In International Journal of Extreme Manufacturing , researchers have created a wearable sensor that reads the chemical signature of human breath to translate silent mouth movements into text. Built on standard silicon wafers, the device uses a microscopic, tree-like structure to capture the rapid water vapor changes that occur when a person mouths words.
By pairing this physical sensor with artificial intelligence, the system achieves a 98.51% accuracy rate in recognizing silent speech. This breakthrough offers a direct, independent way to communicate for the millions of people globally who have lost their voices to conditions like laryngeal cancer or neurological injuries.
Capturing silent speech usually forces patients to choose between two fundamentally flawed systems. Contact sensors glued to the throat are physically uncomfortable, and normal muscle twitches constantly shift the hardware, ruining the data. Camera and ultrasound systems remove the physical discomfort but require large equipment that traps the user in one room.
Monitoring the humidity of someone's breath offers a mobile, touch-free alternative, but traditional humidity sensors hit a physical wall. Their flat surfaces hold onto water for too long. When a person silently speaks a string of words, the moisture does not evaporate fast enough, causing the chemical signals of different syllables to blur together into an unreadable mess.
To solve this water traffic jam, the research team engineered a super-adsorbent "nanoforest". By bombarding a thin polymer layer with oxygen plasma, they carved a microscopic landscape of upright pillars. This three-dimensional shape acts like a highly porous molecular sponge, increasing the surface area for water to interact by 120 times compared to a flat film.
When a user silently mouths a word a few centimeters from the sensor, specific syllables expel precise puffs of vapor. The nanoforest grabs and releases these water molecules incredibly fast, allowing a water droplet to completely spread across the material in just 0.4 seconds. This rapid physical clearing means the sensor resets before the next syllable arrives, allowing it to process complex speech patterns in 0.57 seconds, faster than a human heartbeat.
This design shifts the technology from a laboratory experiment to a commercially viable product. Because the sensors are carved directly onto 8-inch silicon wafers using standard microchip manufacturing tools, they avoid the messy and unreliable results of older methods that simply dropped chemical solutions onto a surface. This means the sensors can be reliably mass-produced.
Furthermore, because the system relies on chemical moisture instead of sound waves, it ignores background noise, maintaining perfect accuracy even in loud environments with up to 79 decibels of acoustic interference. Packaged into a compact, Bluetooth-enabled headset, the system seamlessly sends transcribed text to a mobile screen.
Moving forward, researchers will focus on expanding the software's vocabulary and moving the hardware into clinical trials, providing a highly practical communication tool for those navigating profound vocal loss.
DOI: https://iopscience.iop.org/article/10.1088/2631-7990/ae6c6e
International Journal of Extreme Manufacturing (IJEM, IF: 21.3 ) is dedicated to publishing the best advanced manufacturing research with extreme dimensions to address both the fundamental scientific challenges and significant engineering needs.
Visit our webpage , like us on Facebook , and follow us on Twitter and LinkedIn .
International Journal of Extreme Manufacturing
An ultra-sensitive humidity sensor for silent speech intelligent recognition
28-May-2026