New DiVers dataset boosts AI music version identification robustness

Researchers have introduced DiVers, a new dataset designed to improve musical version identification (VI) systems. Unlike existing datasets that focus on professionally recorded tracks, DiVers includes over 1.1 million musical versions from user-generated and amateur content, addressing a domain mismatch with real-world scenarios. Models trained on DiVers demonstrate enhanced robustness to diverse and noisy audio inputs while maintaining performance on cleaner studio recordings. The dataset, along with its construction code and experimental pipelines, has been released to facilitate reproducibility. AI

IMPACT Enhances AI's ability to identify music versions in diverse, real-world audio, improving applications like music recognition and cataloging.

RANK_REASON The item is a research paper detailing a new dataset and benchmark for a specific AI task (musical version identification). [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.IR (Information Retrieval) →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Da-TACOS
• Discogs
• Discogs-VI-YT
• DiVers
• Gotit.pub
• Hugging Face
• ScienceCast
• SecondHandSongs
• SHS100K

• paper
• other

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