TrueMuse: A Benchmark for Data Attribution in Text-to-Music Models

Jiawei Yu; Jian Liu

cs.LGSep 30, 2026

TrueMuse: A Benchmark for Data Attribution in Text-to-Music Models

Authors: Jiawei Yu, Jian Liu

Organizations: School of Computing University of Georgia

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.

Figures & tables

Figure 1: Overview of the TrueMuse construction pipeline. For each target attribute, selected source clips are used to fine-tune a pre-trained text-to-music model with an association prompt, and the resulting attribute-specific model generates samples from contextual and style prompts with known fine-tuning sources.

Property		Melody	Instrument	Musician	Genre	Total
Attributes		50	25	50	8	133
Total finetuning clips		50	225	450	72	797
Clips per model		1	3/6	3/6	3/6	–
Total fine-tuned models		150	150	300	48	648
Prompts	Contextual	15	15	15	15	–
	Style 1	28/28/8	28/28/8	28/28/8	28/28/28	–

Table 1: Dataset statistics across four attribution sets and three text-to-music models.

Figure 2: Attribution R@10 performance of the standard method and the mapper-enhanced method. The generator names are abbreviated as LDM2 (AudioLDM2), Mstgo (Mustango), and S. Audio (Stable Audio Open).

Figure 3: Attribution performance under different prompt types. R@10 is reported for mappers trained on contextual prompts only (Ctx), style prompts only (Sty), and contextual plus style prompts (Ctx + Sty).

Figure 4: Attribution R@10 comparing a mapper trained on a single attribution set (Single-task) against one trained jointly on all four attribution sets (Joint-task). The generator names are abbreviated as LDM2 (AudioLDM2), Mstgo (Mustango), and S. Audio (Stable Audio Open).

Figure 5: Attribution R@10 performance comparing mapper training with

3

versus

6

ground-truth clips per attribute. The generator names are abbreviated as LDM2 (AudioLDM2), Mstgo (Mustango), and S. Audio (Stable Audio Open).

Appendix figures & tables15 assets

Supplementary material from the paper’s appendix.

Appendix

Generator	Year	Version / Checkpoint	#Params
AudioLDM2 [ 18 ]	2024	cvssp/audioldm2	712M
Mustango [ 22 ]	2024	declare-lab/mustango	1.4B
Stable Audio Open [ 11 ]	2025	stabilityai/stable-audio-open-1.0	1.1B

Appendix

Table 2: Specifications for the three text-to-music generators used in TrueMuse .

Figure 6: Attribution R@10 performance of the standard method and the mapper-enhanced method, using

6

clips per attribute. The generator names are abbreviated as LDM2 (AudioLDM2), Mstgo (Mustango), and S. Audio (Stable Audio Open).

Figure 7: Attribution performance under different prompt types, using $6$ clips per attribute. R@10 is reported for mappers trained on contextual prompts only (Ctx), style prompts only (Sty), and contextual plus style prompts (Ctx + Sty).

Figure 8: Attribution R@10 comparing a mapper trained on a single attribution set against one trained jointly on all four attribution sets, using

6

clips per attribute. The generator names are abbreviated as LDM2 (AudioLDM2), Mstgo (Mustango), and S. Audio (Stable Audio Open).

Model	Raw Output	Aggregation	Dim
MERT-v0 [ 16 ]	$\mathbb{R}^{13\times T\times 768}$	Mean over $T$ ; concat all $L{+}1$ layers; L2-norm	9,984
MERT-v0-pub 1 [ 16 ]
Music2Vec-v1 [ 17 ]	$\mathbb{R}^{13\times T\times 768}$	Mean over $T$ ; concat all $L{+}1$ layers; L2-norm	9,984
DAC-16k [ 15 ]	$\{0,\ldots,1023\}^{12\times T}$	Histogram per codebook; concat; L2-norm	12,288

Appendix

Table 3: Summary of music encoders used in TrueMuse .

Figure 9: Attribution R@1 with and without learned mappers, using

3

clips per attribute. The generator names are abbreviated as LDM2 (AudioLDM2), Mstgo (Mustango), and S. Audio (Stable Audio Open).

Figure 10: Attribution R@1 with and without learned mappers, using

6

clips per attribute. The generator names are abbreviated as LDM2 (AudioLDM2), Mstgo (Mustango), and S. Audio (Stable Audio Open).

