uri":"program://DMOSpeech2/file/src/f5_tts/eval/eval_infer_batch.sh","kind":"file","name":"src/f5_tts/eval/eval_infer_batch.sh","path":"src/f5_tts/eval/eval_infer_batch.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport argparse\nimport time\nfrom importlib.resources import files\n\nimport torch\nimport torchaudio\nfrom accelerate import Accelerator\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\nfrom tqdm import tqdm\n\nfrom f5_tts.eval.utils_eval import (\n get_inference_prompt,\n get_librispeech_test_clean_metainfo,\n get_seedtts_testset_metainfo,","source_hash":"6d5a9ac66798e903d73dc92e658312aa783796fc5cc79bac32bf9d0bc450f353","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/eval/utils_eval.py","uri":"program://DMOSpeech2/file/src/f5_tts/eval/utils_eval.py","kind":"file","name":"src/f5_tts/eval/utils_eval.py","path":"src/f5_tts/eval/utils_eval.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import math\nimport os\nimport random\nimport string\nfrom pathlib import Path\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom tqdm import tqdm\n\nfrom f5_tts.eval.ecapa_tdnn import ECAPA_TDNN_SMALL\nfrom f5_tts.model.modules import MelSpec\nfrom f5_tts.model.utils import convert_char_to_pinyin\n\n\n# seedtts testset metainfo: utt, prompt_text, prompt_wav, gt_text, gt_wav\ndef get_seedtts_testset_metainfo(metalst):\n f = open(metalst)\n lines = f.readlines()\n f.close()","source_hash":"5ca3e0c70d7143d3bf5f23182499b82ddd4bba026d76b9a99d003a65136d7b68","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/eval/eval_infer_batch.py","uri":"program://DMOSpeech2/file/src/f5_tts/eval/eval_infer_batch.py","kind":"file","name":"src/f5_tts/eval/eval_infer_batch.py","path":"src/f5_tts/eval/eval_infer_batch.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport argparse\nimport time\nfrom importlib.resources import files\n\nimport torch\nimport torchaudio\nfrom accelerate import Accelerator\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\nfrom tqdm import tqdm\n\nfrom f5_tts.eval.utils_eval import (\n get_inference_prompt,\n get_librispeech_test_clean_metainfo,\n get_seedtts_testset_metainfo,","source_hash":"6d5a9ac66798e903d73dc92e658312aa783796fc5cc79bac32bf9d0bc450f353","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/eval/ecapa_tdnn.py","uri":"program://DMOSpeech2/file/src/f5_tts/eval/ecapa_tdnn.py","kind":"file","name":"src/f5_tts/eval/ecapa_tdnn.py","path":"src/f5_tts/eval/ecapa_tdnn.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# just for speaker similarity evaluation, third-party code\n\n# From https://github.com/microsoft/UniSpeech/blob/main/downstreams/speaker_verification/models/\n# part of the code is borrowed from https://github.com/lawlict/ECAPA-TDNN\n\nimport os\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\n\n\n\"\"\" Res2Conv1d + BatchNorm1d + ReLU\n\"\"\"\n\n\nclass Res2Conv1dReluBn(nn.Module):\n \"\"\"\n in_channels == out_channels == channels\n \"\"\"\n","source_hash":"3d934824932b71e6ba3824c817367328c5f099335106dc03e6ba68a143da6a0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/eval/eval_seedtts_testset.py","uri":"program://DMOSpeech2/file/src/f5_tts/eval/eval_seedtts_testset.py","kind":"file","name":"src/f5_tts/eval/eval_seedtts_testset.py","path":"src/f5_tts/eval/eval_seedtts_testset.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# Evaluate with Seed-TTS testset\n\nimport argparse\nimport json\nimport os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport multiprocessing as mp\nfrom importlib.resources import files\n\nimport numpy as np\n\nfrom f5_tts.eval.utils_eval import get_seed_tts_test, run_asr_wer, run_sim\n\n\nrel_path = str(files(\"f5_tts\").joinpath(\"../../\"))\n\n","source_hash":"25fd3663207e4448631d3366eec65dae50c3d77def18467925fabd6db2b7a5b0","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/eval/eval_librispeech_test_clean.py","uri":"program://DMOSpeech2/file/src/f5_tts/eval/eval_librispeech_test_clean.py","kind":"file","name":"src/f5_tts/eval/eval_librispeech_test_clean.py","path":"src/f5_tts/eval/eval_librispeech_test_clean.