search

MingjieChen / DYGANVC

demo page https://MingjieChen.github.io/dygan-vc
67 stars 9 forks source link

own data #6

Closed xxbean closed 2 years ago

xxbean commented 2 years ago

hi thanks fot the codes ! i just want to add my own datasets, and could you please tell me how i can do that? thanks :)

MingjieChen commented 2 years ago

Hello,

Since I don't have information about your dataset, so I can only give some points you need to consider

  1. Vocoder I used is specifically for VCC2020 dataset, you may need to use a new vocoder
  2. Mel-spectrogram features should have the same hop size as VQ-Wav2vec, which is 10ms.
  3. For extracting speaker embeddings, you need to set your own sampling rate, which is hard-coded in my code.

If you have more questions, please let me know

csdY123 commented 2 years ago

非常感谢你的代码。 请问运行vqw2v_feat_extract.sh时出现KeyError:'binary_请问运行vqw2v_feat_extract.sh时出现KeyError:'binary_cross_entropy'是什么原因？cross_entropy'是什么原因？

MingjieChen commented 2 years ago

Please have a look at this https://github.com/pytorch/fairseq/issues/3613

非常感谢你的代码。 请问运行vqw2v_feat_extract.sh时出现KeyError:'binary_请问运行vqw2v_feat_extract.sh时出现KeyError:'binary_cross_entropy'是什么原因？cross_entropy'是什么原因？

csdY123 commented 2 years ago

请看一下这个pytorch/fairseq#3613

非常感谢你的代码。请运行vqw2v_feat_extract.sh时出现KeyError:'binary_请运行vqw2v_feat_extract.sh时出现KeyError:'binary_cross_entropy'是什么原因？cross_entropy'是什么原因？

谢谢，问题已经解决，但你给的网址不能解决问题，解决问题可以参考https://github.com/pytorch/fairseq/issues/3256