Towards a Machine Somaesthete: Latent Modeling of EMG Signals in Viola Playing

Somaesthetic experiential qualities can provide a window into process of meaning-making, both human and machinic. We draw such qualities from viola performance into the design-in-progress of a novel interactive performance system. In doing so, we introduce the concept of a Machine Somaesthete that senses and makes sense of these qualities from a second-person perspective. Our system comprises electromyographic (EMG) muscle sensing and a Variational Autoencoder. With a novel dataset, we aim to encode latent representations of performance movement that are meaningful from a somaesthetic perspective. We present our model and our design process, then analyse latent trajectories to interrogate how our system can be considered a Machine Somaesthete, and the nature of its sensitivity to bodily experiences of viola playing. At the intersection of artificial intelligence, music performance and intra-action design, we take a sympoietic (together-making) view of knowledge creation. We and our practices are transformed as we design - and design with - machine learning systems.