ftshijt
commented
2 years ago Many thanks for your suggestion! We are considering supporting GAN-based svs in our third iteration.
Here is a brief plan of the roadmap:
- By end of March, we want to refine the current codebase and support more recipes (as well as vocoders from https://github.com/kan-bayashi/ParallelWaveGAN)
- By end of May, we will start to merge the framework to ESPnet (you may find they are very similar) as a new sub-module ESPnet-Muskit
- By June, we expect to support GAN-based SVS and other frameworks including
- https://www.arxiv-vanity.com/papers/1903.10729/
- https://github.com/SoonbeomChoi/BEGANSing
- VISinger
- HifiSinger
- DiffSinger
If you have more suggestions or comments, please feel free to request upon issues. Given requests, we may also change the priorities.
MaxMax2016
VITS is the best TTS system, I think. So VISinger is attractive. https://github.com/jaywalnut310/vits is the path of VITS code. https://zhangyongmao.github.io/VISinger/ is the demo of VISinger.