5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

Fuzzy Gaussian Mixture Models For Speaker Recognition

Dat Tran, Tu Van Le, Michael Wagner

University of Canberra, Australia

A fuzzy clustering based modification of Gaussian mixture models (GMMs) for speaker recognition is proposed. In this modification, fuzzy mixture weights are introduced by redefining the distances used in the fuzzy c-means (FCM) functionals. Their reestimation formulas are proved by minimising the FCM functionals. The experimental results show that the fuzzy GMMs can be used in speaker recognition and it is more effective than the GMMs in tests on the TI46 database.

Full Paper

Bibliographic reference.  Tran, Dat / Le, Tu Van / Wagner, Michael (1998): "Fuzzy Gaussian mixture models for speaker recognition", In ICSLP-1998, paper 0798.