coqui-ai
/
TTS
mirror of https://github.com/coqui-ai/TTS.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

update Tacotron models for the trainer

pull/506/head
Eren Gölge 2021-05-25 14:36:06 +02:00
parent bdbfc95618
commit 535a458f40
4 changed files with 373 additions and 160 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
TTS/tts/configs/tacotron_config.py
View File

@ -126,6 +126,7 @@ class TacotronConfig(BaseTTSConfig):
use_gst: bool = False use_gst: bool = False
gst: GSTConfig = None gst: GSTConfig = None
gst_style_input: str = None gst_style_input: str = None
# model specific params # model specific params
r: int = 2 r: int = 2
gradual_training: List[List[int]] = None gradual_training: List[List[int]] = None

198
TTS/tts/models/tacotron.py
View File

@ -113,7 +113,8 @@ class Tacotron(TacotronAbstract):
if self.num_speakers > 1: if self.num_speakers > 1:
if not self.embeddings_per_sample: if not self.embeddings_per_sample:
speaker_embedding_dim = 256 speaker_embedding_dim = 256
self.speaker_embedding = nn.Embedding(self.num_speakers, speaker_embedding_dim) self.speaker_embedding = nn.Embedding(self.num_speakers,
speaker_embedding_dim)
self.speaker_embedding.weight.data.normal_(0, 0.3) self.speaker_embedding.weight.data.normal_(0, 0.3)
# speaker and gst embeddings is concat in decoder input # speaker and gst embeddings is concat in decoder input
@ -144,7 +145,8 @@ class Tacotron(TacotronAbstract):
separate_stopnet, separate_stopnet,
) )
self.postnet = PostCBHG(decoder_output_dim) self.postnet = PostCBHG(decoder_output_dim)
self.last_linear = nn.Linear(self.postnet.cbhg.gru_features * 2, postnet_output_dim) self.last_linear = nn.Linear(self.postnet.cbhg.gru_features * 2,
postnet_output_dim)
# setup prenet dropout # setup prenet dropout
self.decoder.prenet.dropout_at_inference = prenet_dropout_at_inference self.decoder.prenet.dropout_at_inference = prenet_dropout_at_inference
@ -181,93 +183,203 @@ class Tacotron(TacotronAbstract):
separate_stopnet, separate_stopnet,
) )
def forward(self, characters, text_lengths, mel_specs, mel_lengths=None, speaker_ids=None, speaker_embeddings=None): def forward(self,
text,
text_lengths,
mel_specs=None,
mel_lengths=None,
cond_input=None):
""" """
Shapes: Shapes:
characters: [B, T_in] text: [B, T_in]
text_lengths: [B] text_lengths: [B]
mel_specs: [B, T_out, C] mel_specs: [B, T_out, C]
mel_lengths: [B] mel_lengths: [B]
speaker_ids: [B, 1] cond_input: 'speaker_ids': [B, 1] and 'x_vectors':[B, C]
speaker_embeddings: [B, C]
""" """
outputs = {
'alignments_backward': None,
'decoder_outputs_backward': None
}
input_mask, output_mask = self.compute_masks(text_lengths, mel_lengths) input_mask, output_mask = self.compute_masks(text_lengths, mel_lengths)
# B x T_in x embed_dim # B x T_in x embed_dim
inputs = self.embedding(characters) inputs = self.embedding(text)
# B x T_in x encoder_in_features # B x T_in x encoder_in_features
encoder_outputs = self.encoder(inputs) encoder_outputs = self.encoder(inputs)
# sequence masking # sequence masking
encoder_outputs = encoder_outputs * input_mask.unsqueeze(2).expand_as(encoder_outputs) encoder_outputs = encoder_outputs * input_mask.unsqueeze(2).expand_as(
encoder_outputs)
# global style token # global style token
if self.gst and self.use_gst: if self.gst and self.use_gst:
# B x gst_dim # B x gst_dim
encoder_outputs = self.compute_gst(encoder_outputs, mel_specs, speaker_embeddings) encoder_outputs = self.compute_gst(encoder_outputs, mel_specs,
cond_input['x_vectors'])
# speaker embedding # speaker embedding
if self.num_speakers > 1: if self.num_speakers > 1:
