SPEAKER IDENTIFICATION BY SPEECH SPECTROGRAMS - SOME FURTHER OBSERVATIONS (FROM SELECTED MATERIALS ON LEGAL AND TECHNICAL ASPECTS OF 'VOICEPRINT' IDENTIFICATION, 1976 - SEE NCJ-44202)

• NCJ Number: 44205
• Author(s): R H BOLT; F S COOPER; E E DAVID; P B DENES; J M PICKETT; K N STEVENS
• Date Published: 1976
• Length: 4 pages
• Corporate Author: National College of Criminal Defense Lawyers and Public Defenders
• Publication Format: Document
• Language: English
• Country: United States of America

Annotation

RESEARCH ON SPEAKER IDENTIFICATION THROUGH ANALYSIS OF SPEECH SPECTROGRAMS (VOICEPRINTS) IS REVIEWED, AND FACTORS AFFECTING THE RELIABILITY OF IDENTIFICATION ARE DISCUSSED.

Abstract

THE REVIEW FOCUSES ON A STUDY OF THE EFFECTS OF FIVE VARIABLES ON THE ACCURACY WITH WHICH OBSERVERS INTERPRET VOICEPRINTS. THE STUDY, IN WHICH OBSERVERS ATTEMPTED TO MATCH VOICEPRINTS UNDER A NUMBER OF EXPERIMENTAL CONDITIONS, CONSIDERED TWO TYPES OF IDENTIFICATION ERRORS: FALSE IDENTIFICATION AND MISSED IDENTIFICATION (FALSE REJECTION). IN THE FORENSIC SITUATION, FALSE IDENTIFICATION COULD LEAD TO THE CONVICTION OF AN INNOCENT PERSON, WHILE FALSE REJECTION COULD RESULT IN THE ELIMINATION OF A GUILTY PERSON FROM CONSIDERATION AS A SUSPECT. THE STUDY RESULTS ARE SAID TO PROVIDE DIRECT EVIDENCE OF THE DETRIMENTAL EFFECT OF INTRASPEAKER VARIABILITY ON VOICE IDENTIFICATION AND OF THE INHERENT DISSIMILARITY BETWEEN VOICE IDENTIFICATION AND FINGERPRINT IDENTIFICATION. THE STUDY FINDINGS DO NOT MAKE CLEAR WHICH SPECTROGRAPHIC FEATURES CORRELATE MOST CLOSELY WITH SPEAKER IDENTITY. CONSIDERABLE DISAGREEMENT WAS FOUND AMONG PANELS OF OBSERVERS ASKED TO PERFORM THE SAME MATCHING TASK, THE INDICATION BEING THAT FURTHER STUDY OF THE DECISION PROCESS USED BY VOICEPRINT INTERPRETERS IS NEEDED. STUDY FINDINGS DEMONSTRATE THE INFLUENCE OF SEVERAL IMPORTANT VARIABLES ON THE ACCURACY OF IDENTIFICATION. IDENTIFICATION ERROR RATES WERE HIGHER WHEN VOICE SAMPLES TO BE COMPARED WERE RECORDED ON DIFFERENT OCCASIONS, WHEN VOICE SAMPLES CONSISTED OF WORDS IN SENTENCES RATHER THAN ISOLATED WORDS, AND WHEN OBSERVERS WERE NOT TOLD WHETHER THE UNKNOWN SPEAKER (I.E., THE 'VOICEPRINT' TO BE MATCHED) WAS ACTUALLY INCLUDED IN THE GROUP OF KNOWN SPEAKERS (I.E., POSSIBLE MATCHES). HOWEVER, THE STUDY NEITHER PROVED NOR DISPROVED THE RELIABILITY OF THE VOICEPRINT IDENTIFICATION TECHNIQUE.