Fusion of MFCC and MPEG-7 attributes for speaker verification

Abstract

Following the anticipation that some of the MPEG-7 audio descriptors hold glottal information differing than those MFCCs hold, possible contribution of MPEG-7 descriptors to speaker verification has been investigated. Both feature level and score level fusion of MFCCs and MPEG-7 descriptors have been studied. Results indicate improvements up to 18 % compared to those obtained by using MFCCs alone; a justification of the anticipation and a novel indication to the community. © 2006 IEEE.