The use of voice in security evaluations

This 2006 overview paper by Harry Hollien and James D. Harnsberger, published in the Journal of Credibility Assessment and Witness Psychology, evaluates the scientific basis for using voice-based stress analysis technologies in security and credibility assessment contexts. Drawing on a program of controlled laboratory experiments, the authors synthesize evidence on whether commercial voice stress analyzers can reliably detect deception or psychological stress. The paper concludes that existing VSA technologies perform at chance levels and lack a validated theoretical or empirical foundation for security use.

The paper is a scientific overview and evaluative synthesis based on a series of controlled laboratory experiments using diverse subject populations, systematically varied deception and stress conditions, double-blind scoring by professional examiners, and sensitivity-based statistical analyses of multiple commercial VSA platforms.

This paper by Harry Hollien provides a targeted scientific review of voice stress analysis technology for security and intelligence communities, based on systematic experimental evaluation at the University of Florida. The research employed controlled laboratory conditions with subjects making deceptive statements under genuine psychological stress, evaluated multiple commercial VSA platforms using double-blind protocols with professional examiners. Results consistently showed chance-level performance across all systems tested, with accuracy rates hovering around 50%. The findings directly challenge the theoretical premise that measurable voice changes can reliably indicate deception or stress.

The findings suggest that security agencies should not rely on voice stress analysis for deception detection, as these technologies perform no better than random chance and lack scientific validity for operational use.

Background & Context

The question of whether the human voice can reveal deception has attracted researchers, law enforcement agencies, and security professionals for decades. The theoretical premise behind voice stress analysis (VSA) is that liars experience more psychological stress when lying than truth-tellers do, and that this stress produces minor changes in blood circulation which subsequently influence voice characteristics — measurable via instruments that detect intensity, frequency, pitch, harmonics, or laryngeal microtremors.

Proponents of vocal stress analysis systems argue that they are able to detect spoken deception by analysis of "stress" in the voice signal. They claim to do so by examining traces made by laryngeal microtremors which are associated with stress and, ultimately, with lying — however, most research seeking to identify relationships between microtremors and laryngeal function has produced negative results, and data on the ability of voice analyzers to detect stress or identify spoken deception have been negative or "mixed."

Harry Hollien, a pioneering forensic phonetician at the University of Florida's Institute for Advanced Study of the Communication Processes (IASCP), spent decades systematically evaluating voice-based deception detection. Law enforcement, security, intelligence, and related agencies have a clear need for an effective method for the detection of deception — and this 2006 paper represents Hollien and Harnsberger's focused effort to brief key professional communities on the scientific status of voice-based tools for use in security evaluation contexts.

Research Design & Methodology

Published in the Journal of Credibility Assessment and Witness Psychology (Volume 7, pp. 74–78), this paper by Harry Hollien and James D. Harnsberger functions as a targeted scientific overview and evaluative analysis directed at security and intelligence communities who were actively considering or deploying voice stress analyzer (VSA) technologies. The authors drew upon an extensive experimental program conducted at the University of Florida.

Key design elements underpinning the analysis include:

• Controlled laboratory speech databases with systematically varied levels of deception and psychological stress
• Evaluation of multiple commercial VSA platforms, including the Psychological Stress Evaluator (PSE), the Computer Voice Stress Analyzer (CVSA), and the Layered Voice Analysis (LVA) system
• Use of double-blind scoring protocols with trained and professional examiners as well as experienced instructors provided by manufacturers
• Assessment of both stress detection and deception detection as distinct but related outcomes
• Statistical evaluation using sensitivity analyses and signal detection frameworks

Subjects in the related experimental work were drawn from a diverse population (age range 18–63 years), all holding strong views about some issue, and required to make intense contradictory statements while believing they would be heard and seen by peers — a design intended to induce genuine psychological stress and realistic high-stakes deception conditions.

Results & Key Findings

The headline finding is unambiguous: voice stress analyzer technologies failed to detect deception or stress at levels above chance. Across the body of experimental work underpinning this overview, VSA instruments consistently performed at or near the 50% accuracy level one would expect from random guessing.

