Repositorio Digital de la UPF
Publicación
Capítulo o parte de un libro (BookPart)
Evaluating disentangled representations for controllable music generation
Comunicació presentada a la IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), celebrada a Barcelona del 3 al 8 de maig de 2026., Recent approaches in music generation rely on disentangled representations, often labeled as structure and timbre or local and global, to enable controllable synthesis. Yet the underlying properties of these embeddings remain underexplored. In this work, we evaluate such disentangled representations in a set of music audio models for controllable generation using a probing-based framework that goes beyond standard downstream tasks. The selected models reflect diverse un-supervised disentanglement strategies, including inductive biases, data augmentations, adversarial objectives, and staged training procedures. We further isolate specific strategies to analyze their effect. Our analysis spans four key axes: informativeness, equivariance, invariance, and disentanglement, which are assessed across datasets, tasks, and controlled transformations. Our findings reveal inconsistencies between intended and actual semantics of the embeddings, suggesting that current strategies fall short of producing truly disentangled representations, and prompting a re-examination of how controllability is approached in music generation., This work has been supported by IA y Musica: Cátedra en Inteligencia Artificial y Música (TSI-100929-2023-1), funded by the Secretaría de Estado de Digitalización e Inteligencia Artificial and the European Union-Next Generation EU, and IMPA: Multimodal AI for Audio Processing (PID2023-152250OB-I00), funded by the Ministry of Science, Innovation and Universities of the Spanish Government, the Agencia Estatal de Investigación (AEI) and co-financed by the European Union.
Proyecto:
ES / 3PE PID2023-152250OB-I00