NCJ Number
32111
Date Published
1974
Length
564 pages
Annotation
REPORT ON THE TECHNICAL STUDIES INVOLVED IN DEVELOPING THE TECHNIQUES USED IN THIS COMPUTER-AIDED SYSTEM DESIGNED TO ANALYZE, PARAMETERIZE, AND COMPARE SELECTED SAMPLES OF CRIMINAL AND SUSPECT UTTERANCES.
Abstract
BOTH MATHEMATICAL AND EXPERIMENTAL METHODS WERE USED TO ACQUIRE A PROCESS A DATA BASE OF OVER 6000 SPECIALLY DESIGNED SENTENCES, RECORDED OVER A TELEPHONE CHANNEL FROM 254 SPEAKERS, YIELDING OVER 35,000 USABLE PHONETIC SEGMENTS. THIS REPORT DESCRIBES THE 20 TASKS (EACH TREATING A PARTICULAR ASPECT OF ALGORITHM DEVELOPMENT) COMPRISING THIS PHASE. INDIVIDUAL CHAPTERS DISCUSS THE DESIGN AND ACQUISITION OF THREE VALID SPEECH DATA BASES, THE LABELING AND SEGMENTING OF PHOENTIC DATA, RESEARCH ON THE EFFECTS OF COARTICULATION ON SPEAKER COMPARISON, FEATURE EXTRACTION, THE REDUCTION OF SYSTEM COMPLEXITY, DISTANCE AND SIMILARITY DESIGN AND OPTIMIZATION, AND THE DEVELOPMENT OF SYSTEM PERFORMANCE TESTING TECHNIQUES. A THREE-PAGE LIST OF REFERENCES, NUMEROUS EXPLANATORY FIGURES, CHARTS, AND TABLES, AND AN EXTENSIVE APPENDIX ARE PROVIDED.