NCJ Number
64187
Journal
International Criminal Police Review Issue: 332 Dated: (NOVEMBER 1979) Pages: 254-270
Date Published
1979
Length
17 pages
Annotation
SYSTEMS COMPONENTS AND EXPERIMENTAL RESULTS ARE DESCRIBED FOR THE AUTOMATIC RECOGNITION OF SPEAKERS (AUROS) WHICH WAS DEVELOPED TO COMPARE THE MERITS OF DIFFERENT SPEAKER IDENTIFICATION AND VERIFICATION USING COMPUTERS.
Abstract
AN OVERVIEW OF THE DEVELOPMENT OF SOFTWARE COMPUTER SYSTEMS TO IDENTIFY VOICES INTRODUCES THE CONCEPTS AROUND WHICH AUROS WAS DESIGNED. ATTENTION IS FIRST GIVEN TO THE TWO FUNDAMENTAL PROBLEMS SUCH A SYSTEM MUST OVERCOME--THE UNIQUENESS OF SPEECH UTTERANCES AND CONSEQUENT NONREPRODUCIBILITY AND, SECONDLY, THE ENORMOUS AMOUNT OF DATA TO BE PROCESSED IN SHORT INTERVALS. GENERAL AUROS EQUIPMENT IS DESCRIBED IN TERMS OF ITS DATA COLLECTION, WHICH OCCURS THROUGH ANALYSIS EITHER IN REAL TIME ON CONSECUTIVE TIME SEGMENTS OF THE SPEECH SIGNAL, OR BASED ON THREE LINEAR TRANSFORMATIONS. THREE PHASES USED IN THE PROCEDURES FOR SPEAKER RECOGNITION AARE EXAMINED: PREPROCESSING THE SPEECH SIGNAL, 'TRAINING' THE SYSTEM, AND TESTING THE SYSTEM'S RELIABILITY; IN EXAMINING THESE PHASES, AN OUTLINE OF SPEECH ANALYSIS PROCEDURES AND CLASSIFICATION ALGORITHMS IS PRESENTED. CATEGORIES DESCRIBED ARE THE MINIMUM-DISTANCE, THE DISTRIBUTION-FREE TOLERANCE-REGION, AND THE MINIMUM-RISK. TEST RESULTS ARE REPORTED THAT RELATE TO AUROS ABILITY TO ACHIEVE SPEAKER IDENTIFICATION AND VERIFICATION AND TO DEAL WITH THE EFFECTS OF VOICE TRANSMISSION BY TELEPHONE LINE. CONTOUR AND SPEECH-SIGNAL ANALYSIS, COMPUTER REPRESENTATIONS OF VOICE WAVES, CLASSIFIERS AND ALGORITHMS ARE ILLUSTRATED IN FIGURES AND DIAGRAMS. REFERENCES ARE PROVIDED. (MHP)