Ashley Noel-HirstWebsite: https://ashleynoelhirst.co.uk/ ProfileProfileProject title: Latent Spaces for Human-AI music generation Theme affiliation: Sonic Interaction Design, music informatics Abstract: Generative music has largely benefited from recent advances in machine learning models, with neural network and deep learning techniques affording engineers increasing levels of complexity and automation. However, such ‘black box’ models are difficult for people, particularly non-experts, to understand and customise. This project centres around the question of how to build generative AI systems which are more understandable, explainable, and interactive for a wider range of end-users. This is achieved by investigating existing RNN approaches to music generation; evaluating the effectiveness of different training sets, latent space regularisation techniques, and user interfaces for real-time interaction. Further research includes an examination of different architectures for AI music generation and their usefulness for human-AI music generation. The use of novel, user-driven mid-level representations also provides opportunities for large-scale e-research in new regions of musicology. Research