Depression Severity Detection Using Read Speech with a Divide-and-Conquer Approach.

Area of Science:

• Computational linguistics
• Clinical psychology
• Machine learning

Background:

• Depression severity assessment is crucial for effective treatment.
• Objective measures for depression are needed to complement subjective reporting.
• Speech analysis offers a non-invasive method for assessing psychological states.

Purpose of the Study:

• To develop and evaluate a novel speech-based method for detecting depression severity.
• To investigate the efficacy of combining acoustic, prosodic, and language features for depression classification.
• To compare the performance of the speech-based model against patient self-reporting.

Main Methods:

• A divide-and-conquer strategy was employed to process speech features.
• Speech features were categorized into acoustic, prosodic, and language attributes.
• A fully connected deep neural network was utilized for feature fusion and classification.
• Experiments were conducted on 76 clinically depressed patients (38 severe, 38 moderate on MADRS).

Main Results:

• The proposed speech analysis method achieved 78% accuracy in detecting depression severity.
• Patient self-reporting scores demonstrated 79% accuracy in classifying their own depression status.
• The study highlights the potential of integrating diverse speech features for robust depression detection.

Conclusions:

• Speech analysis, particularly when combining acoustic, prosodic, and language features, is a viable tool for assessing depression severity.
• The developed deep neural network model shows promise as an objective measure for depression.
• Further research can explore larger datasets and diverse patient populations to refine this approach.