CNN-SENet: A Convolutional Neural Network Model for Audio Snoring Detection Based on Channel Attention Mechanism

Snoring is recognised as an independent risk factor for cardiovascular disease, making its monitoring crucial for disease prevention and management. Existing technologies face challenges due to the diversity of signals caused by individual physiological differences, as well as the non-linearity and multi-dimensional complexity of the signals themselves, making accurate and robust snoring monitoring difficult. To address these issues, this study proposes a hybrid architecture combining Convolutional Neural Networks (CNN) and Squeeze-and-Excitation Networks (SENet). By employing multidimensional nonlinear modelling and Mel-spectrogram feature extraction techniques, this architecture significantly improves the accuracy of snoring and non-snoring signal detection in various complex noise environments. In addition, the introduction of a channel attention mechanism further enhances the model’s focus on multidimensional feature weights, ensuring high robustness and excellent analysis efficiency in the face of environmental noise interference. Experimental results validate the effectiveness of the proposed model, achieving 100% snoring recognition accuracy in noiseless environments and maintaining an average accuracy of 97.17% in noisy conditions, significantly outperforming existing traditional methods and recent advanced baseline models. This research provides an efficient and accurate new method for snoring monitoring.