A Snoring Detector For OSAHS Based On Formant

Obstructive Sleep Apnea Hypopnea Syndrome (OSAHS) is a common sleep-related breathing disorder, which is a frequent in the general population. Frequently apnea and hypopnea during sleep makes patients more likely to lead to cardiovascular disease, kidney disease, hypertension, and other vital organs complications, even death.Polysomnography (PSG) is the gold standard diagnostic tool for assessing OSAHS. But, the high cost, the limited number, and the discomfort of the electrodes attached body of the patients, are the limitations of PSG. It is desirable to have an alternative method to assess OSAHS non-invasively for a mass of snorers, with greater comfort and at a lower cost.The research method in this paper is the use of formant to assess OSAHS. First, the snoring signals were pre-processed by digital speech signal processing method; and we use an improved short-time energy method to detect snoring segments. The snoring model parameters of upper airway were estimated by linear predictive technique, and the formants could be identified by roots finding method. A fixed formant threshold method has been proposed to assess OSAHS, but the detection rate is not very high because the existence of patient's individual personality. In this paper a personalized threshold method is proposed to assess the severity of OSAHS. First, for a certain snorer, the first formants of whole night snoring were divided into two clusters based on K-means clustering algorithm. And the smaller cluster center was considered to be the base frequency; the personalized threshold was set about two double of base frequency.The details of proposed method in this paper is just as follows:Firstly, if the first formant frequency is bigger than the personalized threshold, this formant belongs to abnormal snoring segment; if the duration of the snoring segment is longer than 0.3s, then we think this snoring episode is not an normal one, and this episode also be seen as one hypopnea. Secondly, statisticing the abnormal snoring segments in one hour and this number is considered as the simulated apnea-hypopnea index (AHI). If the value of this simulated AHI is bigger than five, we think this snorer is an OSAHS patient; on the other hand, we think this snorer is a simple one. The sensitivity and specificity of the personalized threshold method are 93.3% and 91.67% respectively that meet the requirements of clinical disease on the screening.