Abstract: In this paper, we evaluate the performance of current state of the art speaker verification (SV) systems against some examples of spoofing and tampering attacks. We understand as spoofing the fact of impersonating another person using, for instance, a recording of his voice. On the contrary, we call tampering to the alteration of somebody's voice in order to prevent being detected by a SV system. These techniques can produce important performance degradations. We show that, for the EER operating point, spoofing can produce false aceptance rates of 68% and tampering misses rates of 50%. This is critical in some security applications which makes necessary to develop methods to detect manipulated speech signals.
Index Terms: speaker verification, forgery, disguise, spoofing, tampering.