MusicBench

dataset·text to music training·active

Music audio-text pairs expanding MusicCaps with extracted musical features and augmentation for controllable generation.

Verified facts

Official site	https://huggingface.co/datasets/amaai-lab/MusicBench ↗
Geography	Singapore
github repository	AMAAI-Lab/mustango
size	test samples: 400 · train samples: 52768 · source samples: 5521
scope	Controllable text-to-music model training and evaluation.
genres	MusicCaps-derived mix
steward	AMAAI Lab / Mustango authors
creators	Jan Melechovsky; Zixun Guo; Deepanway Ghosal; Navonil Majumder; Dorien Herremans; Soujanya Poria
languages	English prompts
modalities	audio; text prompts; chords; beats; tempo; key
geographies	Global
access method	Hugging Face dataset repository.
intended uses	text-to-music generation; controllable generation
related papers	Mustango: Toward Controllable Text-to-Music Generation
annotation method	Audio features extracted and expressed through text templates appended to prompts.
collection method	Expands MusicCaps through pitch, tempo and volume augmentations.
provenance claims	Derived from MusicCaps audio identifiers and captions.
current availability	Dataset card available; actual audio availability follows MusicCaps/YouTube dependencies.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	Augmented samples are not independent recordings.; Source audio inherits MusicCaps/YouTube access and rights constraints.
related models tools benchmarks	Mustango; MusicCaps