Motivations and Methodologies for Automation of the Compositional Process

Abstract

Our aim in this paper is to clarify the range of motivations that have inspired the development of computer programs for the composition of music. We consider this to be important since different methodologies are appropriate for different motivations and goals. We argue that a widespread failure to specify the motivations and goals involved has lead to a methodological malaise in music related research. A brief consideration of some of the earliest attempts to produce computational systems for the composition of music leads us to identify four activities involving the development of computer programs which compose music each of which is inspired by different practical or theoretical motivations. These activities are algorithmic composition, the design of compositional tools, the computational modelling of musical styles and the computational modelling of music cognition. We consider these four motivations in turn, illustrating the problems that have arisen from failing to distinguish between them. We propose a terminology that clearly differentiates the activities defined by the four motivations and present methodological suggestions for research in each domain. While it is clearly important for researchers to embrace developments in related disciplines, we argue that research in the four domains will continue to stagnate unless the motivations and aims of research projects are clearly stated and appropriate methodologies are adopted for developing and evaluating systems that compose music.

Machine learning research that matters for music creation: A case study

“…The uses of such systems span from the analytic, e.g., description and recommendation via music information retrieval (Schedl et al, 2014), to the synthetic, e.g., creative transformation and generation via algorithmic composition (Dannenberg et al, 1997;Pearce et al, 2002;Ariza, 2005;Nierhaus, 2008;Dean, 2018). The latter continues to be a very active research area (Fernández and Vico, 2013;Herremans et al, 2017), especially with deep learning methodologies (Briot et al, 2017), and has growing commercial interest.…”

Taking the Models back to Music Practice: Evaluating Generative Transcription Models built using Deep Learning

“…We thus limit our interrogation of the model to how well it understands the use of or meaning behind the elements of its vocabulary, their arrangement into larger units, and formal operations such as counting, repetition and variation. However, we are ultimately interested in the use of such models to inform and augment the human activity of music, e.g., as a component in the composition of music (Pearce, Meredith, & Wiggins, 2002). Our evaluation thus moves away from quantifying the success of a system in modelling and generating sequences of symbols, and moves toward the results of taking the application of machine learning back to the practice of music -motivated by Wagstaff (2012).…”

Inmamusys: Intelligent multiagent music system

“…However, this is unsatisfactory for two reasons: first, evaluating the music produced by the system reveals little about its utility as a compositional tool; and second, qualitative and subjective evaluation by the designers of the system reveals little about the value of the tool to other composers ( [3]). …”

“…According to [3], the final objective of most of the compositional prototypes is to demonstrate that standard musical techniques could be handled by computer programming, and also to validate generative music theories.…”