A showcase of ÉTS researchers’ publications and other contributions

Detecting depression in Alzheimer’s disease and MCI by speech analysis


Downloads per month over past year

Abdallah-Qasaimeh, Bashar et Ratté, Sylvie. 2021. « Detecting depression in Alzheimer’s disease and MCI by speech analysis ». Journal of Theoretical and Applied Information Technology, vol. 99, nº 5. pp. 1162-1171.

[thumbnail of Ratté-S-22661.pdf]
Ratté-S-22661.pdf - Published Version
Use licence: Creative Commons CC BY-NC-ND.

Download (300kB) | Preview


It is estimated that 30% of Alzheimer’s patients suffer from depression. Since this condition can lead to further cognitive decline and suffering, its detection is essential to alleviate MCI (mild cognitive impairment) or AD symptoms in patients. This paper presents a machine learning method aimed at identifying MCI and Alzheimer’s disease (AD) patients suffering from depression, using different features extracted from their speech. 276 participants (mean age 70.9 years) are selected from DementiaBank’s Pitt Corpus for this research. The interviewer’s voice and the silences are removed from the audio records as a preprocessing task. Several audio features are extracted from the patient's speech to achieve this task. For instance, MFCC’s, Spectral Centroid, Spectral Roll-Off Point, and others. We trained and compared three families of classifiers (SVM, Random Tree, and Random Forest) through two experiments, one using Spectral feature variants and MFCC, and the other using only MFCC features. A third experiment is conducted for comparison with the literature review. In all cases, we used a bootstrapping method to solve the sampling bias, i.e., 70% of the patients not suffering from depression. From the results, the MFCC feature set was more appropriate for tree-based classifiers than SVM, in which the Random Tree classifier reported the highest classification performance (91.3%). Meanwhile, the other feature sets were more appropriate for SVM than the tree-based classifiers, where SVM reported an 89.1 classification accuracy, with 91.1% recall.

Item Type: Peer reviewed article published in a journal
Ratté, Sylvie
Affiliation: Génie logiciel et des technologies de l'information
Date Deposited: 20 May 2021 20:44
Last Modified: 09 Dec 2021 21:03

Actions (login required)

View Item View Item