File size: 6,247 Bytes
a257816 |
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 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 |
import logging
import random
from typing import Dict, Optional
import torch
import torch.nn as nn
from torch.nn import functional as F
from omegaconf import DictConfig
from VietTTS.utils.mask import make_pad_mask
class MaskedDiffWithXvec(torch.nn.Module):
def __init__(self,
input_size: int = 512,
output_size: int = 80,
spk_embed_dim: int = 192,
output_type: str = "mel",
vocab_size: int = 4096,
input_frame_rate: int = 50,
only_mask_loss: bool = True,
encoder: torch.nn.Module = None,
length_regulator: torch.nn.Module = None,
decoder: torch.nn.Module = None,
decoder_conf: Dict = {
'in_channels': 240,
'out_channel': 80,
'spk_emb_dim': 80,
'n_spks': 1,
'cfm_params': DictConfig({
'sigma_min': 1e-06,
'solver': 'euler',
't_scheduler': 'cosine',
'training_cfg_rate': 0.2,
'inference_cfg_rate': 0.7,
'reg_loss_type': 'l1'
}),
'decoder_params': {
'channels': [256, 256],
'dropout': 0.0,
'attention_head_dim': 64,
'n_blocks': 4,
'num_mid_blocks': 12,
'num_heads': 8,
'act_fn': 'gelu'
}
},
mel_feat_conf: Dict = {
'n_fft': 1024,
'num_mels': 80,
'sampling_rate': 22050,
'hop_size': 256,
'win_size': 1024,
'fmin': 0,
'fmax': 8000
}
):
super().__init__()
self.input_size = input_size
self.output_size = output_size
self.decoder_conf = decoder_conf
self.mel_feat_conf = mel_feat_conf
self.vocab_size = vocab_size
self.output_type = output_type
self.input_frame_rate = input_frame_rate
logging.info(f"input frame rate={self.input_frame_rate}")
self.input_embedding = nn.Embedding(vocab_size, input_size)
self.spk_embed_affine_layer = torch.nn.Linear(spk_embed_dim, output_size)
self.encoder = encoder
self.encoder_proj = torch.nn.Linear(self.encoder.output_size(), output_size)
self.decoder = decoder
self.length_regulator = length_regulator
self.only_mask_loss = only_mask_loss
def forward(
self,
batch: dict,
device: torch.device,
) -> Dict[str, Optional[torch.Tensor]]:
token = batch['speech_token'].to(device)
token_len = batch['speech_token_len'].to(device)
feat = batch['speech_feat'].to(device)
feat_len = batch['speech_feat_len'].to(device)
embedding = batch['embedding'].to(device)
# xvec projection
embedding = F.normalize(embedding, dim=1)
embedding = self.spk_embed_affine_layer(embedding)
# concat text and prompt_text
mask = (~make_pad_mask(token_len)).float().unsqueeze(-1).to(device)
token = self.input_embedding(torch.clamp(token, min=0)) * mask
# text encode
h, h_lengths = self.encoder(token, token_len)
h = self.encoder_proj(h)
h, h_lengths = self.length_regulator(h, feat_len)
# get conditions
conds = torch.zeros(feat.shape, device=token.device)
for i, j in enumerate(feat_len):
if random.random() < 0.5:
continue
index = random.randint(0, int(0.3 * j))
conds[i, :index] = feat[i, :index]
conds = conds.transpose(1, 2)
mask = (~make_pad_mask(feat_len)).to(h)
feat = F.interpolate(feat.unsqueeze(dim=1), size=h.shape[1:], mode="nearest").squeeze(dim=1)
loss, _ = self.decoder.compute_loss(
feat.transpose(1, 2).contiguous(),
mask.unsqueeze(1),
h.transpose(1, 2).contiguous(),
embedding,
cond=conds
)
return {'loss': loss}
@torch.inference_mode()
def inference(self,
token,
token_len,
prompt_token,
prompt_token_len,
prompt_feat,
prompt_feat_len,
embedding):
assert token.shape[0] == 1
# xvec projection
embedding = F.normalize(embedding, dim=1)
embedding = self.spk_embed_affine_layer(embedding)
# concat text and prompt_text
token_len1, token_len2 = prompt_token.shape[1], token.shape[1]
token, token_len = torch.concat([prompt_token, token], dim=1), prompt_token_len + token_len
mask = (~make_pad_mask(token_len)).unsqueeze(-1).to(embedding)
token = self.input_embedding(torch.clamp(token, min=0)) * mask
# text encode
h, h_lengths = self.encoder(token, token_len)
h = self.encoder_proj(h)
mel_len1, mel_len2 = prompt_feat.shape[1], int(token_len2 / self.input_frame_rate * 22050 / 256)
h, h_lengths = self.length_regulator.inference(h[:, :token_len1], h[:, token_len1:], mel_len1, mel_len2, self.input_frame_rate)
# get conditions
conds = torch.zeros([1, mel_len1 + mel_len2, self.output_size], device=token.device)
conds[:, :mel_len1] = prompt_feat
conds = conds.transpose(1, 2)
mask = (~make_pad_mask(torch.tensor([mel_len1 + mel_len2]))).to(h)
feat = self.decoder(
mu=h.transpose(1, 2).contiguous(),
mask=mask.unsqueeze(1),
spks=embedding,
cond=conds,
n_timesteps=10
)
feat = feat[:, :, mel_len1:]
assert feat.shape[2] == mel_len2
return feat
|