Voice transformation-based spoofing of text-dependent speaker verification systems

Abstract

In the past few years state-of-the-art text-dependent speaker verification technology has improved significantly in terms of the ability to accept target speakers and reject imposters. As a result, the use of speaker verification systems for real world security is increasing. Real world usage of speaker verification technology raises the issue of spoofing attacks. As part of our efforts for developing countermeasures for such attacks, we describe in this paper a spoofing attack based on simple voice transformation and report an analysis of the vulnerability of several state-of-the-art text-dependent algorithms to such an attack. Copyright © 2013 ISCA.