May 10, 2019
2 min read

Machine learning analysis of speech detects anxiety, depression in children

You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact

Photo of Ellen McGinnis
Ellen McGinnis

Using a machine learning algorithm to analyze audio data, researchers could detect signs of anxiety and depression in the speech patterns of young children, according to research published in Journal of Biomedical and Health Informatics.

‘There are high individual and societal burdens associated with internalizing disorders that highlight the need for effective early assessment,” Ellen W. McGinnis, PhD candidate from the department of child psychiatry, University of Vermont, and colleagues wrote. “New tools that can feasibly and objectively screen children for these internalizing disorders during routine pediatric well-visits would support surrounding adults in understanding the intensity and chronicity of their child’s distress, and connect them with interventions early in development.”

The researchers evaluated a novel approach for detecting internalizing disorders in children aged 3 to 7 years using a 3-minute speech task. During the task, 71 children improvised a 3-minute story that a researcher would judge — giving only neutral or negative feedback — based on how interesting it was, according to a press release. Then, McGinnis and colleagues used a machine learning algorithm to analyze statistical features of the audio recordings. The children were also diagnosed using a structured clinical interview and parent questionnaire.

After performing machine learning analysis of audio data from the task, McGinnis and colleagues could identify children with an internalizing disorder with 80% accuracy (54% sensitivity and 93% specificity). Specifically, they found that affected children had “especially low-pitch voices with repeatable speech inflections and content, but high-pitched response to surprising stimuli,” when compared with children without an internalized disorder.

“A low-pitched voice and repeatable speech elements mirrors what we think about when we think about depression: speaking in a monotone voice, repeating what you're saying,” McGinnis said in the press release.

The results also showed that the machine learning approach outperformed clinical thresholds on parent-reported child symptoms, which detected children with an internalizing disorder with lower accuracy (67% to 77%), but similar specificity and sensitivity. In addition, the algorithm yielded a diagnosis in only a few seconds after the child completed the task, according to the release.

“Overall, this paper describes a methodology requiring very limited computational resources ... which points toward future deployment of this technique for identifying young children with probable internalizing diagnoses using resource-constrained but ubiquitous devices like mobile phones,” McGinnis and colleagues wrote in the full study. “While these results can likely be improved and extended, and should be replicated, this is an important first step in connecting often overlooked children to the help they need to both mitigate their current distress and prevent subsequent comorbid emotional disorders and additional negative sequelae” – by Savannah Demko

Disclosure: Healio Psychiatry was unable to confirm any relevant financial disclosures at the time of publication.