CONVERSATIONAL BIOMETRICS

Abstract

In this paper, we present a new modality for speaker recognition: conversational biometrics. By combining diverse simultaneous conversational technologies, high accuracy transparent speaker recognition becomes possible even in channel or environment mismatches. For speaker identification over very large populations, we combine dialogs to reduce the set of confusable speakers and text-independent speaker identification to pin-point the actual speaker. Similarly, dialogs with personal random or predefined questions are used to perform simultaneously knowledge-based and acoustic-based verifications of the user. Adequate design of the dialog allows to tailor the ROC curves to the needs of most applications.