Method

How we listen to the night

Nefesly summarises the night from three devices: the phone listens to sound, the smartwatch measures heart rate and movement, and the ring measures blood oxygen. Sound is processed on the phone, without being sent to the internet. Our method was developed by comparison with sleep laboratory measurements and is tested again with every version. The results are for information; Nefesly is not a medical device.

Three devices, three jobs

• The phone listens to snoring and breathing sound through the night. How the sound is processed is set out in the four steps below.
• The smartwatch (Apple Watch, Galaxy Watch) measures heart rate, beat to beat interval, the PPG signal and movement. Nefesly reads the watch's sleep stages, heart rate and breathing rate through Apple Health or Health Connect, so that minutes spent awake are set apart and breathing reductions are judged against real sleep time.
• The ring (Wellue O2Ring S) measures blood oxygen (SpO2) and heart rate once a second. We use the ring for oxygen because measurement at the wrist is sparse and less accurate: in a study presented at Sleep Europe 2026, a wrist band missed the moments when oxygen fell below 90%.

On nights without the watch or the ring, the night is assessed from the sound the phone hears; oxygen and sleep stage information for that night is not in the summary.

Sound processing: four steps, on the phone

1. 1 · RecordingAudio stays on the phoneThrough the night, sound is processed in short pieces; the recording does not leave the phone.
2. 2 · ClassificationA probability for every momentA sound model running on the phone separates snoring, speech and cough-like sounds with probabilities; it does not make a firm decision.
3. 3 · EnvironmentRoom noise is set apartContinuous household appliance sounds are marked separately and do not spoil the assessment of the night.
4. 4 · BreathingIntervals where breathing fadesMarked drops in breathing sound are looked for; moments spent awake are left out.

Development and validation

The algorithm was developed on nights recorded in a sleep laboratory at the same time as polysomnography, nights recorded at home with a phone, and recordings made in a controlled setting. Every version is measured on recordings from people never used in development; a new version only goes into the app if it does better than the previous one in this measurement.

Our sleep laboratory validation study is designed to compare the phone, the watch and the ring with polysomnography on the same night: the phone is placed by the bed, the watch on the wrist and the ring on a finger. At Sleep Europe 2026, no abstract tested these three devices together against PSG on the same night.

In development we use open scientific datasets, in line with their licences, and the peer-reviewed literature. Plain summaries of the studies the method rests on are in the Science section.

How we show results

• Because the number for a single night is not reliable at the individual level, the app gives no score or index; it shows a timeline, intervals and the trend over several nights.
• If the phone could not hear well enough, the night is marked "could not be assessed" and the reason is given.
• Sounds that are harder to tell apart, such as coughing, are presented as a "cough-like moment", in the language of probability.

Known limitations

• Airflow and chest movement (breathing effort) are not measured.
• On nights without the watch and the ring, oxygen level and sleep stages are not known.
• Sleep stages from the watch are approximate compared with polysomnography; they are weakest for deep sleep.
• Sensitivity to mild changes in breathing is limited.
• In a shared bed, a single phone cannot tell whose sound it is.
• Loud and changing room noise can affect the result.

We state these limitations clearly in the app as well, and we regularly improve the algorithm in the light of new recordings and current literature. Changes are listed in the release notes (in Turkish).