Detection of Sleep Apnea Based on Snore Signals Using Machine Learning Techniques

People's quality of life is significantly influenced by the quantity and quality of their sleep. Snoring bothers other individuals in the room as well as the snorer, who gets the worst quality sleep as a result. Among the most prominent sleep disorders, Sleep Apnea can have very serious side effects, from elevated blood pressure to heart failure, if it goes undetected. Many people are still unaware of this disease. A hardware part using raspberry pi and a microphone is used for collecting audio sounds at night from a subject. In order to identify snoring, two classifiers namely Long-short Term Memory(LSTM) and Gated Recurrent Unit(GRU) are proposed in this study. First, we gathered a dataset of sounds, both snoring and not from a website and a real-time dataset from hardware prepared for this project. The diverse properties of snoring are represented by the Mel-Frequency Cepstrum Coefficients (MFCC) which are retrieved as distinctive elements of snoring. The LSTM is made to be able to receive different audio cues and synthesize data to recognize snoring. When compared to single-feature network processing, using several feature coefficients can more precisely pinpoint the characteristics of snoring. GRU is capable of working with variable-length input sequences and can record the audio signal's time-varying patterns.