Front-end Diversity in Fused Speaker Recognition Systems
Karen Kua, Julien Epps, Eliathamby Ambikairajah, Mohaddeseh Nosratighods · 2010
Due to the increasing use of fusion in speaker recognition systems, one thread of current research activity focuses on new features to complement MFCCs that can advance the current state of the art in fused systems. In this paper, we investigate some possible variations to the extraction of MFCCs that produce diversity with respect to fused subsystems based on different MFCC-variant features. In particular, the use of different filter shapes is found to provide modified MFCCs that perform promisingly in fused systems, providing the filterbank ripple is minimised. Evaluations on the NIST 2006 SRE database show a relative improvement of 17% in EER when one modified MFCC subsystem is fused with a conventional MFCC-based system, and an improvement of 22% when two modified MFCC subsystems are fused.