MAEST
Music Audio Efficient Spectrogram Transformer; patchout-trained music representations.
About
MAEST (Music Audio Efficient Spectrogram Transformer) brings the transformer era to music tagging without the usual GPU bill. Building on the patchout training trick from PaSST, it trains DeiT-initialized spectrogram transformers on Discogs-annotated audio, and the paper reports that the middle transformer blocks hold the best representations, not the final layer. The ISMIR 2023 paper also claims the best tagging performance among open models on MTG-Jamendo and MagnaTagATune.
The original repo (palonso/MAEST) ships ten pre-trained checkpoint variants (5s to 30s inputs), and the checkpoints are also on Hugging Face under mtg-upf; a community repackaging (maest-infer, OpenMIRLab) strips the research scaffolding for clean inference. Note the licensing split: the research code is AGPL-3.0, so check terms before building it into commercial products, and the MTG-Jamendo results are provider-reported. For a music-specific transformer embedding that runs on reasonable hardware, MAEST is one of the most practical options.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported: best open-model music tagging on MTG-Jamendo and MagnaTagATune per the ISMIR 2023 paper; 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 (verified, metrics-v1). Evidence: ISMIR 2023 paper (arXiv 2309.16418) from MTG-UPF; original repo palonso/MAEST with public checkpoints; HF org mtg-upf; independent maest-infer repackaging verifies bit-identical inference. Strengths: practical music transformer, multiple input lengths, real model zoo. Docked for: (1) small community (72 stars); (2) AGPL-3.0 code license complicates commercial use; (3) headline results are provider-reported. License: AGPL-3.0 (code).
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 4.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 7.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · ISMIR 2023 paper (arXiv 2309.16418), Alonso-Jimenez/Serra/Bogdanov, MTG-UPF: patchout-trained DeiT-init spectrogram transformers on Discogs-annotated audio
- · Repo palonso/MAEST (72 stars) with 10 checkpoint variants (5s-30s); HF org mtg-upf; OpenMIRLab maest-infer wrapper verified bit-identical
- · AGPL-3.0 code license complicates commercial use; tagging results are provider-reported; small community