Membership Inference Attack Against Music Diffusion Models via Generative Manifold Perturbation

TL;DR

This paper introduces LSA-Probe, a novel white-box attack method that assesses the stability of music diffusion models to improve membership inference accuracy for copyright auditing.

Contribution

It proposes a new geometric approach using generative manifold perturbation to enhance membership inference against music diffusion models.

Findings

Members in more stable regions require higher degradation costs.

LSA-Probe outperforms existing loss-based MIAs in low FPR scenarios.

The method effectively identifies training data in generative music models.

Abstract

Membership inference attacks (MIAs) test whether a specific audio clip was used to train a model, making them a key tool for auditing generative music models for copyright compliance. However, loss-based signals (e.g., reconstruction error) are weakly aligned with human perception in practice, yielding poor separability at the low false-positive rates (FPRs) required for forensics. We propose the Latent Stability Adversarial Probe (LSA-Probe), a white-box method that measures a geometric property of the reverse diffusion: the minimal time-normalized perturbation budget needed to cross a fixed perceptual degradation threshold at an intermediate diffusion state. We show that training members, residing in more stable regions, exhibit a significantly higher degradation cost.