We’ve added several features to our PDF/image importer. Here’s what’s new:

A new results page

Previously, after you uploaded an image or PDF, we’d take you to our notation editor. But we came to believe that felt too overwhelming. Now, we take you to a dedicated results page, which is simpler and more streamlined:

On the top, you’ll see the music displayed by our own graphics engine. On the bottom, you’ll see the original image. This lets you quickly see how our importer did.

You can press Play (or press Spacebar) to hear the music played back. You can also click on any area of the notation (in either the top or bottom panes) to move to that moment in the music.

Press the Continue button at upper right for a few options: Practice, Edit and Redo import:

Oooh, what’s “Redo import”? That brings us to the next new feature:

Import settings

There’s a certain kind of music-scanning problem that’s very tedious to fix in a notation editor. For example, our system assumes chord names are always located above the staff, but some publishers put chord names below the staff — in which case our system associates the chord names with the wrong music and fixing it is a huge hassle.

Our new import settings are designed to help with this. It’s a way for you to guide our importer’s behavior.

To access the import settings, click “Redo import” from the Continue menu mentioned above, or click “Problems?” at the top of the results page. Here‘s what it looks like:

Lyrics in the image lets you tell our system how lyrics are positioned in your original image. By default, we assume lyrics are below the staff — but sometimes publishers put lyrics above the staff.

Chord names in the image lets you tell our system how chord names are positioned in your original image. By default, we assume chord names are above the staff — but sometimes publishers put chord names below the staff.

Instrument transposition lets you set the transposition for all instruments we detected in the image. You’d use this, for example, if your music is intended for a B♭ clarinet or if it’s guitar music that’s assumed to be an octave higher than written.

Include the following lets you exclude certain notations from the import: chord names, fingerings or lyrics. This is a handy if you don’t care about those aspects of the music.

After you’ve made changes to these options, click “Update” and we’ll immediately update the preview at the top half of the screen. We’ll also update the slice’s notation.

For more, see our new help page about these import settings.

We have ideas for more settings to add here, and we’d love to hear your own ideas as well. Let us know what you think would be useful!

Ability to ignore notations

Sometimes you just don’t care about certain notations. Fingerings are a good example: perhaps you completely disagree with the fingering suggestions in a PDF you’re scanning.

For this reason, we’ve added a way to ignore stuff. On the upload screen, you can now choose not to include fingerings, lyrics and chord names:

In addition to excluding these notations from the end result, we’ll skip any review questions these notations may have triggered (hence saving you some time).

If you change your mind after the import, no problem. You can tweak the aforementioned import settings to enable the notations after the import is done.

Better handling of multiple voices

Sometimes our scanning system has been too aggressive in deciding music uses multiple voices. This generally happens when notation is spaced very closely together. We’re happy to report this has gotten a lot better now!

We’ve trained a new machine-learning model specifically for the task of deciding whether notes are played at the same time. In our tests, this works much better than our previous approach, which was based on rules we manually came up with.

If that techspeak didn’t make sense, no worries. Just know that our system’s accuracy has improved, especially with tightly spaced music.

You might now encounter a new type of review question, which asks you whether two notes are played at the same time:

We’ll show this question in any case where our system isn’t 100% sure. As with other parts of the system, it will get smarter over time as it sees more music.