2noise / ChatTTS

A generative speech model for daily dialogue.
https://2noise.com
Other
27.01k stars 2.94k forks source link

根据参考音频克隆音色的模型,欢迎试用 Voice Clone #369

Closed hoveychen closed 1 week ago

hoveychen commented 2 weeks ago

训练了几个克隆声音的模型,可以根据参考音频生成ChatTTS使用的音色嵌入。

模型Demo页面: http://region-9.autodl.pro:41137

欢迎大家反馈一下测试效果,加入QQ群474529811给建议或者讨论。


用法:下载声音模型(.pt结尾),记住temperature要设置得非常低,否则声音会不准确。 ` rand_spk = torch.load(f'my_speaker.pt')

params_infer_code = { 'spk_emb': rand_spk, # add sampled speaker 'temperature': .000001, # using custom temperature }

texts = ['hello world', '你好呀,旅行者!']

wavs = chat.infer(texts, params_infer_code=params_infer_code) `

6drf21e commented 1 week ago

效果不错👍

redstoneleo commented 1 week ago

QQ群搜不到啊

ZaymeShaw commented 1 week ago

用自己的声音简短录了两句话测试,用噪声比较少的声音做克隆时能拷贝一个接近的音色出来,多句话之间的音色一致性也比较高,不过生成声音似乎有比较多噪声,音色也不是完全一样

wangqun888 commented 1 week ago

请问一下克隆模型可以在哪里下载?