Polygraph vs voice stress analysis vs layered voice analysis

No voice- or physiology-based method is a definitive proof of truth, and none should be treated as one. Anyone choosing between the polygraph, voice stress analysis (VSA) and layered voice analysis (LVA) is weighing trade-offs in intrusiveness, speed, scale and evidentiary weight, not picking one proven method over two unproven ones. The research record is mixed for all three, and every legal system that has ruled on the question has reached the same conclusion: none of it stands alone.

That shared limit is the right place to start, because it changes what the decision is actually about. If no method settles a question by itself, the useful comparison is which one fits how you need to screen, not which scores highest in a lab: how intrusive it is on the person, whether it runs remotely, how quickly it turns around, how well it holds up across a large intake, how consistently it is scored, and what it tells you at the end. This post is the evidence behind that comparison.

Accuracy

Accuracy matters, and it is worth being straight about where each method sits. However, it’s only one of the axes that should inform a choice, and on the independent evidence the three land closer together than their marketing suggests.

The polygraph

A polygraph exam wires sensors to the subject to track cardiovascular activity, respiration, and skin conductance while an examiner asks a structured sequence of relevant and comparison questions. It takes 90 minutes to three hours and requires a controlled, in-person setting.

The most comprehensive independent review remains the National Research Council’s 2003 study for the U.S. Department of Energy, which spent 19 months examining the evidence base. It found that specific-incident polygraph tests “can discriminate lying from truth telling at rates well above chance, though well below perfection,”, while being equally direct about the limits. The technique’s accuracy “is insufficient to justify reliance on its use in employee security screening” (National Research Council, The Polygraph and Lie Detection, 2003). The industry’s own body disputes the sceptical reading. A 2011 meta-analysis commissioned by the American Polygraph Association, pooling 38 studies and over 11,000 scored examinations, reported high decision accuracy for single-issue testing (American Polygraph Association, 2011). The gap between those conclusions reflects who conducted each review as much as what the underlying studies show, and it is worth holding onto when comparing any of these three technologies.

Voice stress analysis (VSA)

VSA tools read a narrow band of the voice, mainly pitch and micro-tremor, for markers associated with stress. Nothing is attached to the subject, so a test can run over the phone in 30 to 60 minutes, and law enforcement adopted it widely. The strongest independent test of that adoption is a National Institute of Justice-funded field study that compared two commercial VSA tools against urine toxicology results for 319 arrestees at an Oklahoma County jail. The tools performed close to chance at scoring deception about recent drug use, and most arrestees who were lying were scored as truthful (Damphousse et al., Assessing the Validity of Voice Stress Analysis Tools in a Jail Setting, National Institute of Justice).

Layered voice analysis

LVA, developed by the Israeli firm Nemesysco, analyses a wider spectrum of vocal parameters than basic voice stress analysis, and it is built to read involuntary autonomic responses a speaker cannot consciously suppress. It needs only a recording, so testing is remote-capable, and scoring runs through cloud software rather than an examiner’s judgement. We explain the technology in what is layered voice analysis.

Independent testing has been sceptical. In a controlled study of 48 subjects making high-stakes statements, LVA performed close to chance at separating truth from deception, with high false-positive rates (Harnsberger, Hollien, Martin & Hollien, Stress and Deception in Speech, Journal of Forensic Sciences, 2009), and a review of five decades of voice research found no scientific support for the stronger manufacturer claims (Eriksson & Lacerda, Charlatanry in Forensic Speech Science, 2007). In the Oklahoma County jail study, LVA showed higher sensitivity than the competing VSA product, while both sat below any level that would justify relying on the result on its own.

An LVA flag is a starting point, not a finding. It marks where a person’s account and their voice do not sit comfortably together, and that is a signal to look closer, not a verdict to act on. Before any result reaches a client, our team reviews every medium and high-risk flag against the context of the interview and the subject’s circumstances, and delivers it with that context attached. The client then confirms with evidence: a follow-up question, a targeted check, a documented decision. Screen, flag, investigate. A method that is imperfect on its own becomes genuinely useful when it directs attention rather than replaces judgement, and when a person, not an engine, decides what a flag means.

