Robust speaker verification using GFCC and joint factor analysis

In real world situation performance of speaker verification system drops significantly because of mismatched training and test conditions. In this paper we have analyzed three factors namely noise, channel variability and session variability, that are responsible for poor performance of a speaker verification system. The first step towards noise robustness GFCC features were used as recent research has shown better noise robustness of gammatone frequency cepstral coefficients over mel-frequency cepstral coefficients. In the second step robustness towards session and channel variability is achieved by shifting from the classical way of modeling a speaker to a rather new approach of joint factor analysis. Experimental results over different acoustic environment and over different SNR have shown significant improvement in the performance of the system.