Potential of Prosodic Features to Estimate Degree of Parkinson's Disease Severity

Abstract

This paper deals with non-invasive and objective Parkinson’s disease (PD) severity estimation. For this purpose, prosodic speech features expressing monopitch, monoloudness, and speech rate abnormalities were extracted from recordings of stress-modified reading task acquired from 72 patients with idiopathic PD. Using a single feature regression (esimating values of subjective clinical rating scales) with classification and regression algorithm, following performance in terms of root mean squared error was achieved: 10.72 (UPDRS III), 2.16 (UPDRS IV), 4.76 (FOG-Q), 17.89 (NMSS), 2.13 (RBDSQ), 6.43 (ACE-R), 1.41 (MMSE), and 4.82 (BDI). These results show a promising potential of prosodic speech features in the field of objective assessment of PD severity.