Artificial intelligence: using voice analysis to detect depression
Artificial intelligence allows medicine to advance by leaps and bounds, both in detection and treatment. The University of Vermont has conducted experiments on the use of AI in the detection of anxiety and depression disorders, and the results are very encouraging.
The AI needs to listen to a child talk for at least 3 minutes to determine if a child has depression. By way of comparison, a specialized doctor needs a 60 to 90 minute interview to get an idea of the issue.
How does the AI do it? When artificial intelligence studies 8 elements in the child’s speech, three points are especially important to create a diagnosis: a generally low voice, a voice that becomes high-pitched when the child is surprised and the repetition of certain words or sounds. Ellen McGinnis, head of the study, explains that “[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.”
Thus, a group of 71 children aged 3 to 8 years took the test. They had to improvise a three-minute story, the researchers made them believe that only the quality of their story would be judged, while maintaining stressful conditions. A buzzer rang from time to time to surprise the candidates, allowing the AI to detect certain things in the speech.
The success rate is 80% and, over time, should continue to improve. The next major step is, of course, to make this technology available to healthcare providers, for example in the form of an app on smartphones. This will make it easier to identify risks and possibly see problems that parents do not see.
It is important to remember that in medicine the purpose of AI is not really to replace physicians but rather to help them make their diagnoses. The objective is not to let it make the decisions; other diagnostic tests are carried out in parallel (also communicating with the parents is important).