Speech generating devices (SGDs) offer people with difficult to understand speech (dysarthria) a synthesized speech replacement for their natural speech. However, people who require SGDs often prefer using their natural speech because it is perceived as normal and is cognitively less demanding for the speaker than other communication methods. Additionally, natural speech carries with it a sense of personal identity that synthesized speech cannot match. We propose a new genre of communication device, named Real Talk, which augments an individual's natural speech ability instead of replacing it. We accomplish this through our unique speech recognition technology, which we developed specifically for individuals with dysarthria and that integrates tightly with our word prediction software. This combination of speech recognition and word prediction substantially reduces the typing required to produce messages. For this project we will improve our existing speech recognition system to use noise- tolerant microphones that enable accurate recognition in noisy environments. We will supplement natural dysarthric speech with a written transcript for face-to-face conversations and provide synthesized speech output for other communication environments. Individuals with dysarthria will evaluate the technology in laboratory and real-life settings and the impact of this device on communicative interactions will be measured.
The proposed device helps people to communicate effectively when they have physical disability and their speech is difficult or impossible to understand.