Text-constrained speaker verification using fuzzy C means vector quantization

Abstract

The most successful approach to speech and speaker recognition is to treat the speech signal as a stochastic pattern and to use a statistical pattern recognition technique for matching utterances. This paper attempts to study the performance of Text dependent speaker verification system using Delta-Delta Mel Frequency Cepstral Coefficients (MFCC-Δ-Δ) feature vector and Fuzzy C means (FCM) speaker modelling technique. Speaker-specific information which is mainly represented by spectral features, are used in respective models which serves as an important parameter for determining the claim of the speaker. The experimental results performed on microphonic database suggest that accuracy significantly depends on the value of learning parameter of the objective function of FCM. Our work focuses on total success rate or accuracy and the effect of learning parameter of FCM on improving the accuracy.