deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-05 01:49:25 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix bistream bug

Browse Source
This commit is contained in:
lyuxiang.lx
2025-12-12 10:41:25 +00:00
parent b02d7e61f7
commit ca3b054a52
6 changed files with 37 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
cosyvoice/cli/model.py
View File

@@ -413,18 +413,18 @@ class CosyVoice3Model(CosyVoice2Model):
embedding=embedding.to(self.device),
streaming=stream,
finalize=finalize)
tts_mel = tts_mel[:, :, token_offset * self.flow.token_mel_ratio:]
# append mel cache
if self.hift_cache_dict[uuid] is not None:
hift_cache_mel = self.hift_cache_dict[uuid]['mel']
tts_mel = torch.concat([hift_cache_mel, tts_mel], dim=2)
self.hift_cache_dict[uuid]['mel'] = tts_mel
else:
self.hift_cache_dict[uuid] = {'mel': tts_mel, 'speech_offset': 0}
if speed != 1.0:
assert token_offset == 0 and finalize is True, 'speed change only support non-stream inference mode'
tts_mel = F.interpolate(tts_mel, size=int(tts_mel.shape[2] / speed), mode='linear')
tts_speech, _ = self.hift.inference(speech_feat=tts_mel, finalize=finalize)
tts_speech = tts_speech[:, self.hift_cache_dict[uuid]['speech_offset']:]
self.hift_cache_dict[uuid]['speech_offset'] += tts_speech.shape[1]
tts_mel = tts_mel[:, :, token_offset * self.flow.token_mel_ratio:]
# append mel cache
if self.hift_cache_dict[uuid] is not None:
hift_cache_mel = self.hift_cache_dict[uuid]['mel']
tts_mel = torch.concat([hift_cache_mel, tts_mel], dim=2)
self.hift_cache_dict[uuid]['mel'] = tts_mel
else:
self.hift_cache_dict[uuid] = {'mel': tts_mel, 'speech_offset': 0}
if speed != 1.0:
assert token_offset == 0 and finalize is True, 'speed change only support non-stream inference mode'
tts_mel = F.interpolate(tts_mel, size=int(tts_mel.shape[2] / speed), mode='linear')
tts_speech, _ = self.hift.inference(speech_feat=tts_mel, finalize=finalize)
tts_speech = tts_speech[:, self.hift_cache_dict[uuid]['speech_offset']:]
self.hift_cache_dict[uuid]['speech_offset'] += tts_speech.shape[1]
return tts_speech