if not self.embeddings_per_sample: if not self.embeddings_per_sample:
# B x 1 x speaker_embed_dim # B x 1 x speaker_embed_dim
speaker_embeddings = self.speaker_embedding(speaker_ids)[:, None] speaker_embeddings = self.speaker_embedding(cond_input['speaker_ids'])[:,
None]
else: else:
# B x 1 x speaker_embed_dim # B x 1 x speaker_embed_dim
speaker_embeddings = torch.unsqueeze(speaker_embeddings, 1) speaker_embeddings = torch.unsqueeze(cond_input['x_vectors'], 1)
encoder_outputs = self._concat_speaker_embedding(encoder_outputs, speaker_embeddings) encoder_outputs = self._concat_speaker_embedding(
encoder_outputs, speaker_embeddings)
# decoder_outputs: B x decoder_in_features x T_out # decoder_outputs: B x decoder_in_features x T_out
# alignments: B x T_in x encoder_in_features # alignments: B x T_in x encoder_in_features
# stop_tokens: B x T_in # stop_tokens: B x T_in
decoder_outputs, alignments, stop_tokens = self.decoder(encoder_outputs, mel_specs, input_mask) decoder_outputs, alignments, stop_tokens = self.decoder(
encoder_outputs, mel_specs, input_mask)
# sequence masking # sequence masking
if output_mask is not None: if output_mask is not None:
decoder_outputs = decoder_outputs * output_mask.unsqueeze(1).expand_as(decoder_outputs) decoder_outputs = decoder_outputs * output_mask.unsqueeze(
1).expand_as(decoder_outputs)
# B x T_out x decoder_in_features # B x T_out x decoder_in_features
postnet_outputs = self.postnet(decoder_outputs) postnet_outputs = self.postnet(decoder_outputs)
# sequence masking # sequence masking
if output_mask is not None: if output_mask is not None:
postnet_outputs = postnet_outputs * output_mask.unsqueeze(2).expand_as(postnet_outputs) postnet_outputs = postnet_outputs * output_mask.unsqueeze(
2).expand_as(postnet_outputs)
# B x T_out x posnet_dim # B x T_out x posnet_dim
postnet_outputs = self.last_linear(postnet_outputs) postnet_outputs = self.last_linear(postnet_outputs)
# B x T_out x decoder_in_features # B x T_out x decoder_in_features
decoder_outputs = decoder_outputs.transpose(1, 2).contiguous() decoder_outputs = decoder_outputs.transpose(1, 2).contiguous()
if self.bidirectional_decoder: if self.bidirectional_decoder:
decoder_outputs_backward, alignments_backward = self._backward_pass(mel_specs, encoder_outputs, input_mask) decoder_outputs_backward, alignments_backward = self._backward_pass(
return ( mel_specs, encoder_outputs, input_mask)
decoder_outputs, outputs['alignments_backward'] = alignments_backward
postnet_outputs, outputs['decoder_outputs_backward'] = decoder_outputs_backward
alignments,
stop_tokens,
decoder_outputs_backward,
alignments_backward,
)
if self.double_decoder_consistency: if self.double_decoder_consistency:
decoder_outputs_backward, alignments_backward = self._coarse_decoder_pass( decoder_outputs_backward, alignments_backward = self._coarse_decoder_pass(
mel_specs, encoder_outputs, alignments, input_mask mel_specs, encoder_outputs, alignments, input_mask)
) outputs['alignments_backward'] = alignments_backward
return ( outputs['decoder_outputs_backward'] = decoder_outputs_backward
decoder_outputs, outputs.update({
postnet_outputs, 'postnet_outputs': postnet_outputs,
alignments, 'decoder_outputs': decoder_outputs,
stop_tokens, 'alignments': alignments,
decoder_outputs_backward, 'stop_tokens': stop_tokens
alignments_backward, })
) return outputs
return decoder_outputs, postnet_outputs, alignments, stop_tokens
@torch.no_grad() @torch.no_grad()
def inference(self, characters, speaker_ids=None, style_mel=None, speaker_embeddings=None): def inference(self,