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# Evaluate with Librispeech test-clean, ~3s prompt to generate 4-10s audio (the way of valle/voicebox evaluation)\n\nimport argparse\nimport json\nimport os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport multiprocessing as mp\nfrom importlib.resources import files\n\nimport numpy as np\n\nfrom f5_tts.eval.utils_eval import get_librispeech_test, run_asr_wer, run_sim\n\n\nrel_path = str(files(\"f5_tts\").joinpath(\"../../\"))\n\n","source_hash":"9f5b078ca2dbd7ba9c1f0cbfa70a2ce84f990a8e2db8669ad6aef0af0c4f7199","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/eval/eval_utmos.py","uri":"program://DMOSpeech2/file/src/f5_tts/eval/eval_utmos.py","kind":"file","name":"src/f5_tts/eval/eval_utmos.py","path":"src/f5_tts/eval/eval_utmos.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import argparse\nimport json\nfrom pathlib import Path\n\nimport librosa\nimport torch\nfrom tqdm import tqdm\n\n\ndef main():\n parser = argparse.ArgumentParser(description=\"UTMOS Evaluation\")\n parser.add_argument(\"--audio_dir\", type=str, required=True, help=\"Audio file path.\")\n parser.add_argument(\"--ext\", type=str, default=\"wav\", help=\"Audio extension.\")\n args = parser.parse_args()\n\n device = \"cuda\" if torch.cuda.is_available() else \"xpu\" if torch.xpu.is_available() else \"cpu\"\n\n predictor = torch.hub.load(\"tarepan/SpeechMOS:v1.2.0\", \"utmos22_strong\", trust_repo=True)\n predictor = predictor.to(device)\n\n audio_paths = list(Path(args.audio_dir).rglob(f\"*.{args.ext}\"))","source_hash":"852c55bf0b26006faf1ab2a3f74695dd071058944723bdde2b378f85e3a5b6df","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/runtime/triton_trtllm/benchmark.py","uri":"program://DMOSpeech2/file/src/f5_tts/runtime/triton_trtllm/benchmark.py","kind":"file","name":"src/f5_tts/runtime/triton_trtllm/benchmark.py","path":"src/f5_tts/runtime/triton_trtllm/benchmark.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# Copyright (c) 2024 Tsinghua Univ. (authors: Xingchen Song)\n# 2025 （authors: Yuekai Zhang）\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# Modified from https://github.com/xingchensong/S3Tokenizer/blob/main/s3tokenizer/cli.py\n\"\"\" Example Usage\ntorchrun --nproc_per_node=1 \\\nbenchmark.py --output-dir $log_dir \\\n--batch-size $batch_size \\\n--enable-warmup \\\n--split-name $split_name \\","source_hash":"e6ecb3a919bbab27765e4cf2e5a3a0a582c9d21a7a9305b525a3c4185598551e","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/runtime/triton_trtllm/client_grpc.py","uri":"program://DMOSpeech2/file/src/f5_tts/runtime/triton_trtllm/client_grpc.py","kind":"file","name":"src/f5_tts/runtime/triton_trtllm/client_grpc.py","path":"src/f5_tts/runtime/triton_trtllm/client_grpc.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"#!/usr/bin/env python3\n# Copyright 2022 Xiaomi Corp. (authors: Fangjun Kuang)\n# 2023 Nvidia (authors: Yuekai Zhang)\n# 2023 Recurrent.ai (authors: Songtao Shi)\n# See LICENSE for clarification regarding multiple authors\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\"\"\"\nThis script supports to load dataset from huggingface and sends it to the server\nfor decoding, in parallel.\n","source_hash":"feddef6c77b73f44f419571ae75dfae5dec9de0f58da03678cf071a7dc93ddc4","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/runtime/triton_trtllm/client_http.py","uri":"program://DMOSpeech2/file/src/f5_tts/runtime/triton_trtllm/client_http.py","kind":"file","name":"src/f5_tts/runtime/triton_trtllm/client_http.py","path":"src/f5_tts/runtime/triton_trtllm/client_http.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# Copyright 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.