Research On Sleep Monitoring Based On Voice And PPG Signal

Sleep condition is an important factor affecting people's quality of life.Poor sleep can cause a person's mental debility,distraction,and even cause a series of diseases.Scientific and accurate sleep monitoring helps people better understand their sleep quality,and it can also help doctors diagnose sleep disorders.Traditional sleep monitoring uses a variety of sensors to connect to the human body to collect physiological signals,which need to be performed under the supervision of professional medical personnel,which is tedious and expensive,and causes serious interference to sleep,making non-interfering sleep monitoring a hot topic of recent research.Infrared,video,sleep monitoring for smartphones and smart watches.Sleep monitoring based on smart phones and smart watches is more easily accepted because it does not require special equipment.Smartphone-based sleep monitoring mainly uses the accelerometer and microphone in the phone to identify sleep behaviors to derive sleep quality.However,such existing studies only identify sleep events for a single individual.When two individuals are in the same Sleep events in the environment cause mutual interference of sleep events,and the sleep event recognition method for a single individual cannot be applied to two-person scenarios;sleep monitoring based on smart watches mainly uses acceleration sensors and photoplethysmography(PPG)sensors in the watch recognize limb movements and calculate breathing rate to estimate sleep stages.However,existing studies mainly derive sleep stages based on sleep events.The recognition accuracy is low,and the sleep stage information contained in PPG signals is not fully used.The diagnosis of sleep stages in patients with sleep disorders has not been fully studied.In view of the above problems,this thesis uses the two easy-to-collect signals of sound and PPG signal to study the problem of sleep monitoring as follows:(1)A snore detection scheme in a two-person scenario.An algorithm of target individual voice recognition in two-person scene based on Gaussian mixture model(GMM)is proposed.First,the voice characteristics of the target individual are collected,and in the two-person scene,the target individual and the interference individual's voice are modeled as Gaussian functions.Considering that the individual voice has a fixed frequency domain characteristic,a method based on GMM clustering is designed to distinguish the target individual and interfere with individuals.The experimental results based on real data show that the proposed algorithm can stably and accurately identify the target individual's voice in a two-person scene.The recognition of the voice can be used to diagnose apnea symptoms.(2)Sleep staging scheme based on PPG signal heart rate variability(HRV)analysis.A sleep staging algorithm for HRV analysis based on PPG signals is proposed.The algorithm first performs preprocessing according to the spectral characteristics of the PPG signal,and then performs HRV analysis after calculating the heart rate interval.Considering the non-linear separability of the data,the selected kernel function is optimized by a combination of genetic algorithm and grid search,and training and testing are performed using a support vector machine(SVM).The experimental results on the real data set show that compared with ECG signals,the proposed sleep staging method based on PPG signals for HRV analysis can achieve the same high-accuracy sleep staging,which is an efficient alternative to ECG signals for HRV analysis.The effects of apnea symptoms on sleep staging were also explored.