inputs = self.embedding(characters) text_input,
cond_input=None):
inputs = self.embedding(text_input)
encoder_outputs = self.encoder(inputs) encoder_outputs = self.encoder(inputs)
if self.gst and self.use_gst: if self.gst and self.use_gst:
# B x gst_dim # B x gst_dim
encoder_outputs = self.compute_gst(encoder_outputs, style_mel, speaker_embeddings) encoder_outputs = self.compute_gst(encoder_outputs, cond_input['style_mel'],
cond_input['x_vectors'])
if self.num_speakers > 1: if self.num_speakers > 1:
if not self.embeddings_per_sample: if not self.embeddings_per_sample:
# B x 1 x speaker_embed_dim # B x 1 x speaker_embed_dim
speaker_embeddings = self.speaker_embedding(speaker_ids)[:, None] speaker_embeddings = self.speaker_embedding(cond_input['speaker_ids'])[:, None]
else: else:
# B x 1 x speaker_embed_dim # B x 1 x speaker_embed_dim
speaker_embeddings = torch.unsqueeze(speaker_embeddings, 1) speaker_embeddings = torch.unsqueeze(cond_input['x_vectors'], 1)
encoder_outputs = self._concat_speaker_embedding(encoder_outputs, speaker_embeddings) encoder_outputs = self._concat_speaker_embedding(
decoder_outputs, alignments, stop_tokens = self.decoder.inference(encoder_outputs) encoder_outputs, speaker_embeddings)
decoder_outputs, alignments, stop_tokens = self.decoder.inference(
encoder_outputs)
postnet_outputs = self.postnet(decoder_outputs) postnet_outputs = self.postnet(decoder_outputs)
postnet_outputs = self.last_linear(postnet_outputs) postnet_outputs = self.last_linear(postnet_outputs)
decoder_outputs = decoder_outputs.transpose(1, 2) decoder_outputs = decoder_outputs.transpose(1, 2)
return decoder_outputs, postnet_outputs, alignments, stop_tokens outputs = {
'postnet_outputs': postnet_outputs,
'decoder_outputs': decoder_outputs,
'alignments': alignments,
'stop_tokens': stop_tokens
}
return outputs
def train_step(self, batch, criterion):
"""Perform a single training step by fetching the right set if samples from the batch.
Args:
batch ([type]): [description]
criterion ([type]): [description]
"""
text_input = batch['text_input']
text_lengths = batch['text_lengths']
mel_input = batch['mel_input']
mel_lengths = batch['mel_lengths']
linear_input = batch['linear_input']
stop_targets = batch['stop_targets']
speaker_ids = batch['speaker_ids']
x_vectors = batch['x_vectors']
# forward pass model
outputs = self.forward(text_input,
text_lengths,
mel_input,
mel_lengths,
cond_input={
'speaker_ids': speaker_ids,
'x_vectors': x_vectors
})
# set the [alignment] lengths wrt reduction factor for guided attention
if mel_lengths.max() % self.decoder.r != 0:
alignment_lengths = (
mel_lengths +
(self.decoder.r -
(mel_lengths.max() % self.decoder.r))) // self.decoder.r
else:
alignment_lengths = mel_lengths // self.decoder.r
cond_input = {'speaker_ids': speaker_ids, 'x_vectors': x_vectors}
outputs = self.forward(text_input, text_lengths, mel_input,
mel_lengths, cond_input)
# compute loss
loss_dict = criterion(
outputs['postnet_outputs'],
outputs['decoder_outputs'],
mel_input,
linear_input,
outputs['stop_tokens'],
stop_targets,
mel_lengths,
outputs['decoder_outputs_backward'],
outputs['alignments'],
alignment_lengths,
outputs['alignments_backward'],
text_lengths,
)
# compute alignment error (the lower the better )
align_error = 1 - alignment_diagonal_score(outputs['alignments'])
loss_dict["align_error"] = align_error
return outputs, loss_dict
def train_log(self, ap, batch, outputs):
postnet_outputs = outputs['postnet_outputs']
alignments = outputs['alignments']
alignments_backward = outputs['alignments_backward']
mel_input = batch['mel_input']
pred_spec = postnet_outputs[0].data.cpu().numpy()
