SCIENCE / SUMMARY / NIGHT COUGH

Can a night recording tell coughs apart from snores?

SINGLE STUDY · ALGORITHM VALIDATION ■■□□ low evidence: one centre, 26 people in the test set, a lab microphone rather than a phone

Who: 89 adults (62 men, 27 women) referred for polysomnography to the sleep unit of a university hospital in Israel. They had a range of conditions such as asthma, bronchitis and sinusitis and did not use CPAP. Mean age was 50.9 and mean AHI 17.4 per hour.
What they did: Alongside the sleep test, a studio microphone hung 1 metre above the head recorded an average of 7.1 hours. Candidate events were first picked out by sound energy and length (0.15 to 1.1 seconds); two listeners labelled each as cough, snore or noise (93% agreement). This gave 1,533 coughs, 142,972 snores and 158,280 noise events. People were split into training (36), development (27) and test (26) groups with no person in more than one group. Two classifiers were compared.
What they found: The better neural network model caught 86.1% of real coughs in the test set (460 coughs) and correctly set aside 99.87% of other sounds. Of the sounds it called coughs, 78.4% really were coughs. In cross validation with five different splits this share fell to an average of 57.4% and varied widely between splits (±30.8). Raising the catch rate to 90% lowered this share to 67.1%. Coughs were significantly less frequent in sleep than while awake, and in deep sleep than in other stages; the number of coughs had a weak link with AHI in men only (R = 0.278).
Limits: As the authors note: participants were referred with a range of conditions and larger samples are needed; since each system in the field was tested on different data and equipment, direct comparison with other studies is hard. In addition: recordings were made with a high quality lab microphone, not a phone; the catch rate was calculated only over events that passed the energy threshold, so coughs below it are not counted. Funding: Israel Science Foundation. No conflicts of interest reported.

What does this mean for you?

It seems possible for a machine to tell whether a sound in a night recording is a cough, but false marks are far from rare. If your cough at night lasts a long time or worries you, you could talk it over with a doctor. This is not advice.

Link to Nefesly: One of the reasons Nefesly says "a moment that sounds like a cough" rather than "cough"; it led us to keep in mind that accuracy with a lab microphone cannot simply be expected from a phone.
Reference: Eni M, Mordoh V, Zigel Y. Cough detection using a non-contact microphone: A nocturnal cough study. PLOS ONE. 2022;17(1):e0262240. doi:10.1371/journal.pone.0262240 · Open access (Creative Commons Attribution License).
Why chosen: One of 15 sources selected from 534 candidates on the acoustic snoring and apnea axis; a study separating night coughs from snores with a contactless microphone.
Summarised by: Nefesly, from the full text, 11 October 2026. Awaiting clinician check.
Copyright: Summarised from the full text, not translated from the abstract.