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

Bug fix in single speaker emotion embedding training

This commit is contained in:
Edresson Casanova 2022-03-16 20:57:14 +00:00
parent 38027b15c2
commit 10dee54ac3
2 changed files with 17 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
TTS/bin/compute_embeddings.py
View File

@ -7,7 +7,7 @@ from tqdm import tqdm
from TTS.config import load_config
from TTS.tts.datasets import load_tts_samples
from TTS.tts.utils.speakers import SpeakerManager
from TTS.tts.utils.managers import EmbeddingManager
parser = argparse.ArgumentParser(
description="""Compute embedding vectors for each wav file in a dataset.\n\n"""
@ -44,7 +44,7 @@ c_dataset = load_config(args.config_dataset_path)
meta_data_train, meta_data_eval = load_tts_samples(c_dataset.datasets, eval_split=args.eval)
wav_files = meta_data_train + meta_data_eval
encoder_manager = SpeakerManager(
encoder_manager = EmbeddingManager(
encoder_model_path=args.model_path,
encoder_config_path=args.config_path,
d_vectors_file_path=args.old_file,

17
TTS/tts/models/vits.py
View File

@ -936,6 +936,9 @@ class Vits(BaseTTS):
# concat the emotion embedding and speaker embedding
if eg is not None and (self.args.use_emotion_embedding or self.args.use_external_emotions_embeddings):
if g is None:
g = eg
else:
g = torch.cat([g, eg], dim=1) # [b, h1+h2, 1]
# language embedding
@ -1046,8 +1049,11 @@ class Vits(BaseTTS):
eg = self.emb_emotion(eid).unsqueeze(-1) # [b, h, 1]
# concat the emotion embedding and speaker embedding
if eg is not None and g is not None and (self.args.use_emotion_embedding or self.args.use_external_emotions_embeddings):
g = torch.cat([g, eg], dim=1) # [b, h1+h1, 1]
if eg is not None and (self.args.use_emotion_embedding or self.args.use_external_emotions_embeddings):
if g is None:
g = eg
else:
g = torch.cat([g, eg], dim=1) # [b, h1+h2, 1]
# language embedding
lang_emb = None
@ -1614,10 +1620,15 @@ class Vits(BaseTTS):
language_manager = LanguageManager.init_from_config(config)
emotion_manager = EmotionManager.init_from_config(config)
if config.model_args.encoder_model_path:
if config.model_args.encoder_model_path and speaker_manager is not None:
speaker_manager.init_encoder(
config.model_args.encoder_model_path, config.model_args.encoder_config_path
)
elif config.model_args.encoder_model_path and emotion_manager is not None:
emotion_manager.init_encoder(
config.model_args.encoder_model_path, config.model_args.encoder_config_path
)
return Vits(new_config, ap, tokenizer, speaker_manager, language_manager, emotion_manager=emotion_manager)