-
Notifications
You must be signed in to change notification settings - Fork 1.9k
Open
Description
执行下述代码进行保存音色,同时测试使用保存的说话人是没有问题,音色也对的上
# 保存说话人特征
success = cosyvoice.add_zero_shot_spk(prompt_text, prompt_speech_16k, 'tim_spk2') is True
if not success:
print(f"添加说话人 {spk_id} 失败")
cosyvoice.save_spkinfo()
# 使用保存的说话人
for i, j in enumerate(cosyvoice.inference_zero_shot('这是一段生成的声音,咋样呢', '', '', zero_shot_spk_id='tim_spk2', stream=False)):
torchaudio.save('zero_shot2_{}.wav'.format(i), j['tts_speech'], cosyvoice.sample_rate)
cosyvoice.save_spkinfo()
print("声音克隆存储成功~")
然而通过webui.py的预训练音色或者runtime/python/grpc/server.py生成出来的音色相差很大,是怎么回事?
Metadata
Metadata
Assignees
Labels
No labels