CLaMP 2
Multilingual music information retrieval across 101 languages, for ABC notation and MIDI.
About
CLaMP 2 tackles a genuine blind spot in music search: most music-text datasets are English-first. It pre-trains on 1.5 million ABC-notation/MIDI/text triplets and aligns a multilingual text encoder with a multimodal music encoder through contrastive learning, using large language models to clean up and balance multilingual descriptions at scale. The paper reports state-of-the-art results in multilingual semantic search and cross-modal music classification.
The code and benchmark datasets are public on GitHub (sanderwood/clamp2) under MIT, with scripts for feature extraction, classification, and semantic search. One important scope note: CLaMP 2 works on symbolic music (ABC notation and MIDI), not raw audio, so it does not replace audio taggers like the EffNet or transformer families; it complements them. The project is a research artifact from 2024/2025, and community adoption is still small.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported: the NAACL 2025 paper reports SOTA on multilingual semantic search and music classification across modalities; numbers are provider-claimed, no independent replication published.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (metrics, metrics-v1). Evidence: NAACL 2025 paper (arXiv 2410.13267) with named authors; MIT-licensed public repo with benchmarks and search/classification scripts. Strengths: novel multilingual + symbolic coverage, permissive license, peer-reviewed venue. Docked for: (1) small community (65 stars) and thin independent corroboration; (2) all headline numbers are provider-reported; (3) symbolic-only input limits the audience. License: MIT.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 4.0/10
- Maintenance · recency, releases, issue load
- 4.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · NAACL 2025 paper (arXiv 2410.13267) by Shangda Wu et al.; pre-trained on 1.5M ABC-MIDI-text triplets, 101 languages
- · Repo sanderwood/clamp2 (65 stars) with feature-extraction, classification, and semantic-search scripts plus benchmark datasets
- · Symbolic input only (ABC/MIDI, not raw audio); all headline numbers provider-reported; small community