13th International Conference on Signal and Image Processing, SIP 2011, Dallas, TX, Amerika Birleşik Devletleri, 14 - 16 Aralık 2011, ss.177-182
The aim of this study is to analyze the snore related sounds (SRSs) recorded from 12 obstructive sleep apnea/hypopnea (OSAH) patients and 8 healthy subjects during sleep and classify into four groups by using the decision tree classification algorithm. Firstly, SRSs were segmented as apnea,/hypopnea, breathing, silence and normal snore parts from records. After segmentation, model order is evaluated by Akaike's Final Prediction Error (FPE) and parts are fitted to autoregressive (AR) model. We used model order, pitch period and energy of segments for classifier inputs. It is observed from experimental results that training and test accuracy are sequentially %97.17 and %88.07. This results show that the model order, pitch period and energy of parts are efficient parameters to analyze and separate SRSs as apnea/hypopnea, breathing, silence and normal snore segment and can be used for diagnosing snore sound disorders.