This paper presents a new dataset, B-SHARP, which can be used to detect Mild Cognitive Impairment (MCI), an early stage of Alzheimer's Disease. The dataset contains 721 speech recordings from 144 MCI patients and 185 health controls, on three topics about daily activity, room environment and picture description. Given the B-SHARP dataset, several hierarchical transformer models on the text side based on the transcription and multiple speech models with different encoding methods based on acoustic information are developed. And finally, the model performance are evaluated and a comparison is drawn between text models and speech models.
Computer Science / Emory University
BS / Spring 2021
Jinho D. Choi, Emory University (Chair)
Yuanzhe Xi, Emory University
Davide Fossati, Emory University