Research On Snoring Data Analysis Method Based On Artificial Intelligence

Obstructive sleep apnea hypopnea syndrome is the most predominant and morbid disease in sleep-respiratory disease,and snoring is a common feature of OSAHS patients.At present,the treatment of patients with OSAHS usually requires drug-induced sleep endoscopy,which is not only expensive but also is an invasive treatment for the patient.Therefore,It's of great help to use acoustic analysis to determine the source of snoring in curing the OSAHS patients.This paper mainly bases on the National Natural Science Foundation of China:snoring source and upper airway obstruction recognition based on acoustic analysis.First of all,we introduced the snoring signal generation model,based on the classification of doctors labeled snoring source database.Next,the labeled snoring source data was taken as the research object.The linear spectrum and nonlinear spectrum were used as input respectively to analyze the recognition results of convolution neural network classifier for different snoring source feature sets.The nonlinear spectra are divided into Mel frequency spectrum and Mel frequency cepstrum.The results showed that when the linear spectrum was taken as input,the recognition accuracy of convolution neural network model was 89.1%,and the recognition accuracy was 85.7%when Mel frequency spectrum was taken as input.The recognition accuracy rate is 70.7%when Mel frequency cepstrum spectrum was taken as input.Combined with the snoring signal generation model,the time-frequency spectrum and Mel frequency spectrum reflect more information of excitation source,while the cepstrum reflects more upper airway response information,so it is presumed that the snoring source classification should have a high correlation with the excitation source information of snoring characteristics.And we should not choose the upper airway response information.To further study this speculation,the excitation information is represented by the fundamental frequency,and the resonance peak is the upper airway feature.We did lots of parameter statistics in different types of snoring source data and found that the fundamental frequency parameters in all types of snoring sound are higher different.It can be concluded that the characteristics related to excitation should be adopted in the classification of snoring sources.