Key findings reported and synthesized in this work include:

• Chance-level performance: A series of experiments on the detection of relatively high-risk lies by commercial voice analyzers found that correct stress/nonstress identifications occurred only at chance levels, and lie/nonlie identification scores were quite similar.
• Professional examiners performed no better than laypersons: Professional "examiners" performed at about the same level of accuracy as other auditors.
• LVA system: The LVA did not exhibit expected sensitivity to the presence of deception, truth, or high or low stress in experimental speech samples; true positive and false positive rates varied by evaluation team, and the raw data and all statistical analyses suggest only chance-level performance.
• CVSA system: The CVSA system operated at about chance level in controlled evaluations.
• Scientific consensus: Tests have failed to show that VSA devices perform at a level above chance, and it is clear that the CVSA and LVA devices were not able to detect deception or stress.

The paper also addressed the theoretical underpinning of voice-based lie detection. The claim that laryngeal microtremors are associated with stress and lying has not been supported — most research seeking to identify the relationships between microtremors and laryngeal function has produced negative results. This undermines the foundational logic of all commercially marketed VSA devices.

Discussion & Significance

Hollien and Harnsberger's 2006 paper appeared at a particularly consequential moment: VSA technologies were being actively purchased and deployed by law enforcement agencies, military units, and intelligence organizations across the United States, often in high-stakes interrogation and security screening contexts. The core question of how reliable and valid commercial vocal analysis software is for predicting emotion and deception in security screening contexts had direct and immediate operational implications.

The significance of this overview lies in its authoritative synthesis. Hollien brought decades of primary research to bear, situating the negative findings within a consistent pattern stretching back to early experiments. Early researchers working in lab settings failed to find support for VSA theory or technology — and Hollien's own 1987 work in the Journal of Forensic Sciences had already established this pattern nearly two decades prior to this 2006 publication.

The paper also carried important policy implications: it argued against the continued adoption of VSA tools in security and intelligence settings on scientific grounds. Nearly all of the empirical studies reviewed revealed that VSA equipment was able to detect deception no better than chance. Publishing this conclusion in the Journal of Credibility Assessment and Witness Psychology placed it squarely before the professional community most likely to act on it.

Limitations & Considerations

As an overview paper, this article necessarily relied on the experimental record rather than presenting entirely new primary data. Critics of VSA research have sometimes raised the concern that laboratory paradigms may not perfectly replicate the conditions of real-world security interviews. A key caveat to early studies is that the deception in each was relatively minor and no real "jeopardy" was involved — those who deceived had nothing to lose, and if there is no jeopardy, then there is no stress, meaning VSA technology may not have been tested appropriately.

Additional considerations include:

• The paper's relatively brief length (pp. 74–78) limits the depth of quantitative reporting that could be included
• Commercial VSA manufacturers disputed laboratory findings, arguing that field conditions introduce factors not captured in controlled studies
• Some common alternative interpretations of testing results could be used to argue that devices were not adequately tested in a laboratory study
• The fast-moving landscape of VSA technology means that newer commercial systems may require independent re-evaluation

Practical Applications

For polygraph practitioners, security professionals, and policy-makers, this paper delivers a clear evidence-based message: voice stress analysis technologies should not be treated as scientifically validated alternatives to the polygraph. The consistent finding of chance-level performance across multiple VSA platforms and multiple research groups means that decisions made on the basis of VSA readings carry no more evidentiary weight than a coin flip. It is unrealistic to rely completely on the voice to detect deception and hostile intent for all people and all situations.

One practically important nuance, however, involves what has been called the "bogus pipeline" effect: research has shown that individuals who believe they are being monitored by a lie-detection device — even a non-functional one — may be less likely to deceive. Arrestees who thought their interviewers were using lie detectors were much less likely to be deceptive when reporting recent drug use, suggesting that the perceived validity of a device can affect behavior independently of its actual diagnostic accuracy. This distinction matters greatly for how VSA technology is evaluated in the field versus the laboratory.