I have a dataset containing MFCC features as input for a deep learning model. Now when I look at my mfccs they have large varying ranges of values (e.g. (-100,200),(0,5),(-1,1),...). Now I would like to normalize them for my model to be suited for deep learning (e.g. let them all be between (-1,1)).
Here is my problem: If I normalize each mfcc individually I am afraid that I loose relevant information that will distinguish one mfcc from another (like volume for example) If I normalize over a batch I could also loose information because of the large variation in ranges (I think... might be incorrect?).
How would one solve this issue?
