search

NVIDIA / mellotron

Mellotron: a multispeaker voice synthesis model based on Tacotron 2 GST that can make a voice emote and sing without emotive or singing training data
BSD 3-Clause "New" or "Revised" License
854 stars 187 forks source link

Warning Message in yin #47

Closed pneumoman closed 2 years ago

pneumoman commented 4 years ago

For some of my audio I'm getting a warning message:

yin.py:44: RuntimeWarning: invalid value encountered in true_divide cmndf = df[1:] * range(1,N) / np.cumsum(df[1:]).stype(float) #scipy method

Any ideas? The problem children are encoded exactly like their peers.

Thanks

aijianiula0601 commented 4 years ago

I got the same error.

rasenganai commented 2 years ago

Same warning

jinny1208 commented 2 years ago

For some of my audio I'm getting a warning message:

yin.py:44: RuntimeWarning: invalid value encountered in true_divide cmndf = df[1:] * range(1,N) / np.cumsum(df[1:]).stype(float) #scipy method

Any ideas? The problem children are encoded exactly like their peers.

Thanks

Did you solve the problem? If so, may I ask how you did so? If not, did you ignore the warning and continue training? How were the results?

pneumoman commented 2 years ago

My impression is that this is caused by the whole window segment being silence. I think I may have fixed this by just adding a little offset to the denominator to keep it from being a true zero. The other approach is to curate your dataset to remove any excess silence from your recordings

jinny1208 commented 2 years ago

My impression is that this is caused by the whole window segment being silence. I think I may have fixed this by just adding a little offset to the denominator to keep it from being a true zero. The other approach is to curate your dataset to remove any excess silence from your recordings

Thank you. Adding offset to the denominator got rid of the warning.