Assessment of articulatory sub-systems of dysarthric speech using an isolated-style phoneme recognition system

In this work, the variation in the acoustic realizations of phonemes between normal and dysarthric speech is utilized for the assessment of articulatory sub-systems of dysarthric speech. Using a speech recognition system for the assessment of dysarthric speech is well established either by considering the continuous speech as a sequence of phonemes, or by considering isolated words. These systems will provide information about the intelligibility alone. However, the problems associated with the subsystems of speech can be well captured when a dysarthric speaker is asked to speak continuously and the resultant speech is analyzed at phoneme-level in isolation. Considering this aspect, in our work, an isolated-style triphone-based phoneme recognition system is developed for analyzing continuous speech. Acoustic variations of the phonemes based on place of articulation alone provides information associated with the malfunctioning of articulatory sub-system and are correlated with the Frenchay dysarthric assessment (FDA) scores. The correlation error between our system and FDA scores is found to be only 9%.