Speech-like Analysis of Snore Signals for the Detection of Obstructive Sleep Apnea

Correlation between snoring and obstructive sleep apnea (OSA) has been studied over the past years using techniques such as sound intensity computation, power spectrum estimation, counting of snores, and more recently, pitch and jitter analysis. Although these studies have ascertained the relationship between snoring and OSA, they gave mixed outcomes in diagnosing OSA. In this paper, snore signals are analyzed using linear predictive coding (LPC), which is a powerful speech analytic tool. The noisy snore signals were first preprocessed using a Modified Normalized Least-Mean-Square (MNLMS) adaptive filter for noise cancellation, followed by the extraction of formant structures from the denoised snore signals to discriminate between benign and apneic snores. This initial study demonstrates that the formant structures of snore signals contain useful information for OSA detection, and suggests the feasibility of using snore signal analysis as a noninvasive and inexpensive diagnostic tool for mass screening of OSA.