Continuous-time recurrent neural networks for generative and interactive musical performance

Abstract

This paper describes an ongoing exploration into the use of Continuous-Time Recurrent Neural Networks (CTRNNs) as generative and interactive performance tools, and using Genetic Algorithms (GAs) to evolve specific CTRNN behaviours. We propose that even randomly generated CTRNNs can be used in musically interesting ways, and that evolution can be employed to produce networks which exhibit properties that are suitable for use in interactive improvisation by computer musicians. We argue that the development of musical contexts for the CTRNN is best performed by the computer musician user rather than the programmer, and suggest ways in which strategies for the evolution of CTRNN behaviour may be developed further for this context. © Springer-Verlag Berlin Heidelberg 2006.