Jitter (local)
Cycle-to-cycle period variation in the vocal folds. Computed via parselmouth (Praat). The standard measure of laryngeal stability.
band 0.005 — 0.030
A Forensic Instrument for the Human Voice
Logos Probabilis
The Senses of a New Species
Upload a three-second voice sample. Receive four spectral features, a Cartesian projection, and — where the evidence permits — an honest verdict about whether the voice is over-controlled, stressed, or simply itself.
§ ii · method
Four features, one projection.
The instrument is a plain FFT pipeline. No neural network, no opinions — four numbers extracted by methods with decades of phonetic literature behind them.
MFCC Δ-variance
Mean-of-variances across 13 MFCC deltas. A proxy for how much the spectral envelope is permitted to wander frame by frame.
band 0.030 — 0.120
Spectral flux
L2 distance between successive normalised magnitude spectra (Dixon, 2006). The rate at which timbre changes across the signal.
band 0.080 — 0.200
Microtremor 8–12 Hz
RMS of the envelope band-passed through the physiological tremor window (Lippold, 1971). This is the one the speaker cannot suppress.
band 0.002 — 0.005+
§ iii · the ritual
Three samples. Then a verdict.
A personal baseline is required before the instrument may speak honestly. The first sample calibrates; the second and third are compared against it.
Step 01 · Truth
free · does not count
Say something plainly true.
Name the city you are in, or the weather outside your window. Five to ten seconds is enough. This becomes your personal baseline — the instrument has nothing to compare against without it.
Your baseline
jitter · local—
mfcc Δ-variance—
spectral flux—
microtremor—
voiced frame ratio—
Step 02 · Uncertain
one freebie · today
Say something you are unsure of.
A memory you cannot quite retrieve, a name you half-recall, a fact you believe but would not bet on. Natural hesitation is expected here — the instrument is not scoring you, only watching.
Step 03 · Lie
one freebie · today
Now say something you know to be false.
A deliberate fabrication, plainly stated. The purpose is not to catch you — it is to observe the voluntary suppression and involuntary tremor that accompany the act. You already know whether you lied. The instrument is simply reading it off the spectrum.
Step 04 · Reveal
Cartesian projection
—
● over-controlled · tense = deception signature
● natural = unsuppressed voluntary variation
quadrant
—
awaiting sample
confidence · —
features & deltas
jitter — local
—
—
mfcc Δ-variance
—
—
spectral flux
—
—
microtremor
—
—
Features are computed per the methods described in
DOI 10.5281/zenodo.19396809. The deception signature was validated
on n = 3 recordings from a single speaker. Generalisation to
your voice is hypothesised but not proven.
Step 05 · AI Bonus
counts against quota
Now let a machine speak the same words.
Generate the same sentence with any text-to-speech model you prefer — upload it here. The instrument will measure whether synthetic speech lands closer to over-controlled than human truth, or whether modern models have learnt to tremble convincingly.
§ iv · access
Three ways in. No payment required today.
Paid tiers display their promises here but do not collect payment yet. When billing lands in v0.2 these cards will light up with a checkout link.
Free
live€0/forever
The full ritual, unlocked. For anyone curious about their own voice.
- Three analyses per 24 hours
- Ritual freebies (truth + uncertain + lie) do not count
- Session-local baseline · no account
- Raw audio never persisted
Practitioner
v0.2€14/month
For therapists, journalists, security researchers working with many voices.
- Unlimited analyses
- Persistent named baselines per subject
- CSV & PDF export of feature tables
- API key for scripted workflows
Laboratory
v0.3custom
For institutions running studies with informed-consent cohorts.
- Batch upload & analysis pipeline
- Dedicated baseline cohort management
- Self-hosted option — audio never leaves your network
- Citation support & methods review
§ v · the companions
Read the method. Read the book.
Every claim made here is either derived from or disclaimed in these two open-access documents. Honest instruments travel with their own errata.
An Entropy-Oriented FFT Analysis of Voice Stress
Fausto & Claude, 2026. The empirical ground truth for every threshold, range, and quadrant boundary in the instrument above.
AGI ✕ Logos Probabilis — The Senses of a New Species
The broader argument: why voice is a modality a new species of intelligence must learn to read, not merely speak in.