G-Wang/WaveRNN-Pytorch

There are some metallic noises in high frequncy at silence segment

Opened this issue · 2 comments

Hi,
I use your repo to train my model. It works pretty well, but I find there are some metallic noises in high frequency at synthesized silence segment. It exist almost on all models(all steps). An example is as follows:
图片

Hi, I'm not really sure what the issues could be, I haven't encountered these issues in my datasets. Do you have audio samples of this metallic noise? perhaps you can try different hparams setting for audio to see if that improves it.

Same here. Any updates?

PS: 22050Hz, 256 hop_size, 10bit model