2019-07 | journal-article
Speech is a vehicular tool to detect neurological degeneration using certain accepted biomarkers derived from sustained vowels, diadochokinetic exercises or running speech. Classically the Vowel Space Area (VSA) and the Formant Centralization Ratio (FCR) have been proposed to describe dysarthria in Parkinson’s Disease (PD). These features are based in global estimations of the positions of the first two formants in the representation of a vowel triangle. The aim of the paper is to give a description of speech articulation dynamics as a probability density function of the kinematic features derived from the evolution of formants in the time domain. The statistical distribution of the dynamic behavior of articulation features can be used to estimate differences between speech features from subjects with Parkinson’s dysarthria relative to normative subjects. Utterances of vowels [a:, i:, u:] from a subset of 16 subjects with PD (8 males and 8 females), confronted to a subset of 16 normative subjects (8 males and 8 females) have shown that the statistical distributions of dynamic articulation features can be differentiated using information theory based estimations such as Kullback-Leibler and Jensen-Shannon Divergence (JSD). These estimations allow establishing relevant statistical differences between PD and normative subjects both for males and females, improving the differentiation capability of VSA and FCR.