The honest guide to phone-based sleep tracking without a wearable

Wearables dominate the sleep-tracking conversation, but they cost money, need charging, and — critically — many people simply won't wear something on their wrist. Phone-based sleep tracking is the underrated option: the device is already on your nightstand, it does not need charging, and it can actually hear your breathing.

The trade-off is engineering honesty. Here is what phone-based tracking can and cannot measure, and how to do it right.

What the microphone gets you that the accelerometer can't

A phone on the nightstand hears what no wrist sensor can: snoring, gasping, coughing, sleep talking, teeth grinding. These are acoustically rich signals, and they map directly to things people actually care about — like whether their snoring crosses into worrying territory. The full detail on acoustic snore and apnea signal detection is worth a read if you want the signal-processing specifics: snoring and sleep apnea detection on-device.

The honest limits

• Sleep stages are estimates. Without an EEG you are classifying by proxy — movement, breathing rhythm, time-of-night priors. A stage chart is a probability curve, not ground truth. Apps that present it as medical certainty are lying to you.
• Movement detection is coarse. A phone in the bed picks up big movements but not micro-movements. If you need true actigraphy-grade restlessness, the phone is not the tool.
• Roommates and pets are signal. The mic hears your partner, your dog, the heater. A good algorithm knows how to discount a partner who also snores.

How to measure without a wearable

The practical stack that works:

1. Record all night, process on-device. Keep raw audio in memory, never upload it. Process streaming windows so battery stays reasonable.
2. Streaming windows, not full-night files. Detect features every few seconds, keep only the summary vector. You do not need to store 8 hours of PCM.
3. Use time-of-night priors. Deep sleep is statistically more likely in the first third of the night. A Bayesian prior improves classification far more than a bigger model.
4. Validate against a consumer wearable once. You do not need polysomnography-grade accuracy to be useful; you need to know how much your proxy drifts from a reference sensor so you can state your accuracy honestly.

Why "no wearable" is a feature

There is a real, underserved group: people who hate wearing anything to bed, who forget to charge a band, or whose skin reacts to the strap. For them, a phone on the nightstand is not a compromise — it is the only option that fits their life. SleepTrace is built for exactly that group: no wearable, no subscription, everything processed on the iPhone itself.

The bottom line

Phone-based sleep tracking is not a worse version of a wearable. It is a different instrument with different strengths — hearing you breathe beats counting your arm movements for a whole class of questions. Build it honestly, state the accuracy limits plainly, and it earns its place on the nightstand.