\n#\n# Redistribution and use in source and binary forms, with or without\n# modification, are permitted provided that the following conditions\n# are met:\n# * Redistributions of source code must retain the above copyright\n# notice, this list of conditions and the following disclaimer.\n# * Redistributions in binary form must reproduce the above copyright\n# notice, this list of conditions and the following disclaimer in the\n# documentation and/or other materials provided with the distribution.\n# * Neither the name of NVIDIA CORPORATION nor the names of its\n# contributors may be used to endorse or promote products derived\n# from this software without specific prior written permission.\n#\n# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY\n# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE\n# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR\n# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR\n# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,\n# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,\n# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR","source_hash":"d1af65a41400201a85a18404e74cb456991550d79a17c42c106c37be2a18e39f","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","uri":"program://DMOSpeech2/file/src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","kind":"file","name":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/f5_tts_trtllm.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import math\nimport os\nimport time\nfrom functools import wraps\nfrom typing import List, Optional\n\nimport tensorrt as trt\nimport tensorrt_llm\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nfrom tensorrt_llm._utils import str_dtype_to_torch, trt_dtype_to_torch\nfrom tensorrt_llm.logger import logger\nfrom tensorrt_llm.runtime.session import Session\n\n\ndef remove_tensor_padding(input_tensor, input_tensor_lengths=None):\n # Audio tensor case: batch, seq_len, feature_len\n # position_ids case: batch, seq_len\n assert input_tensor_lengths is not None, \"input_tensor_lengths must be provided for 3D input_tensor\"\n","source_hash":"a3bf7d960e10cb304862a69bc653723f5b6a098d75b940208bdcb26ca5548ae7","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","uri":"program://DMOSpeech2/file/src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","kind":"file","name":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","path":"src/f5_tts/runtime/triton_trtllm/model_repo_f5_tts/f5_tts/1/model.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# Copyright 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.\n#\n# Redistribution and use in source and binary forms, with or without\n# modification, are permitted provided that the following conditions\n# are met:\n# * Redistributions of source code must retain the above copyright\n# notice, this list of conditions and the following disclaimer.\n# * Redistributions in binary form must reproduce the above copyright\n# notice, this list of conditions and the following disclaimer in the\n# documentation and/or other materials provided with the distribution.\n# * Neither the name of NVIDIA CORPORATION nor the names of its\n# contributors may be used to endorse or promote products derived\n# from this software without specific prior written permission.\n#\n# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY\n# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE\n# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR\n# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR\n# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,\n# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,\n# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR","source_hash":"36b4bb696585c91cc596932fded5339316d7a0af0d39c9bbdeceb42a7b8f786c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/runtime/triton_trtllm/patch/__init__.py","uri":"program://DMOSpeech2/file/src/f5_tts/runtime/triton_trtllm/patch/__init__.py","kind":"file","name":"src/f5_tts/runtime/triton_trtllm/patch/__init__.py","path":"src/f5_tts/runtime/triton_trtllm/patch/__init__.