AAD-LLM innovates by integrating human-like selective attention into large language models for audio processing, creating more perception-aligned responses in complex sound environments.

• Addresses a key limitation in current auditory foundation models by incorporating selective attention
• Uses neural data (intracranial EEG) to model how humans focus on specific speakers
• Introduces a novel Intention-Informed Auditory Scene Understanding framework
• Enhances AI's ability to process complex audio scenes similar to human perception

Medical Impact: This research bridges neuroscience and AI by using clinical brain recordings to improve machine learning models, potentially advancing both neural signal processing techniques and creating more intuitive AI assistive technologies for patients with hearing or attention disorders.

AAD-LLM: Neural Attention-Driven Auditory Scene Understanding