Somnolyze is a native iOS sleep tracker that detects your phases from sound and motion, wakes you in light sleep with a progressive alarm, and writes sessions to Apple Health. No audio is ever recorded.
Six core capabilities, designed to work together — entirely on your device.
Awake, light, deep, REM — predicted live from audio, accelerometer and elapsed time.
Real-time FFT detects snoring, speech, movement and loud noises. The raw audio never leaves the phone.
Configurable window. Wakes you in a light phase with a 10-minute volume ramp instead of a brutal beep.
Reads existing phases from Apple Watch, writes Somnolyze sessions back, pulls heart rate and HRV.
Phase graph, sound curve, heart rate, detected events and tailored recommendations the next morning.
7 and 30-day views with sleep score, average duration and quick access to every past session.
Somnolyze processes the microphone signal locally with Apple's audio engine and Accelerate framework. No PCM frames, no recordings, no transcripts ever leave the device — only anonymised event metadata is kept for your own summary.
Somnolyze fuses elapsed time, sound level and movement to estimate the phase you're in — second by second. Early in the night, deep sleep dominates; later, REM episodes emerge. HealthKit phases from Apple Watch are layered in when available.
Each frame is run through an FFT to compute RMS in decibels and a dominant frequency. A small heuristic classifier flags snoring, speech, movement, ambience and loud events — and only the timestamps make it into your summary.
Set a target time and a window. Every 60 seconds during the window, Somnolyze checks your phase — if you're in light sleep, it triggers a 10-minute exponential volume ramp from 0.02 to 1.0. No abrupt beep at 7:00 sharp.
100% native, no third-party SDK in the audio path.
Fully declarative interface, dark-mode only, optimized for iPhone.
Reads sleep phases, heart rate and HRV. Writes Somnolyze sessions back to Health.
Microphone tap at 4096 samples, FFT through Accelerate, classification on-device.
Accelerometer signal feeds the agitation score that complements the audio model.
Time-sensitive calendar trigger as a safety net behind the lookahead alarm scheduler.
Optional on-device voice commands — say "commence" or "stop" to start or end a session.
Somnolyze launches in 2026. Want early access, a beta build, or to talk about the project? Drop a line.
Get in touch