Analysis of phonetic markedness and gestural effort measures for acoustic speech-based depression classification

While acoustic-based links between clinical depression and abnormal speech have been established, there is still however little knowledge regarding what kinds of phonological content is most impacted. Moreover, for automatic speech-based depression classification and depression assessment elicitation protocols, even less is understood as to what phonemes or phoneme transitions provide the best analysis. In this paper we analyze articulatory measures to gain further insight into how articulation is affected by depression. In our investigative experiments, by partitioning acoustic speech data based on lower to high densities of specific phonetic markedness and gestural effort, we demonstrate improvements in depressed/non-depressed classification accuracy and F1 scores.