Design and Development of an EMG-Based Interactive Musical Instrument Using the Decision Tree Method

Abstract

Hand motor limitations often hinder individuals from expressing their musical creativity, particularly those affected by neurological disorders, musculoskeletal injuries, or playing-related musculoskeletal disorders. Such impairments restrict access to traditional instruments and highlight the need for alternative modes of musical interaction. This study addresses the problem by designing an interactive musical instrument based on surface electromyography (EMG), enabling the conversion of forearm muscle activity into digital notes via a MIDI controller in real time. The system integrates a Muscle Sensor v3, Arduino Uno, and Python-based software equipped with a graphical user interface. The processing pipeline consists of EMG signal acquisition, feature extraction using three widely adopted time-domain features—Mean Absolute Value (MAV), Root Mean Square (RMS), and Waveform Length (WL)—and gesture classification with a Decision Tree algorithm implemented in scikit-learn. Once classified, the gestures are mapped to corresponding MIDI note values and transmitted to a Digital Audio Workstation (DAW) for sound production. Experimental evaluation was performed on eight distinct hand gesture classes. For each class, 20 repetitions were collected for training, and 10 additional repetitions were used for testing, resulting in 80 independent test trials. The system achieved an overall accuracy of 82.5%, with 66 correct predictions out of 80. Simple gestures such as Hand Open and Index Bend reached 100% accuracy, whereas gestures with overlapping muscle activation patterns, notably Form Number 1 and Form Number 2, achieved only 60% accuracy due to their highly similar EMG features. These results demonstrate that the Decision Tree algorithm, while computationally efficient and interpretable, has limitations when handling non-linearly separable data. Nonetheless, the study establishes the feasibility of using Decision Trees as a lightweight baseline for real-time EMG-based musical interfaces. The findings suggest potential for further development through multi-subject, multi-channel EMG datasets and advanced classifiers such as Support Vector Machines (SVM) or Artificial Neural Networks (ANN). Ultimately, this work contributes to the advancement of inclusive and adaptive digital musical technologies for individuals with motor impairments.