Efficient Electromyography-Based Typing System: Towards a Novel Approach to HCI Text Input.

Abstract

While electromyography (EMG) excels in static gesture recognition and medical diagnoses, its application to real-time interactions like typing is hampered by the difficulty of reconciling continuous EMG signals with discrete output decisions. This paper presents a novel EMG typing system that tackles this challenge by utilizing Connectionist Temporal Classification (CTC) for efficient continuous recognition and a parallel inference approach for improved accuracy. This system enables rapid feedback and accurate word recognition, with experimental results demonstrating a character error rate of 3.8% on the test set, a word error rate of 7.1%, and a response time of less than 100 milliseconds. These results validate the feasibility and potential of EMG-based keyboard-free typing in real-time interactions, with significant implications for human-computer interaction.