NCJ Number
34386
Date Published
1974
Length
105 pages
Annotation
THIS REPORT DESCRIBES THE SEMI-AUTOMATED SPEAKER IDENTIFICATION SYSTEM (SASIS), WHICH COMPARES PAIRS OF SPEECH SAMPLES TO DETERMINE WHETHER THEY WERE SPOKEN BY THE SAME OR DIFFERENT SPEAKERS.
Abstract
THE SYSTEM UTILIZES A MINICOMPUTER AND ASSOCIATED PERIPHERALS TO ANALYZE THE SAMPLES. ANALOG SPEECH IS DIGITIZED AND INPUT TO COMPUTER MEMORY. THE SYSTEM OPERATOR USES AN INTERACTIVE GRAPHIC DISPLAY TERMINAL TO IDENTIFY AND LABEL SPECIFIC PHONETIC EVENTS WHICH WERE DETERMINED EXPERIMENTALLY ON A LARGE DATA BASE TO HAVE A HIGH DEGREE OF DISCRIMINATING POWER. ONCE THIS LABELING IS DONE FOR THE TWO SPEECH SAMPLES WHICH ARE TO BE COMPARED, THE COMPUTER CALCULATES NUMERICAL 'DISTANCES' BETWEEN PAIRS OF LIKE PHONETIC EVENTS IN THE TWO SAMPLES. THESE 'DISTANCES' ARE THEN COMBINED TO FORM A SINGLE NUMERICAL MEASURE OF THE 'SIMILARITY' BETWEEN THE TWO SPEAKERS. THE REPORT DESCRIBES THE METHODS BY WHICH THE COMPARISON TECHNIQUES WERE DEVELOPED, INCLUDING A SUMMARY OF PREVIOUS WORK DONE IN THIS AREA. DETAILED DESCRIPTIONS OF THE SYSTEM DESIGN AND FUNCTIONAL OPERATION ARE GIVEN, AND THE RELATIONSHIP BETWEEN THIS COMPUTER-AIDED APPROACH AND THE 'VOICEPRINT' METHOD IS DISCUSSED. EVALUATION OF THE SASIS TECHNIQUES ON A STATISTICALLY SIGNIFICANT TEST POPULATION GAVE EXCELLENT RESULTS. SEE ALSO NCJ-34383. (AUTHOR ABSTRACT)