A TMs32020-based real time, text-independent, automatic speaker verification system

A fast, reliable, yet inexpensive automatic speaker verification system based on the Texas Instruments TMS32020 digital signal processor (DSP) is described. This system uses a novel speaker verification algorithm which operates in 75% of real time and requires two to three seconds of unconstrained speech to perform accurate authentication. Experimental results on a population of 90 speakers are also given. The rule implemented in the final prototype uses a sequential strategy to arrive at the authentication decision. Hence, the users of the system are required to speak until a decision with high confidence is announced.<
>