A deep tensor-based approach for automatic depression recognition from speech utterances.

Area of Science:

• Psychiatry and Mental Health
• Computational Linguistics
• Digital Health

Background:

• Depression is a major global health burden with diagnostic challenges due to subjective symptom presentation and episodic assessment.
• Current diagnostic methods rely on subjective interviews and manual scoring, lacking continuous monitoring capabilities.
• The rise of wearable technology presents an opportunity for passive, objective mental health assessment.

Purpose of the Study:

• To develop and validate a novel framework for automated depression classification using speech patterns.
• To leverage passive monitoring via wearable devices for real-time depression trait detection.
• To improve diagnostic accuracy and enable continuous monitoring of depressive disorders.

Main Methods:

• A tensor-based approach was applied to the DAIC-WOZ depression dataset for automated depression classification.
• Discriminative features were extracted from speech patterns to identify depression.
• The framework was designed for simplified implementation and onboard deployment on wearable devices.

Main Results:

• The proposed tensor-based method achieved high f1 score and accuracy in depression recognition.
• The approach demonstrated a substantially simpler implementation architecture compared to existing methods.
• The algorithm requires significantly less computational load, facilitating efficient onboard deployment.

Conclusions:

• Automated depression classification using speech patterns is feasible and effective.
• Tensor-based feature extraction offers a promising avenue for objective, continuous mental health monitoring.
• Onboard deployment on wearables can enhance diagnostic accuracy and real-time monitoring of depression.