PlayVoice / vits_chinese

Best practice TTS based on BERT and VITS with some Natural Speech Features Of Microsoft; Support ONNX streaming out!
https://huggingface.co/spaces/maxmax20160403/vits_chinese
MIT License
1.16k stars 168 forks source link

初学者请问作者prosody_model.pt是怎么训练得到的 #107

Open ShangkunTu opened 1 year ago

ShangkunTu commented 1 year ago

作者您好,请问在合成阶段用到的prosody_model.pt是在哪里生成的,您提供的这代码有能生成这个prossody模型文件吗?我尝试找了一下,但是没找到,还是说这个模型文件本代码并没有生成只是拿来用呀。

MaxMax2016 commented 1 year ago

Reference For TTS Microsoft's NaturalSpeech: End-to-End Text to Speech Synthesis with Human-Level Quality

https://github.com/Executedone/Chinese-FastSpeech2 bert prosody

https://github.com/wenet-e2e/WeTextProcessing

https://github.com/jaywalnut310/vits

ShangkunTu commented 1 year ago

Reference For TTS Microsoft's NaturalSpeech: End-to-End Text to Speech Synthesis with Human-Level Quality

https://github.com/Executedone/Chinese-FastSpeech2 bert prosody

https://github.com/wenet-e2e/WeTextProcessing

https://github.com/jaywalnut310/vits

Reference For TTS Microsoft's NaturalSpeech: End-to-End Text to Speech Synthesis with Human-Level Quality

https://github.com/Executedone/Chinese-FastSpeech2 bert prosody

https://github.com/wenet-e2e/WeTextProcessing

https://github.com/jaywalnut310/vits

感谢回复,我研究一下

ShangkunTu commented 1 year ago

Reference For TTS Microsoft's NaturalSpeech: End-to-End Text to Speech Synthesis with Human-Level Quality

https://github.com/Executedone/Chinese-FastSpeech2 bert prosody

https://github.com/wenet-e2e/WeTextProcessing

https://github.com/jaywalnut310/vits

您好,我用您的工程训练到了31w步了,发现比较18w步,发音更清晰了,并且情感上也上扬了一些,比较出来更活泼了一些。发音更清晰了我能理解,但是情感表现为更上扬了,更明亮了是要归功于上面提到了来自于FastSpeech2中文版里面的韵律模型吗?期待您的回复

MaxMax2016 commented 1 year ago

是的,去掉韵律模型会呈现一个平均状态的韵律,没有韵律的起伏变化。

你可以试试修改推理,屏蔽bert韵律向量

def forward(self, x, x_lengths, bert): x = self.emb(x) * math.sqrt(self.hidden_channels) # [b, t, h] b = self.emb_bert(bert) x = x + b x = torch.transpose(x, 1, -1) # [b, h, t]