Can an App Screen for Sleep Apnea From Your Snoring?

Your Phone Listened to You Snore and Now It's Worried

There's a whole category of apps now that record you overnight, run the audio through a model, and hand you a number in the morning. Some of them will tell you that you might have sleep apnea.

If one of those has flagged you, you're probably somewhere between "this is nonsense" and "should I be scared." Both reactions are understandable and both are a little off.

Here's the short version: the technology is real, the studies are small, and a flag is a reason to make a phone call, not a diagnosis. Let's go through why.

This post is educational. It is not medical advice, and nothing here diagnoses anything. If any of it applies to you, the person to talk to is your doctor.

Key Facts

• Sleep apnea is common. An estimated 936 million adults aged 30–69 worldwide have mild-to-severe obstructive sleep apnea; 425 million have moderate-to-severe (Lancet Respiratory Medicine, 2019)
• Sleep doctors say screening tools can't diagnose it. The AASM makes a strong recommendation against using questionnaires or prediction algorithms to diagnose apnea without a sleep study (AASM guideline, 2017)
• Consumer sleep gadgets are explicitly out of scope for diagnosis in the AASM's position statement (AASM)
• Pooled smartphone screening performance: about 90% sensitivity and 88% specificity across 11 studies and 1,644 people — but the authors say it can't replace a sleep study (PLOS ONE, 2022)
• Most of these tools have never been tested outside their own dataset. Only 8 of 41 studies in one review did external validation (JMIR, 2023)
• Loud snoring isn't a reliable stand-in for apnea. In 150 patients, average snoring loudness had no correlation with apnea severity (Lung India, 2020)

How Audio Screening Actually Works

Strip away the marketing and it's fairly simple.

Your phone records the night. Software converts the sound into a spectrogram — a picture of which frequencies were loud at which moments. It looks a bit like a heat map with time going left to right and pitch going bottom to top. Sounds become images.

Then a neural network — a pattern-matching model trained on lots of examples — learns which image patterns went with breathing interruptions in the training data. It looks for the acoustic signature of an airway partially closing, the gasping recovery breath, the silence where breathing should be.

Some apps add the phone's motion sensors on top of the microphone, since the phone on the mattress picks up chest movement and body shifts.

That's the whole idea. It's not magic and it's not fake. It's pattern matching on sound.

The Words You Need, In Plain English

Every paper on this uses four terms, and they're the entire argument. Here they are with no jargon.

You cannot max all four. Push a test to catch more real cases and it will flag more people who are fine. Push it to stop bothering healthy people and it will miss real cases. That's the whole trade, and every screening tool has to pick a side.

Why "Recall-First" Is the Right Choice for a Screener

A screener and a diagnosis have different jobs.

A diagnosis has to be right, because treatment follows from it. A screener only has to decide who's worth looking at more closely. Those jobs deserve different settings.

It helps to know the scale of the problem the screener is aimed at. A 2019 analysis estimated that 936 million adults aged 30 to 69 worldwide have mild-to-severe obstructive sleep apnea, and 425 million have the moderate-to-severe form [^9]. That's a lot of people who have never been tested.

For a screener, a miss is expensive and a false alarm is cheap. If you miss someone with real apnea, they go untreated. If you falsely flag someone, they have an appointment they didn't strictly need. Those costs aren't remotely equal, so a good screener deliberately biases toward catching people.

A 2025 preprint on arXiv — not yet peer reviewed — makes this design choice explicitly. The team built a convolutional neural network on respiratory audio, oversampled the apnea segments so the model wouldn't just learn to say "no," and reported 90.55% recall for apnea events. They're upfront that precision was low, and they argue the tradeoff is right for a low-cost screening tool [^1].

I agree with the reasoning. I also want to be blunt about the scale: that study used 18 audio files. It's a proof of concept, not evidence about how the approach performs in the world. Preprints haven't been through peer review, which is exactly the stage where someone asks hard questions about a sample that size.

What the Better Studies Show

A second arXiv preprint from 2024 — again, not peer reviewed — is more substantial. Researchers tested a smartphone app that uses the phone's microphone plus its motion sensors against full in-hospital sleep studies in 46 patients. For picking out moderate-or-worse apnea, they reported sensitivity of 0.91 and positive predictive value of 0.89. For severe apnea, sensitivity 0.85 and positive predictive value 0.94 ¹.

Those are decent numbers on a real comparison against the gold standard. Forty-six patients is still small, and hospital patients referred for sleep studies are a very different group from the general public.

The peer-reviewed literature is broadly encouraging with heavy caveats:

The pattern: promising in the lab, thin outside it. The JMIR authors put it plainly — these tools "presented promising results," but there's "still a need for quality studies comparing the developed tools with the gold standard and validating them in external populations and other environments before they can be used in clinical settings" ³.

That last part matters more than it sounds. A model tested only on the data it was built from is grading its own homework.

Snoring Is Not Apnea

Worth saying clearly, because a lot of people conflate them.

Snoring is the sound of tissue vibrating in a narrowed airway. Apnea is the airway actually closing enough to interrupt breathing. They're related, they share a mechanism, and they are not the same thing.

