Soft EMG Interface Enables Machine Learning-Powered Silent Speech Recognition

Silent speech recognition (SSR) offers a promising alternative communication method, especially in situations where audible speech is impossible or undesirable. However, existing SSR technologies often face limitations such as the need for constant facial attachment, privacy concerns, and inconsistent signal acquisition. Researchers have developed a novel solution: a soft, active electromyography (EMG) interface designed for silent speech recognition. This innovative device is worn on the hand and features a fingertip electrode that can be precisely positioned near the lips to capture EMG signals only when speech is intended. The interface incorporates liquid metal interconnects, transparent flexible printed circuit electrodes, and elastomer encapsulation, ensuring high mechanical stability even during finger movements. A deep neural network, trained on the stable signals acquired by this device, achieved an impressive mean accuracy of 97.2% across three subjects for a 30-word vocabulary, demonstrating robust linguistic discrimination. Furthermore, the system's practicality was validated through real-time drone control, showcasing its potential for secure and intuitive human-machine interaction in noisy or privacy-sensitive environments where traditional voice recognition systems typically fail.