Silent speech with ultrasound
SMRTR summary
Researchers built a system that reads silent tongue movements using an ultrasound probe placed under the chin, converting them into text without any sound. Trained on just 50 hours of data in one month, it achieved a 15.6% word error rate — approaching lip-reading accuracy — and already works across different speakers, potentially enabling truly private voice computing.
SMRTR provides this summary for quick context. The original article belongs to Hacker News.
Read the original article