IMC Colloquium Series: "Modeling Electronic Music: Intelligent Generation using Human Transcription and Analysis"

Abstract

Artist Philip Galanter proposes that that generative art "refers to any art practice where the artist uses a system, such as a set of natural language rules, a computer program, a machine, or other procedural invention, which is set into motion with some degree of autonomy contributing to or resulting in a completed work of art." Placed within the paradigm of electroacoustic music, generative music is, therefore, music that either uses a computer to autonomously generate the resulting music, and/or electroacoustic means - such as a synthesizer or sampler - to realize its results. One important aspect of generative music is that it is based within composition, rather than improvisation; in fact, I have proposed the concept of real-time composition to further discriminate its compositional basis from the improvisational elements of live electroacoustic music. When creating a generative system, rules are required to limit the possible choices; in most cases, these rules are used to generate new compositions in the style of the composer. One difficulty with generative systems is validating the success of the system - in other words, whether the system has interpreted the rules correctly, or whether the rules themselves accurately model the desired style. In the above mentioned system, it is really only the creator of the system that can make this judgement: listeners can reject the musical result, but the system's creator can argue that they are making aesthetic judgements of the music, rather than the system. However, if the aim of the system is to create music consistent within a given genre, it is possible to judge the success of the system - both artistically and practically - by the relationship of its output to the original corpus.

The Generative Electronica Research Project, part of ongoing research into musical metacreation - the potential of endowing machines with creative behavior - is pursuing the potential of creating software that generates electronic dance music in specific styles. We have selected 100 complete musical examples in the genres of Breakbeat, House, Drum & Bass, and Dubstep, and are using a combination of machine and human analysis of these works to derive rulesets, which, in turn, are used to generate new music consistent within the genres. Unlike the work of David Cope, who used a set corpus of existing music by composers such as Bach, Mozart, Beethoven, and Joplin to create new compositions through recombinance - stitching together music from given examples - we are using generative methods - including probabilistic methods and genetic algorithms - to create new music.

This presentation will discuss how our methods differ from other generative music systems, and other music information retrieval (MIR) programs, and present musical examples of our ongoing research.