No window function used during feature extraction?

DeepSpeech
#1

For the last release 0.8.2, it seems no window function is used during feature extraction. In the function “audio_to_features” from feeding.py, we can see that audio signals are transformed into mfcc by calling contrib_audio.audio_spectrogram and contrib_audio.mfcc.

However, no window function is specified. Can anyone tell me whether there is really no window function used?

#2

contrib_audio.audio_spectrogram uses a periodic Hann window.

1 Like
#3

Thanks for your confirmation.