Publication Date

2024

Document Type

Dissertation/Thesis

First Advisor

Fonseca, Benedito

Degree Name

M.S. (Master of Science)

Legacy Department

Department of Electrical Engineering

Abstract

This paper investigates the performance of the eGeMAPS (extended Geneva Minimalistic Acoustic Parameter Set) feature set in detecting depression from audio samples using subsets of the E-DAIC (Extended Distress Analysis Interview Corpus) database. With depression affecting a significant portion of the U.S. adult population, efficient detection methods are critical for timely diagnosis and treatment. Various classifiers in the WEKA machine learning toolbox are used to evaluate the performance of eGeMAPS features in distinguishing between depressed and nondepressed (D&ND) individuals. Our methodology involves creating balanced subsets of E-DAIC, extracting eGeMAPS features using openSMILE, and testing different machine learning models. This study performs three experiments to answer the following questions: can a machine learning algorithm build a classifier using only eGeMAPS audio features, without considering gender, detect depression in E-DAIC patients?; does the discriminatory power of eGeMAPS improve if a single classifier is built using balanced sets of male and female, D&ND patients?; and can the discriminatory power of eGeMAPS between D&ND E-DAIC patients improve by building two separate classifiers, one exclusively for male patients and another for exclusively female patients? The results indicate that while eGeMAPS features alone struggle to surpass random chance levels in generalization, certain configurations show potential, especially when addressing gender dependencies. The findings highlight the need for further refinement in feature selection and model training to enhance the robustness of depression detection systems using acoustic features.

Extent

70 pages

Language

en

Publisher

Northern Illinois University

Rights Statement

In Copyright

Rights Statement 2

NIU theses are protected by copyright. They may be viewed from Huskie Commons for any purpose, but reproduction or distribution in any format is prohibited without the written permission of the authors.

Media Type

Text

Download

Share

COinS