Typing has always been a bottleneck between what we think and the speed at which we can communicate it to a computer. AI-powered dictation tools like Wispr Flow have improved things significantly. You can essentially go through an entire email or even an article and have it converted to text in seconds.
However, there is a rather unpleasant problem when communicating with your computer. Try it in a quiet office, a library, or even your favorite coffee shop, and in most cases it will attract attention or disturb others. A startup called The Subvocal Company believes it has found a solution to this problem. The company has developed smart glasses that promise to convert silent mouth movements into text at over 200 words per minute.
How does it work?
According to the company, its glasses send radar signals through your face. When you silently form words, tiny movements of your speech muscles change the way those signals are sent back. An AI model then interprets these changes and converts them into text. Since there’s no camera involved, the glasses don’t try to read your lips. However, you still need to say the words consciously.
According to Subvocal, you can use the glasses to write messages, take notes and interact with AI assistants. It is even claimed that two people wearing them could have private conversations in a crowded room without anyone else hearing a word.
When can you get your hands on them?
The startup says it has raised $1.3 million from Y Combinator and Afore Capital and is already accepting reservations in the US. The glasses have an early bird price of $499A refundable deposit of $199 is required for a couple’s reservation. You pay the remaining $300 later. Prescription lenses will be available for an additional fee and the company expects to begin shipping in the first quarter of 2027.
Of course, the glasses are still in development and the impressive typing speeds have not yet been independently verified. I’m also curious to see how well they would handle different accents, unfamiliar words, or long stretches of silent dictation. However, if Subvocal can actually do what it promises, I can imagine using something like this.
