This chapter covers the same instrument as chapter 50 and is placed here deliberately, in the sequence of technologies, because voice stress analysis is the clearest case of the pattern that runs through all of them.
The claim is that involuntary microtremors in the laryngeal musculature, in the eight-to-twelve hertz range, are damped under stress, and that software can detect this and thereby detect deception.
The National Institute of Justice funded a field study at the Oklahoma County jail, testing the two dominant commercial systems — Computer Voice Stress Analyzer and Layered Voice Analysis — on arrestees asked about recent drug use, with urinalysis providing independent ground truth. The systems identified roughly fifteen percent of the deceptive answers. Chance would be fifty. A separate NIJ-funded laboratory evaluation reached compatible conclusions, and independent reviews have not established the microtremor premise itself in laryngeal muscle.
The devices remain in use across American police departments and in insurance and employment screening.
The reason is the bogus pipeline, and it is worth understanding as a general mechanism rather than as an anecdote about one product.
The bogus pipeline is an experimental technique from social psychology: attach subjects to an impressive-looking apparatus, tell them it detects their true attitudes, and they report more honestly than subjects who are simply asked. The device measures nothing. Belief in it does the work, and it works reliably.
Applied to interrogation, the sequence is: a suspect is told a machine has detected deception in their voice; if they believe it, further denial appears pointless and their situation appears already determined; the interrogator's confidence appears evidence-based rather than presumptive; and admission becomes the reasonable move.
Confessions follow. The investigator observes that the machine identified a liar and reports a success. Vendors collect the testimonials. Nothing was measured at any point.
Two consequences follow, and the second is the one that matters.
An instrument that produces confessions through belief produces them from the innocent at the same rate as from the guilty, because the pressure it applies is indifferent to guilt. Everything in chapters 92 through 97 is what that looks like in practice, and a false evidence claim is precisely what Kassin and Kiechel identified as the most potent single ingredient in producing false confessions.
And an agency using the device has no way to discover this. It observes confessions, records them as detections, and never learns which were false. The feedback vacuum of chapter 52 again, now with a machine in the room supplying spurious confirmation.
The general question, applicable to any instrument offered as evidence of a mental state, is the one chapter 50 ends on: what is its measured accuracy against an independent ground truth, published by someone with no commercial interest in the answer.
For this technology, that number exists. It is fifteen percent, and it has been public since 2007.
Cross-ref: polygraph and AI detection chapters