Francesco Ardan Dal Rí, Francesco Manenti
Was presented at:
In this paper, we assess a pipeline for structural analysis and segmentation on live coding performances. Traditional Music Information Retrieval methodologies are primarily evaluated on Western classical or popular music. However, the practice of live coding poses unique challenges for analytical tasks, due to its complexity and absence of standardized forms. Furthermore, given the extemporaneous/improvisational nature of live coding, the literature lacks shared methodologies and tools. To address these problems, we tailored a novelty-based analysis pipeline on live coding performances, focusing on four musical features commonly used in the literature; such features were employed to compute Similarity Matrices and identify boundaries between musical sections via novelty curves. The effectiveness of the proposed methodology was evaluated with the involvement of six experienced live coders, who provided their live performance recordings to be analyzed by the algorithm. The musicians reflected on the accuracy and relevance of the segmentation through semi-structured interviews, offering insights into how the segmentation related to their personal practices and musical perspectives. Additionally, we applied our pipeline to a large dataset, highlighting structural patterns and general trends across performances.