A Speech Dataset for Snoring Detection in Sleep Based on Deep Learning

Obstructive Sleep Apnea Hypopnea Syndrome (OSAHS) is a common sleep disorder that affects the health and quality of life of millions of people worldwide. Traditional methods for diagnosing OSAHS, such as polysomnography (PSG), are widely recognized but have drawbacks such as high cost, operational complexity, and inconvenience to patients. Snoring sound detection methods based on deep learning offer a low-cost and convenient alternative for OSAHS diagnosis. However, these methods often require large amounts of data for training, and the lack of datasets severely hinders the development of snoring recognition. In this study, we aim to construct a comprehensive snoring audio dataset to support research and diagnosis of OSAHS. This dataset includes four categories: non-snoring, normal snoring, abnormal snoring, and breathing sounds. We utilize four signal processing algorithms—Fast Fourier Transform (FFT), Short-Time Fourier Transform (STFT), Continuous Wavelet Transform (CWT), and Mel-Frequency Cepstral Coefficients (MFCC)—to extract features from the audio. These extracted features are then evaluated using 12 state-of-the-art (SOTA) classification models. Experimental results demonstrate that this dataset can achieve high accuracy across most classification models. We believe this dataset can provide valuable reference and support for related research.