search

CorentinJ / Real-Time-Voice-Cloning

Clone a voice in 5 seconds to generate arbitrary speech in real-time
Other
51.65k stars 8.66k forks source link

How to swap in hifigan? #1177

Open pmcanneny opened 1 year ago

pmcanneny commented 1 year ago

Has anyone successfully swapped in hifigan for better inference performance?

raccoonML commented 1 year ago

I have. One obstacle is that the default hifigan models don't match the mel spectrogram definition used in this repo. You'll either need to train a new RTVC model using the hifigan melspec definition, or go the other direction and train a hifigan model to match the RTVC melspecs. I've tried both and a better result is obtained with the latter approach.

When I get a chance, I will clean up the code and commit it to my fork of RTVC. Will keep you posted here.

Raha304 commented 1 year ago

If your results are good with hifigan, can you send your codes to me please? I really really need your help about this. Please answer me. my email: rahaamiri304@gmail.com