Artificial Intelligence Is Being Used To Detect Speech Disorders In Children

PedzSTAR, an AI-driven approach to early speech disorder detection at Northwestern University, is helping to reduce bias, improve accessibility and make screenings more efficient. PHOTO PROVIDED BY MEKKY MEDIA.
PedzSTAR, an AI-driven approach to early speech disorder detection at Northwestern University, is helping to reduce bias, improve accessibility and make screenings more efficient. PHOTO PROVIDED BY MEKKY MEDIA.

Artificial Intelligence Is Being Used To Detect Speech Disorders In Children

By Tia Carol Jones

As Artificial Intelligence gains popularity, the technology is being applied to detect speech disorders in children. Experts say using the technology in this way can also improve accessibility to screenings and make them more efficient, which is helpful to families in marginalized communities who might not have access to those kinds of screenings. Experts also point toward earlier diagnoses as being helpful for children to receive critical interventions.

Dr. Marisha Speights is a leading speech-language pathologist. Speights established Pediatric Speech Technologies and Acoustics Research (PedzSTAR) in 2021. The research lab at Northwestern University’s mission is to advance the early detection of communication disorders in young children. Its aim is to transform clinical practices and ensure equitable access to early intervention, by utilizing innovative research and strategic partnerships, which include Head Start programs and other early childhood programs.

Speights said the work PedzSTAR is doing involves different kinds of tools that support increasing access to care, as well as the accuracy of early identification of children who have speech disorders. In identifying children with potential speech and language delays, speech-language pathologists use instruments that ask questions about speech and language, but they are reliant on a human being able to accurately ask and answer those questions. Speights acknowledged there are challenges with those measures that depend on the measured being used and how it was developed. 

“It could be that it was developed with many children not being represented. We really focus on children from communities who are Black and also Latino, who tend to be the most misidentified by these measures,” Speights said.

Speights said in using AI and pioneering the development of new screening measures, it could be inclusive of groups that were missed in the past and automate the ability to screen children with more accuracy, that would also allow for greater access. Speights added that Black children are often identified with speech and language delays too late, because of access issues – lack of access to a speech-language pathologist or access of an accurate measure. PedzSTAR is hoping to address both of those challenges in early detection with the development of artificial intelligence tools.

The development of the tool is in the early stages, because Speights said there has been a big barrier to the use of AI for children’s speech. She said with speech to recognition tools, like Siri on iPhone or Alexa on Amazon, they were developed using speech for white men and don’t always work for women or young children.

Speights said in developing the AI tool, they are seeking speech from young children and a variety of young children. From there, they worked on a speech database for children. They want to make sure they get the science right, with adequate representation of different speakers and can train the machine learning model using real child speech samples. They are also thinking of the questions to ask caregivers and the children. With all the moving parts, they are focused on taking their time.

Speights said early interventions based on early identification is critically important to education outcomes. She said while one of the first things children learn to do is talk, early intervention can help with language skills and reading, which are foundational to later academic skills.

“When we are able to find those children who would benefit from additional support, early on, we can mitigate some of the issues that they might have down the road in their education,” she said.

Speights added as there is a move toward more AI driven education tools, there needs to be an awareness of the importance of having children of all backgrounds represented in speech and language tools, as well as the importance of information sharing with families on how to use the tools and analyze the results.

For more information, visit http://www.childstudies.northwestern.edu/labs.  


Latest Stories






Latest Podcast

On The 9 Run Crew