Fuzzy Cluster Analysis and its Evaluation Method
Shingo Kuroiwa, Satoru Tsuge, Fuji Ren · 2009
Abstract: Recently, Distributed Speech Recognition (DSR) systems are widely deployed in Japanese cellulartelephone networks. In these systems, personal authentication with voice is strongly desired. In this paper, wepresent several speaker recognition techniques developed in the University of Tokushima for Distributed SpeakerIdentiflcation/Veriflcation (DSI/DSV) systems. Especially, we present recent progress on a non-parametricspeaker recognition system that is robust to quantization in the distributed systems comparing with conventionalspeaker recognition systems based on Gaussian Mixture Model (GMM). Evaluation results using the Japanesede facto standard speaker recognition corpus and CCC Speaker Recognition Evaluation 2006 data developedby the Chinese Corpus Consortium (CCC) show higher performance of the proposed method than GMM andVQ-distortion in the European Telecommunications Standards Institute (ETSI) DSR standard environment.Keywords: Voice Authentication, Speaker Recognition, DSR, DSV, Earth Mover’s Distance