GenoMus: Representing Procedural Musical Structures with an Encoded Functional Grammar Optimized for Metaprogramming and Machine Learning

Abstract

We present GenoMus, a new model for artificial musical creativity based on a procedural approach, able to represent compositional techniques behind a musical score. This model aims to build a framework for automatic creativity, that is easily adaptable to other domains beyond music. The core of GenoMus is a functional grammar designed to cover a wide range of styles, integrating traditional and contemporary composing techniques. In its encoded form, both composing methods and music scores are represented as one-dimensional arrays of normalized values. On the other hand, the decoded form of GenoMus grammar is human-readable, allowing for manual editing and the implementation of user-defined processes. Musical procedures (genotypes) are functional trees, able to generate musical scores (phenotypes). Each subprocess uses the same generic functional structure, regardless of the time scale, polyphonic structure, or traditional or algorithmic process being employed. Some works produced with the algorithm have been already published. This highly homogeneous and modular approach simplifies metaprogramming and maximizes search space. Its abstract and compact representation of musical knowledge as pure numeric arrays is optimized for the application of different machine learning paradigms.