Auditing Training Data in Generative Music Models via Black-Box Membership Inference

2026-05-27

AI Summary

This work addresses the opacity of training data in generative music models and the consequent difficulty in verifying whether a given audio sample was used during training. To tackle this challenge, the authors propose the first black-box membership inference method capable of auditing training data with high accuracy by querying only the deployed model. Without requiring access to model parameters or training metadata, the approach leverages shadow models to construct paired samples and trains a music auditor in a learned feature space. Membership is determined by measuring the semantic and structural alignment between a candidate audio sample and the output generated from its corresponding text prompt. Experimental results demonstrate that the method achieves 98.6% accuracy across multiple state-of-the-art music generation models, with false positive and false negative rates as low as 1.9% and 1.0%, respectively.

Abstract

Recent advances in text-to-music generation enable high-fidelity synthesis of structured musical audio, raising growing concerns about data provenance, consent, and training transparency. These models are typically trained on large-scale corpora with little disclosure, leaving no practical mechanism to verify whether a particular audio sample was included in training. In this paper, we investigate black-box membership inference for generative music models, aiming to determine whether a candidate music sample was used during training, given only query access to the deployed system. Our key insight is that training membership induces systematically stronger semantic and structural alignment between a candidate sample and the model's generation conditioned on its caption. We query the target model with the associated caption and measure the relationship between the candidate audio and the generated output in a learned feature space. To capture features that separate members from non-members, we construct paired examples consisting of each track and its caption-conditioned generation from shadow models, and train a music auditor to classify membership. The auditor captures alignment patterns characteristic of training membership and generalizes to unseen target models in a fully black-box setting without access to model parameters or training metadata. Across multiple state-of-the-art music generators, our method achieves up to 98.6% accuracy, with false-positive and false-negative rates as low as 1.9% and 1.0%, demonstrating that reliable training-data auditing is feasible in realistic deployment scenarios.

Problem

• membership inference
• generative music models
• training data auditing
• black-box setting
• data provenance

Innovation

• black-box membership inference
• generative music models
• training data auditing
• caption-conditioned generation
• feature space alignment