AASD: An Open Non-Invasive EEG Dataset for Spontaneous Auditory Attention Switch Decoding

The Auditory Attention Switching Dataset (AASD) provides a large-scale, open-access collection of electroencephalography (EEG) recordings designed to investigate spontaneous auditory attention switching in natural listening environments. This dataset fills a critical gap in auditory brain–computer interface research by capturing neural dynamics during spontaneous attention switch rather than externally cued transitions. The AASD enables the study of natural auditory attention mechanisms and supports the development of robust neural decoding algorithms for real-world applications.

Participants and Ethical Approval

Eighteen healthy volunteers aged between 18 and 27 years participated in this study. All participants were native Mandarin speakers with normal hearing and no neurological or psychiatric history. The experimental procedures were reviewed and approved by the Ethical Review Board of the Southern University of Science and Technology (Approval No. 2022DZX003). Each participant provided written informed consent and explicitly agreed to the public sharing of their anonymized data. All personally identifiable information was removed to ensure complete data privacy.

Experimental Design and Procedure

Each participant completed 60 trials (60 s each) organized into six randomized attention-switching blocks of 10. Two narrative speech streams, one male and one female, were simultaneously presented through headphones at ±90° azimuth using head-related transfer functions (HRTFs) to simulate spatial separation.

• In half the trials, male speech was presented to the left ear and female to the right; in the remaining half, the configuration was reversed.
• Participants freely chose which stream to attend to and could spontaneously switch attention between them, pressing a button to mark each switch.
• Each session lasted approximately 150 minutes, including setup, training, and rest breaks between blocks.

In addition to the main attention-switching blocks, a passive listening control condition was conducted as the final block (10 trials).  During this block, participants did not perform voluntary auditory attention switching but were instructed to press the response keys alternately at a rate that roughly matched the switching frequency observed during their previous blocks.

The system synchronized all EEG, audio, and behavioral markers to achieve precise temporal alignment of attention labels.

Dataset and Code Release:

This public release provides the complete Auditory Attention Switching Dataset (AASD) for research use. The dataset includes EEG recordings, synchronized auditory stimuli, and behavioral markers collected from all eighteen participants. Each participant completed sixty trials of sixty seconds, involving self-initiated attention switches between two spatialized speech streams.

This release contains:

• EEG recordings from sixty-four channels under spatialized dual-speaker listening conditions.
• Raw EEG data in CNT format and preprocessed EEG data in MAT format, aligned with corresponding event triggers.
• Spatialized two-channel speech stimuli generated from the AISHELL Mandarin corpus (three male and three female speakers).
• Trial-level metadata including attention direction and button-press timestamps marking spontaneous switch events.