I am using MFCC to compare the same speech same recorded with 2
different microphones. When I check the correlation of the MFCC data
it comes back with only a 32% positive correlation. One sample sounds
very clean (headset), the other sounds like it's being played back
through a transistor radio, but they are the same sound sample.
Is there a common technique either during pre-processing of the sample
or during the comparison of the MFCC's to erradicate any difference?
Reply by Raeldor●January 20, 20112011-01-20
Oh, btw I am already applying a high pass filter at the beginning and
then using cepstral mean subtraction on the MFCC's at the end.