Voice risk analysis
|This article needs additional citations for verification. (September 2007)|
Voice Risk Analysis or VRA, not to be confused with Voice Stress Analysis (VSA), is a technology used by financial services companies, governments and law enforcement agencies as a lie detection tool. The software, in addition to advanced training in questioning skills, interviewing techniques and behavioral analysis, is marketed by UK based company Digilog.
VRA technology works by measuring slight, inaudible fluctuations in the human voice that indicate when a subject is under stress and when that stress is generated by an attempt to deceive. The caller's voice patterns are analysed by the software and thereafter assessed as high or low risk of deception. The fundamental belief in voice risk analysis holds that individuals engaged in lying experience added stress through fear of the lie being discovered. Increased fear of detection and a greater potential reward add to the degree of stress experienced. Aside from other physiological responses, particular increases in voice pitch, frequency and intensity detected by the software are used to assess truth or deception.
Marketed to customer contact/call centers across Europe and Latin America by Digilog, voice risk analysis has particular appeal to insurance companies, banks and government agencies involved in processing welfare benefit claims.
Adoption by UK Government & Corporate Clients
- Work and Pensions Secretary John Hutton announces introduction of voice-risk analysis software
- From the Guardian Newspaper: ″Tory-controlled Derbyshire Dales said it had taken part in a county-wide review of council tax in 2011 that had used the technology – a contract worth £280,000 to Capita.″
- Digilog's website lists corporate clients drawn from the insurance and recruitment sectors.
Summary of Research involving VRA
There have been several studies undertaken concerning the reliability of VRA compared to other forms of voice analysis and lie detection systems. Listed below is a summary of some of these studies.
1. A Study Testing the Validity of VRA Technology in a Criminal Field Context -conducted by John Joseph Palmatier Ph.D.
Subjects scheduled for criminal polygraph examinations gave their permission for the recording of their responses to questions asked during the examination, and these responses were analysed using VRA. 100 subjects were used, 25 truthful control subjects, 50 deceptive, and 25 truthful subjects, their status confirmed by either the examinee or another person confessing to the crime at issue. The voice analyst achieved results of 89% accuracy for truthful subjects and 78% accuracy for deceptive subjects. He concluded that these results, along with the others, and the versatility of the product strongly suggests that additional research is needed to explore this ‘new and perhaps promising technology in multiple settings’.
2. Validation Study of VRA within a Claims Investigation Unit -conducted by Dr Colin Linsky
A validation study was set up to evaluate the potential of VRA as a fraud detection, decision support tool in a claim investigation role. 68 tapes from a Globally Operating Financial Organisation Claims Investigation Unit (FOCIU) were analysed using VRA Off-line mode, the results were then compared to those obtained using the Claims Investigation Unit’s standard procedures. Dr Linsky found a concurrence of 73.8% to 78% between both procedures which was statistically significant. He concluded that ‘from the extremely limited analysis of a sub-sample of interviews, evidence suggests the decision-making from the VRA post-hoc analysis has a high concurrence with the contemporaneously assessed claim and claimant credibility ratings’.
3. The VRA Technology Reliability Test -conducted by Shlomo Bruck, Uran Ltd, member of the Israel Polygraph Examiners Association and the American Polygraph Association.
The experiment consisted of a group of 60 individuals in a controlled situation. A mock crime was staged with the subjects creating a situation whereby there were ‘guilty’ subjects and ‘innocent’ subjects. Each subject was examined by 5 polygraph examiners, then interviewed by a VRA examiner. He concluded that VRA ‘offers a degree of reliability that indicates that the developed instrument is suitable for field work’
4. VRA Reliability Test -conducted by Drs Guy and Marie van Damme
Interviews were analysed using VRA Online mode (164 interviews), Interrogation mode (142 interviews) and Offline mode (14 interviews). They found results on average of 89% validity using the online mode, 79% validity using the interrogation mode and 100% validity using the offline mode. They concluded that VRA is a ‘user friendly, versatile and feasible truth verification instrument’ and the accuracy ‘high and more than satisfactory’.
5. US Air Force Study finds voice analysis accurately detects stress 10/23/00 A three-year study by Air Force Research Laboratory Information Directorate, funded by the National Institute of Justice, investigated the ability to detect and classify stress in an individual's voice, and evaluated the effectiveness of commercially available voice stress analyzers. AFRL is interested in voice stress to improve the performance of voice recognition technologies. NIJ's interest in voice stress is for enhanced police investigation. It concluded that several features in an individual's speech pattern are different under stress. The goal of the study was to determine the scientific value and utility of existing voice stress analysis technology for law enforcement applications. Voice stress analysers have been marketed commercially to law enforcement agencies for more than a decade. The systems are cited as being cheaper, easier to use, and less constrained in their operation than polygraph machines that must be physically attached to the speaker's body. "We looked at different types of features and how those features could detect stress," said Darren Haddad, the study leader. "The more voice features you use, the more accurate your results. We also looked at what type of recording media should be used: which ones performed the best and which ones gave corrupt results. "The Department of Defense Polygraph Institute provided us with tapes of investigations from two murder cases where the suspects eventually confessed and were found guilty," Haddad said. "Using voice stress analysers provided by two vendors, both machines were accurate on 45 out of 45 instances."
Along with polygraph testing, VRA technology is used to identify certain changes in physiological responses to deception. Whereas a polygraph requires the physical presence of a subject to measure and record physiological indices such as blood pressure, pulse, respiration, and skin conductivity, VRA only requires the voice of a subject and can therefore be used over the telephone or to analyse a recording. If it is accepted that an altered physiological state is in-fact being detected, other reasons for physiological arousal must be considered within the results of any test such as heightened arousal from individuals experiencing or suffering from situational anxiety, shock (e.g. a road traffic accident) or an anxiety disorder. However, any specific physiological or emotional states can be excluded from the results as the VRA technology must be 'calibrated' to an individual subjects voice before any interview is conducted. This establishes a subjects 'baseline' behaviour and reactions when being asked 'control questions'. The software then looks for micro changes in the voice when the subject is responding to specific 'test questions' which may indicate deception.
The Guardian newspaper reported in 2009 that the UK Department for Work and Pensions analysis of data from trials conducted between May 2007 and July 2008 in various agencies shows accuracy rates no better than chance. Following a Freedom of Information request by the trade union financed campaign body False Economy The Guardian revisited the story in 2014.
- Haddad, D., Walter, S., Ratley, R., & Smith, M. (2001). "Investigation and Evaluation of Voice Stress Analysis Technologies." Rome Laboratory Report (AFRL-IF-RS-TM-2001-7), 18-19.
- http://www.guardian.co.uk/news/datablog/2009/mar/19/dwp-voice-risk-analysis-statistics accessed 4 May 2009