A new peer-reviewed study explores an advanced approach to silent speech interface (SSI) technology, offering a robust solution for communication in high-noise environments.
The research team at Pohang University of Science and Technology incorporated a wearable multiaxial strain sensor with artificial intelligence (AI) to decode and reconstruct speech without relying on sound.
The practical applications of this technology may be more widespread than expected. A significant portion of the global workforce operates in environments where communication equipment is limited due to noise and other environmental factors. Jobs such as construction, military operations, and emergency response scenarios often face communication challenges due to extreme acoustic interference.
Existing SSI technologies—such as electroencephalography (EEG), surface electromyography (sEMG), and single-axis strain sensors—have more limitations than anticipated, including discomfort, limited speech capture, and invasiveness. However, the new system designed by Pohang University addresses these issues head-on.
At the heart of this technology is the Computer Vision-Based Optical Strain (CVOS) sensor embedded in a flexible neck choker. The sensor uses a soft silicone substrate with high-contrast micromarkers, combined with a miniature camera, lens, and LED illumination. This setup allows it to accurately track both the magnitude and direction of throat muscle movements during speech. Unlike conventional sensors, the CVOS captures two-dimensional strain maps, enabling a more comprehensive representation of complex muscle dynamics.
The research team reports that the sensor demonstrates exceptional performance, with a high gauge factor of 3,625, minimal hysteresis, and strong linearity. It can detect extremely small deformations and remains stable over more than 10,000 use cycles. Notably, the technology maintains accuracy in environments with noise levels up to 90 decibels, making it well-suited for real-world applications.
The data is then processed through an AI-driven pipeline designed for rapid and accurate speech decoding. This approach enables the recognition of both localized muscle movements and broader speech patterns in the user.
One of the system’s most significant features is its ability to reconstruct a user’s unique voice using very little training data—approximately 10 minutes of recorded speech. Currently, the system focuses on recognizing the NATO phonetic alphabet, a standardized set of 26 words designed to reduce ambiguity in critical communications.
Laboratory testing has shown promising results. The system achieved 85% accuracy under controlled conditions and maintained strong performance in noisy environments, even during high-intensity scenarios such as rifle firing. It can also leverage minimal training data through fast fine-tuning techniques.
Beyond industrial and military applications, the technology shows strong potential in healthcare. It offers a non-invasive communication method for individuals with speech impairments, including those who have undergone laryngectomy procedures.
Future work will aim to expand the system’s vocabulary, improve resistance to motion-related artifacts, and develop a more refined and wearable design.
This study was published in Cyborg and Bionic Systems.
Chrissy Newton is a PR professional and the founder of VOCAB Communications. She currently appears on The Discovery Channel and Max and hosts the Rebelliously Curious podcast, which can be found on YouTube and on all audio podcast streaming platforms. Follow her on X: @ChrissyNewton, Instagram: @BeingChrissyNewton, and chrissynewton.com. To contact Chrissy with a story, please email chrissy @ thedebrief.org.