A review of hundreds of articles found that it was possible to diagnose conditions such as depression by analysing patients’ speech
“Models that bring together multiple speech features can distinguish speakers with psychiatric disorders from healthy controls with high accuracy." Dr. Rudolf Uher, professor of psychiatry, Dalhousie University, and lead researcher
Automated analysis of speech patterns can be used to aid diagnosis of psychiatric conditions, a review of research has found.
The review, published in the Harvard Review of Psychiatry, looked at literature on the use of speech pattern analysis to manage psychiatric disorders and identified four key areas of application: diagnostic classification; severity assessment; onset prediction; and prognosis and treatment outcomes.
Researchers reviewed hundreds of articles that discussed aspects of patients’ speech. Most studies investigating the use of speech analysis in diagnosis looked at patients with major depression, whose speech is often slow, full of pauses, negative in content, and lacking energy. In these cases, diagnostic accuracy was high, reaching over 80% in one study. “Models that bring together multiple speech features can distinguish speakers with psychiatric disorders from healthy controls with high accuracy,” the paper says.
The study, which was carried out by researchers in Canada, also found that automated analysis is effective in predicting the onset of mental illness, particularly in high-risk populations. Multiple studies that looked at speech semantics, including coherence and complexity, predicted the onset of psychosis in two to two-and-a-half years with accuracy as high as 100% in some cases.
In particular, the use of speech pattern analysis in assessing suicide risk appears to have great potential. One study cited in the review showed that measuring variables such as erratic frequency, hesitations, and jitters identified patients with suicide ideation against healthy patients 73% of the time.
There are, however, numerous factors, that can cause variance in speech patterns, making it more difficult to incorporate speech into objective illness and outcome assessment. These include the effect of medication and demographic variables such as sex and language.
The authors see potential for further investigation into how speech analysis can be used to aid diagnosis: “Convergent progress in speech research and computer sciences opens avenues for implementing speech analysis to enhance objectivity of assessment in clinical practice.”
They recommend that future research should take into account illness states across time, as most of the studies examined in the review looked at currently ill patients rather than at whether similar patterns endure long-term between symptoms. They also emphasise the importance, when creating software algorithms, of avoiding replicating existing bias: “Application of speech analysis will need to address issues of ethics and equity, including the potential to perpetuate discriminatory bias through models that learn from clinical assessment data.”
This review of research shows the potential offered by artificial intelligence in improving diagnostic accuracy for mental health conditions such as depression. While such techniques should never be used as a replacement for traditional diagnostic techniques, they could provide useful as additional tools to aid mental health professionals.