From Mondrian to Modular Synth: Rendering NIME using Generative Adversarial Networks

Abstract

This paper explores the potential of image-to-image translation techniques in aiding the design of new hardware-based musical interfaces such as MIDI keyboard, grid-based controller, drum machine, and analog modular synthesizers. We collected an extensive image database of such interfaces and implemented image-to-image translation techniques using variants of Generative Adversarial Networks. The created models learn the mapping between input and output images using a training set of either paired or unpaired images. We qualitatively assess the visual outcomes based on three image-to-image translation models: reconstructing interfaces from edge maps, and collection style transfers based on two image sets: visuals of mosaic tile patterns and geometric abstract two-dimensional arts. This paper aims to demonstrate that synthesizing interface layouts based on image-to-image translation techniques can yield insights for researchers, musicians, music technology industrial designers, and the broader NIME community.