Vocal teachers and choral conductors: upload sheet music images on the web and turn them into editable MIDI

This sheet-music-to-MIDI tool (HOMR GUI) is an online skill that recognizes sheet music images into editable note data, for piano, vocal, and choral scores. You give it a photo, scan, or screenshot of staff notation, or a PDF score, and it returns downloadable .musicxml and .mid files.

People who only have paper scores or image scores on hand often have no choice but to type them into notation software note by note in front of a screen; not knowing the command line, and not wanting to wrestle with a Python environment and CUDA just to recognize a passage of music, is the reason this step takes the most time.

Using it is straightforward: hand the sheet music image or PDF to this skill, and it will automatically read the staves, notes, and durations, and output a .musicxml and a .mid file for you to download. PDFs are recognized page by page, with no need to export or screenshot in advance. It reads note by note according to the original score, without altering your melody, harmony, or rhythm, and you don't need to install Python, onnxruntime, or CUDA — use it directly on the web.

Once you have the files, the .mid can be imported into a DAW as a track, and the .musicxml can be imported into notation software such as MuseScore, Sibelius, and Finale to continue editing and engraving. What you get is notes that can be transposed, orchestrated, and turned into accompaniment — not just an image.

Who it's for: vocal teachers and choral conductors who need to turn choral and vocal scores into accompaniment, independent musicians and video creators who want to transcribe music or make practice accompaniment, and music students who want to edit their exam pieces in notation software. If what you want is a performance demo or automatic arrangement, that's not what this is for — it is only responsible for reading the score into notes, leaving score editing and orchestration to your notation software.

For the same task, there are usually three paths: manual notation, letting a general-purpose AI chat look at the image and describe it, or setting up an open-source sheet-music recognition tool yourself.

A general-purpose AI chat can read sheet music and state the key signature, chords, and rough structure, but what it gives you is a text description, not a full-page MusicXML you can import into notation software; this skill uses an end-to-end OMR engine to turn a full page of score into note data — the gap from "this passage is roughly like this" to "this is an editable score" is something only you can fill in.

Manual notation is the most reliable, but you have to type it note by note — a single choral score means hours of data entry; this skill first automatically recognizes the notes, and you only need to check and correct them against the original score — what you save is repetitive data entry, and what remains is the proofreading only you can do.

The upstream homr is a set of command-line programs, where installing the environment, configuring dependencies, and switching between CPU / GPU all require knowing commands; this skill moves it onto the web, so handing it a sheet music image produces a result — even people who don't know the command line can use the same recognition engine.

When to choose it: you have a staff-notation image or PDF on hand, your goal is to get an editable note file, and you don't want to install an environment or learn commands. Conversely, if you just want the AI to talk about the harmony of a passage, or you need a performance recording you can listen to directly, general chat and a real performance are more convenient. One more reminder: automatic recognition itself will have errors, and scores with dense chords and many slurs and markings especially need to be checked against the original.