Sixth European Conference on Speech Communication and Technology
Recently, some large-scale text dependent speaker verification systems have been tested. They show that around 1% False Acceptance rate can be obtained on a test set when using an a priori threshold. So far, the majority of impostor tests are performed using speakers who don’t really try to fool the system. This paper explores some ways to generate more realistic impersonations. The first system uses recorded sentences of the speaker which has to be impersonated, then, using an automatic speech recognition system as to align temporally the input speech, phonemes segments are concatenated to generate the password of a registered speaker. The second system describes a simple prototype of a voice transformation vocoder used as an impostor generator. The goal of these impersonations is not only to assess speaker verification systems but also to understand the parameters that define the intra- and inter-speaker variabilities using parametric models of speech.
Full Paper (PDF) Gnu-Zipped Postscript
Bibliographic reference. Genoud, Dominique / Chollet, Gérard (1999): "Deliberate imposture: a challenge for automatic speaker verification systems.", In EUROSPEECH'99, 1971-1974.