Automated Speech Analysis for Risk Detection of Depression, Anxiety, Insomnia, and Fatigue: Algorithm Development and...

Area of Science:

• Speech analysis
• Machine learning
• Mental health assessment

Background:

• Current speech analysis for mental health often focuses on single symptoms, neglecting symptom interactions.
• Predictive models lack assessment of limitations like uncertainty and fairness, hindering clinical deployment.

Purpose of the Study:

• Investigate the potential of mobile-collected speech data for detecting and estimating depression, anxiety, fatigue, and insomnia.
• Focus on factors beyond mere accuracy, including uncertainty and fairness.

Main Methods:

• Utilized data from 865 healthy adults, including self-reported mental and sleep states via questionnaires.
• Developed a novel speech and machine learning pipeline with pretrained deep learning models.
• Evaluated clinical threshold detection, individual score prediction, model uncertainty, and demographic fairness.

Main Results:

• The Whisper M model demonstrated good detection performance for depression, anxiety, insomnia, and fatigue (AUCs ranging from 0.68 to 0.78).
• The system accurately predicted individual symptom scores and performed well in abstaining from uncertain predictions.
• Fairness analysis showed consistency across sex, with moderate fairness for education level and lower fairness for age groups.

Conclusions:

• Speech-based systems show potential for multifaceted mental health assessment, including severity estimation.
• Incorporating fairness and uncertainty estimation is crucial for enhancing clinical utility and responsible implementation.