MM-AAD

Multimodal Auditory Attention Detection Dataset This dataset was collected at the IIP-HCI Lab, School of Computer Science and Technology, Anhui University. We refer to the EEG recorded during each session as a "trial," with 20 trials recorded per subject. The EEG data for each subject is stored in individual subject-specific .npy files, named "Sx", where "x" corresponds to the subject number. This dataset is specifically designed for the EEG-AAD Challenge, particularly focusing on auditory attention decoding (AAD) in both cross-subject and cross-session contexts. Task1 Cross-subject: In the training phase, we provide data for 30 subjects in a 7z file named 'EEG-AAD_audio-visual.7z' for training and validation. Task2 Cross-Session: In the training phase, we provide data for 30 subjects in a 7z file named 'EEG-AAD_audio-only.7z' for training. For validation, we supply the same 30 subjects' data in a 7z file named 'EEG-AAD_audio-visual.7z'. This competition also provides training data for testing the final model on 10 subjects in a 7z file named "EEG-AAD_audio-only_part2.7z". The test data corresponding to each of these 10 subjects will be released during the testing phase. The 7z file contains two folders, named 'raw' and 'preprocessed', where 'raw' represents the original data files and 'preprocessed' represents the preprocessed data files. Each folder contains the corresponding 30 subjects’ data and labels. Related Work [Fan et al., 2025] Cunhang Fan, Hongyu Zhang, Qinke Ni, Jingjing Zhang, Jianhua Tao, Jian Zhou, Jiangyan Yi, ZhaoLv, and Xiaopei Wu. Seeing helps hearing: A multi-modal dataset and a mamba-based dual branch parallel network for auditory attention decoding. Information Fusion, page 102946, 2025.[Yan et al., 2024] Sheng Yan, Cunhang Fan, Hongyu Zhang, Xiaoke Yang, Jianhua Tao, and Zhao Lv. Darnet: Dual attention refinement network with spatiotemporal construction for auditory attention detection. Advances in Neural Information Processing Systems, 37:31688–31707, 2024.