Latent Walking Techniques for Conditioning GAN-Generated Music

Abstract

Artificial music generation is a rapidly developing field focused on the complex task of creating neural networks that can produce realistic-sounding music. As computer-generated music improves in quality, it has potential to revolutionize the multi-billion dollar music industry by providing additional tools to musicians as well as creating new music for consumers. Beyond simply generating music lies the challenge of controlling or conditioning that generation. Conditional generation can be used to specify a tempo for the generated song, increase the density of notes, or even change the genre. Latent walking is one of the most popular techniques for conditional image generation, but its effectiveness on music-domain generation is largely unexplored, especially for generative adversarial networks (GANs). In this paper, latent walking is implemented with the MuseGAN generator to successfully control two semantic values: note count and polyphonicity (when more than one note is played at a time). This shows that latent walking is a viable technique for GANs in the music domain and can be used to improve the quality, among other features, of the generated music.