GAFAM is collaborating with the University of Illinois at Urbana-Champaign on a project aimed in particular at helping people with disabilities that affect speech.
Along with voice assistants or translation tools, voice recognition is a useful technology in everyday life, but it is not accessible to many people. Wishing to remedy this problem, the University of Illinois at Urbana-Champaign (UIUC) has just launched the Speech Accessibility Project, a multi-year research initiative, with the support of GAFAM. This project, which also brings together technologists, researchers and non-profit organizations, aims to make speech recognition more inclusive for people with various forms of speech and disabilities. This includes, in particular, disabilities affecting speech, such as amyotrophic lateral sclerosis, Parkinson’s disease, cerebral palsy and Down syndrome (trisomy 21).
“Voice interfaces should be available to everyone, including people with disabilities”said Mark Hasegawa-Johnson, professor of electrical and computer engineering at UIUC, adding that “this task has been difficult because it requires a lot of infrastructure, ideally the type that can be supported by major technology companies”.
Improving Speech Recognition Using Data
For this project, which will initially focus on American English, the groups will collaborate to collect a set of voice samples from individuals representing a diversity of speech patterns. Paid volunteers will be recruited by UIUC researchers to provide these samples. They will record these by reading texts or by answering questions like “What are your hobbies?” “. These texts and questions will be developed in collaboration with focus groups and community organizations of people with disabilities to ensure that the samples collected will help researchers obtain a data set to more effectively train speech recognition technologies.
The recorded samples will indeed be used to create a private and anonymized dataset, which will be used to train machine learning models to better understand various forms of speech. Being a form of artificial intelligence, machine learning powers speech recognition, but for the University, it is necessary to have diverse and representative data in order to train these models and thus make this technology more accessible.