Design of a dysarthria classifier using global statistics of speech features

Dysarthria is a neurological disorder in which the speech production system is impaired. There are five main types of dysarthrias depending on the location of the lesion in the nervous system. There is evidence suggesting a relationship between the location of the lesion and the resulting speech characteristics. This paper describes a non-intrusive classifier to identify the dysarthria type in a person using global statistics, e.g., mean, variance, etc., of speech features. A tree-based classifier was developed using multiple low-level maximum likelihood classifiers as inputs. An error of 10.5% was achieved in the classification of three types of dysarthrias.