Dysarthric speech: a comparison of computerized speech recognition and listener intelligibility

The purpose of this study was to identify and compare the recognition of dysarthric speech by a computerized voice recognition (VR) system and non-hearing-impaired adult listeners. Intelligibility "functions" were obtained for six dysarthric speakers who varied in severity and six age- and gender-matched controls. Speakers produced 70-item word lists over 5 sessions. VR using the IBM VoiceType and perceptual judgment scores were obtained and functions plotted by session. Data indicate that computerized recognition of both dysarthric and nonimpaired speech was characterized by initially steep increases in correct recognition with more gradual increases noted during the second through fifth sessions. Perceptual recognition by non-hearing-impaired adults indicates generally stable intelligibility scores over time. Severity of dysarthria did appear to influence recognition of target stimuli. Implications of these data to the application of computerized VR technology are presented.