Jordie Shier’s work on percussive timbre remapping contributes novel machine learning techniques and tools for working with timbre in NIME contexts; this technical development is grounded in longitudinal co-design with two percussionists, each reported as an annotated portfolio, confronting the gap between a working demo and a mapping a musician will keep performing with.

Noel-Hirst, Saitis, and Bryan-Kinns (2025) turn the same critical and co-design lens on generative AI (GenAI), drawing on the established but understudied practice of sample-based music to ask how a technology increasingly built on recycling existing sonic/timbral material can support meaningful creativity, and translating an ethnography of two London communities (CreateDefineRelease and ClubWIP) into design directions for generative AI tools.

We found that context was extremely important for making sample-based music, and identified four aspects of sampling context relevant to GenAI: environment, community and identity, reality and fiction, and agency. Subsequent work presented at the Second International Conference in AI Music Studies (AIMS 2026) argued that generative AI systems currently prioritise content over context and are thus ill-equipped to deal with the values and practices of sample-based music making.

Next we want to better understand the role semantically structured generative AI might play in sample-based music making. We created a tool (MALT) for dynamically creating latent controls for generative models based on arbitrary attributes, such as timbre descriptors and raw audio features. By asking sample-based musicians to make music with MALT, we want to understand (i) if semantically mediatiated models are helpful for sample-based music making, (ii) what they miss, and (iii) how the four themes of context in sample-based communities are affected by control interfaces with different levels of semantic mediation.