The science—and the limits—of voice lie detection

What research says

Deception research does not support a single reliable cue that works across every person, language, culture, question, and environment. Pauses, pitch changes, uncertainty, reduced detail, tension, and contradictions can occur for many ordinary reasons.

Performance found in controlled studies may not transfer to an everyday recording where the speaker, stakes, baseline, microphone, and surrounding context are unknown. That is why LieTone AI does not display an accuracy percentage or claim to prove a lie.

What LieTone AI actually does

1. Transcription and speaker timingMaps spoken content to time and separates speakers where the audio allows it.
2. Acoustic observationReads changes in timing, pauses, energy, pitch, and spectral patterns as descriptive audio features.
3. Language and contextExamines expression, detail, corrections, uncertainty, and consistency within the supplied recording and optional background.
4. Structured synthesisProduces a directional reading while keeping observations, possible interpretations, and alternative explanations distinct.

What a result means

Leans truthful and Leans deceptive describe the direction of patterns found in the recording. Signal strength describes how consistently those patterns point in the same direction. Neither is a probability, accuracy score, factual verdict, or measure of a person’s character.

Responsible use

Stress, fatigue, context, noise, accent, culture, illness, and personal speaking habits may produce similar signals. Use the result as one prompt for reflection—not as evidence against another person.

LieTone AI is not designed for law enforcement, courts, employment, lending, insurance, medicine, education discipline, or other decisions with significant consequences.