Group-level acoustics do carry signal — that meta-regression of 3,153 adults found a pooled correlation of about 0.71 between snoring sound features and apnea severity ⁴. But zoom into individuals and it gets messy. In a study of 150 patients, the average loudness of snoring had no statistically detectable relationship with apnea severity. Peak loudness and snoring rate did correlate, but weakly — around r = 0.36 and r = 0.34 ⁵.

Translation: you can snore like a chainsaw and have a normal sleep study. You can be a quiet sleeper with significant apnea. Sound is a clue, not a verdict. Position plays into this too, which we cover in snoring and sleep position.

What Sleep Medicine Actually Says

The American Academy of Sleep Medicine has been direct about this.

Their 2017 diagnostic guideline carries a strong recommendation: "clinical tools, questionnaires and prediction algorithms not be used to diagnose OSA in adults, in the absence of polysomnography or home sleep apnea testing" ⁶. A phone app running a prediction algorithm on your snoring sits squarely inside that sentence.

The same guideline adds something people often miss: if a home sleep apnea test comes back negative, inconclusive, or technically inadequate, they recommend a full in-lab sleep study anyway ⁶. That's the professional standard applied to medical home testing. A consumer app is several steps further from the gold standard than that.

And their position statement on consumer sleep technology says these devices "cannot be utilized for the diagnosis and/or treatment of sleep disorders at this time," while allowing that they "may be utilized to enhance the patient–clinician interaction when presented in the context of an appropriate clinical evaluation" ⁷.

That second half is the useful part. Bring the data to the appointment. Don't try to be the appointment.

So What Should You Actually Do?

If an app flagged you: make an appointment. Don't spiral, and don't buy anything. Bring the app's output, and bring the things that actually matter clinically — whether you snore loudly, whether anyone has seen you stop breathing or gasp, whether you're exhausted during the day, whether you wake with headaches or a dry mouth. Those symptoms plus a screening flag are a reasonable reason to get tested properly. If you end up with a sleep study, we wrote a plain-English guide to reading a sleep report.

If an app cleared you but you feel terrible: a clean result is not an all-clear. A tool with high sensitivity in one study population can still miss you. If you have the symptoms, get evaluated regardless of what your phone said. The AASM's own logic — a negative home test in a high-suspicion patient still warrants a full study — applies here with room to spare.

What not to do: don't self-treat. Don't buy a device off the internet because an app said a number. Untreated apnea has real consequences and so does treating something you don't have.

Where Komori Fits

Let me be direct about our own product, because this is exactly the topic where a company would be tempted to overclaim.

Komori is a contactless wellness device we're building. It is not a medical device, it does not diagnose or screen for any condition, and it is not being built to detect sleep apnea. It is not designed to measure breathing, respiration, heart rate, or blood oxygen.

What it is designed to do: track sleep position and how it changes, movement and restlessness, bed-exit events, room conditions like temperature, humidity, CO2, light, and noise, and general sound events — contactless, no camera, nothing worn.

If you're already working with a doctor on a sleep issue, night-after-night position and movement history from your own bed is context a single lab night can't give you. That's the honest scope. It's context for a clinical conversation, and nothing more. For related reading, we've written about what position data can and can't tell you and why we refuse to reduce a night to one score.

Screening apps are a useful nudge for a condition that goes undiagnosed in enormous numbers of people. Treat them as a nudge. The diagnosis belongs to your doctor, and so does the decision about what to do next.

---

Footnotes

1. Frija J, Millet J, Bequignon E, et al. "Validation of a new, minimally-invasive, software smartphone device to predict sleep apnea and its severity: transversal study." arXiv preprint 2406.16953, submitted 20 June 2024. Preprint — not peer reviewed. 46 patients. ↩

2. Kim DH, Kim SW, Hwang SH. "Diagnostic value of smartphone in obstructive sleep apnea syndrome: A systematic review and meta-analysis." PLOS ONE, 2022. 11 studies, 1,644 participants. ↩

3. Duarte M, Pereira-Rodrigues P, Ferreira-Santos D. "The Role of Novel Digital Clinical Tools in the Screening or Diagnosis of Obstructive Sleep Apnea: Systematic Review." Journal of Medical Internet Research, 2023. 41 studies. ↩ ↩²

4. Chiang JK, et al. "Correlation between snoring sounds and obstructive sleep apnea in adults: a meta-regression analysis." Sleep Science, 2022. 13 studies, 3,153 adults. ↩ ↩²

5. Kallel S, et al. "Snoring time versus snoring intensity: Which parameter correlates better with severity of obstructive sleep apnea syndrome?" Lung India, 2020. 150 patients. ↩

6. Kapur VK, et al. "Clinical Practice Guideline for Diagnostic Testing for Adult Obstructive Sleep Apnea." Journal of Clinical Sleep Medicine, 2017. American Academy of Sleep Medicine. ↩ ↩²

7. American Academy of Sleep Medicine. "Consumer Sleep Technology: AASM Position Statement." ↩