Empirical studies in end-user computer-generated music composition systems

Hunt, Samuel

Abstract

Computer music researchers dream of the perfect algorithm, in which the music generated is indistinguishable from, or even superior to, that composed by the world’s most talented composers. However, the fulfilment of this aim remains ambitious. This thesis pursues a different direction, proposing instead that computer-generated music techniques can be used as tools to support human composers, acting as a catalyst for human creativity, rather than a replacement.

Computer-generated music remains a challenge. Techniques and systems are abundant, yet there has been little exploration of how these might be useful for end-users looking to compose with generative and algorithmic music techniques. User interfaces for computer-generated music systems are often inaccessible to non-programmers as they frequently neglect established composition workflow and design paradigms that are familiar to composers in the digital age.

For this research, the Interactive Generative Music Environment (IGME) was developed for studying interaction and composition; building on the foundations established in modern music sequencing software, whilst integrating various computer-generated music techniques.

Three original studies are presented, based on participatory design principles, and evaluated with a mix-methods approach that involved studying end-users engaged with the IGME software. Two studies were group sessions where 54 participants spent an hour with IGME, in either a controlled (lab) environment or remotely as part of a conference workshop. The third study provided users more time with the software, with interactions studied and analysed with the use of screen recording technologies. In total, over 80 hours of interaction data was captured.

It was discovered that users need to understand several threshold concepts before engaging with computer-generated music, and have the necessary skills to debug musical problems within the generative output. The ability to do this requires pre-existing knowledge of music theory. The studies support the conclusion that computer-generated music is used more as a catalyst for composition than as a replacement for it.

A range of recommendations and requirements for building computer-generated music systems are presented, and summarise the contributions to knowledge, along with signposts for future work.

• Thesis Type: Thesis
• Deposit Date: Mar 29, 2021
• Publicly Available Date: Apr 13, 2022
• Public URL: https://uwe-repository.worktribe.com/output/7239594
• Award Date: Apr 13, 2022