Set side by side, none of the three is proof, and each has independent evidence that limits how much weight a single result can carry. That is exactly why accuracy is not where the choice gets made. The differences that decide a screening programme are elsewhere, and that is what the rest of this comparison covers.

Intrusiveness and reach

A polygraph wires cardiovascular, respiration and skin-conductance sensors to the subject and has to run in person in a controlled setting. VSA attaches nothing and can run over the phone. LVA also attaches nothing and needs only a recording, so an assessment runs over a normal video or phone call from wherever the subject feels comfortable or if they are in transit. For a shortlist spread across a province, or a driver at a depot two towns over, that difference often decides whether an assessment happens at all. It is also why the experience is easier on the person being assessed, with no wires and no room to sit in. The mechanics are covered in how layered voice analysis works.

Speed and scale

This is where the gap is widest, and it is the claim the scalability comparison leads with. A polygraph exam takes 90 minutes to three hours, and a trained examiner can run only four to six in a day. An LVA assessment is a short structured interview, scored by software, with the report usually back within 24 hours, and Integrigate completes around 50 assessments a day across its team of operators, adding operators when volume requires it. When you are screening a whole intake, or testing a fleet on a schedule, that throughput is the difference between a programme you can run and one you cannot. It is also what makes ongoing integrity maintenance practical rather than a one-off event, which matters most in logistics, where the risk does not end at hire.

Consistency of scoring

Polygraph and VSA both depend on a human reading the output, and that reading varies. The Oklahoma County study found novice and expert interpreters of the same VSA recordings agreeing at correlations between 0.11 and 0.52, which means the same recording could yield different conclusions depending on who read it. LVA is scored by software on the same basis every time, so two people put through the same questions are assessed the same way, and a person then reviews every medium and high-risk flag in context. That consistency is a large part of what makes a result defensible when it is questioned later.

What you get back

A polygraph or a VSA tool returns a stress or deception call. LVA returns more: a risk score for each question and an emotional profile mapped across the interview, so you can see where the signal is strongest. Before it reaches you, our team reviews every medium and high-risk flag against the context of the interview, so the output points you to where a closer human look is needed, rather than handing down a verdict. This is the difference between a single light that turns red and a read you can act on, and we walk through what it looks like in what to expect from an assessment.

Resistance to coaching

LVA reads involuntary signals produced by the autonomic nervous system, the responses that sit below conscious control and are hard to coach or rehearse. That is a real advantage over methods a prepared subject can manage. No method is impossible to game, which is exactly why any result belongs alongside other evidence rather than standing on its own.

Behaviour changes when people know they are assessed

There is a second effect worth naming, because it’s true regardless of the accuracy debate. In the Oklahoma County study, arrestees who believed their answers were being verified by a voice-analysis tool were far less likely to answer deceptively than those in an earlier study run without it (14% versus 40%). People disclose more when they believe they are being verified, and those disclosures are real evidence. For integrity screening, and especially for ongoing testing of the same people over time, that effect is a real part of the value, separate from what any single flag says.

What holds across all three

Wherever a legal system has weighed in, the answer has been the same regardless of which technology produced the result. None of it is proof on its own. In the United States, the Employee Polygraph Protection Act of 1988 bars most private employers from requiring a polygraph for pre-employment screening. South Africa has no equivalent statute, but the Labour Court and CCMA have held consistently, in matters such as DHL Supply Chain (Pty) Ltd v De Beer NO (2014) and Goldplat Recovery (Pty) Ltd v CCMA (2021), that a failed polygraph is never enough on its own to prove misconduct and must be corroborated, and the same reasoning applies to any comparable voice-based result.

That shared limit is how these tools should be used responsibly. A result is one input that sits alongside other evidence and human judgement, applied consistently, and never the sole basis for a high-stakes decision about someone’s job or freedom. That is the model we run: screen, flag, investigate. Used that way, the question stops being which method is a magic answer and becomes which one you can run honestly, at the scale you need, on the people you need to screen. For high-volume integrity work across hiring, logistics and tenant vetting, that is where LVA earns its place.

See how an assessment runs from booking to report, or book a test and we will set it up.