Note: the sounds in this entry do not contain actual words, and often are not identifiable sounds. This post is inaccessible to the Deaf or Hard of Hearing. I have done my best to describe the sounds in the entry itself, however I recognize that any level of inaccessibility to any person for any reason is unacceptable. In an attempt to offset this, I have donated $75 to the Canadian Cultural Society of the Deaf. I strongly encourage anyone who listens to any of the sounds on this page to do the same.

There has been a good deal of discussion about how well AI represents people with disabilities, both in image and text generation. However, as AI has continued to advance, more mediums have become available. How well does state of the art AI represent the sounds of disability?

In order to find out, I asked Eleven Labs to generate sounds that are part of the fabric of the daily lives of people with disabilities. I kept all of the settings (length, influence, etc) at there defaults.

Text To Speech

If you're a blind screenreader user, the sound of text to speech is an enormous part of your every day reality. But how well does AI understand it?

I started off by asking it for arguably the most well-known and historically significant text to speech system: the votrax. Created by Federal Screw Works and featured in everything from the Q*bert video game to the original Braille 'N Speak, the Votrax was what really got consumer accessible text to speech started. If you've never heard it, it sounds like this.

The prompt: "The Votrax text to speech synthesizer".

Hmmm, pretty darn close! I didn't ask it to say anything in particular, so I can't ding it for saying gibberish.

But can it tell the difference between different text to speech systems? Created by Denis Klatt and used by Steven Hawking, the DECtalk synthesizer is easily the second most influential text to speech system. If you've never heard it before, it sounds a bit like this:

So what does AI make of it? Well...

Amusingly, it was obviously trying to introduce itself as DECtalk. But it sounds nothing like it! In fact, it sounds just like the Votrax. Seems like Eleven Labs was only trained on one text to speech system.

The White Cane

If you've ever been to a camp or conference widely attended by blind people, or a school for the blind, the sound of white canes clacking down the hallway or sidewalk is a familiar one. If you've never heard it before, it's a rhythmic tapping, in time with the traveler's footsteps. But is it familiar to AI? I asked Eleven Labs to generate "the sound of someone using the two point touch technique with a white cane".

Apparently...no! That's obviously a cow walking in traffic.

Maybe that was too specialized? What about just "Using a white cane".

This is the sound of a Ninja whacking something. If they are blind, whatever they're doing is not appropriate usage of a white cane.

Unfortunately, none of my other prompts did any better. Asking for people Writing Braille, the sound of a Braille embosser, the alarm from a talking watch, and every other sound I could think of resulted in either garbled speech or clips of music.

Wheelchairs

Maybe AI can do better with "An electric wheelchair stopping and starting"? Back in my university days, I dated a person who used an electric wheelchair, and was friends with another. The sounds of the beeps as they use the joystick to control the chair, and the humming of the electric motor and wheels, are something I'm quite familiar with.

And I can tell you from personal experience that the sound of some kind of toy car hitting a ramp is not at all close.

Diabetes

What about the sound of injecting the latest wonder drug, Ozempic? This is something my own father does every week, so I'm quite familiar with the unique clicking sound it makes as he uses it.

This is just what the Eleven Labs AI does when it has no idea: music and someone speaking in gibberish.

Okay, maybe that was too specialized. What about the sound of an insulin pump? I'm far less familiar with that sound, but a girl who attended my church as a teenager had one, and one of my current co-workers also has one. So I am, at least, pretty sure what they don't sound like.

Yeah, they don't sound at all like a toy shotgun clicking.

Conclusion

Most AI starts off as a strange novelty that people make fun of; it happened with machine translation, captions, and text and image generation. But the same AI quickly gains mainstream acceptance over time, with only minor improvements. I suspect sound effect generation is on the same trajectory. I can only hope that Hollywood doesn't start getting disability related sound effects from AI any time soon.

Can you think of any disability related sound effects I missed? Post them in the comments! Also, don't forget to smash that bell icon and...oh gosh...I feel dirty even writing this.