xinwen-cs / AudioDVP

AudioDVP:Photorealistic Audio-driven Video Portraits
https://github.com/xinwen-cs/AudioDVP
296 stars 49 forks source link

how to train with chinese? #14

Closed ghost closed 3 years ago

ghost commented 3 years ago

how to train with chinese? not english.

ghost commented 3 years ago

German Speech Results. Although the high-dimensional feature extraction from audio in our Audio2Expression network is only trained on an English corpus, it generalizes to some other languages, such as German.

so how to train with chinese language of Audio2Expression network?

xinwen-cs commented 3 years ago

English and German have similar alphabet, so it can works. For Chinese, you need to use a big audio-video corpus to train from scratch:)

ghost commented 3 years ago

how to use a big audio-video corpus to train from scratch in detail? the code and method.

ghost commented 3 years ago

@xinwen-cs