File size: 2,523 Bytes
11b4f27 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 | import torch
import torch.nn as nn
import torch.optim as optim
from .qwen import LLM, topk_sampling
import pytorch_lightning.core.module as pl
import matplotlib.pyplot as plt
plt.switch_backend('agg')
class Wav2TTS(pl.LightningModule):
def __init__(self, hp):
super().__init__()
self.model = LLM(audio_token_size=hp.vocab_size)
self.lr = hp.lr
def configure_optimizers(self):
print('train with lr ', self.lr)
optimizer = optim.AdamW([*self.parameters()], lr=self.lr, betas=(0.9, 0.95), weight_decay=0.1)
return optimizer
def training_step(self, batch, batch_idx):
acc, loss = self.step(batch)
self.log("train/loss", loss, on_step=True, prog_bar=True)
# self.log("train/quantize_loss", quantize_loss, on_step=True, prog_bar=True)
self.log("train/acc", acc, on_step=True, prog_bar=True)
return loss
def validation_step(self, batch, batch_idx):
acc, loss = self.step(batch)
self.log("val/loss", loss, on_step=True, prog_bar=True)
self.log("val/acc", acc, on_step=True, prog_bar=True)
def step(self, batch):
# speaker_embedding = self.spk_proj(batch['spk_emb'])
# speaker_embedding, commitment_loss, codebook_loss, indices, z_e = self.spk_vq(speaker_embedding.unsqueeze(-1))
# quantize_loss = codebook_loss + 0.25*commitment_loss
# speaker_embedding = speaker_embedding.squeeze(-1)
# speaker_embedding = self.spk_emb(batch['spk_emb'])
# Deal with phone segments
phone_features = batch['phone']
lengths = batch['length']
bert_feature = batch['bert_feature']
speaker_embedding = batch['spk_emb']
target = batch['quantization']
output = self.model(phone_features, bert_feature.transpose(1, 2), speaker_embedding, target, lengths)
loss = output['loss']
acc = output['acc']
return acc, loss
@torch.inference_mode()
@torch.no_grad()
def inference(self, phonemes, bert_feature, spk_emb, prompt=None):
# speaker_embedding = self.spk_proj(spk_emb)
# speaker_embedding, commitment_loss, codebook_loss, indices, z_e = self.spk_vq(speaker_embedding.unsqueeze(-1))
# speaker_embedding = speaker_embedding.squeeze(-1)
# print('quantized spk code:', indices)
output = self.model.inference(phonemes, bert_feature.transpose(1, 2), spk_emb, prompt, sampling_method=topk_sampling,sampling_kwargs={'top_k': 10})
return output |