Utilizing Large Language Models for Optimizing Sound Selection in Dementia Care Environments

The effect of soundscape on people with dementia has been studied, with established links between acoustic environments and Behavioral and Psychological Symptoms of Dementia.Previous research has demonstrated promising results in using soundscape augmentation for dementia care interventions.However, selecting appropriate sound segments remains challenging due to the heterogeneous nature of dementia and varying individual needs.This study explores the innovative application of Large Language Models (LLMs) in selecting appropriate sound segments for soundscape augmentation.By analyzing responses from Claude 3.5 and GPT-4 to systematically designed prompts, the potential to recommend suitable sound segments based on the specific auditory deficits associated with different types of dementia was investigated.The LLMs were provided with semantic complexity ratings and affective information of pre-labelled sound segments and then tasked with matching appropriate sounds to various dementia types.Results demonstrate LLMs' capability to consider multiple factors, including semantic complexity, emotional impact, and specific auditory processing challenges when making recommendations.Key findings indicate that LLMs can effectively differentiate between suitable sound segments for various dementia types.This research suggests potential benefits in using AI-assisted sound selection to enhance personalized soundscape design in dementia care while acknowledging the importance of human oversight and individual patient preferences.