NCJ Number
88848
Date Published
1981
Length
60 pages
Annotation
A review of the available studies of the reliability of sound spectrography (voiceprint) results indicates they do not establish the technique's accuracy in realistic, forensic situations approximating the problems a criminalist encounters during a criminal investigation.
Abstract
A voiceprint is a pictorial representation of the acoustical energy output of a speaker as a function of time, frequency, and amplitude. It is a 'picture' (spectrogram) made by a machine (spectrograph). The voiceprint technique is essentially a visual comparison of such pictures of known and unknown speakers to determine their relative similarities and the exercise of subjective judgments to proclaim that the pictures are or are not similar enough to be considered the same voice. Although the odds of two fingerprints being identical are estimated to be 1 in 64 billion, similar proof is lacking for the distinctiveness of voice patterns. To supplement the inadequacies of a theory based on differences in bodily characteristics alone, proponents of the voiceprint technique theorize that such physical differences, in combination with use-patterns developed during growth, produce the uniqueness that characterizes a given speaker. This theory is unsupported by scientific data and runs contrary to general principles concerning the need to communicate and the phenomenon of dialect. Speech scientists do not yet know how to separate those features on a sound spectrogram thought to reflect speaker characteristics from the word characteristics, let alone identify such speaker characteristics or classify them. Thus, given the state of the research on voiceprint, it cannot satisfy the 'Frye' criterion of general acceptance of the underlying theory by the scientific community. A total of 151 footnotes are provided.