Title: Selective Attention Tracking - Neural Markers in Listeners Across Hearing Conditions Opponent: Professor Edmund C. Lalor, University of Rochester

Abstract: Understanding how the human brain selectively attends to a single speaker in multi-speaker environments is central to both auditory neuroscience and the development of hearing technologies. A promising approach for addressing these challenges, and learning how the brain encodes speech, is auditory attention decoding (AAD), which aims to infer a listener’s attended speech stream from measurements of neural activity, such as electroencephalography (EEG). 

Challenges within the field of AAD includes developing reliable and usable measure for selective attention in wearers. These methods should be both applicable in wearable EEG and robustly decode attention. Additionally, these methods should not bias decisions without necessity. Furthermore, the spectrum of listening scenarios is broad. AAD methods should be applicable to a wide variety of languages, talkers, acoustic listening conditions, and listening tasks such as switching attention between conversations of interest. Solutions should also work for both normal hearing and hearing aid users, as their acoustic experiences differ. 

This thesis develops and evaluates methods for tackling some of these challenges. The work investigates how preprocessing choices affect AAD performance, proposes model-based methods for quantifying relationships between continuous speech signals and concurrent neural recordings. The work also evaluates spatial feature extraction approaches for decoding the directional focus of auditory attention. Furthermore, a novel dataset is introduced, capturing more ecologically valid listening conditions including conversational speech, attention switching, and mobile EEG recordings. The generalization of state-of-the-art AAD methods to these settings is examined. Finally, methods for tracking auditory attention dynamically over time are explored as a step toward real-time applicability. The results highlight both the potential and current challenges of integrating neural attention decoding into future hearing aid technology, with the long-term goal of understanding and enabling selective attention in varied acoustic scenes.