The Voice Under Pressure: Pitch, Latency and Speech Rate

The voice is the channel where the evidence is best and the temptation to over-read is greatest, so the two findings that matter should be stated together.

First: vocal pitch really does rise under stress. Fundamental frequency is determined by the rate of vocal fold vibration, which depends on the tension of the laryngeal musculature, and sympathetic arousal increases that tension. This is not disputed, it has a clear physiological mechanism, and it appears in Bella DePaulo and colleagues' 2003 meta-analysis of deception cues as one of the few behaviors reliably associated with lying across studies.

Second, from the same meta-analysis: the effect is small. DePaulo's team pooled 158 cues across more than a hundred independent samples, and the overall picture was that behavioral cues to deception are weak. Pitch was among the more consistent, with an effect size that would be described as small by any standard convention. Response latency and speech errors showed effects, though inconsistent ones, sometimes in opposite directions depending on whether the lie was planned or spontaneous.

Both of those findings are true at once, and holding them together is the whole skill.

A pitch rise on a particular answer means the speaker's arousal increased. It does not mean much on its own, because the difference between liars and truth-tellers on this measure is small relative to the difference between people, and relative to the difference within one person across topics of varying emotional weight. A person discussing their divorce and a person lying about a theft will both show it.

Which is why the analytic method that works is within-speaker comparison rather than comparison to any norm. You do not ask whether this voice sounds stressed. You ask whether this speaker's pitch, latency and rate on this topic differ from the same speaker's pitch, latency and rate on structurally similar topics — questions of comparable difficulty, comparable emotional weight, comparable length of answer. That is a demanding comparison to construct, and it is the reason serious analysis of recorded material takes hours per interview.

Latency deserves separate mention because it behaves in a way people find counterintuitive. The folk expectation is that liars pause longer while inventing. In practice, prepared lies often come faster than the truth, because the answer is retrieved from rehearsal rather than reconstructed from memory, while genuine recall of a real event takes time and arrives with corrections and false starts. The suspicious pattern is often the answer that is too smooth for the question asked — an unexpectedly detailed, fluent response to a question the speaker could not have anticipated.

The application worth taking away is not detection but interviewing. If within-speaker change across matched questions is the signal, then the interview has to be designed to produce the comparison: neutral questions interleaved with substantive ones, similar in form and difficulty. That is what the structured protocols in Part 4 are for, and it is why an interrogator who simply accuses and then watches is generating noise.