Voice Stress Analysis Stirs Controversy & Debate

BY STEPHEN COX

In an announcement in the British parliament in April 2007, the Work and Pensions Secretary, John Hutton, said that "voice-risk analysis software" would be trialed by Harrow Council, in north London, from May, for the purpose of detecting fraudulent housing benefit and council tax benefit claims. "Voice risk analysis software" is another name for "voice stress analysis" (VSA), a technology that has been marketed by several companies since the 1970s. It is based on the proposition that "telling a significant lie will produce some degree of psychological stress" [1], and detecting the presence of stress in the voice. There has been a certain amount of research on the detection of stress in speech, and it appears it can be done - for instance, one study showed that conversational and read speech of 31 male speakers in neutral and stressed states could be distinguished with 88% accuracy [2], and a 2002 evaluation report by the US Department of Justice of two VSA systems said in its conclusion that "... it could be stated that these two VSA units do recognize stress." [3].

However, crucially, the same report went on to say: "Although these systems state they detect deception, this was not proven." [3]. This point is taken up by Aldert Vrij, professor of social psychology at the University of Portsmouth, UK: "It [VSA] is not measuring lying; it's measuring stress. The assumption is that liars are more stressed than truth-tellers and that's not always the case. A skilled liar will not be stressed." [4]. And the converse also holds: many people making legitimate claims will be subject to stress, especially if they know that VSA is being used to analyze their voice. A more recent evaluation of VSA [5] showed that commercial VSA systems fared better than chance at detecting deception in recordings where "ground truth" was known, but only just: the overall accuracy was 63% in 516 tests, and the number of false positives was 23%. In these tests, the VSA systems only provided data, and the deception/non-deception decision was made by analysts with years of training - it seems unlikely that Harrow council, or the several other UK councils and insurance companies that are beginning to trial VSA, would be employing such personnel.

More disturbing is the case of Michael Crowe of Escondido, Calif., whose 12-year-old sister, Stephanie, was found stabbed to death in her bedroom. He was given a VSA test which he was told he had "failed". He then began to doubt his own memory and wonder whether he might have killed his sister. "I didn't want to go to prison, and I just wanted to be out of that room," Crowe recalled. "So my only option was to say, 'Yeah, I guess I did it,' and then hope for the best." However, a week before the start of his trial, the police found DNA evidence that led to the real killer, who is now in prison for killing Crowe's sister [6]. In the aftermath of this case, an expert who testified stated that VSA was used routinely as a scare tactic to elicit confessions.

Some bloggers have termed VSA junk-science, but this is incorrect: it is a classical example of decent scientific research that is being misapplied, perhaps unintentionally, with potentially devastating consequences for individuals. As professionals working in this area, the best we can do is to emphasize that VSA is just that; stress analysis, and stress does not equate to deception. Or alternatively, campaign to have all statements made by politicians subject to VSA: that would soon see its withdrawal.

[I] Allan Bell Enterprises, Comparisons of Existing Lie-Detection Equipment, Unpublished.

[2] M. Sigmund and T. Dostal, Detection of psychological stress by analysing of glottal pulse waveform, Artificial Intelligence and Applications, 2007, pp 561-566.

[3] D. Haddad, et. al, Investigation and Evaluation of Voice Stress Analysis Technology. Final Report for National Institute of Justice, Interagency Agreement 98- LB-R-013. Washington, DC, 2002. NCJRS, NCJ 193832. Available at https://antipolygraph.org/documents/193832.pdf

[4] http://www.guardian.co.uk/g2/story/0,,2171410,00.html

 [5] C.S. Hopkins, R.J. Ratley, D.S. Benincasa and J.J. Grieco Evaluation of Voice Stress Analysis Technology Proceedings of the 38th Annual Hawaii International Conference on System Sciences, 2005.

[6] http://abcnews.go.com/Primetime/Story?id=1786421&page=1