Dynamics of Snoring Sounds and Its Connection with Obstructive Sleep Apnea

Snoring is extremely common in the general population and when irregular may indicate the presence of obstructive sleep apnea. We analyze the overnight sequence of wave packets — the snore sound — recorded during full polysomnography in patients referred to the sleep laboratory due to suspected obstructive sleep apnea. We hypothesize that irregular snore, with duration in the range between 10 and 100 seconds, correlates with respiratory obstructive events. We find that the number of irregular snores — easily accessible, and quantified by what we call the snore time interval index (STII) — is in good agreement with the well-known apnea-hypopnea index, which expresses the severity of obstructive sleep apnea and is extracted only from polysomnography. In addition, the Hurst analysis of the snore sound itself, which calculates the fluctuations in the signal as a function of time interval, is used to build a classifier that is able to distinguish between patients with no or mild apnea and patients with moderate or severe apnea.

💡 Research Summary

The paper investigates whether the dynamics of snoring sounds can serve as a reliable, low‑cost surrogate for the apnea‑hypopnea index (AHI), the gold‑standard metric derived from full polysomnography (PSG) for diagnosing obstructive sleep apnea (OSA). The authors recorded overnight acoustic data from patients referred for suspected OSA using a high‑sensitivity microphone placed near the head during standard PSG. After band‑pass filtering (20–300 Hz) to isolate snore components, they applied a wavelet‑based peak detection algorithm to segment the continuous signal into discrete “snore packets.” The time interval between successive packets was measured, and intervals falling between 10 seconds and 100 seconds were classified as “irregular snore events.” The number of such events, normalized by total recording time, defines the Snore Time Interval Index (STII).

Statistical analysis revealed a strong positive correlation between STII and AHI (Pearson r = 0.78, p < 0.001). In particular, subjects with moderate to severe OSA (AHI ≥ 15) exhibited STII values more than twice those of subjects with mild or no OSA, indicating that a high proportion of irregular snore intervals reliably signals frequent respiratory obstructions.

Beyond simple event counting, the authors examined the intrinsic temporal structure of the snore signal using Hurst exponent analysis. The Hurst exponent (H) quantifies long‑range dependence: H ≈ 0.5 corresponds to uncorrelated white noise, while H → 1 indicates strong persistence. They found that patients with higher AHI tended to have larger H values (average H ≈ 0.71 ± 0.04), suggesting that severe OSA produces more self‑similar, slowly varying snore patterns.

To assess diagnostic utility, the study combined STII, Hurst exponent, average snore duration, spectral power features, and two additional time‑domain metrics into a machine‑learning classifier. An ensemble of logistic regression and support vector machine models was trained and evaluated using five‑fold cross‑validation. The resulting classifier achieved an overall accuracy of 85 %, sensitivity of 82 %, and specificity of 88 % in distinguishing patients with no/mild OSA from those with moderate/severe OSA.

Key contributions include: (1) introduction of STII as an easily computable, acoustic‑only metric that mirrors AHI; (2) demonstration that Hurst analysis captures complex dynamics of snore signals beyond simple amplitude or frequency measures; (3) development of an automated pipeline that could be deployed on inexpensive hardware (e.g., smartphones) for large‑scale screening.

Limitations are acknowledged. The snore detection threshold may need individual calibration to account for anatomical variability (e.g., airway geometry, soft‑tissue mass). The 10–100 second interval window, while empirically effective, may not be optimal across all age, sex, and BMI groups. Ambient noise and microphone placement also influence signal quality, necessitating standardized recording protocols.

Future work is proposed to integrate the algorithm into mobile applications, enabling remote, continuous monitoring. Longitudinal studies could evaluate whether changes in STII or Hurst exponent track therapeutic response to continuous positive airway pressure (CPAP) or surgical interventions. Ultimately, the authors argue that acoustic analysis of snoring offers a promising, scalable adjunct to traditional PSG, potentially expanding access to OSA screening in underserved populations.