๐ŸŽผ LEGATO: Large-scale End-to-end Generalizable Approach to Typeset OMR

๐Ÿ“„ Paper ยท ๐Ÿ”— GitHub ยท ๐Ÿ“œ License

Optical Music Recognition (OMR) converts images of typeset sheet music into machine-readable notation.

LEGATO is a large-scale, end-to-end model for typeset OMR. It takes a score image ๐Ÿ“ธ and outputs ABC notation ๐ŸŽต directly, which can be converted to MusicXML for further use.

From: "LEGATO: Large-Scale End-to-End Generalizable Approach to Typeset OMR"

  1. Upload image or PDF โ€” Use the uploader below to provide a score image or PDF (typeset sheet music). PDFs are limited to 3 pages and are concatenated into one image.
  2. Click "Run LEGATO" โ€” The model will transcribe the image into ABC notation.
  3. Fix the errors in the editor and check the realtime renderer โ€” Edit the ABC text in the transcription box; the rendered notation updates live. Correct any recognition mistakes and verify the result in the "Rendered ABC notation" panel. You can fix mistakes before generating MusicXML, or after downloading (e.g. in notation software that imports MusicXML).
  4. Download MusicXML โ€” Click "Generate MusicXML" to create the file; a download link will appear below for you to download it.

ABC notation reference: ABC notation standard / grammar

โœจ Try it

Model disclaimer: This is a demo of the LEGATO model. LEGATO works best with typeset, high-quality scores. It is not a perfect model and may not be able to transcribe all scores correctly. Please use with caution.

Privacy: Our team does not store the files uploaded to the demo, nor do we use any of the uploaded data for further model training. Files are processed temporarily on the server to generate the MusicXML output and are then discarded. The demo is hosted on Hugging Face, so processing runs on their infrastructure; we have not configured the tool to save or harvest user inputs.

Example scores