gt_spec = mel_input[0].data.cpu().numpy()
align_img = alignments[0].data.cpu().numpy()
figures = {
"prediction": plot_spectrogram(pred_spec, ap, output_fig=False),
"ground_truth": plot_spectrogram(gt_spec, ap, output_fig=False),
"alignment": plot_alignment(align_img, output_fig=False),
}
if self.bidirectional_decoder or self.double_decoder_consistency:
figures["alignment_backward"] = plot_alignment(
alignments_backward[0].data.cpu().numpy(), output_fig=False)
# Sample audio
train_audio = ap.inv_spectrogram(pred_spec.T)
return figures, train_audio
def eval_step(self, batch, criterion):
return self.train_step(batch, criterion)
def eval_log(self, ap, batch, outputs):
return self.train_log(ap, batch, outputs)

308
TTS/tts/models/tacotron2.py
View File

@ -1,12 +1,15 @@
# coding: utf-8
import numpy as np
import torch import torch
from torch import nn from torch import nn
from TTS.tts.utils.measures import alignment_diagonal_score
from TTS.tts.utils.visual import plot_alignment, plot_spectrogram
from TTS.tts.layers.tacotron.gst_layers import GST from TTS.tts.layers.tacotron.gst_layers import GST
from TTS.tts.layers.tacotron.tacotron2 import Decoder, Encoder, Postnet from TTS.tts.layers.tacotron.tacotron2 import Decoder, Encoder, Postnet
from TTS.tts.models.tacotron_abstract import TacotronAbstract from TTS.tts.models.tacotron_abstract import TacotronAbstract
# TODO: match function arguments with tacotron
class Tacotron2(TacotronAbstract): class Tacotron2(TacotronAbstract):
"""Tacotron2 as in https://arxiv.org/abs/1712.05884 """Tacotron2 as in https://arxiv.org/abs/1712.05884
@ -43,69 +46,52 @@ class Tacotron2(TacotronAbstract):
speaker_embedding_dim (int, optional): external speaker conditioning vector channels. Defaults to None. speaker_embedding_dim (int, optional): external speaker conditioning vector channels. Defaults to None.
use_gst (bool, optional): enable/disable Global style token module. use_gst (bool, optional): enable/disable Global style token module.
gst (Coqpit, optional): Coqpit to initialize the GST module. If `None`, GST is disabled. Defaults to None. gst (Coqpit, optional): Coqpit to initialize the GST module. If `None`, GST is disabled. Defaults to None.
gradual_trainin (List): Gradual training schedule. If None or `[]`, no gradual training is used.
Defaults to `[]`.
""" """
def __init__(self,
def __init__( num_chars,
self, num_speakers,
num_chars, r,
num_speakers, postnet_output_dim=80,
r, decoder_output_dim=80,
postnet_output_dim=80, attn_type="original",
decoder_output_dim=80, attn_win=False,
attn_type="original", attn_norm="softmax",
attn_win=False, prenet_type="original",
attn_norm="softmax", prenet_dropout=True,
prenet_type="original", prenet_dropout_at_inference=False,
prenet_dropout=True, forward_attn=False,
prenet_dropout_at_inference=False, trans_agent=False,
forward_attn=False, forward_attn_mask=False,
trans_agent=False, location_attn=True,
forward_attn_mask=False, attn_K=5,
location_attn=True, separate_stopnet=True,
attn_K=5, bidirectional_decoder=False,
separate_stopnet=True, double_decoder_consistency=False,
bidirectional_decoder=False, ddc_r=None,
double_decoder_consistency=False, encoder_in_features=512,
ddc_r=None, decoder_in_features=512,
encoder_in_features=512, speaker_embedding_dim=None,
decoder_in_features=512, use_gst=False,
speaker_embedding_dim=None, gst=None,
use_gst=False, gradual_training=[]):
gst=None, super().__init__(num_chars, num_speakers, r, postnet_output_dim,
): decoder_output_dim, attn_type, attn_win, attn_norm,
super().__init__( prenet_type, prenet_dropout,
