The following is the summary of “Detecting subtle signs of depression with automated speech analysis in a non-clinical sample” published in the December 2022 issue of Psychiatry by König, et al.
The use of automated speech analysis to aid in the diagnosis of depression is gaining popularity. However, most previous research only compared the speech of individuals with major depressive disorder and that of healthy volunteers. One possible approach is to look for early and sensitive signs of depression risk by correlating speech with depressive symptoms in a non-clinical sample. Researchers surveyed (n=118) healthy young adults (mean age: 23.5 ± 3.7 years; 77% women) and had them reflect on a recent pleasant and bad experience. Then, they used a self-report questionnaire that assigned a score between 0 and 60 to the severity of the aforementioned depressed symptoms.
To extract acoustic and linguistic elements from speech data, we transcribed it. Then, investigators compared speech characteristics between people with and without clinically significant depressive symptoms. They then used both the cutoff and the individuals’ depression questionnaire scores to predict whether they would be below or over the threshold. In light of the fact that depression is linked to slowing of cognitive processes and attention problems, they compared depression scores with Trail Making Test results. A total of (n=93) participants scored below the threshold for clinically significant depression symptoms, whereas (n=25) participants scored over the threshold. Individuals over the cut-off spoke more than those below it, both in the positive and negative stories, but this difference was not significant for most aspects of speech. Slower Trail Making Test scores were also associated with greater depression scores in that group.
Who would go below or over the threshold could be predicted with 93% precision. Furthermore, they used a support vector machine to predict the individuals’ depression scores with a low mean absolute error (3.90). Their data show a correlation between linguistic shifts and increased depression scores, even in a population where the disorder was not formally diagnosed. A deeper dive into this is needed. Whether the speech traits identified in their study serve as early and sensitive markers for eventual depression in those at risk might be evaluated in a longitudinal investigation.