What Your Snoring Sound Actually Tells You
Machine-learning studies can classify where an airway narrows from the snoring sound alone, reaching 92% accuracy in development. They also mostly fail external validation — which tells you both that the sound carries real information and that no app can read it reliably yet.
Snoring is a vibration, and what vibrates determines what you hear. That is not folk wisdom — it is the premise behind a substantial research effort to work out the site of airway obstruction from sound alone.
The results are genuinely interesting in both directions.
The sound does carry information
A 2026 systematic review searched five databases and included 42 studies applying machine learning to snoring sounds, assessed for risk of bias with PROBAST and reported to PRISMA standards.
The target was the VOTE classification — identifying whether obstruction is at the Velum (soft palate), Oropharynx, Tongue base or Epiglottis, which is the same scheme a surgeon uses after sleep endoscopy.
On VOTE classification, baseline approaches reached unweighted average recalls of 49% to 75%. With targeted acoustic features and simpler classifiers, studies reported up to 87.5–94.7% recall or 92.2–95.5% accuracy.
The features doing the work are things like mel-frequency cepstral coefficients and wavelet transforms — essentially fine-grained descriptions of the sound's frequency structure over time. Related work has looked at whether snoring sound can predict apneic events and whether a smartphone alone can identify upper airway obstruction.
So the premise holds. Different obstruction sites really do produce acoustically different snores.
And the models mostly do not generalise
The same review is blunt about why none of this is in your pocket yet. The recurring problems were limited external validation, dataset imbalance, and noise interference, and the authors conclude that future work needs real-world validation and model generalisability "to support clinical translation."
Translated: models were often trained and tested on the same small, clean, unbalanced datasets. A classifier reporting 95% accuracy in that setting frequently collapses on a different population, a different microphone, or a bedroom with a fan running.
Evidence tier 2 that snoring acoustics encode the obstruction site. Tier 3 for any current consumer tool claiming to tell you where your airway narrows — the research models that perform well have not been externally validated, and the apps have not been validated at all.
What you can usefully tell by ear
Formal acoustic analysis is out of reach, but the crude version is still worth something, because the mechanisms are distinct.
Nasal snoring tends to be lower, steadier and more "snuffly," and it usually changes if you open your mouth or block one nostril. It often disappears when congestion clears.
Palatal snoring is the classic fluttering, rattling, variable-pitch noise. The soft palate is a loose membrane and it behaves like one.
Tongue-base snoring tends to be deeper and more intermittent, and it is characteristically much worse on your back, because that is when the tongue falls backward.
Silence punctuated by gasping is not snoring at all and it is the pattern that matters most. A fully obstructed airway makes no sound — so getting quieter can mean getting worse, not better.
None of those are diagnostic. They are hypotheses to test, and the tests are free: does a nasal strip or the cheek-pull manoeuvre change it, does lying on your side change it, does pushing your jaw forward change it. Where your airway actually narrows, and how to test for it.
How to use this practically
Record yourself and listen, rather than relying on an app's verdict. The recording is the valuable part; the app's classification is not validated. What you want from a recording is your own observation of pattern, timing and whether position changes it. How to record your snoring properly.
Treat app "snore scores" as trend data, not measurement. They are reasonable for comparing Tuesday with Wednesday on the same phone in the same room. They are not measuring your airway. What snoring apps can and cannot do.
Be most interested in the silences. If a recording contains pauses followed by gasps or snorts, the sound analysis question becomes irrelevant and the priority becomes a sleep study. Witnessed breathing pauses, gasping arousals, waking unrefreshed, or sleepiness that more sleep does not fix all point there. Snoring versus apnea.