Hiroshiba / become-yukarin

Convert your voice to favorite voice
https://hiroshiba.github.io/blog/became-yuduki-yukari-with-deep-learning-power/
MIT License
576 stars 88 forks source link

When extracting acoustic features, process goes into sleep indefinitely #72

Open VincentThiemann opened 4 years ago

VincentThiemann commented 4 years ago

Hello All!

I have been trying to get Become Yukarin working for the past few weeks, and I've been running into a consistent problem.

Whenever I try to run extract_acoustic_feature.py, the process runs indefinitely. It seems to run normally for a few minutes (ps aux outputs R on extract_acoustic_feature, the processes are consuming resources), but after a while, all the processes go to sleep (s+ or Sl+) and quit consuming resources. I've let it sit for a day and a half, but nothing ever happens, so I just end it early.

This is what I run: python3 scripts/extract_acoustic_feature.py \ -i1 './input_wav/' \ -i2 './target_wav/'\ -o1 './input_feature/' \ -o2 './target_feature/'

Inside of input_wav is a 1 hour 50 minute wav file, and in target_wav is a 2 hour and 54 minute wav file.

The only exception to the issue I have above is when I run extract_acoustic_feature using the 'dio' estimation method, and with very short audio files (<10 minutes). In that case, it does output the feature, mean, and var files, but they do not work in training, and are always the exact same size no matter what wav I use: 9.9 mb for feature, and 615 kb for mean and var.

Does anyone have any idea what may be the issue?