Detection of mimicry attacks on speaker verification system for cartoon characters’ dataset

Abstract

Automatic speaker verification (ASV) is the process of automatically verifying the identity of the claimant speaker using his voice biometric. In this paper, we have made an attempt to investigate the effect of conventional short-term features on the performance of speaker verification system in the presence of Mimicry attacks. We have prepared a small dataset for some popular cartoon characters. Initial results show that the best false acceptance rate (FAR) achieved is 22.90, with the use of stand-alone linear predictive coding (LPC) features.