Auditing Training Data in Generative Music Models via Black-Box Membership Inference
Yi Chen Liu 1 , Jiawei Yu 1 , Kexin Cao 2 , Syed Irfan Ali Meerza 3 , Trishika Movva 1 , Jian Liu 1 · 2026
Read Auditing Training Data in Generative Music Models via Black-Box Membership Inference by Yi Chen Liu 1 , Jiawei Yu 1 , Kexin Cao 2 , Syed Irfan Ali Meerza 3 , Trishika Movva 1 , Jian Liu 1 (2026) 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-2605-29202.txt.
Description
Yi Chen Liu and colleagues audit generative music models for training-data membership using black-box query access. Their insight: training creates stronger semantic and structural alignment between a candidate track and its caption output. They query the target model with the caption, gauge candidate-generation alignment, and train a music auditor on paired shadow cases. The auditor captures membership alignment patterns and generalizes to unseen models without parameters or training metadata. Across systems it reaches nearly 99 percent accuracy, with false rates near 2 and 1 percent.
Catalog record
| CREATOR | Yi Chen Liu 1 , Jiawei Yu 1 , Kexin Cao 2 , Syed Irfan Ali Meerza 3 , Trishika Movva 1 , Jian Liu 1 |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 37140 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.29202