Is is necessary to pretrain the speaker encoder using GE2E loss on VCTK?

auspicious3000 / autovc

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

https://arxiv.org/abs/1905.05879

MIT License

987 stars 205 forks source link

Is is necessary to pretrain the speaker encoder using GE2E loss on VCTK? #15

Closed KinglittleQ closed 5 years ago

KinglittleQ commented 5 years ago

@auspicious3000 Can I train the model directly on VCTK without pretrain the speaker encoder using GE2E loss?

auspicious3000 commented 5 years ago

Yes. Just use one-hot speaker embedding.

KinglittleQ commented 5 years ago

@auspicious3000 I want to use the speaker encoder instead of one-hot speaker embedding, but pre-training the speaker encoder using GE2E loss will take much time. So I want to train my speaker encoder within the whole model without pre-train it. Would it work?

auspicious3000 commented 5 years ago

No. But you can train your own speaker encoder instead of using GE2E