Watermarking Training Data of Music Generation Models
On the map: LLM safety and security → Watermarking generated content and models
Abstract (arXiv)
Generative Artificial Intelligence (Gen-AI) models are increasingly used to produce content across domains, including text, images, and audio. While these models represent a major technical breakthrough, they gain their generative capabilities from being trained on enormous amounts of human-generated content, which often includes copyrighted material. In this work, we investigate whether audio watermarking techniques can be used to detect an unauthorized usage of content to train a music generation model. We compare outputs generated by a model trained on watermarked data to a model trained on non-watermarked data. We study factors that impact the model's generation behaviour: the watermarking technique, the proportion of watermarked samples in the training set, and the robustness of the watermarking technique against the model's tokenizer. Our results show that audio watermarking techniques, including some that are imperceptible to humans, can lead to noticeable shifts in the model's outputs. We also study the robustness of a state-of-the-art watermarking technique to removal techniques.
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in Watermarking generated content and models, one of 12 directions in LLM safety and security. Builds on 22 of its 31 references that are in the map. First posted on arXiv 11 Dec 2024.
Builds on
22 of 31 references.
References · 22
- 2016Membership Inference Attacks against Machine Learning Modelscited by 2,221
- 2018The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networkscited by 847
- 2021Membership Inference Attacks From First Principlescited by 839
- 2023Extracting Training Data from Diffusion Modelscited by 723
- 2016CNN Architectures for Large-Scale Audio Classificationcited by 701
- 2022High Fidelity Neural Audio Compressioncited by 673
- 2023Simple and Controllable Music Generationcited by 480
- 2023High-Fidelity Audio Compression with Improved RVQGANcited by 423
- 2022Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Modelscited by 418
- 2023Scalable Extraction of Training Data from (Production) Language Modelscited by 415
- 2022Deduplicating Training Data Mitigates Privacy Risks in Language Modelscited by 309
- 2021Enhanced Membership Inference Attacks against Machine Learning Modelscited by 237
10 more
- 2018Fr'echet Audio Distance: A Metric for Evaluating Music Enhancement Algorithmscited by 221
- 2023Understanding and Mitigating Copying in Diffusion Modelscited by 205
- 2017Machine Learning Models that Remember Too Muchcited by 165
- 2024DE-COP: Detecting Copyrighted Content in Language Models Training Datacited by 72
- 2020Radioactive data: tracing through trainingcited by 67
- 2024Proactive Detection of Voice Cloning with Localized Watermarkingcited by 67
- 2024Blind Baselines Beat Membership Inference Attacks for Foundation Modelscited by 65
- 2023VampNet: Music Generation via Masked Acoustic Token Modelingcited by 57
- 2024Copyright Traps for Large Language Modelscited by 36
- 2024Exploring Musical Roots: Applying Audio Embeddings to Empower Influence Attribution for a Generative Music Modelcited by 13
Built on by
Citing papers per year: 2025: 1.
1 paper.