Speaker identity verification over telephone lines: where we are and where we are going

Abstract

The ability to verify the identity of a speaker by analyzing speech, or Speaker Identity Verification (SIV), is an attractive means of providing security in the telephone network. As with other biometric security techniques, SIV uses a quality inherent in the user being verified, rather than a possession (e.g. badge) or piece of knowledge (e.g. password). SIV has a further advantage since it does not require elaborate customer premises equipment: only a telephone is needed. We will discuss potential applications of voice-based security and present accuracy data from a series of evaluative experiments with a Bellcore developed verifier. In addition, we will offer suggestions on how verification performance might be further improved, based on data collected using human listeners. Finally, although SIV system performance is quite good, there remain sonic difficult problems. Foremost among these is that individuals' voices are known to. change over time in ways that are poorly understood. Until more is known of the causes behind such changes, they must be acknowledged and dealt with in any potential application. We will discuss empirical and theoretical considerations in template management and criterion determination, particularly as they relate to changes that occur over time.