num_chars, prenet_dropout_at_inference, forward_attn,
num_speakers, trans_agent, forward_attn_mask, location_attn, attn_K,
r, separate_stopnet, bidirectional_decoder,
postnet_output_dim, double_decoder_consistency, ddc_r,
decoder_output_dim, encoder_in_features, decoder_in_features,
attn_type, speaker_embedding_dim, use_gst, gst, gradual_training)
attn_win,
attn_norm,
prenet_type,
prenet_dropout,
prenet_dropout_at_inference,
forward_attn,
trans_agent,
forward_attn_mask,
location_attn,
attn_K,
separate_stopnet,
bidirectional_decoder,
double_decoder_consistency,
ddc_r,
encoder_in_features,
decoder_in_features,
speaker_embedding_dim,
use_gst,
gst,
)
# speaker embedding layer # speaker embedding layer
if self.num_speakers > 1: if self.num_speakers > 1:
if not self.embeddings_per_sample: if not self.embeddings_per_sample:
speaker_embedding_dim = 512 speaker_embedding_dim = 512
self.speaker_embedding = nn.Embedding(self.num_speakers, speaker_embedding_dim) self.speaker_embedding = nn.Embedding(self.num_speakers,
speaker_embedding_dim)
self.speaker_embedding.weight.data.normal_(0, 0.3) self.speaker_embedding.weight.data.normal_(0, 0.3)
# speaker and gst embeddings is concat in decoder input # speaker and gst embeddings is concat in decoder input
@ -176,16 +162,24 @@ class Tacotron2(TacotronAbstract):
mel_outputs_postnet = mel_outputs_postnet.transpose(1, 2) mel_outputs_postnet = mel_outputs_postnet.transpose(1, 2)
return mel_outputs, mel_outputs_postnet, alignments return mel_outputs, mel_outputs_postnet, alignments
def forward(self, text, text_lengths, mel_specs=None, mel_lengths=None, speaker_ids=None, speaker_embeddings=None): def forward(self,
text,
text_lengths,
mel_specs=None,
mel_lengths=None,
cond_input=None):
""" """
Shapes: Shapes:
text: [B, T_in] text: [B, T_in]
text_lengths: [B] text_lengths: [B]
mel_specs: [B, T_out, C] mel_specs: [B, T_out, C]
mel_lengths: [B] mel_lengths: [B]
speaker_ids: [B, 1] cond_input: 'speaker_ids': [B, 1] and 'x_vectors':[B, C]
speaker_embeddings: [B, C]
""" """
outputs = {
'alignments_backward': None,
'decoder_outputs_backward': None
}
# compute mask for padding # compute mask for padding
# B x T_in_max (boolean) # B x T_in_max (boolean)
input_mask, output_mask = self.compute_masks(text_lengths, mel_lengths) input_mask, output_mask = self.compute_masks(text_lengths, mel_lengths)
@ -195,94 +189,176 @@ class Tacotron2(TacotronAbstract):
encoder_outputs = self.encoder(embedded_inputs, text_lengths) encoder_outputs = self.encoder(embedded_inputs, text_lengths)
if self.gst and self.use_gst: if self.gst and self.use_gst:
# B x gst_dim # B x gst_dim
encoder_outputs = self.compute_gst(encoder_outputs, mel_specs, speaker_embeddings) encoder_outputs = self.compute_gst(encoder_outputs, mel_specs,
cond_input['x_vectors'])
if self.num_speakers > 1: if self.num_speakers > 1:
if not self.embeddings_per_sample: if not self.embeddings_per_sample:
# B x 1 x speaker_embed_dim # B x 1 x speaker_embed_dim
speaker_embeddings = self.speaker_embedding(speaker_ids)[:, None] speaker_embeddings = self.speaker_embedding(cond_input['speaker_ids'])[:,
None]
else: else:
# B x 1 x speaker_embed_dim # B x 1 x speaker_embed_dim
speaker_embeddings = torch.unsqueeze(speaker_embeddings, 1) speaker_embeddings = torch.unsqueeze(cond_input['x_vectors'], 1)
encoder_outputs = self._concat_speaker_embedding(encoder_outputs, speaker_embeddings) encoder_outputs = self._concat_speaker_embedding(
encoder_outputs, speaker_embeddings)
encoder_outputs = encoder_outputs * input_mask.unsqueeze(2).expand_as(encoder_outputs) encoder_outputs = encoder_outputs * input_mask.unsqueeze(2).expand_as(
