SUMMARY REVIEW OF PROCEDURES FOR SPEAKER RECOGNITION

• NCJ Number: 1520
• Author(s): ANON
• Date Published: 1970
• Length: 20 pages
• Grant Number(s): NI-70-004
• Sponsoring Agency: US Dept of Justice; US Dept of Justice NIJ Pub
• Corporate Author: Stanford Research Institute
• Sale Source: National Institute of Justice/
• Language: English
• Country: United States of America

Annotation

THREE GENERAL METHODS OF SPEAKER RECOGNITION, LISTENING, COMPARISON OF SPECTROGRAMS, AND BY MACHINE ARE REVIEWED.

Abstract

SPEAKER RECOGNITION BY LISTENING HAS BEEN STUDIED FOR A LONGER PERIOD OF TIME AND APPEARS TO BE MORE ACCURATE AND RELIABLE THAN EITHER OF THE OTHER METHODS AS THEY ARE NOW PRACTICED. A POSSIBLE LIMITATION OF THIS METHOD IS THAT IT IS ENTIRELY SUBJECTIVE. SPEAKER RECOGNITION BY VISUAL COMPARISON OF SPECTROGRAMS IS CONSIDERED TO BE A MORE OBJECTIVE METHOD. SPECTROGRAMS ARE VISUAL DISPLAYS OF THE SPEECH SIGNAL. THEY EXHIBIT GRAPHIC FEATURES THAT CAN BE DISCUSSED IN A FAIRLY OBJECTIVE MANNER. BUT THESE FEATURES ARE STILL INTERPRETED SUBJECTIVELY IN ARRIVING AT AN OVERALL DECISION. FOR THIS REASON THERE HAS BEEN MUCH INTEREST IN A THIRD METHOD, SPEAKER RECOGNITION BY MACHINE. ALTHOUGH MACHINE DECISIONS ARE INHERENTLY OBJECTIVE, THEY ARE, AS OF NOW, OFTEN LESS ACCURATE FOR SPEAKER RECOGNITION PURPOSES THAN COMPARABLE HUMAN DECISIONS. CURRENT RESEARCH EFFORTS IN SPEAKER RECOGNITION BY MACHINE ARE SPECIFICALLY DIRECTED TOWARD OVERCOMING THIS LIMITATION. REFERENCES. AUTHOR ABSTRACT