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 6f33506d89
commit e409f3588b
1 changed files with 15 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -1030,6 +1030,9 @@ class Vits(BaseTTS):
# concat the emotion embedding and speaker embedding # 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 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] g = torch.cat([g, eg], dim=1) # [b, h1+h2, 1]
# language embedding # language embedding
@ -1145,8 +1148,11 @@ class Vits(BaseTTS):
eg = self.emb_emotion(eid).unsqueeze(-1) # [b, h, 1] eg = self.emb_emotion(eid).unsqueeze(-1) # [b, h, 1]
# concat the emotion embedding and speaker embedding # 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): if eg 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 g is None:
g = eg
else:
g = torch.cat([g, eg], dim=1) # [b, h1+h2, 1]
# language embedding # language embedding
lang_emb = None lang_emb = None
@ -1780,10 +1786,15 @@ class Vits(BaseTTS):
language_manager = LanguageManager.init_from_config(config) language_manager = LanguageManager.init_from_config(config)
emotion_manager = EmotionManager.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( speaker_manager.init_encoder(
config.model_args.encoder_model_path, config.model_args.encoder_config_path 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) return Vits(new_config, ap, tokenizer, speaker_manager, language_manager, emotion_manager=emotion_manager)