Analysis of Sleeping Respiratory Signal Utilizing Frequency Energy Features

Abstract

The sleeping respiratory sounds can reflect the condition of the airway, which is meaningful for the diagnosis and therapeutic of sleep-related disorders. Snoring is an essential vital sign for monitoring obstructive sleep apnea (OSA) during all-night sleep. This study presents a spectral division method for analyzing respiratory sounds, such as stable breathing, snoring, etc. The sleep respiratory sound signal is acquired by a portable, wearable sound device. After segmentation, the spectrum of each segmented data is computed. The frequency energy features are extracted to display the spectrum distribution more clearly and applied to classify the different respiratory statuses. A set of data is used to validate the efficiency of the proposed features by the cubic SVM. The accuracy rate for identifying snoring and breath sounds is more than 80.0%. The proposed spectral division method shows good performance for sleep respiratory analysis and has potential for sleep health monitoring.