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.\n# SPDX-License-Identifier: Apache-2.0\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\nfrom .baichuan.model import BaichuanForCausalLM\nfrom .bert.model import (\n BertForQuestionAnswering,\n BertForSequenceClassification,\n BertModel,\n RobertaForQuestionAnswering,\n RobertaForSequenceClassification,","source_hash":"3646bc4561f11285ed5c20c5f10458d5f6086db6062bfeb9adb6889f560a566a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/runtime/triton_trtllm/patch/f5tts/model.py","uri":"program://DMOSpeech2/file/src/f5_tts/runtime/triton_trtllm/patch/f5tts/model.py","kind":"file","name":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/model.py","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/model.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"from __future__ import annotations\n\nimport os\nimport sys\nfrom collections import OrderedDict\n\nimport tensorrt as trt\nfrom tensorrt_llm._common import default_net\n\nfrom ..._utils import str_dtype_to_trt\nfrom ...functional import Tensor, concat\nfrom ...layers import Linear\nfrom ...module import Module, ModuleList\nfrom ...plugin import current_all_reduce_helper\nfrom ..modeling_utils import PretrainedConfig, PretrainedModel\nfrom .modules import AdaLayerNormZero_Final, ConvPositionEmbedding, DiTBlock, TimestepEmbedding\n\n\ncurrent_file_path = os.path.abspath(__file__)\nparent_dir = os.path.dirname(current_file_path)\nsys.path.append(parent_dir)","source_hash":"d67c74e53ee7cb463f0df30b2b0461c72b3535beee99192466313eb247dc4c9c","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","uri":"program://DMOSpeech2/file/src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","kind":"file","name":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","path":"src/f5_tts/runtime/triton_trtllm/patch/f5tts/modules.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"from __future__ import annotations\n\nimport math\nfrom typing import Optional\n\nimport numpy as np\nimport torch\nimport torch.nn.functional as F\nfrom tensorrt_llm._common import default_net\n\nfrom ..._utils import str_dtype_to_trt, trt_dtype_to_np\nfrom ...functional import (\n Tensor,\n bert_attention,\n cast,\n chunk,\n concat,\n constant,\n expand,\n expand_dims,\n expand_dims_like,","source_hash":"c6bfcc23fd0f3b3d57aae6b7c61f61a1fff64548204ec2739f43b0ac1f37021d","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/runtime/triton_trtllm/scripts/conv_stft.py","uri":"program://DMOSpeech2/file/src/f5_tts/runtime/triton_trtllm/scripts/conv_stft.py","kind":"file","name":"src/f5_tts/runtime/triton_trtllm/scripts/conv_stft.py","path":"src/f5_tts/runtime/triton_trtllm/scripts/conv_stft.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# Modified from https://github.com/echocatzh/conv-stft/blob/master/conv_stft/conv_stft.py\n\n# Copyright (c) 2024, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\n# MIT License\n\n# Copyright (c) 2020 Shimin Zhang\n\n# Permission is hereby granted, free of charge, to any person obtaining a copy","source_hash":"9284255817c07d01ef0c1f72ccdcdc6f0fd6cb65a00ba9170b0bfa46f2c614d2","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/runtime/triton_trtllm/scripts/export_vocoder_to_onnx.py","uri":"program://DMOSpeech2/file/src/f5_tts/runtime/triton_trtllm/scripts/export_vocoder_to_onnx.py","kind":"file","name":"src/f5_tts/runtime/triton_trtllm/scripts/export_vocoder_to_onnx.py","path":"src/f5_tts/runtime/triton_trtllm/scripts/export_vocoder_to_onnx.