Doubiiu / CodeTalker

[CVPR 2023] CodeTalker: Speech-Driven 3D Facial Animation with Discrete Motion Prior
MIT License
515 stars 57 forks source link

Is it possible to use facial models trained in Chinese #72

Open huang-1030 opened 4 months ago

huang-1030 commented 4 months ago

Thank you for the surprising project of open source. Currently, the second phase uses Facebook's wav2vec2-base-960h model. Can I replace this part with another Chinese large model?

Doubiiu commented 4 months ago

Hi, I am afraid that this would not work, and it may involve additional fine-tuning.