emg2qwerty: A Large Dataset with Baselines for Touch Typing using Surface Electromyography

emg2qwerty is a large-scale surface electromyography (sEMG) dataset comprising 1,135 sessions from 108 participants performing touch typing on a QWERTY keyboard. The dataset contains 5.26 million keystroke events recorded from wrist-mounted 32-channel sEMG devices, totaling 346.4 hours of data. Designed to enable keyboard-free text input through direct decoding of typing intent from neuromuscular activity, this dataset supports research in sequence-to-sequence learning, cross-user generalization, domain adaptation, and high-bandwidth brain-computer interfaces for AR/VR and accessibility applications.