Multiple recognition engines
StaveWave compares several OMR readings, then builds a MusicXML draft from the strongest results.
StaveWave OMR Studio
Turn scanned sheet music into editable scores you can trust.
Upload a PDF or image score. StaveWave reads it, highlights uncertain measures, and gives you a MusicXML draft ready for careful review.
- 50 MB
- Max file size
- 5
- Input formats
- 100
- Confidence scale
Confidence by measure
Each measure gets a confidence signal so the spots that need attention are easy to find.
Download after processing
You can download the converted MusicXML when processing finishes, with confidence notes available for review.
How It Works
A clearer path from scan to MusicXML.
Instead of treating OMR as one all-or-nothing attempt, StaveWave checks the source, prepares the pages, compares recognition results, validates the draft, and packages confidence notes with the exported MusicXML.
File check
StaveWave checks what you uploaded before any recognition starts. It verifies the file type, size, pages, and image details so weak scans are caught early.
- PDFs are classified as vector, hybrid, raster, or unknown from their fonts and image data.
- Images are checked for frame count, resolution, and basic readability.
- Each job gets a readiness score, quality notes, and a recommended path.
Preparing the pages
The system creates clean recognition inputs. Raster pages are turned into several versions so the recognizers can compare independent reads.
- Pages are normalized for staff detection, deskewing, dewarping, and symbol contrast.
- Staff checks catch unusual page geometry before it turns into notation.
- Vector-like PDFs can take a faster path, while scanned pages go through OMR voting.
Reading the notation
Each recognizer produces a possible MusicXML draft and supporting evidence. StaveWave compares the results and tracks where the engines agree or disagree.
- Audiveris, homr, and oemer use the same run format across source and preprocessing variants.
- If an external engine fails, the error is logged and a fallback keeps the job reviewable.
- Elements keep measure, staff, voice, beat, confidence, source, and optional location data.
Checking the draft
Before release, the draft is checked from several angles. Any issue becomes a clear finding tied to the measure it affects.
- MusicXML schema, rhythm, and structure checks look for internal notation problems.
- When source images are available, visual diff compares the rendered draft against the original page.
- Language, specialist, corpus, and critic checks look for unlikely notation and missing symbols.
Confidence and decision
StaveWave turns the evidence into score-level and measure-level confidence, then decides whether the job is ready to review or needs closer attention.
- Medium and high severity findings lower confidence for the measures they touch.
- Alternative readings are kept so reviewers can see why a measure was flagged.
- Jobs finish as ready for review or needing human attention.
Export package
The final package includes the editable MusicXML file and the supporting files needed to understand the conversion.
- final.musicxml contains the editable score for Sibelius, MuseScore, Dorico, and Finale.
- confidence-map.json stores measure confidence, verdicts, findings, and review state.
- audit-log.json stores pipeline version, cache events, layer evidence, and stage timings.
Supported Inputs