encoder_outputs)
# B x mel_dim x T_out -- B x T_out//r x T_in -- B x T_out//r # B x mel_dim x T_out -- B x T_out//r x T_in -- B x T_out//r
decoder_outputs, alignments, stop_tokens = self.decoder(encoder_outputs, mel_specs, input_mask) decoder_outputs, alignments, stop_tokens = self.decoder(
encoder_outputs, mel_specs, input_mask)
# sequence masking # sequence masking
if mel_lengths is not None: if mel_lengths is not None:
decoder_outputs = decoder_outputs * output_mask.unsqueeze(1).expand_as(decoder_outputs) decoder_outputs = decoder_outputs * output_mask.unsqueeze(
1).expand_as(decoder_outputs)
# B x mel_dim x T_out # B x mel_dim x T_out
postnet_outputs = self.postnet(decoder_outputs) postnet_outputs = self.postnet(decoder_outputs)
postnet_outputs = decoder_outputs + postnet_outputs postnet_outputs = decoder_outputs + postnet_outputs
# sequence masking # sequence masking
if output_mask is not None: if output_mask is not None:
postnet_outputs = postnet_outputs * output_mask.unsqueeze(1).expand_as(postnet_outputs) postnet_outputs = postnet_outputs * output_mask.unsqueeze(
1).expand_as(postnet_outputs)
# B x T_out x mel_dim -- B x T_out x mel_dim -- B x T_out//r x T_in # B x T_out x mel_dim -- B x T_out x mel_dim -- B x T_out//r x T_in
decoder_outputs, postnet_outputs, alignments = self.shape_outputs(decoder_outputs, postnet_outputs, alignments) decoder_outputs, postnet_outputs, alignments = self.shape_outputs(
decoder_outputs, postnet_outputs, alignments)
if self.bidirectional_decoder: if self.bidirectional_decoder:
decoder_outputs_backward, alignments_backward = self._backward_pass(mel_specs, encoder_outputs, input_mask) decoder_outputs_backward, alignments_backward = self._backward_pass(
return ( mel_specs, encoder_outputs, input_mask)
decoder_outputs, outputs['alignments_backward'] = alignments_backward
postnet_outputs, outputs['decoder_outputs_backward'] = decoder_outputs_backward
alignments,
stop_tokens,
decoder_outputs_backward,
alignments_backward,
)
if self.double_decoder_consistency: if self.double_decoder_consistency:
decoder_outputs_backward, alignments_backward = self._coarse_decoder_pass( decoder_outputs_backward, alignments_backward = self._coarse_decoder_pass(
mel_specs, encoder_outputs, alignments, input_mask mel_specs, encoder_outputs, alignments, input_mask)
) outputs['alignments_backward'] = alignments_backward
return ( outputs['decoder_outputs_backward'] = decoder_outputs_backward
decoder_outputs, outputs.update({
postnet_outputs, 'postnet_outputs': postnet_outputs,
alignments, 'decoder_outputs': decoder_outputs,
stop_tokens, 'alignments': alignments,
decoder_outputs_backward, 'stop_tokens': stop_tokens
alignments_backward, })
) return outputs
return decoder_outputs, postnet_outputs, alignments, stop_tokens
@torch.no_grad() @torch.no_grad()
def inference(self, text, speaker_ids=None, style_mel=None, speaker_embeddings=None): def inference(self, text, cond_input=None):
embedded_inputs = self.embedding(text).transpose(1, 2) embedded_inputs = self.embedding(text).transpose(1, 2)
encoder_outputs = self.encoder.inference(embedded_inputs) encoder_outputs = self.encoder.inference(embedded_inputs)
if self.gst and self.use_gst: if self.gst and self.use_gst:
# B x gst_dim # B x gst_dim
encoder_outputs = self.compute_gst(encoder_outputs, style_mel, speaker_embeddings) encoder_outputs = self.compute_gst(encoder_outputs, cond_input['style_mel'],
cond_input['x_vectors'])
if self.num_speakers > 1: if self.num_speakers > 1:
if not self.embeddings_per_sample: if not self.embeddings_per_sample:
