Vulnerability in speaker verification - a study of technical impostor techniques

Abstract

This paper reports on some ways to try to deceive a state-of-the-art speaker verification (SV) system. In order to evaluate the risk in SV systems one has to take into account the possible intentional impostors who know whom they are attacking. We defined a worst case scenario where the impostor has extensive knowledge about the person to deceive as well as the system to attack. In this framework we tested our SV system against concatenated client speech, re-synthesis of the client speech and diphone synthesis of the client.