Building Guitar Strum Models for an Interactive Air Guitar Prototype

Abstract

In this work-in-progress, we propose the design of an interaction that allows a guitar player to air guitar with the use of forearm Electromyogram (EMG). We integrate results from our previous study where we have used the same medium in training a classifier to recognize standard guitar chords. In this paper, we aim to train a classifier this time to recognize the different types of strums when playing the guitar. We collected data from ten (10) participants using the Myo armband doing strum repetitions for at least fifty (50) times. The strumming EMG data was then pre-processed and fed into a machine learning task to build a model. A k-Nearest Neighbor (k=11) classifier was trained and yielded an accuracy of at least 46% accuracy with a kappa statistic of 0.3712. Model results de- scribe that data size needs to be improved while considering equally the same set of features. Additionally, user insights and feedback on the armband usage as an alternative creative medium was gathered from our target respondents. Different views and insights are stated which opened opportunities for the improvement of the actual air guitar concept as a creativity tool.