AI-Based Dementia Detection in Lebanese Arabic

This project develops an attention-based AI model for early dementia detection in Lebanese Arabic by combining linguistic and acoustic speech features with direct audio analysis via YAMNet, evaluated using precision-focused metrics suited for imbalanced clinical data.

Abstract

Dementia’s impact on language skills, including prolonged pauses and reduced verbal fluency, has been well-documented, highlighting the potential of speech analysis in detecting various neurological conditions. This pr​​oject proposes to develop an AI model tailored for the Lebanese Arabic dialect, a demographic area largely overlooked by current models predominantly focused on English. We explore attention-based models and integrate linguistic and acoustic features. Additionally, we employ direct audio analysis with YAMNet. Our evaluation strategy is focused on precision, recall, and F1 score, which are tailored to address the imbalanced nature of the dataset.