search

thuhcsi / VAENAR-TTS

The official implementation of VAENAR-TTS, a VAE based non-autoregressive TTS model.
MIT License
144 stars 20 forks source link

pytorch implementation #3

Open thepowerfuldeez opened 3 years ago

thepowerfuldeez commented 3 years ago

hi! nice work and finally non-autoregressive tts with no explicit durations as labels. Do you have any plans to release pytorch implementation?

light1726 commented 3 years ago

hi! nice work and finally non-autoregressive tts with no explicit durations as labels. Do you have any plans to release pytorch implementation?

Thanks. I'm sorry that I don't have time to reimplement it in Pytorch. I actually heavily referenced this wonderful NAR-MT model XuezheMax/flowseq, which is implemented in Pytorch, when I was doing VAENAR-TTS implementation. I think you can find most code snippets (e.g. the flow part) needed to implement a Pytorch version VAENAT-TTS.

thepowerfuldeez commented 3 years ago

I’ll go get it a try, thank you! Most of the modules are already implemented (eg encoder and decoder from transformer-tts or postnet from tacotron2)

keonlee9420 commented 3 years ago

Hi @thepowerfuldeez, I just published the PyTorch implementation of VAENAR-TTS, and you can check it here: https://github.com/keonlee9420/VAENAR-TTS I'm training the model and going to release pre-trained models as well. Please feel free to play with it, and I hope it helps! Any reports are appreciated.

And thanks @light1726 for the great work. It was exciting to implement the PyTorch version of VAENAR-TTS, and if you have room for it, I want to ask you about sharing my implementation in README.md. I think it's not a big deal since my work is hugely based on yours and can be seen as another implementation in another computer language. I am looking forward to your answer! Thanks.