search

r9y9 / wavenet_vocoder

WaveNet vocoder
https://r9y9.github.io/wavenet_vocoder/
Other
2.33k stars 500 forks source link

Using DeepVoice3 mels spectrogram as wavenet vocoder's input #39

Closed yishaicohe closed 5 years ago

yishaicohe commented 6 years ago

I hope anyone can help me with a problem I got while trying to run both DeepVoice3 and Wavenet systems.

When I run TTS with Deepvoice3 on LJSpeech, I got a robotic sound. I know that Wavenet can predict a better reasults. I extract the mel spectrogram from DeepVoice3 system (hopefully I did it right) and tried to set it as an input for the Wavenet system. The results weren't so good, as the voice still has a robotic sound and lots of background whistling.

I am using UBUNTU 16.04 os. The model I used is LJ 410K iters

Hopefully anyone has any advice how can I improve results.

Thank in advance, Yishai

SvenShade commented 6 years ago

Yeah, I'm currently training on solo instrumental audio, and experiencing this same background whistling! Anyone else experienced the same thing, or know what might be causing it?

Shridharpawar77 commented 5 years ago

I am also facing same problem. Did anyone solved the problem?

stale[bot] commented 5 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.