EMG-based Silent Speech Interface for Human-Robot Collaboration.

This project focuses on developing unobtrusive silent speech interfaces using advanced materials and deep learning algorithms. The system features conformal, transparent, and self-adhesive electromyography electrode arrays to capture speech-relevant muscle activities. Temporal convolutional networks are employed for speaker identification and silent speech content recognition, achieving high accuracy with minimal electrodes. The interface integrates with an optical hand-tracking system and a robotic manipulator, enabling control of the robot through silent speech and enhancing the hand-over process with hand motion detection. This technology aims to facilitate natural robot control in noisy environments and support collaborative tasks involving multiple human operators.