jameslyons/python_speech_features

missing frequencies in Mel FillterBank

saddlekiller opened this issue · 2 comments

saddlekiller commented

Hi,

When I tried to extract fbank features I found missing frequencies. So, I checked the filter bank in your code and found this:
image
I am curious about why remove values in 2nd and 6th rows? Is there any trick or that's a bug actually?

tadangkhoa1999 commented

Hi, i have same problem. Have you fixed it?

jameslyons commented

Hi, i have same problem. Have you fixed it?

This is because of rounding in the mel.filterbank calculation. Increase NFFT to fix it. In any case, it wont affect the accuracy of derived features to any significant degree.