Source microphone identification using multitaper MFCC features

Abstract

In this paper, we show studies of Multitaper spectrum estimation techniques to obtain the Mel-frequency cepstral coefficients (MFCCs) for source microphone identification systems. In the previous works, MFCCs were computed commonly from a single windowed (Hamming) DFT spectrum for source microphone identification. Emphasis on Multitaper MFCC features led us to extract intrinsic source microphone features obtained from non-speech segments. Experimental results on a set of 16 microphones yield that the Multitaper MFCC features show better performance compared to the Hamming windowed MFCC features.