July 10, 2026
Talk is cheap. Neck gel isn't
Silent speech with ultrasound
Your tongue can now text in secret — but commenters are already roasting the neck patch
TLDR: Researchers built a system that reads tongue movements with ultrasound so people can "speak" silently, and it worked better than many expected, even on new users. Commenters are torn between calling it life-changing and joking that nobody wants to glue a weird patch to their neck every day.
A new experiment says your tongue might become your next keyboard: researchers trained a system to turn ultrasound videos taken under the chin into words, even when the person says nothing out loud. The flashy part? It reportedly got surprisingly close to lip-reading accuracy after training on just 50 hours of data, and it even worked on new people right away — as long as they had an American accent. That little detail alone had the comment section instantly side-eyeing the whole thing, with one person wondering whether other languages might actually work better than famously messy English.
But the real popcorn moment was the "would you actually wear this?" debate. The researchers dream of a future sticky patch or wearable probe under your chin, and commenters immediately split into two camps: the sci-fi optimists and the absolutely-not crowd. One summed up the vibe perfectly by saying they were not sure they'd want an adhesive patch on their neck every morning just to speak silently. Brutal, but relatable.
Others saw huge potential. Some imagined life-changing help for people who can't speak because of vocal cord problems, while another went full movie trailer and suggested special forces sneaky-talk missions. Meanwhile, history buffs chimed in with a NASA subvocal speech throwback, basically saying: haven't we been flirting with mind-reading gadgets forever? So yes, the tech is wild — but the comments are asking the real question: private future, or just another weird thing to glue to your body before coffee?
Key Points
- •The article reports a model that predicts speech from ultrasound recordings of the tongue during silent speech.
- •The system achieved a 15.6% word error rate on open-vocabulary speech.
- •The article compares this result with lip-reading at 12.5% word error rate on a 1 million hour dataset, versus 50 hours for the ultrasound system.
- •The team collected its own dataset and chose audible speech for training data collection so audio could be used to verify that participants read prompts correctly.
- •The dataset includes 50 hours of ultrasound tongue imaging from people reading synthetically generated short stories aloud, with real-time quality checks for probe positioning and recording quality.