AI news ·
Wearable neck sensor reads muscle movements to reconstruct speech without sound
POSTECH researchers built a neck-worn sensor that reads muscle movements to reconstruct a person's natural voice without sound. It's aimed at laryngeal surgery patients and workers in high-noise environments.
Researchers develop wearable sensor that reconstructs speech from silent neck movements
A team at POSTECH has created a wearable device that reads microscopic movements in neck muscles and skin to reconstruct speech without sound. The sensor combines a miniature camera with reference markers on soft silicone, worn on the neck to detect strain patterns. An AI system then analyzes these patterns and synthesizes them into the wearer's actual voice.
The technology addresses a real limitation in existing voice restoration methods. Previous approaches using electromyography (EMG) or electroencephalography (EEG) required complex equipment and uncomfortable wearables that didn't work well in daily life. This sensor is designed for real-world conditions, including noisy factories and moving subjects.
How it works
When someone speaks, the muscles and skin around the neck move in specific patterns. These movements contain information about what the person intends to say-even before sound is produced. The sensor captures these strain maps through computer vision, and an algorithm automatically corrects errors when the device is repositioned.
The reconstructed speech combines two components: a text prediction based on the strain patterns, and a voice synthesis model trained on the individual's vocal characteristics. This produces natural-sounding speech that matches the user's original voice, not a robotic approximation.
Practical applications
The device works in environments where traditional microphones fail. In factories and construction sites, the sensor reads muscle movement rather than sound waves, delivering clear communication in extreme noise.
For patients who have undergone laryngeal surgery, the technology offers something existing devices cannot: restoration of their pre-surgery voice. An electronic larynx produces a buzzing, artificial sound and requires the user to hold it against the throat. This wearable is hands-free and generates natural speech.
The technology also enables silent communication in libraries, theaters, and sensitive settings where audio cannot be used. Military and industrial applications include relaying complex instructions without microphones or radios.
Technical details
The sensor uses what researchers call a "Multiaxial Strain Mapping Sensor." The soft silicone substrate contains micromarkers tracked by a built-in camera. An inference pipeline incorporates physics-based automated calibration and temporal attention mechanisms to handle anatomical differences between users and variations in how individuals speak.
The system requires subvocalization-physical movement of neck muscles to form words-to function. It does not read thoughts or monitor brain activity. The device only decodes intent through muscle action.
Researchers validated the system in real-world noisy scenarios. The findings appear in Cyborg and Bionic Systems, a journal focused on biomedical engineering.
Current limitations and next steps
The research team solved wearability problems that plagued earlier signal-based voice interfaces. However, the technology remains in the research phase. Clinical trials and further refinement will be needed before widespread deployment.
Professor Sung-Min Park, who led the study, said the work aims to help patients with speech disorders reclaim their voices and enable communication in environments where traditional audio fails.
The research received support from South Korea's Ministry of Education and Ministry of Science and ICT through multiple funding programs focused on biomedical and convergence technologies.