Install & use · 繁體中文
Turn PDF sheet music into files MuseScore can open. Drag a PDF in, every page is recognised, and multi-page scores are stitched into one piece.
Four things to check. If they all hold, carry on.
The two installers are for different machines, and the wrong one will not open. Apple menu → About This Mac, look at the “Chip” or “Processor” row.
The warning in step 3 is expected, and does not mean anything is wrong with the file — this app has no paid Apple certificate, so macOS blocks it once by default. You only have to do this the first time.
Double-click it in your Downloads folder. A window opens: drag Pdf2Score onto the Applications folder beside it.
Do not double-click for this first launch. Right-click Pdf2Score (or Control-click it) and choose Open from the menu.
You will see something like “Apple could not verify … it may contain malware”. Click Open.
If the dialog only offers “Move to Trash” and “Cancel”, with no Open:
Click Cancel, then open System Settings → Privacy & Security and scroll to the bottom. A line says Pdf2Score was blocked — click Open Anyway next to it and enter your password.
The app opens to an empty window, waiting for PDFs. Nothing to configure, no account, no download to wait for — everything it needs is already inside.
Three moves: drop it in, press Convert, open the result in MuseScore.
As many at once as you like, or a whole folder — it finds every PDF inside. The Add PDFs button in the toolbar does the same thing.
About 5 seconds per page on Apple Silicon (around 15 with Audiveris); Intel Macs are slower again. You can press Stop at any time. One file failing does not affect the others — that row turns red and the rest carry on.
Each PDF produces an .mxl of the same name
in the folder the PDF came from. Double-click it to open
it in MuseScore, or right-click the row and choose Show in Finder.
When a result comes out badly, switch engines before you touch anything else — it helps more than any setting.
A neural model. Copes better with photographs, skewed scans and soft printing. A little faster.
Classical structural analysis. Very steady on clean, regularly engraved scores, and it reads multi-page pieces as one book itself. A little slower.
The engine switch is in the toolbar, right next to Convert. Pick one before converting.
For a file that has already been converted, right-click its row → Re-recognise with …. That runs the same PDF through the other engine and writes a second file, so you can compare them.
Neither is always better. On the same score the two may differ by a bar or two, or by a lot — compare them and decide which one to proofread.
Optical music recognition is never perfect. That is the state of the technology, not a fault in this app. Cleanly printed scores usually come out well; skewed scans, repeated photocopies, handwriting and dense piano writing produce noticeably more wrong notes and rhythms.
Think of it as having typed most of the notes for you. Check it against the original in MuseScore and fix what is wrong — still far faster than entering the whole thing by hand.
The defaults are fine; these are the ones you might want to change. Open them from Pdf2Score → Settings in the menu bar.
By default results land next to the source PDF. To collect everything in one place, switch to “A folder I choose”.
Only meaningful on Apple Silicon; leave it off on Intel. The first run has to compile the model, so it pays off across many files rather than one.
300 dpi by default. If a scan looks soft, 400 dpi may read it better, at the cost of speed; 200 dpi is faster when you are in a hurry.
On by default: one PDF, one file. If a page's part layout does not line up with the others, it writes the pages out separately and flags the row rather than forcing together a score that would be wrong.
That is how macOS treats an app without a paid Apple certificate; the file itself is fine. Follow install steps 2 and 3: right-click → Open, or System Settings → Privacy & Security → Open Anyway.
On the first launch after installing, macOS checks every file inside the app. This one bundles two complete recognition engines, so there are a lot of files and that check can take several minutes. It only happens once. “Starting the recognition engine…” in the queue means it is working, not stuck.
Leave the app open for a moment and it finishes this in the background; conversions after that run at normal speed.
Normal speed is about 5 seconds per page (around 15 on Intel).
Click that row and the Processing log below it expands with the reason. The two common ones:
No noteheads found — the page may be a cover, a table of contents or a lyrics page, or the scan may be too faint. The whole file failed — check the PDF really is sheet music and not a heavily compressed image scan.
Trying again at 400 dpi is worth a shot.
When one page is recognised with a different number of parts than the others (a page misread as two staves, say), forcing them together would produce a garbled score, so the pages are written out separately and the row is flagged.
You can join them by hand in MuseScore, or change the resolution and try again to see whether that page reads correctly.
Start with the quality of the PDF. A photograph or a low-resolution scan makes a large difference — get a cleaner copy if you can. Skewed scans can be straightened in Preview first.
If the original is already clean and the result is still poor, that score is beyond what the current engines handle. Try the other engine before giving up.
Check you have MuseScore 4 (musescore.org,
free). .mxl is the standard compressed MusicXML format, so
Sibelius, Finale and Dorico read it too.
No. The recognition models and both engines are inside the app; everything happens on your machine and no network connection is needed at any point.