Evolving the Living Looper: Artistic Research, Online Learning, and Tentacle Pendula

Victor Shepardson; Halla Steinunn Stefánsdóttir; Thor Magnusson

Image credit: Victor Shepardson; Halla Steinunn Stefánsdóttir; Thor Magnusson

• Format: oral
• Session: papers-2
• Paper PDF link
• Presence: in person
• Duration: 10
• Type: medium

Abstract:

The Living Looper is a neural audio synthesis looper system for live input. It combines online learning with pre-trained neural network models to resynthesize incoming audio into “living loops” that transform over time. This paper describes new features of the Living Looper and musician perspectives on its use. A new graphical interface facilitates use of the instrument by non-programmers and visualizes each loop to aid performers in tracking which loop is making which sound. We also describe a new living loop algorithm including incremental learning with partial least squares regression. Finally, we report on an artistic project using the Looper and lessons learned, resulting in an increased importance of training data and a developing sense of relationality.