Remove cache_source kwarg from HiFT inference in CosyVoiceModel

cosyvoice/cli/model.py

@@ -158,7 +158,7 @@ class CosyVoiceModel:
         if finalize is False:
             self.mel_overlap_dict[uuid] = tts_mel[:, :, -self.mel_overlap_len:]
             tts_mel = tts_mel[:, :, :-self.mel_overlap_len]
-            tts_speech, tts_source = self.hift.inference(speech_feat=tts_mel, cache_source=hift_cache_source)
+            tts_speech, tts_source = self.hift.inference(speech_feat=tts_mel)
             if self.hift_cache_dict[uuid] is not None:
                 tts_speech = fade_in_out(tts_speech, self.hift_cache_dict[uuid]['speech'], self.speech_window)
             self.hift_cache_dict[uuid] = {'mel': tts_mel[:, :, -self.mel_cache_len:],
@@ -169,7 +169,7 @@ class CosyVoiceModel:
             if speed != 1.0:
                 assert self.hift_cache_dict[uuid] is None, '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, tts_source = self.hift.inference(speech_feat=tts_mel, cache_source=hift_cache_source)
+            tts_speech, tts_source = self.hift.inference(speech_feat=tts_mel)
             if self.hift_cache_dict[uuid] is not None:
                 tts_speech = fade_in_out(tts_speech, self.hift_cache_dict[uuid]['speech'], self.speech_window)
         return tts_speech