AI for AAC: Can AI actually enhance self-expression?

Erin was finishing an internship at a veteran’s affairs hospital the first time she heard a synthetic voice. It belonged to a client of hers who needed an AAC device. “The voice sounded very robotic and mechanical. It was understandable, but lacked realism,” she said. This was in 2000. Back then, people who had lost the ability to speak had only a few voice options. Kids with conditions like cerebral palsy were limited to a single child voice. Special education teachers with multiple AAC users in their class often had to guess who was communicating because so many students sounded the same.  

Today, AI is making it possible for digital voices to capture accent and emotion. Our voice partners Acapela, Microsoft and Eleven Labs offer the means to personalize a voice to sound unique and expressive. 

Erin is thrilled about it. A speech language pathologist by background, she’s a specialist on our clinical team, whose insights help to shape our products. We called her up to find out how AI is helping to create more personalized communication experiences for people with disabilities.  

Note: Some answers have been shortened for conciseness.

Our mission is to help people with disabilities do what they never thought possible. What kinds of things is AI making possible that wouldn't be otherwise? 

A big thing is voices. Even just being able to create a child voice that has a certain accent. Being able to create a voice for someone that sounds like their community creates a big impact for a person's confidence and self-identity.  

[Acapela’s] My-Own-Voice 5--the fact that we offer an AI voice offering that kids are able to use--is huge. 

What parts of their identity do people want their voice to capture? 

I think a lot is in emotion, in how you say something. Did I say it sarcastically? Did I say it like a question? We tend to say a lot with the tone of our voice. 

I know people really like how with Acapela you can program phrases to say in a certain tone. For example, “Mom” regular, “Mom” in a whiny voice, “Mom” in an angry voice. 

What concerns do you hear about AI being used in AAC? 

I think a primary one is that you could possibly take away the person's ability to have their own self-expression. If it’s predicting things for them to say, if it’s phrases in a system, and the user doesn't have a way to modify it or not select it, then they're kind of forced into this choice. Can the user still have agency? Can they decide what they want to say? 

Do we give people the option to reject phrase suggestions? 

Yes, everywhere. In TD Talk, there is phrase prediction, but you don't have to select it, and you can move on and say whatever it is you want to say. We do that even with word prediction. You can easily not pick it or change it. 

What do you hope AI will do in the future to make communication more personal? 

I definitely think there's opportunities to help people speed up communication with AI. To be able to respond more quickly than they normally would.  

How can we ensure efficiency doesn’t take away from someone’s personality? 

It's an interesting conversation. There was a research study done where they looked at using AI in humorous situations. And some people would choose not sounding like themselves to be faster. So they would choose efficiency over self-identity where other people would choose self-identity over efficiency. 

I think we just need to make sure that people are always allowed to not use what is AI generated easily. You don't want to have to do a million steps to get out of what it's suggesting. 

CTA: Learn more about how we’re using AI to enhance assistive communication