NCJ Number
56185
Date Published
1977
Length
7 pages
Annotation
THIS ARTICLE FROM THE PROCEEDINGS OF THE 1977 INTERNATIONAL CONFERENCE ON CRIME COUNTERMEASURES DISCUSSES TEST RESULTS OF THE COMPUTER-OPERATED AUTOMATIC SPEAKER RECOGNITION SYSTEM.
Abstract
THE NEED FOR REPRODUCIBLE, OBJECTIVE ANSWERS CONCERNING A VOICE COMPARISON LED TO RESEARCH IN THE FIELD OF AUTOMATIC SPEAKER RECOGNITION BY COMPUTER. SPEAKER IDENTIFICATION, AND SPEAKER VERIFICATION WERE EXAMINED IN THE GERMAN GOVERNMENT-SPONSORED PROJECT FOR AUTOMATIC RECOGNITION OF SPEAKERS BY COMPUTERS (AUROS). THE AUROS SYSTEM CONSISTED OF A GENERAL PURPOSE COMPUTER AND A SET OF SIGNAL ANALYSIS PROCESSORS, FORMING A POWERFUL RESEARCH TOOL. TECHNICAL INFORMATION IS PROVIDED CONCERNING THE ASSEMBLY OF THE EQUIPMENT, AND ITS FUNCTION. TOPICS FOR THE DISCUSSION OF THE SYSTEMS OPERATION INCLUDE THE PREPROCESSING PHASE, THE SHORT-TIME ANALYSIS, AND CONTOUR ANALYSIS, AND THE STATISTICAL SPEECH-SIGNAL ANALYSIS. FOR SYSTEM PERFORMANCE EVALUATIONS THREE LARGE VOICE DATA BASES EACH CONTAINING 2,500 UTTERANCES WERE USED. IT WAS DEMONSTRATED THAT WITH THE LARGE VOICE DATA BASES, BOTH SPEAKER IDENTIFICATION AND SPEAKER RECOGNITION COULD BE PERFORMED AUTOMATICALLY AT AN ERROR RATE OF LESS THAN 1 PERCENT. ADDITIONAL INFORMATION IS PRESENTED CONCERNING THE SAMPLE SIZE FOR TRAINING THE SYSTEM, AND THE ELIMINATION OF TELEPHONE LINE DISTORTIONS. TABULAR DATA, GRAPHIC DISPLAYS, EQUATIONS, AND REFERENCES ARE PROVIDED. (TWK)