arXiv：cs.LG· Fotis Koutsikos, Ioannis Prokopiou, Spyridon Kantarelis, Vassilis Lyberatos, Pantelis Vikatos, Athanasios Aidinis, Themos Stafylakis, Athanasios Voulodimos, Giorgos Stamou·· 4 小时前AI 评分41
面向 AI 生成音乐抄袭检测的版本识别研究：COPYCAT 基准发布
Towards AI-Generated Music Plagiarism Detection as a Version Identification Problem
AI 导读
研究将人类翻唱领域的音乐版本识别架构迁移至人-AI 抄袭检测场景，并发布 COPYCAT 基准，包含 350,654 个评估对，由真实抄袭案例经生成式重合成与数字信号处理混淆扩充而来。结果显示，标量距离阈值法在生成式重合成下失效，而利用坐标级嵌入向量偏移的监督框架能恢复分散的抄袭信号，将整体 F0.5 从 0.612 提升至 0.803。
正文
Abstract:The rapid expansion of text-to-music generative models challenges traditional paradigms of music creation and intellectual property. Plagiarism in this context is rarely an absolute mathematical binary, but an ambiguous threshold negotiated over harmonic structure, melodic contours, or overall perceived stylistic character. In this work, we test the transferability of state-of-the-art music version identification architectures from the human-to-human cover domain to the human-to-AI plagiarism setting. To evaluate this task, we introduce COPYCAT, a benchmark derived from real-world plagiarism cases and extended through generative re-synthesis and digital signal processing obfuscations, yielding 350,654 evaluation pairs. We show that scalar distance thresholding collapses under generative re-synthesis, while a supervised framework leveraging coordinate-wise embedding shifts recovers the dispersed plagiarism signal, raising overall $F_{0.5}$ from $0.612$ to $0.803$.
| Comments: | 5 pages, 3 figures, submitted to 2027 IEEE International Conference on Acoustics, Speech, and Signal Processing |
| Subjects: | Sound (cs.SD); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS) |
| Cite as: | arXiv:2610.09075 [cs.SD] |
| (or arXiv:2610.09075v1 [cs.SD] for this version) | |
| https://doi.org/10.48550/arXiv.2610.09075 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Fotis Koutsikos [view email]
[v1]
Tue, 6 Oct 2026 20:18:23 UTC (284 KB)
来源：arXiv：cs.LG · arxiv.org