Spotting the traces of depression in read speech: An approach based on computational paralinguistics and social signal processing

This work investigates the use of a classification approach as a means to identify effective depression markers in read speech, i.e., observable and measurable traces of the pathology in the way people read a predefined text. This is important because the diagnosis of depression is still a challenging problem and reliable markers can, at least to a partial extent, contribute to address it. The experiments have involved 110 individuals and revolve around the tendency of depressed people to read slower and display silences that are both longer and more frequent. The results show that features expected to capture such differences reduce the error rate of a baseline classifier by more than 50% (from 31.8% to 15.5%). This is of particular interest when considering that the new features are less than 10% of the original set (3 out of 32). Furthermore, the results appear to be in line with the findings of neuroscience about brain-level differences between depressed and non-depressed individuals.