iVitals

Each frame's average green-channel intensity over your forehead forms a photoplethysmographic (PPG) waveform — haemoglobin absorbs green light, so the signal pulses with your heartbeat. We detrend, band-pass filter to 0.7–4 Hz, and take an FFT; the dominant peak is your pulse. Beat-to-beat intervals give HRV (RMSSD, SDNN, pNN50) and a rough autonomic stress proxy. Respiration rate comes from the slow breathing modulation of the same pulse signal (the 0.1–0.5 Hz band). Blood pressure is inferred from HR + HRV features with a transparent heuristic.

Calibration. Take a reading with a real arm cuff, then run a scan and enter the cuff numbers. We store the difference and apply it to future estimates (single-point offset calibration). This personalizes the baseline and is the single biggest accuracy improvement, but it is still an estimate, not a diagnosis. Re-calibrate periodically. History and calibration are stored only in this browser on this device.

Blood-pressure categories (American Heart Association): Normal below 120/80 · Elevated 120–129 and below 80 · Stage 1 is 130–139 or 80–89 · Stage 2 is 140+ or 90+. The category is set by whichever number is higher, so a raised bottom (diastolic) number alone can read as Stage 1. It's a rough camera estimate, not a diagnosis — confirm with a real cuff.

Best results: bright, even, flicker-free light; hold still; bare forehead filling the box with the bottom edge just above your eyebrows (no eyes or glasses inside it — blinks and lens reflections add noise); ~40 s scan.

Your ~12 seconds of speech are analyzed entirely on your device — the recording is decoded to a raw waveform and discarded; nothing is uploaded. From the waveform we measure a handful of acoustic features: pitch and its variation (autocorrelation), brightness (spectral centroid — how much high-frequency energy your voice carries), steadiness (jitter & shimmer — tiny cycle-to-cycle wobbles in pitch and loudness), resonance (harmonics-to-noise ratio), energy, and your pause rhythm (how speech and silence alternate).

The six states. A transparent, deterministic classifier compares those features against a baseline and lands on one of six everyday states — centered, charged, scattered, dimmed, expanded or misaligned. Until you have saved 5 check-ins it compares against typical population values; after that it uses your own baseline, which is when readings get personal and meaningfully more accurate.

Best results: somewhere quiet, phone at a normal distance, and just talk naturally — read anything or describe your day. Checking in around the same time each day makes your trend easier to read. This reflects how your voice sounds, not a mental-health measurement or diagnosis — if something feels genuinely wrong, talk to a person, not an app.