Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

如何用训练好的模型合成语句 #9

Open
lower-fish opened this issue Dec 5, 2020 · 10 comments
Open

如何用训练好的模型合成语句 #9

lower-fish opened this issue Dec 5, 2020 · 10 comments

Comments

@lower-fish
Copy link

您好,老师

我按照您的教程已经训练好了模型,但不懂如何使用。
比如我要用此模型合成一句自己准备的文本句子,而非预先准备的mel谱。我无从下手,盼赐教!

谢谢

@wqt2019
Copy link
Owner

wqt2019 commented Dec 5, 2020

你需要先了解下整个语音合成的流程。如果t2和melgan都训练好了,那么给t2输入你要合成的文字的拼音或者音素,取决你用哪个标签训练的。得到mel谱后给到melgan就可以合成声音了

@lower-fish
Copy link
Author

感谢赐教

@yyHMA
Copy link

yyHMA commented Dec 14, 2020

你好,tacotron-2的readme说的是吧audio-xxx.npy和mel-xxx.npy复制到training/validing。那training/validing的数据和melgan里的config/default.yml中的 train: './data/train/wav/' 和validation: './data/valid/wav/'有什么关系。melgan的输入难道不是.npy文件,而是处理后的.mel文件?

@yyHMA
Copy link

yyHMA commented Dec 14, 2020

您好,老师

我按照您的教程已经训练好了模型,但不懂如何使用。
比如我要用此模型合成一句自己准备的文本句子,而非预先准备的mel谱。我无从下手,盼赐教!

谢谢

你好,melgan的训练,输入的数据你是怎么处理?输入数据是npy文件还是mel文件?

@wqt2019
Copy link
Owner

wqt2019 commented Dec 14, 2020

melgan的输入是mel谱,mel谱是二维数组,保存格式为.npy

@yyHMA
Copy link

yyHMA commented Dec 15, 2020

melgan的数据集格式可以提供一下吗?

@wqt2019
Copy link
Owner

wqt2019 commented Dec 15, 2020

原始音频和mel谱对应起来就可以了, https://github.com/seungwonpark/melgan

@yyHMA
Copy link

yyHMA commented Dec 15, 2020

Step (0): python preprocess.py ,process the audios for t2 and melgan training .
Step (1): cp audio-xxx.npy and mel-xxx.npy(real mel) to melgan's training/validing data-path .

这个Step (1),把audio-xxx.npy和 mel-xxx.npy分别放在train: './data/train/wav/' 和validation: './data/valid/wav/的路径里,这样的做法对吗

@wqt2019
Copy link
Owner

wqt2019 commented Dec 15, 2020

是的,你试一下,我好几个月没弄melgan了。主要是音频数据和mel对应起来就可以了

@wjMY
Copy link

wjMY commented Dec 29, 2020

Step (0): python preprocess.py ,process the audios for t2 and melgan training .
Step (1): cp audio-xxx.npy and mel-xxx.npy(real mel) to melgan's training/validing data-path .

这个Step (1),把audio-xxx.npy和 mel-xxx.npy分别放在train: './data/train/wav/' 和validation: './data/valid/wav/的路径里,这样的做法对吗

应该不是这样的,从dataloader.py的代码看,train和validation是两个分开的数据,audio*.npy和mel*.npy需要一起存在,在https://github.com/seungwonpark/melgan中说明了validation是验证集。

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants