MeloBottleneck framework extracts melody skeletons using self-supervised learning

Researchers have developed MeloBottleneck, a novel self-supervised framework for extracting melody skeletons. This method represents a melody skeleton as a controlled-length, order-preserving latent subsequence, distinguishing it from prior approaches that relied on hand-crafted rules or pseudo-labels. MeloBottleneck incorporates a hard-bottleneck extractor, a rhythmic-closure operator, and a re-ornamentation decoder, trained using reconstruction, a melody prior, and consistency objectives. The framework demonstrates improved transferability to out-of-distribution datasets compared to imitation-based methods and enhances retrieval performance in music fragment retrieval tasks. AI

IMPACT This self-supervised approach to melody skeleton extraction could improve music information retrieval and analysis tools.

RANK_REASON The cluster contains a single academic paper detailing a new methodology. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• MeloBottleneck

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →