Interactive Machine Learning: Strategies for live performance using Electromyography

We present use of the electromyogram (EMG) for sensing musical gesture and discuss interactive machine learning for designing complex muscle-music inter- actions. We propose a signal flow for musical interaction with body movement sensed by EMG and other sensing modalities, feature extraction, and interactive machine learning that result in the manipulation of sound synthesis parame- ters. We discuss ways to capture the EMG for musical use including electrode placement and the use of multiple EMG channels. Techniques for extracting meaningful data and features from electromyographic signals are discussed. We present a composite feature, a multi-channel EMG vector sum. Signal pro- cessing and machine learning are demonstrated. We frame classification and regression as cases of recognising and mapping. We finish by providing a series of resources for musicians wishing to work with the techniques presented here. These tools are free and open source, and can be implemented easily in new applications, projects or products.