Improved accuracy of speaker verification system in noisy environment

Abstract

Automated speaker verification is the computing task of validating a user's claimed identity using characteristics or we can say features extracted from their voices. In automated speaker verification the speaker's voice signal is processed to extract speaker-specific characteristics. We have proposed a method for speaker verification system in noisy environment using the auditory features Gammatone Frequency Cepstral Coefficient (GFCC) and Mel Frequency Cepstral Coefficient (MFCC). The score level fusion of both the feature improves the matching accuracy and reliability of this speaker verification system in noisy and unconstrained environment. Also as the system is challenge response based, it verifies the spoken phrase with the prompted phrase making the system more robust against recorded imposter samples. The experimental results and its comparative study show the robustness and reliability of our proposed approach.