Watermarking Training Data of Music Generation Models

Pascal Epple, Igor Shilov, Bozhidar Stevanoski, Yves-Alexandre de Montjoye · 2024

Read Watermarking Training Data of Music Generation Models by Pascal Epple, Igor Shilov, Bozhidar Stevanoski, Yves-Alexandre de Montjoye (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-08549.txt.

Description

Pascal Epple, Igor Shilov, Bozhidar Stevanoski, and Yves-Alexandre de Montjoye ask whether watermarked songs can be traced inside a music generator in two thousand twenty-four. AI music tools learn from vast stores of human work, much of it under copyright. If training songs carry hidden marks, the model's output may betray their use.

Two models were trained side by side, one on marked songs and one on clean ones, and their outputs compared. Marks that ears cannot hear still shifted what the model made. The size of the shift hinged on the marking method and on how much marked data was in the mix. The team also tested how well a top marking method resists removal tricks.

Catalog record

CREATORPascal Epple, Igor Shilov, Bozhidar Stevanoski, Yves-Alexandre de Montjoye
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE43 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-08549

Browse the full library · About the Tower of Babel