Traditional Human-Machine Interfaces (HMIs), such as keyboards and mice, have long facilitated efficient digital communication through hand movements. To address this limitation, electromyography (EMG)-based inter- faces offer intuitive control through muscle activity with high signal amplitude and ease of acquisition. This study proposes an EMG-based Chunjiin speller system designed for users with limited hand mobility. The system incor- porates a directional input method (up, down, left, right, and select) and a Korean keyboard layout to support acces- sible and efficient character input. EMG signals were collected using four surface electrodes attached to the extensor digitorum and flexor carpi radialis muscles on both forearms. After real-time preprocessing, three time-domain fea- tures—root mean square, slope sign change, and peak amplitude—were extracted to determine user intent. In Exper- iment 1, five discrete input commands were classified with an average accuracy of 94.67%. In Experiment 2, which involved continuous input for actual word construction, the system maintained an average accuracy of 93.87%. Nota- bly, these performances were achieved without the use of deep learning models, relying solely on simple time-domain features, making it viable for real-time use in low-resource environments. The proposed system demonstrates prac- tical usability and real-time performance, highlighting its potential for augmentative and alternative communication (AAC) applications. Its lightweight architecture and direction-based design further support flexible deployment for users with motor impairments across diverse contexts.