Voice stress analysis

Voice risk or voice stress analysis (VSA) technology is said to record psychophysiological stress responses that are present in the human voice when a person suffers psychological stress in response to a stimulus (e.g., a question), and the consequences of the person's answer may be dire.^[1] This is based on the belief that non-verbal content of the voice carries, among other things, information about the physiological and psychological state of the speaker.^[2] Manufacturers of VSA typically claim it can be employed to detect deception in a variety of settings such as police interviews, insurance claims, and social benefit claims.

Principle and origins

In 1970, and prior to the publishing of Lippold's article in 1971, three military officers retired from the U.S. Army and formed a company they named Dektor Counterintelligence and Security (CIS). The three officers were Alan Bell, Bill Ford and Charles McQuiston. Bell's expertise was in counterintelligence, Ford's was in electronics, and McQuiston's was in polygraphy. Ford had invented an electronic device that utilized the theory of Lippold, Halliday and Redfearn in which he tape-recorded the human voice, slowed it down to one-third or one-fourth its normal rate, and fed it through several low pass filters which then fed the signal into an EKG strip chart recorder. The strip chart recorder then made chart tracings on heat-sensitive paper. They named their device the Psychological Stress Evaluator (PSE). According to Allan Bell Enterprises, "All lie-detection examinations or evaluations are predicated upon the fact that telling a significant lie will produce some degree of psychological stress. Psychological stress, in turn, causes a number of physiological changes."^[3]

Academic analysis

The use of Voice Stress Analysis (VSA) technology to detect deception is highly controversial and use of these machines has been referred to as 'charlatanry' by academics.^[4] Discussions about its application revolve around whether this technology can indeed reliably detect stress, and if so, whether deception can be inferred from this stress. The latter is a logical inference problem referred to as the Othello error. It has even been argued that - even if stress could reliably be measured from the voice - this would be highly similar to measuring stress with for example the polygraph, and that all critiques voiced towards polygraph testing apply to VSA as well.^[5]

Several studies published in peer reviewed journals showed VSA to perform at chance level when it comes to detecting deception. Horvath, McCloughan, Weatherman, and Slowik (2013),^[6] for example, tested VSA on the recordings of interrogation of 74 suspects. Eighteen of these suspects later confessed, making the deception the most likely ground truth. With 48% accurate classification, VSA performed at chance level. Damhousse and colleagues interviewed over 300 arrestees about recent drug use. Their responses and the VSA output were compared to a subsequent urinalysis to determine if the VSA programs could detect deception. The results showed the programs were not able to detect deception at a rate any better than chance.^[7][8] Other peer reviewed studies showed similar results.^[9][10] In 2003, the National Research Council concluded “Overall, this research and the few controlled tests conducted over the past decade offer little or no scientific basis for the use of the computer voice stress analyser or similar voice measurement instruments”.^[11]

Research published in the Proceedings of the 2005 Hawaii International Conference on System Sciences, identified that VSA technology correctly classified 73% of deceptive examinees. Yet, for truth tellers - with 52% accurate classification - it performed as well as a coin flip.^[12] A report from the U.S. Air Force Research Laboratory in Rome New York concluded that the voice stress units tested were indeed able to recognize stress in the spoken voice. Yet, the authors also tested VSA as a deception detector on material of 2 convicted murderers and concluded that 'Although these systems state they detect deception, this was not proven'.^[3]

Support for VSA as a deception detector typically comes from poorly executed studies published in journals with a doubtful status. A new Indian open access Journal, for example, (the International Journal of Electrical, Electronics and Computer Engineering (IJEECE)) published a study claiming that Voice Stress Analysis (VSA) technology can identify emotional stress better than polygraph.^[13] This study did not, however, make any experimental comparison between VSA and the polygraph, meaning that the evidence for this claims remain mysterious. Also the statement 'better than polygraph' is not very informative, as it does not contain any information on the effect size of either of the techniques.

Similarly, an Ukrainian Journal issued by Kyiv Scientific Research Institute of Court Expertise published a study by James Chapman and Marigo Stathis, evaluating the use of the voice stress analysis technology for the detection of stress associated with possible deception. VSA detected stress associated with criminal activities in 95% of the cases in which a confession was obtained, and there no cases wherein a confession was obtained in the absence of stress.^[14] Importantly, such a study suffers from sampling bias. Suspects will only be interrogated when they fail the test. Consequently, confessions will only follow in those cases where the test outcome was deception indicated. When one subsequently rounds up all the cases in which a confession was obtained, it should come as no surprise that the confessions match deception indicated test outcomes to a high degree. This would happen even when a test performs at chance level.^[15] Using lie detectors as a prop to elicit confessions is referred to as the bogus pipeline effect, and is not without danger. In the Murder of Stephanie Crowe, for example, three suspects confessed following an interrogation with a VSA, but these confessions were later ruled false.^[16]

Notable examples of use

VSA played a decisive role in the three false confessions in the Murder of Stephanie Crowe^[16] The manufacturer of the VSA equipment later settled a lawsuit that alleged that the voice analyzer's manufacturers were liable for the harm the three defendants suffered.^[17] In a similar case, Donovan Allen falsely confessed to killing his mother after failing a VSA test. He was acquitted 15 years later based on exonerating DNA evidence.^[18]

Following the 2012 case in which George Zimmerman fatally shot Florida teenager Trayvon Martin, Zimmerman was given a voice stress analysis test by the police department of Sanford, Florida. He passed the test.^[19] A videotape of the test was publicly released in June 2012.^[20]

Federal judge approves computer voice stress analysis to monitor sex offenders. A recent ruling from a U.S. Federal court judge may require sex offenders to submit Computer Voice Stress Analyzer (CVSA) examinations throughout the post-release supervision process.^[21]

Confession made following a voice stress examination was allowed to be used as evidence in a rape trial.^[22]

Work and Pensions Secretary John Hutton announced the introduction of voice-risk analysis software to root out benefit cheats,^[23] but in 2010, the government dropped plans to introduce the controversial lie detector tests to after trials found that the technology is not sufficiently reliable.^[24] The Guardian newspaper reported in 2009 that the UK Department for Work and Pensions^[25] analysis of data from trials conducted between May 2007 and July 2008 in various agencies shows accuracy rates no better than chance.^[26] Following a Freedom of Information request by the trade union financed campaign body False Economy The Guardian revisited the story in 2014.^[27]

References