Inspiration
Two months ago, I invented a prototype that went viral and allowed you to talk to your computer without talking. A family then called me and offered to fly to wherever I would be in the world to make it work on their niece, who couldn't speak. So today I made a wearable that allows people who cant speak, their voice back.
What it does
People who lost their ability to speak can now just move their tongue and other articulators and decode that movement into text, which gets turned into their own personal voice.
How we built it
Noninvasive wearable sensors pick up how you articulate. A local Python bridge streams the signal to a lightweight model that trains on a laptop in minutes. ElevenLabs gives it a voice, and a simple web app shows it all live.
Challenges we ran into
The signals are tiny, and accuracy drops when the sensors come off and go back on. A few quick recalibration takes fix it.
Accomplishments that we're proud of
It learned four new phrases from about 3 minutes of recording at 98 % accuracy, and kept working after re-wearing.
What we learned
Phrases that differ in rhythm and length are far easier to tell apart than phrases that differ only in their words.
What's next for Outer Voice
More phrases, then open vocabulary, and getting it to people who’ve lost their voice, starting with the family who reached out.
Built With
- claude
