MusicBERT

Learn note-level representations for melody, style and other symbolic analysis tasks.

Symbolic musicOpen-source code
Original model card
Start here
Use the documented MIDI preprocessing and start with one analysis task. Keep the public Lakh reproduction separate from the paper's MMD results.
Training data
The paper describes Million MIDI Dataset (MMD) pretraining; the public reproduction guide uses Lakh MIDI (LMD-full), with separate downstream datasets. These are different training setups.
Training rights
Web availability is not a composition-rights clearance. The model's training collection is not a uniformly licensed musical commons.
License & output terms
Muzic code: MIT; model and corpus terms must be checked separately.
Limitations
OctupleMIDI encoding and quantization impose musical assumptions. A pretrained encoder is not a complete MIDI analysis application.
Technique
Masked symbolic transformer / OctupleMIDI
Runtime
Local research inference with preprocessing and downstream task heads.
Parameters
Not itemized for this release/family. Parameter sizes shown elsewhere may exclude encoders, decoders or other runtime components.
Code & reproduction data (opens in a new tab)Paper (opens in a new tab)

Sources reviewed . Training and rights notes summarize provider disclosures. A model license does not clear every source recording or generated output. Parameter counts refer to the named variant or component, not total runtime memory.