Tech Titans are teaming up with the University of Illinois to launch a new language accessibility project

0

Tech heavyweights Amazon, Apple, Google, Meta, and Microsoft have banded together in a way that would make Voltron blush. The companies are making a concerted effort to make voice-first user interfaces more accessible for people with speech disabilities. The five industry titans are supporting the University of Illinois with the school’s Speech Accessibility Project. The project, part of the Beckman Institute for Advanced Science & Technology, is described on its website as “a new research initiative to make speech recognition technology more useful for people with a range of different speech patterns and disabilities.”

Engadget’s Steve Dent was the first to report the news earlier this week.

ADVERTISEMENT

Using a “private, unidentified data set” of speech samples collected from paid volunteers, the University of Illinois will take that information and use it to develop stronger, more adept machine learning models to better understand more diverse speech patterns. The project is currently focused on American English and is designed to accommodate people with Parkinson’s, Down’s Syndrome and other conditions where typical fluency may be affected. The big goal is to make language-centric products like digital assistants more user-friendly and accessible for millions of Americans with a speech impairment.

“Amazon, Apple, Google, Meta, and Microsoft each have longstanding accessibility commitments that encompass providing products, services, and experiences to people from a variety of communities and backgrounds. This unique collaboration is rooted in the belief that inclusive speech recognition should be a universal experience,” the project consortium said in a statement provided to me. “Collaborating on the Speech Accessibility Project is an opportunity to find the best and most expedient path to inclusive speech recognition services.”

They continued, “Each of the companies aims to use project data to make improvements in their respective speech recognition products and services. The project is fully funded by equal contributions from all companies committed to supporting UIUC [University of Illinois Urbana-Champaign] in leading the project for at least two years.”

ADVERTISEMENT

This week’s news is similar in spirit to a report from last year in which Katie Deighton wrote for the Wall Street Journal about tech companies Amazon and Apple, which have invested significant chunks of their respective war chests in improving their digital assistants for those who are ready become friendlier with speech delays.

As smart speakers like Apple’s HomePod and Amazon’s Echo line have risen in popularity in recent years, their usefulness has been stifled — and the entire product category shut out as a whole — because digital assistants Siri and Alexa are inherently under the Accepting normal language are built. But as the Illinois researchers rightly point out on their Speech Accessibility Project FAQ page, the lack of a diverse dataset means the software can’t learn other ways of speaking. Artificial intelligence and machine learning models are only as good as the information people feed them; Software engineers and researchers are required to collect (and thus use) data that is as representative as possible.

“Speech technologies like smart speakers, digital assistants, and speech-to-speech translation systems have gotten much, much better in the last five years. The great success of speech technology is the result of great advances in computing coupled with great advances in the public availability of sharable data: Researchers have figured out how to use large, sharable audio databases to teach computers to speak,” said Mark Hasegawa- Johnson, professor of electrical and computer engineering at the University of Illinois who leads the Speech Accessibility Project, in a recent email interview with me.

ADVERTISEMENT

He added: “As technology gets better, those of us who work in language technology can pay more attention to the people it’s failing on. In general, speech technology fails in people whose speech is a little different than normal, for example due to a neuromotor disorder such as Parkinson’s or Lou Gehrig’s disease [known medically as ALS]. To solve this problem, we need to form a communication channel that amplifies the voices of people with neuromotor disorders so that the sounds of their voices can be an effective part of the data used to develop future language technologies. ”

Speech disabilities, like other disabilities, must be recognized and accessibility technically prioritized. “There are millions of Americans who have language differences or disabilities. Most of us interact with digital assistants fairly seamlessly, but for people with less understandable speech there can be a barrier to entry,” said Clarion Mendes, clinical professor of speech and hearing sciences and speech-language pathologist. “This initiative [the Speech Accessibility Project] narrows the digital divide for people with disabilities. Improving access and breaking down barriers mean an improved quality of life and greater independence. As we begin this project, the voices and needs of people in the disability community will be paramount as they share their feedback.”

Speech is about communication, and that’s what the Speech Accessibility Project is all about. Voice-first user interfaces should be democratized and egalitarian, and not reserved specifically for those who are linguistically privileged. “As a speech pathologist, I’ve been privileged to have many patients and their caregivers share with me that the ability to communicate effectively can be life-giving and potentially life-saving,” Mendes said.

ADVERTISEMENT

Mendes said the group is “excited” to partner with organizations like the Parkinson’s-focused Davis Phinney Foundation and ALS-centric Team Gleason as the Speech Accessibility Project gets underway. Both organizations are “eager to contribute,” according to Mendes, with other stakeholders potentially participating as well. “We will continue to bring in other etiologies as part of our focus on collecting language data that will benefit most people,” she said.

Mendes is proud to contribute to the Speech Accessibility Project. Your longtime dream? “Successful, individualized communication for everyone who is looking for it,” she said.

Share.

Comments are closed.