deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-04 17:39:25 +08:00
Code Issues Packages Projects Releases Wiki Activity

add speech fade in out

Browse Source
This commit is contained in:
lyuxiang.lx
2024-09-19 18:02:42 +08:00
parent cd26f11859
commit f65eca6723
2 changed files with 5 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
cosyvoice/cli/model.py
View File

@@ -40,6 +40,8 @@ class CosyVoiceModel:
# hift cache
self.mel_cache_len = 20
self.source_cache_len = int(self.mel_cache_len * 256)
# speech fade in out
self.speech_window = np.hamming(2 * self.source_cache_len)
# rtf and decoding related
self.stream_scale_factor = 1
assert self.stream_scale_factor >= 1, 'stream_scale_factor should be greater than 1, change it according to your actual rtf'
@@ -50,7 +52,6 @@ class CosyVoiceModel:
self.llm_end_dict = {}
self.mel_overlap_dict = {}
self.hift_cache_dict = {}
self.speech_window = np.hamming(2 * self.source_cache_len)
def load(self, llm_model, flow_model, hift_model):
self.llm.load_state_dict(torch.load(llm_model, map_location=self.device))
@@ -117,10 +118,9 @@ class CosyVoiceModel:
tts_speech, tts_source = self.hift.inference(mel=tts_mel, cache_source=hift_cache_source)
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:],
'source': tts_source[:, :, -self.source_cache_len:],
'speech': tts_speech[:, -self.source_cache_len:]}
self.hift_cache_dict[uuid] = {'mel': tts_mel[:, :, -self.mel_cache_len:],
'source': tts_source[:, :, -self.source_cache_len:],
'speech': tts_speech[:, -self.source_cache_len:]}
tts_speech = tts_speech[:, :-self.source_cache_len]
else:
if speed != 1.0: