PDF to MIDI
How sheet music PDFs actually become MIDI files
PDF to MIDI isn't a simple file conversion — it requires OMR (Optical Music Recognition) to read every symbol on the page, identify pitch, rhythm, dynamics, and repeats, and reconstruct them as MIDI events on a timeline. This page explains how OMR works, what it can and can't read, and why some PDFs convert cleanly while others fail. Use it to decide whether your score is a good candidate for automatic MIDI conversion.
Want sound now, not theory?
Convert a recording instead. Our Audio to MIDI tool is live and works great on MP3 and WAV.
OMR is a three-stage pipeline
- 01
Page segmentation & deskew
The OMR detects staff systems on every page, corrects skew, and measures staff height — establishing the coordinate system everything else depends on.
- 02
Symbol recognition
Deep-learning models classify noteheads, stems, rests, key/time signatures, dynamics, slurs, ornaments, and repeat marks.
- 03
Music reconstruction
Symbols are grouped into voices and time, repeats and jumps are unfolded, accidentals are propagated, and the result is exported as MIDI.
Why this page is worth your time
Decide if your score is convertible
Scan quality, engraving style, and complexity all change the success rate.
Understand why conversions fail
Long ties, voice crossings, and handwritten manuscripts have predictable failure modes.
Make OMR-friendly scans
300+ dpi, grayscale, upright A4, no shadows — these basics matter.
Pick the right tool combination
Know when PureMIDI fits, when MuseScore's built-in OMR is enough, and when manual entry is faster.
What controls PDF-to-MIDI accuracy
- Resolution: 300 dpi is the practical floor; 600 dpi handles almost any modern engraved score.
- Engraving style: Sibelius/Finale font output is the easiest to read; handwritten manuscripts are still hard.
- Voice count: solo, duet, and grand-staff piano are most reliable; 4+ voices need a manual review pass.
- Skew: more than 2° tilt drops staff detection sharply — straighten before upload.
- Background: colored paper, stamps, watermarks, and translucent annotations all interfere with symbol detection.
What our OMR engine actually reads
These elements are written into the MIDI file. Visual-only markings stay in the MusicXML export so you can edit them in notation software.
- Noteheads and stems (whole through 64th notes), dotted notes, rests
- Key signatures and time signatures (including 7/8, 5/4, and irregular meters)
- Dynamics (pp through ff, crescendo / decrescendo)
- Slurs, ties, and ornaments (trills, mordents, turns)
- Repeats (||: :||, D.C., D.S., Coda, 1st/2nd voltas)
- Tempo and tempo changes (rit., accel., a tempo)
- Voice separation (high and low voices on a piano grand staff)
PDF to MIDI vs traditional OMR tools
How PureMIDI compares with the OMR tools developers and musicians already know.
| Tool | Type | Strengths | Weaknesses |
|---|---|---|---|
| PureMIDI | Online + AI post-processing | Free, browser-based, repeats unfolded, MusicXML export | Jianpu and TAB still on the roadmap |
| MuseScore OMR | Free desktop | Free and integrated with engraving | Sensitive to scan quality; high failure on dense notation |
| PhotoScore Ultimate | Commercial desktop | High accuracy on engraved print | $249, macOS/Windows only |
| ScanScore | Commercial desktop + iOS | Snap a phone photo | Subscription, multi-page handling is weak |
| Soundslice OMR | Commercial API | Easy API access, stable accuracy | Per-page billing, self-hosting unfriendly |
When we recommend skipping OMR
If you only need MIDI for one specific piece and a recording exists, audio-to-MIDI is usually faster and more accurate — you skip the hardest stage entirely.
What you should know about PDF to MIDI
What's the difference between PDF to MIDI and PDF to MusicXML?+
MIDI only encodes pitch, on/off, and velocity — it's playback data. MusicXML preserves the full visual and structural score for re-engraving. The OMR effort is the same; only the export differs.
Why do some PDFs come out as gibberish?+
Most online sheet music PDFs are wrappers around scanned images, so the OMR has to re-recognize pixels. True vector engraved PDFs are rare and could in principle be parsed semantically, but you can't tell which is which without inspecting.
Are OMR and OCR the same thing?+
Conceptually similar but OMR is harder. Music is a 2D symbol network where the relative positions encode rhythm itself, so OMR is years behind OCR in accuracy.
Why is piano music easier than a full orchestral score?+
Grand staves have only two staves and few crossing voices. Orchestral scores have 10+ systems with dense vertical alignment, multiplying the difficulty.
Will lyrics be captured?+
OMR identifies lyric text but the MIDI standard has no lyric channel, so they don't appear in your downloaded .mid. They will appear in MusicXML output.
Are repeats and jumps unfolded?+
Yes. PureMIDI will expand D.C., D.S., Coda, Fine, and volta brackets in the MIDI output — a notorious pain point for naive OMR exports.
Why does my own handwriting fail?+
Handwritten symbol variation is huge. Every commercial OMR currently lands under 60% accuracy on handwritten scores. Re-engrave in notation software first.
Will historical or non-standard notation work?+
Modern Western standard notation is the launch target. Square notation, neumes, jianpu, TAB, and shape-note are on the longer-term roadmap.