MERIT: Factor-Controlled Music Triplet Dataset

MERIT is a dataset of audio triplets designed for training a framework that learns three independent music similarity spaces: melody, rhythm, and timbre. It was created by the AMAAI-Lab and is hosted on Hugging Face. The dataset page was last updated on 2026-05-26.

Use Cases

• Training models for melody similarity based on the described factor-controlled triplets.
• Training models for rhythm similarity based on the described factor-controlled triplets.
• Training models for timbre similarity based on the described factor-controlled triplets.
• Evaluating disentangled music representation learning frameworks.
• Benchmarking audio similarity algorithms on structured, factor-specific data.

Strengths

• Dataset is explicitly designed for learning three distinct music similarity spaces.
• Each triplet is structured around a single controlled musical factor.
• Associated with a published paper and pre-trained models from AMAAI-Lab.

Limitations

• The exact number of triplets, audio files, and dataset size is unknown.
• Column-level documentation and sample data are unavailable, requiring manual inspection after download.
• License information is not provided, which may restrict usage.

Provenance

Source

AMAAI-Lab

Collection Method

Likely created as training data for the MERIT framework.

Freshness

Last updated 2026-05-26 17:38:10; freshness should be verified.