The PAAD Dataset: Decoding Passive Auditory Attention from EEG in Dichotic Natural Soundscapes

Decoding human auditory attention is essential for developing intelligent systems capable of adaptive speech perception in complex acoustic environment. While significant progress has been made in decoding active auditory attention using electroencephalography (EEG) devices, research on passive auditory attention has primarily focused on temporal saliency detection within a single stream, rather than identifying which sound source among multiple competing sounds captures attention. Public EEG benchmarks for this selective passive attention decoding task are currently lacking. To bridge this gap, we introduce the Passive Auditory Attention Dichotic (PAAD) dataset. This dataset was collected using a dichotic listening paradigm with natural, real-world sounds, where subjects indicate which of two simultaneous streams passively captured their attention. This paper details the experimental paradigm, stimulus design, and data pre-processing. We also provide baseline results using both machine learning and neural network-based EEG decoding methods, demonstrating the feasibility of identifying the attended sound source from neural signals. By providing this dataset and benchmark, we aim to foster further research into passive attention decoding for next-generation attention-aware hearing technologies.