Voice Recognition Over the Telephones

• NCJ Number: 80101
• Journal: Journal of Police Science and Administration Volume: 9 Issue: 3 Dated: (September 1981) Pages: 280-284
• Author(s): H A Rathborn; R H Bull; B R Clifford
• Date Published: 1981
• Length: 5 pages
• Publication Format: Article
• Language: English
• Country: United States of America

Annotation

This paper reports results of a British study designed to test the commonsense belief that if during a crime the criminal's voice is heard over the telephone, then testing for voice recognition without using the telephone is inappropriate and error prone.

Abstract

Subjects comprised 140 members of the general public, including 49 males and 91 females, ranging in age from 18 to 75 years. The subjects were randomly assigned to four experimental groups, which were defined by the tapes that they heard. Five tapes were heard, including a practice tape which every subject heard. The four experimental tapes all had the same format and consisted of a target voice (the voice to be later recognized), followed by an alerting tone and the recognition set consisting of six numbered voices, one being the same person talking as the target. Tape 1 was a target voice taped naturally, with the recognition set taped naturally. Tape 2 had a target voice taped naturally, and the recognition set taped over the telephone. Tape 3 had a target voice taped over the telephone and the recognition set taped naturally. Tape 4 had a target voice taped over the telephone and the recognition set taped over the telephone. All tapes used the same speakers, each uttering the same 13-word sentence in identical order. The results showed that the conditions involving the telephone rrecordings (tapes 2, 3, and 4) all resulted in similar mean correct identification scores. Increased recognition accuracy under the naturally recorded recognition set condition was only experienced when the presentation mode was similarly recorded. In addition, subjects were found to give higher ratings of confidence (in their answers) when making correct recognitions than when making erroneous ones. Implications of these findings are that if the original criminal's voice is heard over the telephone, testing for recognition using the telephone will not result in greater accuracy than not using the telephone. Moreover, these findings show that in the case of voice recognition, a significant positive relationship exists between witness accuracy and witness confidence in their accuracy. Two tables, 1 figure, and 14 references are included.