Using Speech Analysis to Detect Mental Health Issues
Author Information
Author(s): Eysenbach Gunther, Coristine Andrew, Yang Lulu, Martin Vincent, Riad Rachid, Denais Martin, de Gennes Marc, Lesage Adrien, Oustric Vincent, Cao Xuan Nga, Mouchabac Stéphane, Bourla Alexis
Primary Institution: Callyope, Paris, France
Hypothesis
Can mobile-collected speech data effectively detect and estimate depression, anxiety, fatigue, and insomnia in the general population?
Conclusion
The study shows that speech-based systems can effectively assess various mental health symptoms and their severity in the general population.
Supporting Evidence
- The best model achieved an AUC of 0.78 for detecting depression using the Beck Depression Inventory.
- Fairness analysis showed consistent performance across sex but less so for age and education level.
- Individual symptom scores were accurately predicted with significant correlations.
Takeaway
This study found that analyzing how people speak can help doctors figure out if someone is feeling depressed, anxious, or tired, just by listening to their voice.
Methodology
The study involved 865 healthy adults who completed self-assessment scales and recorded speech tasks, which were analyzed using machine learning models.
Potential Biases
The model's performance varied across demographic groups, particularly by age, indicating potential bias.
Limitations
The study's dataset is limited to French speakers and lacks longitudinal data, which may affect generalizability.
Participant Demographics
Participants were healthy adults from the general population, with a balanced representation of sex and varying education levels.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website