Design of a Text Independent Speaker Recognition System

Abstract

This paper presents the design of a text independent speaker recognition system based on Mel-Frequency Cepstrum Coefficients and Gaussian Mixture Models. HTK speech recognition toolkit is used in the design of speaker models. The system is aimed to use it as a biometric authentication system. The experiments were performed on speech data consist of 134 speakers from YOHO database for different training conditions. The increase of the proposed system performance is observed with the decrease of Equal Error Rate. Experiment results show that the system gives the best recognition performance for Gaussian mixture model with 64 mixtures.