Accession Number : ADA307810
Title : Speaker Identification Technology.
Descriptive Note : Final rept. Dec 92-Feb 94,
Corporate Author : BOOZ-ALLEN AND HAMILTON INC MCLEAN VA
Personal Author(s) : Ricart, Richard
PDF Url : ADA307810
Report Date : JAN 1996
Pagination or Media Count : 134
Abstract : The purpose of this effort was to explore speaker identification algorithms for military applications. Previously developed algorithms have not been robust to the short, noisy transmissions generally found in military operational communications. New and emerging technologies were evaluated and some basic assumptions of current speaker identification technology were reevaluated. The most important notion challenged was that high performance speaker identification is not possible with very short training utterances (e.g., 5 seconds).
Descriptors : *ALGORITHMS, *NEURAL NETS, *SPEECH RECOGNITION, DATA BASES, MATHEMATICAL MODELS, SIGNAL PROCESSING, OPTIMIZATION, REAL TIME, PARAMETERS, SIGNAL TO NOISE RATIO, LEARNING MACHINES, IDENTIFICATION, MATHEMATICAL FILTERS, MATHEMATICAL PREDICTION, MILITARY APPLICATIONS, ACOUSTIC SIGNALS, PATTERN RECOGNITION, SPEECH ANALYSIS, SPEECH TRANSMISSION, SYSTEMS ANALYSIS, ACOUSTIC FILTERS, ACOUSTIC DATA, TACTICAL COMMUNICATIONS, BANDPASS FILTERS, STRATEGIC COMMUNICATIONS, AUDITORY PERCEPTION, PITCH DISCRIMINATION.
Subject Categories : Cybernetics
Distribution Statement : APPROVED FOR PUBLIC RELEASE