Figure 11: Attribution R@5 with and without learned mappers, using

3

clips per attribute. The generator names are abbreviated as LDM2 (AudioLDM2), Mstgo (Mustango), and S. Audio (Stable Audio Open).

Figure 12: Attribution R@5 with and without learned mappers, using

6

clips per attribute. The generator names are abbreviated as LDM2 (AudioLDM2), Mstgo (Mustango), and S. Audio (Stable Audio Open).

Figure 13: Attribution R@100 with and without learned mappers, using

3

clips per attribute. The generator names are abbreviated as LDM2 (AudioLDM2), Mstgo (Mustango), and S. Audio (Stable Audio Open).

Figure 14: Attribution R@100 with and without learned mappers, using

6

clips per attribute. The generator names are abbreviated as LDM2 (AudioLDM2), Mstgo (Mustango), and S. Audio (Stable Audio Open).

Figure 15: Attribution Mean Average Precision (MAP) with and without learned mappers, using

3

clips per attribute. The generator names are abbreviated as LDM2 (AudioLDM2), Mstgo (Mustango), and S. Audio (Stable Audio Open).

Figure 16: Attribution Mean Average Precision (MAP) with and without learned mappers, using

6

clips per attribute. The generator names are abbreviated as LDM2 (AudioLDM2), Mstgo (Mustango), and S. Audio (Stable Audio Open).

Figure 17: Attribution performance under different prompt types broken down by generator, using 3 clips per attribute. Each row corresponds to one generator; each panel corresponds to one attribution set.

Figure 18: Attribution performance under different prompt types broken down by generator, using 6 clips per attribute. Each row corresponds to one generator; each panel corresponds to one attribution set.

Explore similar work

Academic Text-to-Music Grand Challenge: Datasets, Baselines, and Evaluation Methods
May 20, 2026Fang-Chih Hsieh, Wei-Jaw Lee, Chun-Ping Wang +3Text-To-Music
Artificial Intelligence Center of Research Excellence, National Taiwan University, Taipei, Taiwan · Department of Performing Arts Technology, University of Michigan, Ann Arbor, MI, United States
Instrumental Text-to-Music Generation with Auxiliary Conditioning Branches
May 20, 2026Junyoung KohText-To-Music
Department of Artificial Intelligence, Yonsei University MAAP KRAFTON Seoul, Republic of Korea
Auditing Training Data in Generative Music Models via Black-Box Membership Inference
May 28, 2026Yi Chen Liu, Jiawei Yu, Kexin Cao +3Text-To-Music Membership Inference
University of Georgia, Athens, GA, USA · Independent Researcher · University of Tennessee, Knoxville, TN, USA

May 20, 2026, cs.SDJ/K move · Enter open · S save

Academic Text-to-Music Grand Challenge: Datasets, Baselines, and Evaluation Methods

Fang-Chih Hsieh, Wei-Jaw Lee, Chun-Ping Wang, Hung-yi Lee+2

Artificial Intelligence Center of Research Excellence, National Taiwan University, Taipei, Taiwan · Department of Performing Arts Technology, University of Michigan, Ann Arbor, MI, United States

Text-To-Music

Open paper AskPDF arXiv

This paper presents an overview and the technical framework of the ICME 2026 Grand Challenge on Academic Text-to-Music Generation (ATTM). Despite the rapid progress in text-to-music generation (TTM) systems, the field is currently dominated by models trained on massive proprietary datasets with industrial-scale computational resources, creating a significant barrier for academic research. To address this, the ATTM Challenge establishes a fair-play benchmark that requires participants to train generative models strictly from scratch using a standardized, CC-licensed subset of the MTG-Jamendo dataset containing only instrumental music. The challenge is divided into two tracks: the Efficiency Track (limited to 500M parameters) and the Performance Track (no parameter limit). Submissions are evaluated through a multi-stage process involving objective metrics, including Frechet Audio Distance, CLAP score, and a novel Concept Coverage Score (CCS), followed by a subjective listening test. By providing open-source baselines, preprocessing pipelines, reference captions, and public evaluation code for computing FAD and CLAP, this challenge aims to facilitate and promote TTM research in academic contexts.

Abstract

Explore similar work

Academic Text-to-Music Grand Challenge: Datasets, Baselines, and Evaluation Methods

Instrumental Text-to-Music Generation with Auxiliary Conditioning Branches

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