Supporting Creative Ownership through Deep Learning-Based Music Variation

1 Introduction

• AI creativity support tools range from automating technical tasks to generating or extending musical ideas, while prompt-based systems now offer complete song generation.The introduction distinguishes transcription and stem-splitting tools from systems for controllable MIDI, harmonisation, and phrase infilling.Source: §1 · p. 1
• Prior research identifies risks in AI ideation, including design fixation and homogenized outputs, and argues that humans should remain active co-creators rather than editors.Active involvement is linked to intrinsic motivation in the cited work.Source: §1 · pp. 1–2
• The authors’ earlier co-design study found that musicians valued owning the creative process and were unwilling to relinquish control to a machine.The paper frames supporting rather than replacing creative skills as an important design question.Source: §1 · p. 2
• A four-week evaluation of Rhapsody Refiner examines ownership and active co-creation in a system that varies musicians’ phrases rather than generating complete ideas.The musician provides the initial phrase and remains active in developing the final song.Source: §1 · p. 2

2 Rhapsody Refiner

• Rhapsody Refiner varies user-provided MIDI phrases by using MusicBert to predict masked sections of music.Users select aspects to vary, such as melody pitch or performance dynamics, and can generate multiple variations from an input.Source: §2 · p. 2
• Users guide variation by setting parameters that determine which song sections are masked and predicted.The system was packaged as executable software that participants could run without dependencies.Source: §2 · p. 2
• A Digital Audio Workstation plugin would be needed for a complete musician-facing tool, but implementation was beyond the study’s scope.The interface is shown in Figure 1.Source: §2 · pp. 2–3

3 Ecological Evaluation

• Eight practising musicians took part in the four-week ecological evaluation, which let them use the software in their own creative environments over an extended period.The study defined practising musicians as people actively making music professionally or as a hobby, unlike novices who rarely make music.Source: §3 · p. 3
• Participants were asked to compose a song with Rhapsody Refiner and use it consistently throughout the four weeks.A single instrument was recommended but not required; participants also kept journals, and the researchers collected system logs.Source: §3 · p. 3
• After the study, semi-structured interviews explored participants’ tool use.Two researchers independently coded transcripts using inductive thematic analysis, then discussed common themes.Source: §3 · p. 3

4.1 A Tool for Moments

• Participants valued fragments within variations as inspiration, though randomness could overwhelm them under some parameter settings and support exploration under others.The useful output was often a moment or idea rather than a complete variation.Source: §4.1 · pp. 3–4

4.2 A Tool for Ownership

• Participants often felt control of the creative process, but views on ownership of the final artefact differed.Some attributed the composition to their own motif and effort; others described a degree of joint ownership.Source: §4.2 · p. 4

4.3 A Tool for Creative Involvement

• Participants wanted their own initial idea and the final say, and valued shaping imperfect variations rather than receiving ready-to-use outputs.They contrasted this control over musical details with prompt-based systems that offered less control over minutiae.Source: §4.3 · pp. 4–5

4.4 A Tool for Support

• Participants valued unexpected ideas and input on specific creative goals, while describing the system as a sounding board rather than a source of complete songs.They also said it could help clarify what they disliked, but could not fix a problem or complete the MIDI for them.Source: §4.4 · p. 5

5.1 Importance of Creative Ownership

• Participants connected ownership of the process and song to musical identity, effort invested in shaping tracks, and reluctance to use tools that create for them.Some said they would feel bad passing AI work off as their own.Source: §5.1 · p. 5
• For musicians who value authorship, requiring effort and suggesting rather than automating may help preserve creative agency.The authors note that not every musician or system needs to prioritize ownership.Source: §5.1 · p. 5

5.2 Imperfect Systems Encourage Active Co-Creation

• Imperfect outputs often provide moments, not complete variations, so musicians use their expertise to develop them; participants also valued randomness and ideas arising beyond the system.This contrasts with AI development that prioritizes consistent, clean outputs.Source: §5.2 · pp. 5–6
• The reliance on musicians’ skill may make Rhapsody Refiner harder for novices to use, as they need more predictability and have less ability to turn moments into finished songs.The authors distinguish these needs from those of practising musicians, some of whom accept imperfect outputs when they lead to better ideas.Source: §5.2 · p. 6

5.3 Maintaining Humanity in Music Composition

• Participants’ anxieties about AI and musical identity were not extended to Rhapsody Refiner, whose supportive role left creative control with the composer.Participants supplied an idea understood as a musical vision already in their minds, rather than merely a linguistic blueprint.Source: §5.3 · p. 6

• Lower-than-planned compensation made recruitment difficult and limited the study’s ability to run quantitative evaluations.Future work will compare whether Rhapsody Refiner enables composers to produce higher-quality music, as judged by experts, than systems offering more automation.Source: §6 · p. 6

• Tools that preserve creative ownership, encourage co-creation, and maintain a human role can foster meaningful and satisfying artistic engagement with music AI.Source: §7 · p. 6
• Requiring effort and embracing imperfections helped participants preserve personal identity while still finding the tool useful in their practice.Source: §7 · p. 6