Detect mimicry by enhancing the speaker recognition system

Abstract

Mimicry voice sample is a potential challenge to the speaker verification system. The system performance is highly depended on the equal error rate. If the false accept to reduce, then the equal error rate decrease. The speaker verification process, verifies the claim voice is originally produced by the said speaker or not. The verification process is highly depended upon the biometric features carried out by the acoustic signal. The pitch count, phoneme recognition, cepstral coefficients are the major components to verify the claim voice signal. This paper shows a novel frame work to verify the mimicry voice signal through the two-stage testing. The first stage is GMM based speaker identification. The second stage of testing filters the identification through the various biometric feature’s comparisons.