Exploring Auditory Attention Decoding to go with mobile and portable hardware.

Auditory attention decoding (AAD) has emerged as a promising approach for neuroadaptive hearing technology. However, its feasibility in naturalistic, mobile settings remains underexplored. In this study, we investigated AAD using a minimal and wearable setup comprising of around-the-ear EEG and portable hearing aid research hardware. Participants performed an auditory attention task under both controlled (seated) and naturalistic (walking) conditions with varying auditory scene complexity (single- and dual-speaker conditions, cafeteria background noise). We applied both backward (decoding) and forward (encoding) models to assess attention-dependent neural tracking of continuous speech. Decoding results replicated established trends, with the highest reconstruction accuracy for the single speaker, followed by dual-speaker attend, and then ignored. These trends were extended to the seated and walking contexts, initially demonstrating the potential of cEEGrids for mobile AAD. Post hoc forward modeling revealed an artifactual response in the walking condition, biasing the reconstruction and prediction accuracy. Further analysis indicated that this effect could reflect hardware-related interference, possibly from contact between hearing aid cables and electrodes. Importantly, the artifact, while speech-locked, was detectable only through the forward model. Despite these constraints, our results demonstrate the feasibility of decoding auditory attention using a lightweight, unobtrusive EEG system in stationary scenarios. This work emphasizes the need for robust hardware integration, real-time validation, and user-centered design in future AAD applications. Our findings add a critical step towards practical brain-computer interface solutions for hearing support in everyday environments and highlight the importance of interpretable methods in application-driven research.