Update cosyvoice/cli/model.py

cosyvoice/cli/model.py

@@ -323,10 +323,11 @@ class CosyVoice2Model(CosyVoiceModel):
             self.tts_speech_token_dict[this_uuid], self.llm_end_dict[this_uuid] = [], False
             self.hift_cache_dict[this_uuid] = None
         if source_speech_token.shape[1] == 0:
-            p = threading.Thread(target=self.llm_job, args=(text, prompt_text, llm_prompt_speech_token, llm_embedding, this_uuid))
+            # p = threading.Thread(target=self.llm_job, args=(text, prompt_text, llm_prompt_speech_token, llm_embedding, this_uuid))
+            self.llm_job(text, prompt_text, llm_prompt_speech_token, llm_embedding, this_uuid)
         else:
             p = threading.Thread(target=self.vc_job, args=(source_speech_token, this_uuid))
-        p.start()
+        # p.start()
         if stream is True:
             token_offset = 0
             prompt_token_pad = int(np.ceil(flow_prompt_speech_token.shape[1] / self.token_hop_len) * self.token_hop_len - flow_prompt_speech_token.shape[1])
@@ -347,7 +348,7 @@ class CosyVoice2Model(CosyVoiceModel):
                     yield {'tts_speech': this_tts_speech.cpu()}
                 if self.llm_end_dict[this_uuid] is True and len(self.tts_speech_token_dict[this_uuid]) - token_offset < this_token_hop_len + self.flow.pre_lookahead_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.tensor(self.tts_speech_token_dict[this_uuid]).unsqueeze(dim=0)
             this_tts_speech = self.token2wav(token=this_tts_speech_token,
@@ -360,7 +361,7 @@ class CosyVoice2Model(CosyVoiceModel):
             yield {'tts_speech': this_tts_speech.cpu()}
         else:
             # deal with all tokens
-            p.join()
+            # p.join()
             this_tts_speech_token = torch.tensor(self.tts_speech_token_dict[this_uuid]).unsqueeze(dim=0)
             this_tts_speech = self.token2wav(token=this_tts_speech_token,
                                              prompt_token=flow_prompt_speech_token,