Performance of Automatic Speech Analysis in Detecting Depression: Systematic Review and Meta-Analysis

Area of Science:

• Mental Health
• Artificial Intelligence
• Speech Analysis

Background:

• Depression is prevalent and often underdiagnosed.
• Automatic speech analysis (ASA) offers a potential solution for depression assessment.
• A comprehensive evaluation of ASA's diagnostic accuracy is needed.

Purpose of the Study:

• To systematically review and meta-analyze the diagnostic performance of ASA for depression detection.
• To evaluate both machine learning and deep learning approaches in ASA for depression assessment.

Main Methods:

• Systematic search of 8 databases (Jan 2013-Apr 2025) for English-language studies on ASA for depression.
• Inclusion of studies reporting diagnostic accuracy metrics.
• Quality assessment using a modified QUADAS-R.
• 3-level meta-analysis to estimate pooled accuracy, sensitivity, specificity, and precision.
• Meta-regression and subgroup analyses to explore heterogeneity.

Main Results:

• 105 studies met inclusion criteria from 1345 records.
• Pooled highest accuracy: 0.81 (95% CI 0.79-0.83).
• Pooled highest sensitivity: 0.84 (95% CI 0.81-0.86).
• Pooled highest specificity: 0.83 (95% CI 0.79-0.86).
• Pooled highest precision: 0.81 (95% CI 0.77-0.84).
• Pooled lowest accuracy: 0.66 (95% CI 0.63-0.69).

Conclusions:

• ASA demonstrates potential for depression detection.
• Clinical application as a standalone tool is currently limited.
• ASA is best utilized as a complementary method across various settings.
• Further high-quality research is required for robust and generalizable models.