A Deep Snore Recognition Approach Integrating Attention Mechanism and Spatio-Temporal Feature Extraction

Snoring, a common sleep issue, can indicate serious sleep-related breathing disorders. To aid in detection and diagnosis, this paper proposes a snore recognition model that integrates attention mechanisms and spatio-temporal feature extraction, known as the Dual Convolutional Layer with Convolutional Block Attention, Channel Splitting, and Gated Recurrent Network (DCSA-GRU), which integrates attention mechanisms and spatio-temporal feature extraction for snore recognition. The model uses dual convolutional layers with different kernel sizes and channel-spatial attention to enhance key information, improving robustness. A channel slicing method captures time-frequency dependencies, while two GRU layers dynamically model feature sequences. This approach yields better classification results compared to existing methods.