Abstract
Ten speakers with cerebral palsy and dysarthria and 13 nondisabled speakers were tested using a Shadow VET 2 speech recognition system installed in an Apple Ile computer. The speakers each produced four types of stimuli: 12 consonants followed by a neutral vowel, 12 vowels in an h-d environment, 12 words defined as easy and 12 words defined as difficult for dysarthric speakers. Using all four types of stimuli, the dysarthric speakers had scores above chance, but they had significantly fewer stimuli recognized by the computer than the nondisabled speakers. For the dysarthric group, vowels in an h-d environment and both types of words were recognized significantly more often than the consonants followed by a neutral vowel. For the nondisabled group, the vowels in an h-d environment were recognized significantly more often than consonants followed by a neutral vowel, and both types of words were recognized more often than vowels. Confusion matrices and a discussion of item errors were used to further compare the nondisabled and dysarthric speakers. Suggestions were made for improving computer recognition of dysarthric speech.