MEL frequency cepstral coefficients (MFCC) of original speakers and their imitators

Abstract

The results of intra- and interspeaker distances between MFCC vectors obtained from speech samples of eight well-known Polish personalities and their imitations performed by cabaret entertainers are presented and discussed. The intraspeaker distances between MFCC vectors representing normal and disguised speech samples of 10 speakers are also presented. The analysis of the measurement results indicated that, utilizing Euclidean distance between MFCC vectors, it is possible to differentiate the original speakers from the imitators. On the other hand, the MFCC vectors cannot be used to confirm the speaker's identity in the case of voice disguise.