Latent Watermarking of Audio Generative Models

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Roman, Robin San, Fernandez, Pierre, Deleforge, Antoine, Adi, Yossi, Serizel, Romain
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866913491589267456
author Roman, Robin San
Fernandez, Pierre
Deleforge, Antoine
Adi, Yossi
Serizel, Romain
author_facet Roman, Robin San
Fernandez, Pierre
Deleforge, Antoine
Adi, Yossi
Serizel, Romain
contents The advancements in audio generative models have opened up new challenges in their responsible disclosure and the detection of their misuse. In response, we introduce a method to watermark latent generative models by a specific watermarking of their training data. The resulting watermarked models produce latent representations whose decoded outputs are detected with high confidence, regardless of the decoding method used. This approach enables the detection of the generated content without the need for a post-hoc watermarking step. It provides a more secure solution for open-sourced models and facilitates the identification of derivative works that fine-tune or use these models without adhering to their license terms. Our results indicate for instance that generated outputs are detected with an accuracy of more than 75% at a false positive rate of $10^{-3}$, even after fine-tuning the latent generative model.
format Preprint
id arxiv_https___arxiv_org_abs_2409_02915
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Latent Watermarking of Audio Generative Models
Roman, Robin San
Fernandez, Pierre
Deleforge, Antoine
Adi, Yossi
Serizel, Romain
Sound
Audio and Speech Processing
The advancements in audio generative models have opened up new challenges in their responsible disclosure and the detection of their misuse. In response, we introduce a method to watermark latent generative models by a specific watermarking of their training data. The resulting watermarked models produce latent representations whose decoded outputs are detected with high confidence, regardless of the decoding method used. This approach enables the detection of the generated content without the need for a post-hoc watermarking step. It provides a more secure solution for open-sourced models and facilitates the identification of derivative works that fine-tune or use these models without adhering to their license terms. Our results indicate for instance that generated outputs are detected with an accuracy of more than 75% at a false positive rate of $10^{-3}$, even after fine-tuning the latent generative model.
title Latent Watermarking of Audio Generative Models
topic Sound
Audio and Speech Processing
url https://arxiv.org/abs/2409.02915