Data attribution benchmark assesses text to music model training impact
TrueMuse: A Benchmark for Data Attribution in Text-to-Music Models
Machine Learning
Summary
Text-to-music models learn from huge music collections, but it’s hard to tell which parts of the training data influence the output. The authors created TrueMuse, a special dataset and test setup to measure how well methods can attribute generated music back to specific training samples. They fine-tuned models on known pieces and tested several attribution techniques in different scenarios, showing that current methods struggle and vary widely in accuracy. This work helps identify the challenges and sets a path for better data attribution in music generation.
What this means in practice
- •For music software developers: Use TrueMuse to test how well attribution methods identify training data influence in text-to-music generation tools.
- •For digital rights managers: Evaluate which training samples contribute to generated music to help manage artist rights and licensing claims.
Authors
Jiawei Yu, Jian Liu
Abstract
Text-to-music generation models are trained on massive music collections, creating a growing need for data attribution methods that can quantify the contribution of individual training samples. However, existing attribution methods are difficult to rigorously evaluate due to the lack of reliable ground truth, making it challenging to reliably assess their actual effectiveness. To address this gap, we introduce TrueMuse, a controlled dataset and benchmark for text-to-music data attribution. TrueMuse is constructed by fine-tuning three diffusion-based text-to-music models on carefully curated attribution samples, whose known inclusion in fine-tuning provides controlled attribution targets for evaluation. The benchmark covers four attribution settings, spanning melodic structure, timbral characteristics, artist-level stylistic signatures, and genre-level shared patterns, and includes 133 attributes, 648 fine-tuned models, and 95,456 generated samples across two prompt types. Using TrueMuse, we systematically evaluate existing black-box attribution methods along four dimensions: fine-tuning improvement, prompt-type difficulty, multi-task training, and fine-tuning data size. Our results show that attribution remains challenging, with existing methods exhibiting substantial variation across evaluation settings, highlighting the need for more reliable and generalizable attribution methods for text-to-music generation. Code and Dataset will be released upon acceptance.