VoxCheck
In betaAn iOS app that tracks vocal load — cumulative stress on the vocal folds — and turns it into a daily Readiness Score for singers, teachers, and public speakers who rely on their voice for a living.
The problem
Professional voice users have no easy way to know how much vocal capacity they have left on a given day, or how long they actually need to recover after a heavy one. VoxCheck measures it directly from the microphone, with no wearable required.
How it's built
Real acoustic measurement
Sound pressure level, vocal register, ambient noise, and CPP (Cepstral Peak Prominence, a clinical phonation-quality metric) — measured live from the iPhone microphone, on-device.
A model that learns the user
Personal recovery time and overload sensitivity are fitted from each user's own morning check-ins, not a fixed formula — so the Readiness Score gets more accurate the longer someone uses it.
On-device AI coach
Daily recommendations generated on-device with Apple's Foundation Models (iOS 26+), with a safe, deterministic fallback when the model isn't available.
Live Activity & Dynamic Island
Session load and elapsed time visible on the Lock Screen without unlocking the phone.
Shadow-tested formulas
New scoring logic ships as a shadow variant and has to earn predictive accuracy against real check-ins before it ever affects a live score — no formula change ships on a hunch.
Privacy by construction
Raw audio and PCM buffers are processed in memory and discarded. Only derived numbers — never audio — ever leave the device.
Stack
React Native for the app shell and scoring logic, native Swift for the audio DSP (FFT-based pitch and loudness detection, CPP, Core ML) and platform integrations, SQLite for a fully local-first data model, and Supabase for optional cross-device sync.