Learnable Window

[gunikavashisht13472]

Could you please elaborate why you have not used Learnable_window in STFT , Mel Spectrograms and MFCC but used in their inverse counterparts?

[KinWaiCheuk]

The learnable kernels are, by default, disable in all STFT, Mel spectrograms, MFCC, and their inverse counterparts.

If you are referring to the argument refresh_win as shown below, it is not for learnable kernels. It is for recalculating the window_sumsquare for different audio lengths, which is essential to obtained a correct inverse. If all of your audio clips are of the same length, you can make refresh_win=False to speed up the calculation a little bit.

nnAudio/Installation/nnAudio/Spectrogram.py

Line 47 in e3ad18d

def inverse_stft(self, X, kernel_cos, kernel_sin, onesided=True, length=None, refresh_win=True):

[gunikavashisht13472]

Thank you for the answer. I want to use RNNS instead of CNNs in my maodel. Will this code work for RNNs too?

[KinWaiCheuk]

Thank you for the answer. I want to use RNNS instead of CNNs in my maodel. Will this code work for RNNs too?

Yes it will work. nnAudio is just for spectrogram extraction using GPU. Once you have that spectrogram, you can use any model of your choice. The CNN in the example is just a demonstration on how to use nnAudio together with a pytorch model.