Enhanced depression detection from speech using Quantum Whale Optimization Algorithm for feature selection.

Area of Science:

• Speech analysis
• Computational psychiatry
• Machine learning for healthcare

Background:

• Accurate and non-intrusive depression detection is crucial.
• Existing methods may be invasive or computationally expensive.
• Speech signals contain valuable biomarkers for mental health assessment.

Purpose of the Study:

• To propose a simple, efficient, and non-invasive unimodal depression detection approach using speech.
• To develop a feature selection method for identifying relevant and non-redundant speech features.
• To evaluate the proposed model's performance against existing techniques.

Main Methods:

• Extraction of spectral, temporal, and spectro-temporal features from speech signals.
• Application of a two-phase Quantum-based Whale Optimization Algorithm (QWOA) for wrapper-based feature selection.
• Comparison with univariate filtering techniques and evolutionary algorithms on the DAIC-WOZ dataset.

Main Results:

• The proposed QWOA-based model significantly outperformed univariate filters and other evolutionary algorithms.
• Achieved superior performance compared to existing unimodal and multimodal depression detection models.
• Demonstrated high accuracy with F1-scores of 0.846 (depressed) and 0.932 (non-depressed), and an error rate of 0.094.
• Selected acoustic features were statistically proven to be non-redundant and discriminatory.

Conclusions:

• The proposed speech-based depression detection model is accurate, efficient, and computationally inexpensive.
• The QWOA feature selection method effectively identifies discriminatory speech biomarkers.
• This approach offers a promising non-invasive tool for automated depression screening.