Music Arena Dataset

dataset·human music preferences·active

Rolling human preference data from live pairwise comparisons of text-to-music model outputs.

Verified facts

Official site	https://huggingface.co/datasets/music-arena/music-arena-dataset ↗
Geography	United States
arxiv	2507.20900
huggingface repository	music-arena/music-arena-dataset
size	rows snapshot: 5704 · repository size gb: 19
scope	Live human preference evaluation of text-to-music systems.
genres	open-ended prompts
license	unknown/mixed by model-output agreement
steward	Music Arena collaboration
creators	Yonghyun Kim; Wayne Chi; Anastasios N. Angelopoulos; Wei-Lin Chiang; Koichi Saito; Shinji Watanabe; Yuki Mitsufuji; Chris Donahue
languages	user-provided multilingual prompts; distribution not normalized
modalities	text prompts; model outputs; pairwise votes; listening events; natural-language feedback; system metadata
geographies	Global
access method	Hugging Face Git LFS repository.
intended uses	text-to-music evaluation; reward modeling; preference analysis
consent claims	Project reports CMU IRB approval under STUDY2024_00000489.
related papers	Music Arena: Live Evaluation for Text-to-Music
annotation method	Pairwise preference options plus listening telemetry and optional natural-language feedback.
collection method	Consenting users submit prompts, compare anonymized model outputs and vote.
provenance claims	Official card documents model-specific exclusions from public audio release.
current availability	Dataset available; some model audio excluded by release agreements.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	Public arena users are not representative of the global population.; Some models' audio cannot be released.; Preferences may favor quality over diversity.
related models tools benchmarks	Music Arena live benchmark