An automatic and efficient method of snore events detection from sleep audio recordings

snores are respiratory sounds produced during sleep. they are reported to be a risk factor for various sleep disorders, such as obstructive sleep apnea syndrome (OSA). diagnosis of OSA relies on the expertise of the clinician that inspects whole night polysomnographic recording. this inspection is time consuming and uncomfortable for the patients. thus, there is a strong need for a tool to analyze snore sounds automatically. nocturnal respiratory sounds are composed of two kind of events: "silence" episodes and "sound" episodes that include breathing, snoring and "other" sounds. in this paper a new method to detect snoring episodes from full night audio recordings is proposed. signal analysis is performed in three steps: Pre-processing, automatic segmentation, extraction of features and classification. With the segmentation step, only the "sound" parts of the audio signal are extracted using a short-term energy and the otsu thresholding method. the aim of classification step is the detection of snore episodes only, using two neural artificial network applied to four features (length, maximum amplitude, standard deviation and energy). data from 24 subject are analyzed using the proposed method; on the dataset, a sensitivity of 86,2% and specificity of 86,3% are obtained