kan-bayashi / ParallelWaveGAN

Unofficial Parallel WaveGAN (+ MelGAN & Multi-band MelGAN & HiFi-GAN & StyleMelGAN) with Pytorch
https://kan-bayashi.github.io/ParallelWaveGAN/
MIT License
1.54k stars 339 forks source link

Analysis or paper of hubert-discrete-hifigan? #392

Closed seastar105 closed 1 year ago

seastar105 commented 1 year ago

388 was merged last week and it seems this PR is for reconstruct speech from hubert discrete symbols, right?

i think this work could be applied directly for any-to-{one, many} voice conversion. is there any reference paper or anlaysis for this work?

ftshijt commented 1 year ago

There are a few related works on using discrete units:

Ours is mostly following the one in https://arxiv.org/pdf/2107.05604