Identification and monitoring of voice information has been seen to have multiple DOD and commercial applications. The proposed program will develop a non-realtime prototype of a text-independent speaker identification system, which is based on a nonparametric classifier, and spectral features which incorporate speech parameter dynamics. The proposed approach builds upon previous experience in fielding operational speaker identification systems for experimental purposes. The Phase I effort will result in a performance assessment against the Switchboard Corpus, and a design for a real-time Phase II demonstration. The proposed technology should find uses in voice verification for security, law enforcement and intelligence. Commercial applications are anticipated in the area of credit card fraud reduction.
