Surface EMG signal classification for unsupervised musical keyboard learning application

Learning is an integral part of human growth and it has evolved with the advent of personal computers. The current technologies provide enough flexibility for a learner to learn on his own and does not mandate a mentor to be physically present while learning. However, in learning musical instruments we still have not reached this stage. Hence, it mandates a trainer to be present with the learner to guide. The paper aims to address this issue by classifying finger movements of a person by tracking muscle movements using Electromyogram (EMG) signal that will help in evaluating the learner's performance. EMG signal is an effect of the human brain signaling the muscular neuron to perform an action. For this work, EMG signal was recorded at surface of the forearm, by an inhouse surface-EMG (sEMG) signal acquisition system. sEMG signal was recorded for 10 seconds for four finger keypress and classification algorithm was implemented to identify individual key pressed for music playing application. With the collected data, eleven features including time and frequency domain parameters were extracted. The feature data was standardized using a standard scaler approach and vector dimensions were reduced using Linear Discriminant Analysis (LDA) method. Random forest (RF) algorithm was applied to the reduced dimension feature vector for successful identification of key press. A maximum classification accuracy of 100% for two finger key press and 65.85% for four finger key press was achieved using the sEMG signal acquisition system for the first time. The sEMG integrated wearable system should enhance the unsupervised learning of musical instruments using the proposed system.