m-toman / tacorn

2018/2019 TTS framework integrating state of the art open source methods
MIT License
47 stars 4 forks source link

bits=9 -> 16, Is it OK? #6

Closed chkwon1 closed 5 years ago

chkwon1 commented 5 years ago

I modified bits=9 to 16 in wavernn/preprocess.py and model.py. By this modification your code is working. Do you think the synthesized speech quality can be improved?

m-toman commented 5 years ago

Well, the model is created by https://github.com/fatchord/WaveRNN and not me, but I suspect that there are probably too few model parameters for that. Still, might be worth a try.

chkwon19 commented 5 years ago

close #6

m-toman commented 5 years ago

Will be solved once we have #8