speaker_embeddings = self.speaker_embedding(speaker_ids)[:, None] x_vector = self.speaker_embedding(cond_input['speaker_ids'])[:, None]
speaker_embeddings = torch.unsqueeze(speaker_embeddings, 0).transpose(1, 2) x_vector = torch.unsqueeze(x_vector, 0).transpose(1, 2)
encoder_outputs = self._concat_speaker_embedding(encoder_outputs, speaker_embeddings) else:
x_vector = cond_input
decoder_outputs, alignments, stop_tokens = self.decoder.inference(encoder_outputs) encoder_outputs = self._concat_speaker_embedding(
encoder_outputs, x_vector)
decoder_outputs, alignments, stop_tokens = self.decoder.inference(
encoder_outputs)
postnet_outputs = self.postnet(decoder_outputs) postnet_outputs = self.postnet(decoder_outputs)
postnet_outputs = decoder_outputs + postnet_outputs postnet_outputs = decoder_outputs + postnet_outputs
decoder_outputs, postnet_outputs, alignments = self.shape_outputs(decoder_outputs, postnet_outputs, alignments) decoder_outputs, postnet_outputs, alignments = self.shape_outputs(
return decoder_outputs, postnet_outputs, alignments, stop_tokens decoder_outputs, postnet_outputs, alignments)
outputs = {
'postnet_outputs': postnet_outputs,
'decoder_outputs': decoder_outputs,
'alignments': alignments,
'stop_tokens': stop_tokens
}
return outputs
def inference_truncated(self, text, speaker_ids=None, style_mel=None, speaker_embeddings=None): def train_step(self, batch, criterion):
"""Perform a single training step by fetching the right set if samples from the batch.
Args:
batch ([type]): [description]
criterion ([type]): [description]
""" """
Preserve model states for continuous inference text_input = batch['text_input']
""" text_lengths = batch['text_lengths']
embedded_inputs = self.embedding(text).transpose(1, 2) mel_input = batch['mel_input']
encoder_outputs = self.encoder.inference_truncated(embedded_inputs) mel_lengths = batch['mel_lengths']
linear_input = batch['linear_input']
stop_targets = batch['stop_targets']
speaker_ids = batch['speaker_ids']
x_vectors = batch['x_vectors']
if self.gst: # forward pass model
# B x gst_dim outputs = self.forward(text_input,
encoder_outputs = self.compute_gst(encoder_outputs, style_mel, speaker_embeddings) text_lengths,
mel_input,
mel_lengths,
cond_input={
'speaker_ids': speaker_ids,
'x_vectors': x_vectors
})
if self.num_speakers > 1: # set the [alignment] lengths wrt reduction factor for guided attention
if not self.embeddings_per_sample: if mel_lengths.max() % self.decoder.r != 0:
speaker_embeddings = self.speaker_embedding(speaker_ids)[:, None] alignment_lengths = (
speaker_embeddings = torch.unsqueeze(speaker_embeddings, 0).transpose(1, 2) mel_lengths +
encoder_outputs = self._concat_speaker_embedding(encoder_outputs, speaker_embeddings) (self.decoder.r -
(mel_lengths.max() % self.decoder.r))) // self.decoder.r
else:
alignment_lengths = mel_lengths // self.decoder.r
mel_outputs, alignments, stop_tokens = self.decoder.inference_truncated(encoder_outputs) cond_input = {'speaker_ids': speaker_ids, 'x_vectors': x_vectors}
mel_outputs_postnet = self.postnet(mel_outputs) outputs = self.forward(text_input, text_lengths, mel_input,
mel_outputs_postnet = mel_outputs + mel_outputs_postnet mel_lengths, cond_input)
mel_outputs, mel_outputs_postnet, alignments = self.shape_outputs(mel_outputs, mel_outputs_postnet, alignments)
return mel_outputs, mel_outputs_postnet, alignments, stop_tokens # compute loss
loss_dict = criterion(
outputs['model_outputs'],
outputs['decoder_outputs'],
mel_input,
linear_input,
outputs['stop_tokens'],
stop_targets,
mel_lengths,
outputs['decoder_outputs_backward'],
