This branch aims to integrate a modified VAE into fish-diffusion. It will help us to replace mel spectrogram with an embedding and thus improve the robustness of diffusion-based svc and svs tasks.
Note: Auto Vocoder repo is not used AT ALL. It can't create any helpful representation.
This branch aims to integrate a modified VAE into fish-diffusion. It will help us to replace mel spectrogram with an embedding and thus improve the robustness of diffusion-based svc and svs tasks.
Note: Auto Vocoder repo is not used AT ALL. It can't create any helpful representation.