Development and validation of a machine learning-based vocal predictive model for major depressive disorder.

Area of Science:

• Computational psychiatry
• Speech analysis
• Machine learning in mental health

Background:

• Speech intonation variations correlate with mental state changes.
• Behavioral vocal analysis offers objective biomarkers for psychiatric assessment.
• Remote assessment tools are crucial, especially for conditions like major depressive disorder (MDD), and were highlighted during the COVID-19 pandemic.

Purpose of the Study:

• To design and validate an automated speech analysis prototype for classifying speech features related to MDD.
• To utilize a remote assessment system combining a mobile app for speech recording and cloud processing for prosodic vocal patterns.

Main Methods:

• Machine learning algorithms were employed to compare vocal patterns.
• Vocal data from 40 MDD patients were analyzed against 104 non-clinical participants.
• Vocal patterns of MDD patients in acute and remission phases were compared.

Main Results:

• A vocal depression predictive model was successfully developed.
• MDD patients exhibited significantly higher vocal depression scores than controls (p < 0.0001).
• Vocal depression scores were significantly higher in the acute phase of MDD compared to remission (p < 0.02).

Conclusions:

• Computerized analysis of prosodic speech changes can serve as biomarkers for early MDD detection.
• This technology supports remote monitoring of patients with major depressive disorder.
• Automated speech analysis aids in evaluating treatment response in MDD.