Upload 2 files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+reference.wav filter=lfs diff=lfs merge=lfs -text

F5TTS_Base.yaml

@@ -0,0 +1,52 @@
+hydra:
+  run:
+    dir: ckpts/${model.name}_${model.mel_spec.mel_spec_type}_${model.tokenizer}_${datasets.name}/${now:%Y-%m-%d}/${now:%H-%M-%S}
+
+datasets:
+  name: Emilia_ZH_EN  # dataset name
+  batch_size_per_gpu: 38400  # 8 GPUs, 8 * 38400 = 307200
+  batch_size_type: frame  # frame | sample
+  max_samples: 64  # max sequences per batch if use frame-wise batch_size. we set 32 for small models, 64 for base models
+  num_workers: 16
+
+optim:
+  epochs: 11
+  learning_rate: 7.5e-5
+  num_warmup_updates: 20000  # warmup updates
+  grad_accumulation_steps: 1  # note: updates = steps / grad_accumulation_steps
+  max_grad_norm: 1.0  # gradient clipping
+  bnb_optimizer: False  # use bnb 8bit AdamW optimizer or not
+
+model:
+  name: F5TTS_Base  # model name
+  tokenizer: pinyin  # tokenizer type
+  tokenizer_path: null  # if 'custom' tokenizer, define the path want to use (should be vocab.txt)
+  backbone: DiT
+  arch:
+    dim: 1024
+    depth: 22
+    heads: 16
+    ff_mult: 2
+    text_dim: 512
+    text_mask_padding: False
+    conv_layers: 4
+    pe_attn_head: 1
+    checkpoint_activations: False  # recompute activations and save memory for extra compute
+  mel_spec:
+    target_sample_rate: 24000
+    n_mel_channels: 100
+    hop_length: 256
+    win_length: 1024
+    n_fft: 1024
+    mel_spec_type: vocos  # vocos | bigvgan
+  vocoder:
+    is_local: False  # use local offline ckpt or not
+    local_path: null  # local vocoder path
+
+ckpts:
+  logger: wandb  # wandb | tensorboard | null
+  log_samples: True  # infer random sample per save checkpoint. wip, normal to fail with extra long samples
+  save_per_updates: 50000  # save checkpoint per updates
+  keep_last_n_checkpoints: -1  # -1 to keep all, 0 to not save intermediate, > 0 to keep last N checkpoints
+  last_per_updates: 5000  # save last checkpoint per updates
+  save_dir: ckpts/${model.name}_${model.mel_spec.mel_spec_type}_${model.tokenizer}_${datasets.name}

reference.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d6db1e038c67df75cdde9ad1e43ba05f660eebc9346a30617d9b2f3892a5b201
+size 1058478