Silent speech with ultrasound (alephneuro.com)

Researchers have developed a groundbreaking silent speech recognition system that uses ultrasound technology to detect tongue movements, potentially revolutionizing human-computer interaction in public and noisy environments. This innovative approach captures tongue positioning via a probe placed under the chin, translating silent speech into text with a word error rate (WER) of 15.6%—a significant achievement given the limited 50-hour training dataset used. Unlike traditional methods such as lip-reading, which have a higher word error rate, this ultrasound technique provides direct measurements, presenting a cleaner signal and improved accuracy in speech recognition. The implications for the AI and machine learning community are substantial. This technology not only offers a more private means of communication between users and computers but also showcases the potential for cross-speaker generalization, enabling new users to quickly adopt the system without extensive recalibration. Although challenges remain, including the need for lighter hardware and broader accent recognition, the success of the initial prototype indicates a promising future for silent speech interfaces. As researchers continue to collect data and enhance model performance, this development could lead to wearable devices that offer seamless, private communication, reshaping our interaction with technology in everyday settings.