SSF-DST: A Spectro-Spatial Features Enhanced Deep Spatiotemporal Network for EEG-Based Auditory Attention Detection

In a cocktail party scenario, humans can selectively attend to different speakers. EEG-based auditory attention detection (AAD) aims to identify the attended speaker from brain signals. Existing AAD methods enhance features through temporal-frequency fusion, but frequency features incur heavy computational cost with limited contribution. Furthermore, many methods utilize attention mechanisms to capture temporal dynamics, which may not fully capture the complex interplay of spatial and temporal information. To address these issues, we propose a spectro-spatial features enhanced deep spatiotemporal network (SSF-DST). SSF-DST efficiently extracts spectro-spatial features from frequency domain to enhance spatial patterns of EEG, and constructs deep spatiotemporal features to capture long-term dependencies and spatial patterns. Experimental results demonstrate that SSF-DST outperforms the state-of-the-art model in classification accuracy while reducing the trainable parameters by 50%.