How layered voice analysis works

Layered voice analysis works by reading involuntary changes in the voice that follow the body’s automatic response to a perceived threat. The words a person chooses can be prepared and smoothed. The physiological layer underneath them is much harder to stage, and that is what LVA measures.

New to the technology? Start with our plain-English guide to what layered voice analysis is. This post goes a layer down, into the mechanism itself.

What signal does layered voice analysis read?

It reads the trace left in the voice by the autonomic nervous system. When someone hears a question that carries risk for them, the brain registers a perceived threat and signals the autonomic nervous system, the same system behind a faster heartbeat or a dry mouth. That system is not under voluntary control, and it also affects the fine motor control of the vocal apparatus. The result is a set of micro-changes in the voice that the speaker does not choose and cannot easily suppress.

This is the point that matters for credibility. Because the signals are produced automatically, they are far harder to coach, mask or rehearse than the answer itself.

How is LVA different from voice stress analysis?

LVA analyses a broad spectrum of the voice; basic voice stress analysis (VSA) reads a narrow band. VSA tools tend to track a single measure of stress and treat stress as a proxy for deception. That is a blunt instrument, because stress is not the same as dishonesty. A nervous but honest person and a calm but deceptive one can look alike to a stress-only tool.

By analysing many parameters across the whole signal, LVA produces a fuller emotional and cognitive profile rather than a single stress flag, which is what lets it separate ordinary nerves from a genuine risk signal. We line the methods up side by side, with the evidence for each, in our comparison of the polygraph, VSA and LVA.

What does the LVA engine produce?

Each interview is captured as high-quality audio and processed on secure servers, where the engine scores every response. The output covers three things: a risk read for each answer, an overall risk value for the session, and a profile of the cognitive and emotional markers mapped across the interview.

What matters technically is how that score is reached. It is examiner-independent, so it does not depend on one operator’s judgement, and two subjects put through the same questions are assessed on the same basis every time. That consistency is a large part of what makes the result defensible. Before anything reaches a client, our team then 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. We walk through what the report looks like in practice in what to expect from an assessment.

Is it proven? What are the limits?

An honest account has to include the limits, and they are real. Independent, peer-reviewed testing of voice-based methods is mixed, and some studies have found them operating close to chance. The algorithms are proprietary, which makes independent replication harder. The responsible position, and the one every legal system that has ruled on these methods has taken, is that a voice-based result is one input alongside other evidence, never proof on its own. We cover this in depth in the comparison post.

Why a profile is more useful than a red or green light

The value is a consistent, hard-to-game signal that shows where the risk sits. It informs the decision without making it. The model is screen, flag, investigate: the interview screens, the profile flags where a closer look is warranted, and you investigate with evidence. That lets a manager, a fleet operator or a letting agent spend their attention where it counts, rather than reading a light that turns red. To see how that works in practice, read what to expect from an assessment.

Frequently asked questions

How does layered voice analysis detect stress or risk? It analyses involuntary micro-changes in the voice caused by the autonomic nervous system’s automatic response to a perceived threat. These signals are not consciously controlled, so they are hard to fake.

What is the difference between LVA and VSA? VSA reads a narrow band of the voice for a single stress measure. LVA analyses a broad spectrum of parameters and returns an emotional and cognitive profile as well as a stress reading.

Is the scoring subjective? The scoring is examiner-independent, run by software rather than one person’s judgement, so the same questions are assessed on the same basis every time.

Can layered voice analysis be trusted as proof? No single result should be treated as proof. It is designed to be used as one defensible input alongside other evidence and human judgement.