Drawing Music

Abstract

This dissertation presents a potential solution to the challenge of allowing users to control music generation systems, by the way of two different implementations of a novel constraint algorithm which is interfaced via a free-form drawing interface. A survey was carried out to evaluate the constraints and interface. The drawing interface's visual representation of constraints is shown to be identifiable in close to 60% of cases by all participants, and participants with high levels of musical expertise prove themselves to perform significantly better in identifying constraints. Participants showed a significant preference for the output music of one the two constraint implementations, but also showed a significant preference for unconstrained over constrained music generation. Results showed promise for the presented method and that more testing and fine-tuning could be done for the current implementation, and that more exploration within controlled music generation is very likely to yield even better results and perhaps allow for closing the gap between constrained and unconstrained music generation.