How to generate the pretrained model/checkpoint files?

auspicious3000 / autovc

AutoVC: Zero-Shot Voice Style Transfer with Only Autoencoder Loss

https://arxiv.org/abs/1905.05879

MIT License

983 stars 207 forks source link

How to generate the pretrained model/checkpoint files? #65

Closed KevinHua closed 3 years ago

KevinHua commented 3 years ago

There are three pretrained files: autovc.ckpt: autovc model. 3000000-BL.ckpt: D-VECTOR. checkpoint_step001000000_ema.pth: wavenet

The first one could be saved in solver_encoder.py as described in an previous issue. And how to generate the last two files? Thanks a lot.

auspicious3000 commented 3 years ago

The 2nd one is for the speaker verification model, and the 3rd one is for the vocoder, neither are part of this project.

KevinHua commented 3 years ago

Gotted it. The file checkpoint_step001000000_ema.pth is r9y9 wavenet_vocoder's model and I'll try it. And could you please give some tips more about the the speaker verification model, git repo or papers? Thanks.

auspicious3000 commented 3 years ago

Yes. that model is described and cited in our paper.