Silent speech smart glasses sound like a sci-fi prop: you mouth the words, the frames do the listening, and text lands on your screen at the speed of speech. That's the pitch from The Subvocal Company, a two-person team of Georgia Tech CS students named Aarush and Arnav, now in Y Combinator's F26 batch. Their entire product is one promise — you can talk to your computer without talking out loud.
It's a small idea aimed at a big problem. Over the past couple of years, dictation has quietly become the fastest way to get words into a machine. The catch is the setting. You can't dictate in a library, an open-plan office, or on a crowded bus without sharing your grocery list and your investor update with everyone in earshot.
What happened: a 36-hour hackathon bet became a company
At the start of the year, Aarush and Arnav were studying CS at Georgia Tech with zero background in silent speech or sensing. Both were AI-native builders and heavy users of Wispr Flow, a dictation tool. What bugged them wasn't accuracy — it was location. They couldn't use these tools in the library, in the office, or in public.
So at Stanford's TreeHacks in February, they decided to build the most technically ridiculous thing they could think of. They'd read about AlterEgo, a silent speech system out of MIT's Media Lab, and set out to recreate it in 36 hours.
The first prototype taped electrodes to a face
The early build was crude: EMG electrodes stuck onto Aarush's face. EMG just means reading the tiny electrical signals your muscles give off when they move — think lie detector, but for your jaw. It kind of worked at the letter level, in a very janky way. It didn't win the hackathon. But the first time Aarush typed a full sentence with nobody around him noticing, he felt the magic of the technology. The founders raised a pre-seed from Afore Capital and went all in.
Then the hardware pushed back
EMG broke in four specific ways. Sweat changed the signal. Electrical interference polluted it. Electrode placement varied from person to person. And the real killer: nobody wants electrodes stuck to their face.
So they switched to radio waves
The current glasses emit radio waves, similar to Bluetooth, that pass through your face. As those waves cross your tongue, lips and soft palate, silent mouthing perturbs them. Picture how your hand changes the shape of light shining through a lampshade — same bulb, different pattern. The company trains foundation models on that data to decode the perturbations into text at the speed of speech.
What it means for you: a private keyboard made of air
Strip away the engineering and here's the plain version. Every input method you've ever used forces a trade-off. Typing is quiet but slow. Talking is fast but public. Silent speech input is the first serious attempt to give you fast and private at the same time — which changes where you can use AI, not just how.
At home
Think about the last time you wanted to ask an AI something at 11pm with a sleeping kid in the next room. Today your options are thumb-typing a half-sentence or waking the house. Mouthing the question and having it appear on screen removes that friction entirely. It's the difference between whispering and thinking out loud in your head.
At work
Open-plan offices are the worst place for voice tools. You either annoy forty people or you don't use the tool at all. Silent input means you can draft a reply, search a doc, or fire off a prompt during a meeting without anyone clocking it. It's a private channel at your desk, and that's a bigger upgrade than it sounds.
Running a business
Founders spend their days in transit — walking to a warehouse, waiting in a supplier's lobby, standing on a factory floor where it's too loud to dictate. Silent input turns that dead time into writing time. If you want a feel for how fast this shift is moving, browse the free AI tools people are already pairing with dictation today; the input layer is changing faster than most teams realize.
Studying
Libraries are exactly where voice tools die, and exactly where students need to capture thoughts fast. Talking through an idea out loud is one of the best ways to understand it. Mouthing it silently, at roughly the speed of speech, gets you most of the benefit without getting you shushed. Same for anyone in a shared study room at midnight.
Creating
Writers don't think in sentences they can type — they think in bursts, out loud. Most people speak three to four times faster than they type, so the gap between the idea and the page is usually a keyboard. Silent speech closes that gap without turning your coffee shop into a recording studio.
Extra income
Anyone doing client work — transcription, captions, notes, freelance writing — knows the bottleneck is capture, not polish. Getting the raw material down quickly while you're walking the dog is where the hours hide. Faster capture means more finished work per week, and that's the whole game.
How to try it right now (you can't buy a pair yet)
1. Watch the demo first. Dave from YC tested the glasses in a short clip — watch the demo. It's free and it takes less time than reading this section.
2. Build the habit with a free tool. Go to mykreatool.com and pick a free AI tool that accepts voice or text input. Use it out loud for one full day and count how many times you'd rather not be overheard in the room.
3. Test the awkward part yourself. Open a blank doc, mouth a sentence silently, and watch what happens. Nobody reads your lips reliably, and that's the honest preview of what the product feels like in public.
4. If you already dictate, map your dead zones. Wispr Flow users already know where they stop using voice: the library, the office, the train. Write those three places down — that list is your personal case for silent input.
5. Don't wait on hardware. There's no public price or ship date in the announcement, so treat this as a direction, not a purchase. The tooling you can use today is the part that already works.
Upsides and what changes
The upside isn't that the glasses are cool. It's that they remove the last social cost of using AI. Voice input made talking to a computer normal; silent input makes it invisible. Once that happens, people stop treating AI as something you sit down for and start treating it as something you wear. The founders' own story is the proof: they were already power users, and the missing piece wasn't a better model — it was a place to use the model where nobody could hear them.
Limitations
Be clear-eyed about where this stands. The first version used EMG and never got past letter-level output before the team abandoned it, and the radio-wave version is shown so far in a short demo clip, not in a shipping product. There's no price, no release date, and no independent accuracy data. Radio sensing through a face also has to survive the real world — different face shapes, movement, background signals — and the company hasn't published how it handles those cases. And there's a human factor nobody has tested at scale: mouthing words silently in a meeting looks strange, and people may not want to be seen doing it even if the glasses work perfectly.
Conclusion
Silent speech smart glasses are one of those ideas that sounds gimmicky until you remember how much of your day you spend unable to talk to your computer. The Subvocal Company has a working demo, funding, and a clear reason to exist: fast input that doesn't broadcast itself. You can't buy a pair today, but the underlying shift is real.
One action for today: open a free AI tool, dictate one paragraph out loud, and write down the exact moment you'd want to mouth it instead. That list is your benchmark for whatever ships next.



Comments 0