LargeSHS: A large-scale dataset of music adaptation

2025-11-19

AI Summary

Current AI-based music generation research predominantly focuses on text-conditioned modeling, while reference-based song adaptation—such as cover generation—remains severely constrained by the lack of large-scale, structured datasets. Method: We introduce LargeSHS, the largest publicly available music adaptation dataset to date, comprising 1.7 million metadata entries and approximately 900,000 accessible audio links, systematically curated and enriched from the SecondHandSongs platform. We further propose the first structured modeling framework for musical adaptation relationships, enabling adaptation tree construction, performance clustering, and cover family identification to support cross-work semantic mapping. Contribution/Results: LargeSHS bridges critical data gaps in reference-guided music generation and adaptive Music Information Retrieval (MIR), establishing a foundational resource and novel research paradigm for cover generation, style transfer, and music evolution analysis.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Recent advances in AI-based music generation have focused heavily on text-conditioned models, with less attention given to reference-based generation such as song adaptation. To support this line of research, we introduce LargeSHS, a large-scale dataset derived from SecondHandSongs, containing over 1.7 million metadata entries and approximately 900k publicly accessible audio links. Unlike existing datasets, LargeSHS includes structured adaptation relationships between musical works, enabling the construction of adaptation trees and performance clusters that represent cover song families. We provide comprehensive statistics and comparisons with existing datasets, highlighting the unique scale and richness of LargeSHS. This dataset paves the way for new research in cover song generation, reference-based music generation, and adaptation-aware MIR tasks.

Problem

• Addresses limited research on reference-based music generation models
• Provides structured adaptation relationships between musical works
• Enables research in cover song generation and adaptation-aware MIR

Innovation

• LargeSHS dataset with structured adaptation relationships
• Contains 1.7 million metadata entries and audio links
• Enables cover song generation and adaptation research