Three-channel impairment screening
Prove you're sober in 90 seconds, not a lab visit.
SoberSense reads your eyes, listens to your voice and measures your reflexes, then fuses all three into a single timestamped sobriety score.
Beta research build — eye, voice and reaction are measured live on your device; scores are not medical or legal evidence
CH-01
Ocular
Pupil diameter drift, gaze stability and nystagmus proxies captured from a 12-second front-camera pass.
CH-02
Vocal
Articulation rate, slur onset and pitch variance measured from a randomly chosen read-aloud sentence.
CH-03
Motor
Simple reaction time across randomized stimuli, benchmarked against your own sober baseline.
Fusion model
Single-signal tools miss compensating behaviour — a steady voice can mask ocular drift. SoberSense weights ocular at 40%, vocal at 35% and motor at 25%, and flags any result where the channels disagree as inconclusive rather than clear.
The science
Why the results can't be faked
Each channel in the scan targets a behaviour the brain runs automatically — not one you can perform "well" on purpose. Impairment doesn't just slow these systems down; it changes how they fail, and that failure signature is not under conscious control.
INV-01 · Ocular
Involuntary pursuit
Following a moving target with a smooth glide is run by brainstem circuits you cannot consciously control. Alcohol and sedatives break that glide into corrective jerks within minutes — sober testers can't fake the failure, and impaired testers can't suppress it.
INV-02 · Vocal
Automatic speech motor control
Speech is an over-learned motor sequence. Alcohol slows articulation and loosens stop consonants before a drinker notices it themselves. Trying to speak carefully produces over-correction that appears as extra pitch and timing variance, not improvement.
INV-03 · Cognitive
Failed inhibition
The Go/No-Go task asks the brain to cancel an action it has just been trained to make. That cancellation runs in the prefrontal cortex — the first circuit impairment degrades. Reaction speed can be faked by slowing down; tapping a forbidden shape cannot.
INV-04 · Identity
Geometric face lock
Handing the phone to a sober friend won't pass. At baseline enrollment SoberSense maps invariant facial ratios, and a scan only unlocks when the live face matches that map — computed entirely on-device, with no photo stored or uploaded.
The fusion weights above reflect this: ocular signals are the hardest to fake, so they carry the most influence — and any result where the channels disagree is reported as inconclusive rather than clear. And because every scan is checked against the face enrolled at baseline, a clear result is proof about one specific person — not proof that someone sober happened to be holding the phone.