Abstract

Music is an inherently creative field and therefore fundamentally human. Recent breakthroughs in deep learning have given rise to novel forms of applications of machine learning in music. We focus on interactivity in a human context, examining the role of machine learning in creative work, and ways of designing interactive musical machine-learning systems. We introduce NornsVAE, a novel interactive musical instrument powered by machine learning. NornsVAE is based on “latent spaces” of drum patterns, and various operations that allow the user to change the output of the system by navigating the latent space. The machine learning model powering the latent space is a Variational Autoencoder (VAE). The results of our user study indicate that users desire the role of machine learning in creative work to be more of a supportive nature, bringing in tools that facilitate certain tasks but offering enough means to let users maintain their own fingerprint in the output. We also find the concept of latent spaces to be very powerful in creative contexts, and further research can be done into the exploration and visualization of such spaces, as well as other machine learning models and interaction concepts.

Results/Implementation/Project Description

Conclusion

[ PDF (optional) ] 

[ Slides Kickoff/Final (optional)]