librosa.feature.melspectrogram is not equal as kaldi's mfcc extraction

[KnowBetterHelps]

the same audio, when I use librosa.feature.melspectrogram to extracte log mfcc, says Matrix A(43,N). But it turns out that Matrix A is not equal as kaldi's mfcc extraction, says Matrix B(43,M), even their dims, N != M. Got very confused, Could any help me?

[zhuimin]

The method of dividing frames in librosa and Kaldi is different, and you can refer this issue:
librosa/librosa#595

[xjwla]

Is it because melspectrogram equal to fbank?not mfcc?