Detecting Depression Using an Ensemble Logistic Regression Model Based on Multiple Speech Features.

Haihua Jiang,Gang Wang,Bin Hu,Huanyu Kang,Zhenyu Liu,Lan Zhang,Xiaoyu Li

doi:10.1155/2018/6508319

Abstract

Early intervention for depression is very important to ease the disease burden, but current diagnostic methods are still limited. This study investigated automatic depressed speech classification in a sample of 170 native Chinese subjects (85 healthy controls and 85 depressed patients). The classification performances of prosodic, spectral, and glottal speech features were analyzed in recognition of depression. We proposed an ensemble logistic regression model for detecting depression (ELRDD) in speech. The logistic regression, which was superior in recognition of depression, was selected as the base classifier. This ensemble model extracted many speech features from different aspects and ensured diversity of the base classifier. ELRDD provided better classification results than the other compared classifiers. A technique for identifying depression based on ELRDD, ELRDD-E, was here suggested and tested. It offered encouraging outcomes, revealing a high accuracy level of 75.00% for females and 81.82% for males, as well as an advantageous sensitivity/specificity ratio of 79.25%/70.59% for females and 78.13%/85.29% for males.

Highlights

Worldwide, over 300 million people of different ages have clinical depression [1]. e rise in the prevalence of this disease has been connected to a group of important outcomes [2]
analysis of variance (ANOVA) and least significant difference (LSD) tests were carried out on the four speech feature subspaces (MFCC, PROS, SPEC, and GLOT) over the ten-fold cross validation outcomes utilizing support vector machine (SVM), Gaussian mixture model (GMM), and logistic regression (LR) classifiers. e accuracy, sensitivity, and specificity significantly varied between the four feature subspaces (p < 0.05). e accuracy and sensitivity of GLOT were worse in comparison to melfrequency cepstrum coefficients (MFCC), PROS, and SPEC (p < 0.05), and the accuracy and specificity of SPEC and PROS were greater than MFCC (p < 0.05)
It was noted that utilizing SPEC, PROS, and MFCC features offered significantly better classification outcomes for females compared to utilizing GLOT features

Summary

Introduction

Over 300 million people of different ages have clinical depression [1]. e rise in the prevalence of this disease has been connected to a group of important outcomes [2]. To halt the onset of clinical depression, advance intervention can offer a pivotal action to ease the burden of the disease. Current depression diagnosis methods rely on self-report of patient and clinical opinion [4], which risk several subjective biases. Erefore, a convenient and objective method for detecting depression is of primary importance. E acoustic qualities of speech can be affected by the emotional state of a person with depression [6]. Erefore, depression can be detected by analyzing changes in the acoustical characteristics of speech. To improve the effect of classification, many features were extracted in early studies. It is still unclear which acoustic features are most effective for detecting depression especially in Mandarin speech. An objective method based on speech is still in need

Methods

Results

Discussion

Conclusion