NCJ Number
42628
Date Published
1977
Length
187 pages
Annotation
A FINAL REPORT ON EFFORTS TO DEVELOP A COMPUTER-ASSISTED SPEAKER IDENTIFICATION SYSTEM FOR USE IN INVESTIGATION, AS WELL AS IN COURTROOM TESTIMONY, AND TO INVESTIGATE OTHER APPLICATIONS OF SPEAKER IDENTIFICATION TECHNOLOGY.
Abstract
BASED ON RESEARCH FUNDED BY THE NATIONAL INSTITUTE OF LAW ENFORCEMENT AND CRIMINAL JUSTICE, A SPEAKER COMPARISON ALGORITHM OPERATING ON THE STEADY STATE FEATURES OF SPEECH WAS SYNTHESIZED. THE SPEAKER IDENTIFICATION SYSTEM WAS THEN EVALUATED USING LABORATORY-DERIVED DATA AND FOUND TO PRODUCE SPEAKER IDENTIFICATION ACCURACIES IN EXCESS OF 97 PERCENT ACROSS A BROAD VARIETY OF TEXTS AND VOICE CHARACTERISTICS WHENEVER TEN PHONETIC EVENTS WERE PRESENT IN THE VOICE RECORDINGS. THE COMPUTER-AIDED SPEAKER IDENTIFICATION SYSTEM WAS DESIGNED TO BE OPERATED BY THE VOICEPRINT EXAMINER OR ONE WHO IS SIMILARLY TRAINED TO LOCATE SOUNDS FROM VOICE SPECTROGRAMS AND TO PHONETICALLY DESCRIBE SOUNDS ACCURATELY BY LISTENING. AFTER LIMITED TESTING UNDER EXTENDED LABORATORY CONDITIONS, THE SEMI-AUTOMATIC SPEAKER IDENTIFICATION SYSTEM APPEARED INVARIANT ACROSS THE STANDARD ENGLISH, BLACK URBAN, AND CHICANO FEMALE DIALECTS, SPEECH WITH SIMULATED STRESS, AND SPEECH WITH NASAL DISGUISES. THERE WAS SOME INCONSISTENCY IN THE RESULTS WHEN COMPARING SPEAKERS WITH THE BLACK URBAN MALE DIALECT, CHICANO MALE DIALECT, OR BLACK URBAN MALE DISGUISES. THE MOST SIGNIFICANT RESULT WAS THAT THE TELEPHONE CHANNEL RESPONSE WAS FOUND TO DOMINATE SPEAKER COMPARISONS. THE SEMI-AUTOMATIC SPEAKER IDENTIFICATION SYSTEM THEN UNDERWENT A PILOT TEST IN A VOICEPRINT LABORATORY WHERE THREE VOICEPRINT EXAMINERS, TRAINED TO OPERATE THE SYSTEM, WHO PROCESSED CRIMINAL EVIDENCE AND SIMULATED RECORDINGS ON THE SYSTEM. THE PILOT TEST DEMONSTRATED SHORTCOMINGS IN THE OPERATION OF THE SPEAKER IDENTIFICATION SYSTEM, EACH ACCOMPANIED BY RECOMMENDED DESIGN IMPROVEMENTS. THE MAJOR PROBLEM ENCOUNTERED WAS, AGAIN, THAT THE TELEPHONE CHANNEL RESPONSE DOMINATES THE SPEAKER COMPARISON. THE OTHER PROBLEMS ENCOUNTERED WERE OPERATOR VARIABILITY AND INEFFICIENT MAN-MACHINE INTERACTION. A DESIGN OPTIMIZATION REQUIREMENTS DEFINITION WAS DEFINED WITH A RECOMMENDED SHORT-TERM STUDY TO BE MADE ON THE SEMI-AUTOMATIC SPEAKER IDENTIFICATION SYSTEM TO DETERMINE IF THERE WAS A FEASIBLE SOLUTION TO THE PROBLEMS OF NOISE AND CHANNEL EFFECTS AND OTHER UNDESIRABLE PARAMETERS SUCH AS PHASE DISTORTION AND RECORDER CLIPPING. THE SPEAKER IDENTIFICATION PROGRAM WAS TERMINATED BY THE LAW ENFORCEMENT ASSISTANCE ADMINISTRATION BEFORE THE FEASIBILITY STUDY WAS COMMENCED. (AUTHOR ABSTRACT) ...ELW