- A machine learning algorithm was used to analyze children’s speech and identify depression and/or anxiety with 80% accuracy.
- Three features proved to indicate these illnesses: low pitch, repeated speech inflections and subject, and a higher pitch when surprised.
- Kids came up with a three-minute story on the spot, of which was judged after; a buzzer sounded 90 seconds in and again when 30 seconds remained.
- In addition, researchers interviewed the kids and analyzed questionnaires completed by parents, which helped them to diagnose the kids.
- When the researchers analyzed the kids’ speech (during their stories) using the algorithm, it successfully identified depression and anxiety 80% of the time.
- The research team plans to create a screening test, perhaps in the form of a smartphone app, which will help to detect these illnesses.
A new study “Giving Voice to Vulnerable Children: Machine Learning Analysis of Speech Detects Anxiety and Depression in Early Childhood,” uses a machine learning algorithm to analyze recordings of children’s speech and identify depression and anxiety with an 80% accuracy rate. This algorithm found several features to be indicative of these disorders: a lower pitch, repeated speech inflections, and a higher pitch in response to being surprised. Researchers plan to use this technology to create a screening test, perhaps a smartphone app, which can help to detect depression and anxiety in children.
Around 20% of kids suffer from anxiety and depression. However, children under eight years old aren’t always able to communicate their feelings and emotional needs—which means adults need to pay particularly close attention to young children, specifically their mental and emotional health. The researchers of this study hoped to invent a fast, objective test for detecting when kids are suffering, so they can get the help they need as early as possible.
The research team tweaked a mood induction task, the Trier-Social Stress Task, which is designed to induce feelings of stress and anxiety in the participant. They asked 71 children between the ages of three and eight to come up with a short three-minute story, of which would be judged. The judge (one of the researchers) maintained a stern demeanor throughout and gave neutral as well as negative feedback. Additionally, a buzzer sounded at 90 seconds and again when there were only 30 seconds left to alert the participant of the time.
Researchers also interviewed the kids and analyzed questionnaires completed by the parents, which helped them to diagnose the kids with depression and/or anxiety. Next, the researchers put the machine learning algorithm to the test. It analyzed audio recordings of each kid’s story and then delivered a prognosis.
The algorithm successfully diagnosed children 80% of the time. Furthermore, it identified three audio features to be indicative of depression and/or anxiety: low-pitched voices, repeated speech inflections and subject, and a higher pitch in response to a surprise.
The researchers plan to advance the speech analysis algorithm and use it to create a screening tool for depression and anxiety. This screening tool could equal a smartphone app that analyzes one’s speech and provides feedback on the spot.
McGinnis, E. W., Anderau, S. P., Hruschak, J., et al. (2019, April 26). Giving voice to vulnerable children: machine learning analysis of speech detects anxiety and depression in early childhood. IEEE Journal of Biomedical and Health Informatics. Retrieved from https://ieeexplore.ieee.org/document/8700173