The research investigates the use of deep Self-Supervised (SS) representations for Auditory Attention Decoding (AAD) using EEG data. AAD algorithms are significant for isolating desired sound sources within challenging acoustic environments directly from brain activity. The study compares the performance of linear decoders across 12 deep and 2 shallow representations, applied to EEG data from multiple studies spanning 57 subjects and various languages. The results show deep features’ superiority in decoding background speakers, suggesting a possible nonlinear encoding of unattended signals in the brain. The research also examines the impact of different layers of SS representations and window sizes on AAD performance.
Publication date: 3 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.00814
