The proposed research project supports the investigation and development of technology to provide better access to speech recognition-based systems for individuals with speech impairments due to dysarthria or other medical condition. With current automatic speech recognition software, typical error rates are unnacceptably high for speakers whose speech differs significantly from the "standard" language. In the U.S. alone, more than 500,000 U.S. residents currently suffer from dysarthria. Many of these dysarthric speakers are commonly diagnosed with "cattle car" diagnosis codes and procedures that incorporate little to no advanced technology. Inasmuch as these individuals are unable to use commonly available speech-based tools due to their speech impairment, they are denied access to information, services available to "standard speakers." Moreover, these impaired speakers are unable to take advantage of advanced technology for improved therapy or self-practice. In essence, the design of current speech technologies excludes the group of people that would benefit most by using speech-based applications. Phase I of the proposed project will demonstrate the feasibility of improving speech recognition results for many individuals with dysarthria by analyzing the prosodic features that characterize their speech and using the results to enhance the performance of the speech recognition engine. The technology developed by this research project will open access to speech-based applications for many impaired speakers. In addition to the potential of improved diagnosis, this technology will be more accessible for under-served populations and can quickly become economically feasible, given advances in the improvement of the recognizer. The proposed technology will most importantly enable a new generation of user-friendly training and assistive devices for dysarthric and other medically impaired speakers and holds great potential for the speech therapy practice