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# Copyright (c) 2024, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nimport argparse\n\nimport torch\nimport torch.nn as nn\nfrom conv_stft import STFT\nfrom huggingface_hub import hf_hub_download\nfrom vocos import Vocos","source_hash":"72c9953957b8be407255e0642519f0817158c4fdb5550dc8ccc57b124828bd98","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","uri":"program://DMOSpeech2/file/src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","kind":"file","name":"src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","path":"src/f5_tts/runtime/triton_trtllm/scripts/convert_checkpoint.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import argparse\nimport json\nimport os\nimport re\nimport time\nimport traceback\nfrom concurrent.futures import ThreadPoolExecutor, as_completed\n\nimport safetensors.torch\nimport torch\nfrom tensorrt_llm import str_dtype_to_torch\nfrom tensorrt_llm.mapping import Mapping\nfrom tensorrt_llm.models.convert_utils import split, split_matrix_tp\n\n\ndef split_q_tp(v, n_head, n_hidden, tensor_parallel, rank):\n split_v = split(v, tensor_parallel, rank, dim=1)\n return split_v.contiguous()\n\n\ndef split_q_bias_tp(v, n_head, n_hidden, tensor_parallel, rank):","source_hash":"453fc4ae05851022942ced5d7df412f8d66f1d0fe85850978205a2c35becd32b","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/runtime/triton_trtllm/scripts/fill_template.py","uri":"program://DMOSpeech2/file/src/f5_tts/runtime/triton_trtllm/scripts/fill_template.py","kind":"file","name":"src/f5_tts/runtime/triton_trtllm/scripts/fill_template.py","path":"src/f5_tts/runtime/triton_trtllm/scripts/fill_template.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"#! /usr/bin/env python3\nfrom argparse import ArgumentParser\nfrom string import Template\n\n\ndef main(file_path, substitutions, in_place, participant_ids):\n with open(file_path) as f:\n pbtxt = Template(f.read())\n\n sub_dict = {\"max_queue_size\": 0}\n sub_dict[\"participant_ids\"] = participant_ids\n for sub in substitutions.split(\",\"):\n key, value = sub.split(\":\")\n sub_dict[key] = value\n\n pbtxt = pbtxt.safe_substitute(sub_dict)\n\n if in_place:\n with open(file_path, \"w\") as f:\n f.write(pbtxt)\n else:","source_hash":"2341448e22389ebd542a52552ba5c093ebd6096c2cc7d5e00c9c598e7b062654","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/scripts/count_max_epoch.py","uri":"program://DMOSpeech2/file/src/f5_tts/scripts/count_max_epoch.py","kind":"file","name":"src/f5_tts/scripts/count_max_epoch.py","path":"src/f5_tts/scripts/count_max_epoch.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"\"\"\"ADAPTIVE BATCH SIZE\"\"\"\n\nprint(\"Adaptive batch size: using grouping batch sampler, frames_per_gpu fixed fed in\")\nprint(\" -> least padding, gather wavs with accumulated frames in a batch\\n\")\n\n# data\ntotal_hours = 95282\nmel_hop_length = 256\nmel_sampling_rate = 24000\n\n# target\nwanted_max_updates = 1200000\n\n# train params\ngpus = 8\nframes_per_gpu = 38400 # 8 * 38400 = 307200\ngrad_accum = 1\n\n# intermediate\nmini_batch_frames = frames_per_gpu * grad_accum * gpus\nmini_batch_hours = mini_batch_frames * mel_hop_length / mel_sampling_rate / 3600","source_hash":"43c0b6aa0a0a924397c53b71cb0b00eac662736f7daa94790e413147a8bf05de","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/scripts/count_params_gflops.py","uri":"program://DMOSpeech2/file/src/f5_tts/scripts/count_params_gflops.py","kind":"file","name":"src/f5_tts/scripts/count_params_gflops.py","path":"src/f5_tts/scripts/count_params_gflops.