UOT-IR: Optimal Transport for Fixed-Budget Music Compression
A recent paper on arXiv (2608.00576) presents UOT-IR (Unbalanced Optimal Transport for Information Routing), which is a framework that does not require training for compressing high-polyphony symbolic music into representations with a fixed budget. This method reinterprets the compression challenge as a structured routing issue, utilizing constrained unbalanced optimal transport to maintain structural roles, compatibility in orchestration, and playability. UOT-IR integrates an orchestration prior, adaptive marginal relaxation, temporal decoding, and projections that consider playability. It overcomes the drawbacks of heuristic simplification and broad representation-space reduction. Classified as a cross-type submission, the paper is accessible on arXiv and is significant for tasks in music generation, analysis, and arrangement that necessitate fixed-track representations.
Key facts
- Paper arXiv:2608.00576 introduces UOT-IR framework.
- UOT-IR is training-free and based on constrained unbalanced optimal transport.
- It addresses compression of high-polyphony symbolic music into fixed-budget representations.
- The method preserves structural roles, orchestration compatibility, and playability.
- UOT-IR combines orchestration prior, adaptive marginal relaxation, temporal decoding, and playability-aware projection.
- The paper is announced as a cross-type submission.
- The work is relevant for music generation, analysis, and arrangement.
- The paper is available on arXiv.
Entities
Institutions
- arXiv