outputs['alignments'],
alignment_lengths,
outputs['alignments_backward'],
text_lengths,
)
# compute alignment error (the lower the better )
align_error = 1 - alignment_diagonal_score(outputs['alignments'])
loss_dict["align_error"] = align_error
return outputs, loss_dict
def train_log(self, ap, batch, outputs):
postnet_outputs = outputs['model_outputs']
alignments = outputs['alignments']
alignments_backward = outputs['alignments_backward']
mel_input = batch['mel_input']
pred_spec = postnet_outputs[0].data.cpu().numpy()
gt_spec = mel_input[0].data.cpu().numpy()
align_img = alignments[0].data.cpu().numpy()
figures = {
"prediction": plot_spectrogram(pred_spec, ap, output_fig=False),
"ground_truth": plot_spectrogram(gt_spec, ap, output_fig=False),
"alignment": plot_alignment(align_img, output_fig=False),
}
if self.bidirectional_decoder or self.double_decoder_consistency:
figures["alignment_backward"] = plot_alignment(
alignments_backward[0].data.cpu().numpy(), output_fig=False)
# Sample audio
train_audio = ap.inv_melspectrogram(pred_spec.T)
return figures, train_audio
def eval_step(self, batch, criterion):
return self.train_step(batch, criterion)
def eval_log(self, ap, batch, outputs):
return self.train_log(ap, batch, outputs)

26
TTS/tts/models/tacotron_abstract.py
View File

@ -1,10 +1,12 @@
import copy import copy
import logging
from abc import ABC, abstractmethod from abc import ABC, abstractmethod
import torch import torch
from torch import nn from torch import nn
from TTS.tts.utils.generic_utils import sequence_mask from TTS.tts.utils.data import sequence_mask
from TTS.utils.training import gradual_training_scheduler
class TacotronAbstract(ABC, nn.Module): class TacotronAbstract(ABC, nn.Module):
@ -35,6 +37,7 @@ class TacotronAbstract(ABC, nn.Module):
speaker_embedding_dim=None, speaker_embedding_dim=None,
use_gst=False, use_gst=False,
gst=None, gst=None,
gradual_training=[]
): ):
"""Abstract Tacotron class""" """Abstract Tacotron class"""
super().__init__() super().__init__()
@ -63,6 +66,7 @@ class TacotronAbstract(ABC, nn.Module):
self.encoder_in_features = encoder_in_features self.encoder_in_features = encoder_in_features
self.decoder_in_features = decoder_in_features self.decoder_in_features = decoder_in_features
self.speaker_embedding_dim = speaker_embedding_dim self.speaker_embedding_dim = speaker_embedding_dim
self.gradual_training = gradual_training
# layers # layers
self.embedding = None self.embedding = None
@ -216,3 +220,23 @@ class TacotronAbstract(ABC, nn.Module):
speaker_embeddings_ = speaker_embeddings.expand(outputs.size(0), outputs.size(1), -1) speaker_embeddings_ = speaker_embeddings.expand(outputs.size(0), outputs.size(1), -1)
outputs = torch.cat([outputs, speaker_embeddings_], dim=-1) outputs = torch.cat([outputs, speaker_embeddings_], dim=-1)
return outputs return outputs
#############################
# CALLBACKS
#############################
def on_epoch_start(self, trainer):
"""Callback for setting values wrt gradual training schedule.
Args:
trainer (TrainerTTS): TTS trainer object that is used to train this model.
"""
if self.gradual_training:
r, trainer.config.batch_size = gradual_training_scheduler(trainer.total_steps_done, trainer.config)
trainer.config.r = r
self.decoder.set_r(r)
if trainer.config.bidirectional_decoder:
trainer.model.decoder_backward.set_r(r)
trainer.train_loader = trainer.setup_train_dataloader(self.ap, self.model.decoder.r, verbose=True)
trainer.eval_loader = trainer.setup_eval_dataloder(self.ap, self.model.decoder.r)
logging.info(f"\n > Number of output frames: {self.decoder.r}")