Home [Music] Research {Software} [Teaching] Contact

Autocousmatic

(2009-2011) Autocousmatic by Nick Collins: a program for automatic generation of electroacoustic works, incorporating machine listening

Autocousmatic generates electroacoustic music intended for acousmatic presentation. Based only on a seed directory of source sound files, a desired duration and number of output channels, the program creates multi-channel spectromorphological tape pieces. Audio analysis capabilities are used to discover 'useful' portions of sound files and assess processed files. The project investigates a deeper relation from machine listening to algorithmic composition tasks, and welcomes feedback.

A few further details

A great challenge to the automated production of musical works is the critical role of the human auditory system within the design cycle. Human apprehension of musical form provides for continual feedback at multiple timescales, from selecting and refining momentary material, to the control of flow between sections and across the whole work. Much prior algorithmic composition has not extensively engaged with issues in the audition of the generated music including the psychology of musical form (Collins 2009). Yet as machine listening technology advances (Klapuri 2006), there are opportunities to build artificial listening capabilities into algorithmic works. Although this has been a feature of recent work in live interactive systems, it has not taken place to any great degree within the iterative design cycles and out of time algorithmic production of fixed works. With its focus on intensive listening and compositional design, electroacoustic art music seems a natural context for an algorithmic investigation of this nature.

Precedents should be acknowledged in any exploration of creativity! Because electroacoustic composers, inevitably in the current age, are computer-savvy, there have been a number of previous attempts to render electroacoustic works automatically. Precedents include automated mixing strategies, such as Vaggione's work on Octuor (1982) and subsequent hard drive and network harvesting sound file mixing programs, the use of Csound particularly with external programming routines to create scores, and of course generative works created with SuperCollider. Yet the extent to which auditory analysis has been incorporated, and evaluation sought on a par with human composed works, is questionable. This project is very open to any feedback people wish to give, through trying the software, responding to the short affiliated questionnaire, or just emailing comments through. It is hoped that in due course, a fuller publication will follow.