Analysis of Voice Biomarkers for the Detection of Cognitive Impairment

Abstract

The objective of this work is to determine whether speech obtained from interactions with a smart speaker can be used to predict the level of cognitive impairment (CI). We use a voice assistant to administer a cognitive test in Spanish, and we record the conversations in order to extract features that could potentially be used as voice biomarkers. A total of 21 participants (14 patients and 7 healthy controls) between the ages of 68 and 86 are included in the study (15 were women). Using just speech we are able to perform a regression with machine learning models, in order to predict the Global Deterioration Scale (GDS) of cognitive functions. Then, we measure the performance of the estimations with standard metrics - an $R^{2}$ of 0.74 was obtained in the best case using Support Vector Machine (SVM) algorithms. Despite needing a bigger sample of participants in future studies, this is a positive and promising result for such a non-intrusive procedure, which could potentially be used as a screening tool for automatic cognitive impairment assessment.

Keywords

• Biomarkers
• cognitive impairment
• dementia
• regression
• voice