Obstructive Sleep Apnea Diagnosis in Awake Subjects Using Speech Signal Processing Methods - ISMBE 2016 - Biomedical Engineering 2016, International Convention Center Haifa, 24.2.2016

Dvir Ben Or ¹ Ariel Tarasiuk ² Yaniv Zigel ¹

¹Dept. of Biomedical Engineering, Ben-Gurion University of the Negev
²Sleep-Wake Disorders Unit, Soroka University Medical Center, and Department of Physiology, Faculty of Health Sciences,, Ben-Gurion University of the Negev

Obstructive sleep apnea (OSA) is a common sleep-related breathing disorder which is increasingly recognized as an important cause of medical morbidity. It is also associated with anatomical and functional deviations of the upper respiratory tract. We hypothesize that the abnormal vocal tract geometry of OSA patients will be reflected in the acoustic parameters of speech. Hence, our aim is to use speech signals recorded from awake subjects in order to detect and evaluate OSA. The study included 198 men and 105 women with a variety of OSA severities, who were recorded, right before undergoing full polysomnography study, while reading a one-minute speech protocol. In order to exploit the hidden information within the signals, descriptive speech features were extracted, while Gaussian mixture models (GMM) and adaptive boosting (Adaboost) classifiers were designed, achieving accuracy rates of 81% and 67% for men and women, respectively. The suggested system may serve as an automatic, fast and low complexity tool for screening of potential OSA patients.