Audiveris: Open-Source Optical Music Recognition Explained

Audiveris Workflow: From PDF Scores to Editable MuseScore Files

This article walks through a complete, practical workflow to convert scanned or PDF sheet music into editable MuseScore files using Audiveris (open-source Optical Music Recognition). Steps include preparation, running Audiveris, cleaning output, and importing into MuseScore with tips to maximize accuracy.

1. What you’ll need

• A PDF or scanned image of the score (preferably high-contrast, 300 DPI or higher).
• Audiveris (latest stable release).
• Java Runtime Environment (JRE) compatible with Audiveris.
• MuseScore (latest stable release) to open MusicXML files.
• Optional: image editor (GIMP, Photoshop) for preprocessing.

2. Preparing the PDF/image

1. Scan quality: If scanning, set 300–600 DPI and save as PDF or TIFF.
2. Contrast and cleanup: Increase contrast, remove speckles and skew; use deskew tools.
3. Split multi-page PDFs: If the PDF contains multiple scores or parts, split into logical sections to reduce recognition errors.
4. Monochrome conversion: For best OMR accuracy, convert to clean black-and-white where possible, but keep anti-aliasing off to avoid blurring staff lines.

3. Running Audiveris

1. Install Audiveris: Unzip or install per project instructions and ensure Java is available.
2. Command-line recommended: Use the CLI for batch processing and more control. Example command pattern:

bash
Copy CodeCopied
audiveris -batch -export -outputDir /path/to/output /path/to/input.pdf 

3. GUI option: For single files, use the Audiveris GUI to load pages, run recognition, and inspect results.
4. Export format: Choose MusicXML (usually .musicxml or .xml) as the export format — this interoperates well with MuseScore.

4. Inspecting and correcting Audiveris output

1. Open MusicXML in MuseScore: File → Open → select the exported MusicXML.
2. Check layout and playback: Look for missing accidentals, rhythm discrepancies, or misplaced notes.
3. Common issues to fix:
   • Missing or extra rests/beams.
   • Incorrect note durations (especially with beamed divisions).
   • Clef, key, or time signature misreads.
   • Articulations and dynamics often require manual re-entry.
4. Use MuseScore’s Inspector and Palettes: Adjust beams, articulations, and dynamics efficiently.

5. Iterating for better accuracy

1. Adjust Audiveris settings: If using CLI, try different preprocessing flags or page segmentation options.
2. Preprocess images further: Remove staff lines or isolate systems if staff detection fails.
3. Split staves manually: For complex scores, extract single-staff pages and run separately.

6. Automating batch conversions

• Use shell scripts to loop over PDFs and call Audiveris with consistent options.
• Post-process MusicXML files with scripts (e.g., Python + music21) to fix recurring issues before opening in MuseScore.

7. Tips for best results

• Start with the best possible scan: clean, straight, high-resolution.
• Simple, single-staff, single-voice scores convert most accurately.
• Expect to spend time proofreading — Audiveris greatly speeds conversion but rarely produces perfect scores for complex layouts.
• Keep backups of original MusicXML exports so you can revert if edits introduce errors.

8. Quick checklist (summary)

• Scan at 300–600 DPI, high contrast.
• Preprocess images for cleanliness and straightness.
• Use Audiveris CLI for control and export MusicXML.
• Open in MuseScore, proofread, and correct musical/formatting errors.
• Iterate preprocessing or Audiveris settings for stubborn pages.

This workflow turns PDFs into editable MuseScore files efficiently: clean input, careful recognition with Audiveris, and methodical proofreading in MuseScore are the keys to good results.