Can You Understand Me, Siri?

By ANNA FLANAGAN

Speech recognition software such as Alexa, Google Assistant, Amazon Echo, Cortana and Siri allow anyone to access information and use smart home technologies through spoken questions and commands. At least, that’s what they’re supposed to do. Unfortunately, these devices typically don’t recognize speech that is affected by a disability.

Mark Hasegawa-Johnson, professor of electrical and computer engineering at Illinois, wants to change that. He launched the Speech Accessibility Project (SAP), which aims to amass a database of audio recordings from people with disabilities that affect their speech. Volunteers with Parkinson’s disease, amyotrophic lateral sclerosis (ALS), stroke-related disabilities, cerebral palsy and Down syndrome record responses to three different types of prompts to capture commands, phonetically diverse speech such as one might produce when reading aloud, and conversational speech. 

SHS Associate Professors Laura Mattie and Marie Moore Channell are leading the Illinois Down Syndrome Team. “People with Down syndrome have intelligibility issues so it’s common for them to not be understood at all or to be misunderstood by voice recognition systems,” Mattie said. And it’s not just that it’s the hot new thing, as Channell observed. “These systems are among the strategies we put into place to make life easier for people with disabilities,” she said. One can imagine the frustration that results from being unable to use technologies that are supposed to improve your life.

Mattie said she and Channell put a lot of effort into developing the prompts for the recordings “…to make them representative of the kinds of things for which individuals would be using the software.” Added Channell, “It’s about making sure on the front end that the recordings are valuable and representative so that what goes into the database is relevant.”

The Speech Accessibility Project database initially will be available to the consortium of technology companies that are funding the project, including Amazon, Apple, Google, Meta and Microsoft, before becoming widely available to the public.