Feature Extraction Method for Predicting Depression by Frequency Domain Analysis

DOI

10.2991/cmes-15.2015.164How to use a DOI?

Keywords

Depression, Speech Analysis, Random Forest, Frequency Domain

Abstract

In this paper, we propose a feature extraction method, which subdivides feature vectors into three frequency regions of 300-1000Hz, 1000-2000Hz, and 2000-3000Hz. The range and mean of intensity are extracted for each frequency region. By so doing, we can compensate the defect of increasing the intensity value, when a person intentionally increases his or her vocalization. Previous studies extracted the slope and correlation of the glottal flow spectrum from the 300-3000Hz region. But, we extract the slope and correlation from each separated frequency region. The overall experimental results show 92.85% for men, and 92.08% for women. The proposed method enhances the respective classification accuracy by 6.73% for men, and 8.09% for women.

Copyright

© 2015, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Eun-Joo Seo
AU  - Kwang-Seok Hong
PY  - 2015/04
DA  - 2015/04
TI  - Feature Extraction Method for Predicting Depression by Frequency Domain Analysis
BT  - Proceedings of the 2nd International Conference on Civil, Materials and Environmental Sciences
PB  - Atlantis Press
SP  - 600
EP  - 603
SN  - 2352-5401
UR  - https://doi.org/10.2991/cmes-15.2015.164
DO  - 10.2991/cmes-15.2015.164
ID  - Seo2015/04
ER  -