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import os\nimport sys\n\n\nsys.path.append(os.getcwd())\n\nimport thop\nimport torch\n\nfrom f5_tts.model import CFM, DiT\n\n\n\"\"\" ~155M \"\"\"\n# transformer = UNetT(dim = 768, depth = 20, heads = 12, ff_mult = 4)\n# transformer = UNetT(dim = 768, depth = 20, heads = 12, ff_mult = 4, text_dim = 512, conv_layers = 4)\n# transformer = DiT(dim = 768, depth = 18, heads = 12, ff_mult = 2)\n# transformer = DiT(dim = 768, depth = 18, heads = 12, ff_mult = 2, text_dim = 512, conv_layers = 4)\n# transformer = DiT(dim = 768, depth = 18, heads = 12, ff_mult = 2, text_dim = 512, conv_layers = 4, long_skip_connection = True)\n# transformer = MMDiT(dim = 512, depth = 16, heads = 16, ff_mult = 2)\n\n\"\"\" ~335M \"\"\"","source_hash":"ea250747154d3900b617f90c24c9447c9c992397a3b6e08f774bb84a21846af9","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/infer/utils_infer.py","uri":"program://DMOSpeech2/file/src/f5_tts/infer/utils_infer.py","kind":"file","name":"src/f5_tts/infer/utils_infer.py","path":"src/f5_tts/infer/utils_infer.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# A unified script for inference process\n# Make adjustments inside functions, and consider both gradio and cli scripts if need to change func output format\nimport os\nimport sys\nfrom concurrent.futures import ThreadPoolExecutor\n\n\nos.environ[\"PYTORCH_ENABLE_MPS_FALLBACK\"] = \"1\" # for MPS device compatibility\nsys.path.append(f\"{os.path.dirname(os.path.abspath(__file__))}/../../third_party/BigVGAN/\")\n\nimport hashlib\nimport re\nimport tempfile\nfrom importlib.resources import files\n\nimport matplotlib\n\n\nmatplotlib.use(\"Agg\")\n\nimport matplotlib.pylab as plt","source_hash":"a2ad94c6e0a5174d628d1318f351bb6b57467ce8baee76b256213a7385e9cb0a","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/infer/speech_edit.py","uri":"program://DMOSpeech2/file/src/f5_tts/infer/speech_edit.py","kind":"file","name":"src/f5_tts/infer/speech_edit.py","path":"src/f5_tts/infer/speech_edit.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import os\n\n\nos.environ[\"PYTORCH_ENABLE_MPS_FALLBACK\"] = \"1\" # for MPS device compatibility\n\nfrom importlib.resources import files\n\nimport torch\nimport torch.nn.functional as F\nimport torchaudio\nfrom cached_path import cached_path\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\n\nfrom f5_tts.infer.utils_infer import load_checkpoint, load_vocoder, save_spectrogram\nfrom f5_tts.model import CFM\nfrom f5_tts.model.utils import convert_char_to_pinyin, get_tokenizer\n\n\ndevice = (\n \"cuda\"","source_hash":"b91b2f03f44460701eea899837e71da92b4d92a6bf44f4e56e5f1eacd86e6540","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/infer/infer_cli.py","uri":"program://DMOSpeech2/file/src/f5_tts/infer/infer_cli.py","kind":"file","name":"src/f5_tts/infer/infer_cli.py","path":"src/f5_tts/infer/infer_cli.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"import argparse\nimport codecs\nimport os\nimport re\nfrom datetime import datetime\nfrom importlib.resources import files\nfrom pathlib import Path\n\nimport numpy as np\nimport soundfile as sf\nimport tomli\nfrom cached_path import cached_path\nfrom hydra.utils import get_class\nfrom omegaconf import OmegaConf\nfrom unidecode import unidecode\n\nfrom f5_tts.infer.utils_infer import (\n cfg_strength,\n cross_fade_duration,\n device,\n fix_duration,","source_hash":"5d9b324c18a7021852710940ce4e98cece7c097b2a96d66a9417339fa7dda25b","truncated":false} {"repo_id":"DMOSpeech2","entity_id":"file:src/f5_tts/infer/infer_gradio.py","uri":"program://DMOSpeech2/file/src/f5_tts/infer/infer_gradio.py","kind":"file","name":"src/f5_tts/infer/infer_gradio.py","path":"src/f5_tts/infer/infer_gradio.py","language":"python","start_line":1,"end_line":1,"context_start_line":1,"context_end_line":21,"code":"# ruff: noqa: E402\n# Above allows ruff to ignore E402: module level import not at top of file\n\nimport gc\nimport json\nimport os\nimport re\nimport tempfile\nfrom collections import OrderedDict\nfrom functools import lru_cache\nfrom importlib.resources import files\n\nimport click\nimport gradio as gr\nimport numpy as np\nimport soundfile as sf\nimport torch\nimport torchaudio\nfrom cached_path import cached_path\nfrom transformers import AutoModelForCausalLM, AutoTokenizer\n","source_hash":"1d4a5107f0b04e64058ca79f01cb79bcd2b9f0068c94315bba0d22ae5c056383","truncated":false}