Universal Access (UA) Speech Corpus Development

Universal Access Automatic Speech Recognition Project for Talkers with Dysarthria.

The goal was to develop automatic speech recognition (ASR) systems for dysarthric speech, a neuromotor speech disorder. Our initial focus was on cerebral palsy (CP)-associated spastic dysarthria given its common etiology: i.e., severe congenital dysarthria is frequently caused by CP, and spastic dysarthria is the most common type of dysarthria. The key problem that motivated our work was that ASR development has provided a useful human-computer interface especially for people who have difficulties in typing with a keyboard, but individuals with a neuromotor disorder—such as cerebral palsy, Parkinson’s disease, amyotrophic lateral sclerosis, and traumatic brain injury—have not been able to utilize the benefit of these advances, mainly because their symptoms include dysarthria.

A major challenge in developing automatic speech recognition systems for dysarthric speech is the need for a great amount of training data. With a team of collaborators, I have built the Universal Access (UA) Speech Corpus of dysarthric speech, which has been distributed to research institutions worldwide from 2008 to the present. The UA speech corpus has been a vital resource for developing assistive technologies for people with motor disabilities, as shown by the publication of hundreds of papers by numerous research teams that utilize the corpus. Speech recognition error rates using the UA Speech Corpus have reduced to almost 1/3 of the prior error rates since the first release of the corpus.

My collaborators included Dr. Mark Hasegawa Johnson (Electrical & Computer Engineering), Dr. Adrianne Perlman (Speech and Hearing Science), and Dr. Jon Gunderson (Coordinator of Assistive Communication and Information Technology Accessibility in the Division of Disability Resources and Education Services).