deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-05 09:59:23 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'inference_streaming' into flow_tensorrt

Browse Source
This commit is contained in:
Xiang Lyu
2024-08-30 14:20:06 +08:00
committed by GitHub
parent 18599be8d5 662012999a
commit ee988420f3
16 changed files with 754 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
cosyvoice/cli/model.py
View File

@@ -137,7 +137,6 @@ class CosyVoiceModel:
self.tts_speech_token_dict[this_uuid], self.llm_end_dict[this_uuid], self.mel_overlap_dict[this_uuid], self.hift_cache_dict[this_uuid] = [], False, None, None
p = threading.Thread(target=self.llm_job, args=(text, prompt_text, llm_prompt_speech_token, llm_embedding, this_uuid))
p.start()
p.join()
if stream is True:
token_hop_len = self.token_min_hop_len
while True:
@@ -158,7 +157,7 @@ class CosyVoiceModel:
token_hop_len = min(self.token_max_hop_len, int(token_hop_len * self.stream_scale_factor))
if self.llm_end_dict[this_uuid] is True and len(self.tts_speech_token_dict[this_uuid]) < token_hop_len + self.token_overlap_len:
break
# p.join()
p.join()
# deal with remain tokens, make sure inference remain token len equals token_hop_len when cache_speech is not None
this_tts_speech_token = torch.concat(self.tts_speech_token_dict[this_uuid], dim=1)
with self.flow_hift_context:
@@ -171,7 +170,7 @@ class CosyVoiceModel:
yield {'tts_speech': this_tts_speech.cpu()}
else:
# deal with all tokens
# p.join()
p.join()
this_tts_speech_token = torch.concat(self.tts_speech_token_dict[this_uuid], dim=1)
with self.flow_hift_context:
this_tts_speech = self.token2wav(token=this_tts_speech_token,