Characterization of atypical vocal source excitation, temporal dynamics and prosody for objective measurement of dysarthric word intelligibility

Objective measurement of dysarthric speech intelligibility can assist clinicians in the diagnosis of speech disorder severity as well as in the evaluation of dysarthria treatments. In this paper, several objective measures are proposed and tested as correlates of subjective intelligibility. More specifically, the kurtosis of the linear prediction residual is proposed as a measure of vocal source excitation oddity. Additionally, temporal perturbations resultant from imprecise articulation and atypical speech rates are characterized by short- and long-term temporal dynamics measures, which in turn, are based on log-energy dynamics and on an auditory-inspired modulation spectral signal representation, respectively. Motivated by recent insights in the communication disorders literature, a composite measure is developed based on linearly combining a salient subset of the proposed measures with conventional prosodic parameters. Experiments with the publicly-available 'Universal Access' database of spastic dysarthric speech (10 patient speakers; 300 words spoken in isolation, per speaker) show that the proposed composite measure can achieve correlation with subjective intelligibility ratings as high as 0.97; thus the measure can serve as an accurate indicator of dysarthric speech intelligibility. (C) 2011 Elsevier B.V. All rights reserved.