PDF to MIDI converter

Aki's converter reads the printed notes inside a sheet music PDF and exports a MIDI file, so you can open the score in a DAW or notation editor instead of typing every note in by hand. Turning a sheet music PDF into MIDI relies on optical music recognition, a technique that estimates rather than copies, so the result always needs proofreading. The tool is still being built and is not open to users yet.

What is PDF to MIDI conversion?

PDF to MIDI conversion is a machine looking at the notes printed on a page and rebuilding them as note data a computer can play. The technique is called optical music recognition, usually shortened to OMR. It runs in four stages: straighten and clean the page image, find the staff lines and temporarily erase them, classify each remaining symbol as a particular note or marking, then assemble those symbols into voices with pitch and duration along a timeline. What comes out is a MIDI file, which is simply a list of which note sounds when and for how long.

Why is this not a file format conversion?

A sheet music PDF contains pictures of notes, not the identity of those notes. To a computer, a page of music and a photograph of a landscape are the same kind of data: pixels. So no format conversion can turn a PDF into MIDI; something has to actually look at the page and make a judgement. That is why the generic "convert anything" sites hand back an empty or broken MIDI file, and it is why this job is priced per page while converting an audio file between formats is a free audio converter that runs in your browser.

Where does music recognition go wrong?

Handwritten manuscript is the worst case: stroke, spacing and notehead shape differ per writer, so the model misreads constantly, to the point where handwriting should be treated as outside this tool. On printed scores, the recurring failures are several voices sharing one staff, tuplets and cross-rhythms, repeat marks and jump instructions, and dense two-hand piano writing. A scan tilted by a few degrees, or a photo of a curved book page, is enough to bend the staff lines and drive the error rate up sharply. Even under the best conditions the output is an estimate to be checked, not an accurate transcription. This tool shortens the typing, it does not replace someone who can read the score.

How is the price worked out?

The unit is the page, not the run: a 3-page file spends 3 pages of the quota, not one turn. The free allowance isn't published yet, because the recognition engine is still being built (status: soon); once it runs, the weekly free page count for guests and for signed-in users will show right here. The paid lane is already locked in: anything past the free allowance is billed at 10 Energy per page, only available once signed in, with a hard cap of 9 pages per file even on the paid lane. Splitting a long file into several smaller ones does not get around the cap, because the allowance counts total pages, not total files.

What is the MIDI file actually good for?

A MIDI file opens in most DAWs and notation editors, which makes it a starting point rather than a finished product. Learners use it to hear a piece slowly and practise section by section. Arrangers use it to swap instruments, change key, or pull a single part out and build a new arrangement around it. Teachers use it to put together accompaniment for students quickly. In every case you are editing something that already exists, which is far faster than entering an entire score by hand.

How does this compare to typing it in or working from a recording?

Three routes for three different situations. Typing the score into notation software is perfectly accurate if you can read music, at the cost of hours on a long piece. Machine reading is many times faster but returns a draft with errors in it, which suits you when you want a skeleton and are willing to correct it. And if what you have is a recording rather than a printed page, neither route applies: in that case find the chords from the audio instead, because listening to sound and reading a page are entirely different problems.

Who needs sheet music converted to MIDI?

• Piano and guitar learners who want to hear a piece slowly before practising it
• Arrangers who need a ready note skeleton to change instruments or rebuild an arrangement
• Music teachers preparing accompaniment and practice material for students
• Anyone with a shelf of paper scores who wants them digitised, editable and reprintable
• Composers checking a score by having a machine play it back

The tool is not open yet. These are the steps it is being built to have.

1. 1

   Upload the sheet music PDF

   Pick a PDF containing the score. A PDF exported straight from notation software gives the best result; a photo or scan needs to be straight, well lit and at least 300 DPI.

2. 2

   Choose which pages to read

   Mark only the pages that actually carry staves. Cover pages, lyric sheets and programme notes contribute nothing to the MIDI and should not be billed, so leave them out.

3. 3

   Wait for note recognition

   The system finds the staves, clefs and key signatures, then the individual notes and rests, and arranges them into voices in the right order in time.

4. 4

   Proofread before downloading

   Compare the preview against the original, paying attention to the first bar of each system, unexpected accidentals and passages where several voices share a staff. This step is required, not optional.

5. 5

   Download the MIDI file

   The MIDI file opens in most DAWs and notation editors, where fixing a handful of wrong notes is far faster than typing the whole score in by hand. The result is not kept on the server indefinitely, so download it promptly once it is ready.

• Machine reading of notation always makes mistakes. Even on a clean printed score, treat the output as a draft to correct, not as an accurate transcription.
• Handwritten manuscript is the weakest case by a wide margin: stroke, spacing and notehead shape vary per writer, so recognition rates fall sharply.
• The things most often read wrong: repeat marks and jump instructions, tuplets, several voices sharing one staff, and dense two-hand piano writing.
• Lyrics, chord symbols, fingerings and performance directions do not travel into the MIDI file, because MIDI records only notes and timing.
• A skewed scan or a photo of a curved page bends the staff lines and drives the error rate up fast. Shoot square on, or straighten the page before uploading.