VOCAL BIOMARKERS OF CLINICAL DEPRESSION: WORKING TOWARDS AN INTEGRATED MODEL OF DEPRESSION AND SPEECH

Speech output has long been considered a sensitive marker of a person’s mental state. It has been previously examined as a possible biomarker for diagnosis and treatment response for certain mental health conditions, including clinical depression. To date, it has been difficult to draw robust conclusions from past results due to diversity in samples, speech material, investigated parameters, and analytical methods. Within this exploratory study of speech in clinically depressed individuals, articulatory and phonatory behaviours are examined in relation to psychomotor symptom profiles and overall symptom severity. A systematic review provided context from the existing body of knowledge on the effects of depression on speech, and provided context for experimental setup within this body of work. Examinations of vowel space, monophthong, and diphthong productions as well as a multivariate acoustic analysis of other speech parameters (e.g., F0 range, perturbation measures, composite measures, etc.) are undertaken with the goal of creating a working model of the effects of depression on speech. Initial results demonstrate that overall vowel space area was not different between depressed and healthy speakers, but on closer inspection, this was due to more specific deficits seen in depressed patients along the first formant (F1) axis. Speakers with depression were more likely to produce centralised vowels along F1, as compared to F2—and this was more pronounced for low-front vowels, which are more complex given the degree of tongue-jaw coupling required for production. This pattern was seen in both monophthong and diphthong productions. Other articulatory and phonatory measures were inspected in a factor analysis as well, suggesting additional vocal biomarkers for consideration in diagnosis and treatment assessment of depression—including aperiodicity measures (e.g., higher shimmer and jitter), changes in spectral slope and tilt, and additive noise measures such as increased harmonics-to-noise ratio. Intonation was also affected by diagnostic status, but only for specific speech tasks. These results suggest that laryngeal and articulatory control is reduced by depression. Findings support the clinical utility of combining Ellgring and Scherer’s (1996) psychomotor retardation and social-emotional hypotheses to explain the effects of depression on speech, which suggest observed changes are due to a combination of cognitive, psycho-physiological and motoric mechanisms. Ultimately, depressive speech is able to be modelled along a continuum of hypo- to hyper-speech, where depressed individuals are able to assess communicative situations, assess speech requirements, and then engage in the minimum amount of motoric output necessary to convey their message. As speakers fluctuate with depressive symptoms throughout the course of their disorder, they move along the hypo-hyper-speech continuum and their speech is impacted accordingly. Recommendations for future clinical investigations of the effects of depression on speech are also presented, including suggestions for recording and reporting standards. Results contribute towards cross-disciplinary research into speech analysis between the fields of psychiatry, computer science, and speech science