I have a developed a keras lstm model to audio emotion recognition (using audio MFCC\'s for training and adding a padding to maintain the same mfcc size for all files) but w