This paper explores the automatic continuation of melod-ic passages, based on analyses of trends in existing musi-cal corpora, and the challenges of integrating generative processes with manual composition practices and aesthet-ics. While music composition is a personal creative ex-ploration, probability can be used to model the develop-ment of musical pieces.
The papers central theory, referred to as continua-tion, looks to find likely notes that can continue a given input sequence, with the results bound to a probabilistic model. Three methods of continuation are used to gener-ate new content. The first considers composing music using a call-and-response model of interaction with single note prediction. The second considers how the output from the initial process can be fed back as input to con-tinue generating sequences of arbitrary length. The third is a conceptual model that addresses some of the issues with the first two models, also highlighting a powerful extension that utilizes the time domain. A user survey is presented that gauges the musicality of these methods, evaluating the perceived levels of computer-influenced composition. The results of this are discussed along with proposals for future research in this field.
Hunt, S. (2016). Continuation- Musical Note Prediction Through Analysis. UWE