jameslyons / python_speech_features

This library provides common speech features for ASR including MFCCs and filterbank energies.
MIT License
2.38k stars 617 forks source link

Std of log mel-filterbank will be close to zero in some dimension when nfilt == 80. #85

Closed TeaPoly closed 4 years ago

TeaPoly commented 5 years ago

import python_speech_features

y, sr = soundfile.read(file, dtype='float32') log_mfbank = python_speech_features.logfbank(y, samplerate=16000, nfilt=80, winlen=0.025, winstep=0.01, nfft=512, lowfreq=125, highfreq=7600, preemph=0.97, winfunc=np.hanning) std = np.std(features, axis=0) print(np.min(std))

tadangkhoa1999 commented 3 years ago

I have same problem. Have you fixed it?