Rishikesh (ऋषिकेश)

Hi @sungjae-cho this implementation is incomplete, I have mentioned this in Note section. I stop working on this project due to some other engagement s.

It’s pretrain for 100k, means only generator run after that disc training starts. https://github.com/rishikksh20/HiFi-GAN/blob/d044dbcdf799f0fdfbfc1920e57e95ac6a05f91b/configs/default.yaml#L9

Nope this code is the implementation of model not training or other stuffs. I don't have spare GPUs for training and testing.

Hi @shaktikshri this implementation is incomplete, I have mentioned this in `Note` section. I stop working on this project due to some other engagement if you are wiling to change...

@shahuzi yes because this way I am able to train a model easily. But in this repo I am sampling from Gaussian distribution.

@paklau99988 wait it is still in Work in progress, although you can run checkpoint on this https://github.com/rishikksh20/HiFi-GAN/tree/d044dbcdf799f0fdfbfc1920e57e95ac6a05f91b tree commit.

Ljspeech

@DanielSWolf espeak-ng is forked from this repo (espeak) and modify speak bit to support 102 languages whereas this repo support 101.

@bshall Can we also replace Encoder and Decoder with Transformers ?

yes I feel the same with my training