Hackaday0 Hot2 phút
Speak Silently With An Ultrasound Probe
A prototype silent speech system uses an ultrasound probe placed under the chin to capture tongue movements, which are then decoded into speech using machine learning. Built in about a month with roughly 50 hours of training data, it achieves a 15.6% word error rate and works across different speakers with similar accents. The approach is immune to noisy environments since it doesn't rely on sound waves. Current limitations include the bulky off-the-shelf probe, but a miniaturized adhesive patch form factor is considered feasible as the technology matures.