Voice-based Grading of Psychiatric Disorders

It is well recognized that many psychiatric disorders and illnesses present with acoustic markers that manifest in the speech of the affected people. Studies continue to uncover positive relationships between the human voice and a widening range of mental conditions. There is not much literature, however, on the use of voice in the grading of psychiatric illnesses. Here the term grading refers to automatically estimating the degree of an illness measured in terms of the severity of multiple associated symptoms to gauge a patient’s condition. This is currently done manually by trained clinicians based on human observations within doctor-patient sessions. In this paper, we address the problem of psychiatric grading of patients who are under treatment for active mental illnesses, entirely through their voice. Using acoustic data from clinician interviews with sixteen patients recorded within a hospital setting, we use nonparametric models to learn the relations between manually assigned ratings and the articulatory-phonetic units of speech. In addition to discovering interesting relations that point to the physical effects of mental illnesses on specific articulators, our studies highlight the merits of using this approach in the development of voice-based automatic grading mechanisms for many symptoms associated with psychiatric illnesses.