You can now import PDFs and photographs of sheet music into Soundslice!

Over the past year, we’ve developed our own music-scanning technology that extracts musical information from PDFs and photographs.

It analyzes an image, figures out which parts are notes, what the notes’ pitches/rhythms are, where the bar lines are, and so on — all while trying not to get fooled by skewed photos, bad lighting and crud on the paper.

If all goes well, you get a full-fledged slice — meaning you can listen to the music, edit it, loop it, sync audio/video, add to a practice list and do all of the other lovely Soundslice things. Nice!

We built this because getting music into Soundslice has long been our biggest barrier to entry. Yes, we have a good MusicXML importer, and yes, we have a full-featured sheet music editor — but many musicians still only have stuff on paper or PDF. We want to take you “from zero to slice” as fast as possible, whether it’s for learning, practicing or teaching.

How it works

If you’re in a paid plan, you’ll see a new option “Start with a scanned image” when creating a slice:

Click that, then upload a photo or PDF. You’ll see a “Pending image uploads” section in your slice manager while it processes:

When it’s ready, you’ll get an email with a link to review your upload. In this review screen, our system will ask you some questions about what it detected.

For example, here’s a question the system asked me when it wasn’t sure what side a note’s stem was on:

Here’s another example. The tie is pretty close to the bottom note — so our system is making sure the tie is in fact pointing at the other note.

Some of our questions might seem laughably simple or stupid. This will improve over time, because the system is based on machine learning — indeed it’s the first part of our product that’s powered by AI. Each of your answers helps give the system a deeper, more nuanced knowledge of some very specific aspect of music notation.

You know those “Prove you’re a human” things you see online, where you have to click every bicycle, or identify the street lights? This is like that, except instead of helping a global megacorp build a self-driving car system, you’re helping a small music-education company improve its music-scanning system. (And obviously you’re helping ensure your particular music gets imported as accurately as possible.)

Eventually our system will reach a point where it won’t need to ask you any questions at all. Or perhaps only questions about things that might not have a clear answer or otherwise require human judgment. We’re not sure how long it’ll take to get there, but we’ve already seen the system become significantly smarter during its development.

Anyway, after you’ve completed the review process by answering those questions, we’ll create a slice and immediately bring you into our notation editor. This is our normal editor, with one exception: you’ll see your original upload in the bottom half of the screen:

This lets you cross-reference between the generated slice and your original image, to spot-check for any errors or omissions. You can toggle the display of the original image by clicking “Show image” toward the top of the screen.

From here, you have a full-blown slice — so you can edit it, practice it, share it with students, etc.

Limitations

We’re still actively developing our engine, and there’s a lot it doesn’t yet do. But we’ve decided to go forth and launch it as a public beta, because we think it’s already useful enough for many Soundslice users. And due to the way its AI is set up, the more people that use it, the faster it learns and improves.

One significant thing it doesn’t yet handle is lyrics. And while it already supports tablature, the tab support is limited to standard guitar tuning or four-string bass tuning — and we require tab to have a corresponding standard notation staff (so alas your screenshot of a 1990s ASCII tab will not work).

For more information on the system’s limitations, see our detailed help page, which outlines specifics all the way down to the level of individual notations. We’ll update this page as the system improves.

During our beta period, usage is deliberately limited, while we keep an eye on how things are working. If you’re in the Plus plan, you can do five imports per month. If you’re in the Teacher or Licensing plans, you can do 15 imports per month. Eventually we’ll increase these limits.

For more information on this whole thing, see our new help page. Happy scanning!