Scanning the current OMR landscape

 

Optical music recognition (OMR) software — or “music scanning software” as it is more casually known — has a wide audience, appealing to hobbyists, professionals and everything in between. Like every other corner of the music notation software universe, there seem to be endless opinions about the best available options, dependent upon personal preferences and use cases.

Moreover, since we last visited this topic in early 2021, the technology and the products have significantly changed. Four years can seem like a lifetime or more in this area! No doubt, these products will continue to evolve, and others will come and go . So, while this article is a review of some of the most widely used options, it is not intended to rate what the “best” OMR product on the market is — only to report on how the different products performed on a specific set of tests, what to expect as a user, and observations about the experience and results.

I tested the following products (click on the link to be taken straight to the review):

Newzik
Soundslice
PlayScore 2
Sheet Music Scanner
SmartScore 64 Pro 64 NE
ScanScore Professional

Notably not included here is PhotoScore & NotateMe Ultimate 2020 by Neuratron, since it has not been updated since the 2021 Scoring Notes OMR review.

A convergence of technologies and approaches

Because there is so much to cover, I limited most of my comments to high-level observations are made on the results. If you wish to make further comparisons, I have posted all the files from the results for you to explore further as much as you like.

Some applications I tested include an editor, but this functionality is not included as part of the evaluation. Some of these applications allow you to change settings to look for specific elements in the scan as well, but the focus of this review is to see how well the different OMRs do without having to select any specific options before conversion.

These six products I tested can be grouped into three categories of two each:

Machine learning-based:

• Newzik
• Soundslice

Mobile applications:

• PlayScore 2 (also available as a Windows application)
• Sheet Music Scanner

Desktop applications:

• SmartScore Pro 64 NE
• ScanScore Professional

This is hardly an exhaustive list of products on the market, but they are representative of the different approaches taken to the challenge of taking an existing piece of sheet music and generating results. Each category exhibited some general advantages and limitations.

The machine learning applications generally performed the best “out-of-the-box”, with little or no configuration required. However, they took a relatively long time to process scans as they worked their “magic” behind the scenes with considerable processing power required.

The mobile applications were the quickest and most convenient — nothing beats grabbing your phone, firing up an app and getting to work — generating usable results that could be played back nearly instantaneously. But this generally came at the expense of accuracy and configurability.

The desktop applications were the best at providing the user with the ability to customize scanning settings as well as edit results within the scanning software environment after processing, before subjecting them to MusicXML transfer, although as I just mentioned, I did not include these features in my testing process.

In essence, this review will most directly appeal to the user who wants to get MusicXML from the OMR applications as accurately and quickly as possible, export them to their own scoring software of choice (Dorico, Sibelius, MuseScore, etc.) and does not want to spend any further time learning about any specific settings within the scanning application itself. Like many other great products, some of these applications can benefit from a user becoming seasoned in it, while other products are designed to require no learning curve in order to operate to their full potential. The goal here is not to judge one approach or the other, but to highlight this common use case and guide users accordingly.

General remarks, and imagining the future

The OMR industry may not be in its infancy, but is still perhaps in kindergarten, despite the fact that commercial OMR products have existed since 1991. The big things like simple rhythm and note recognition are quite solid, with only an occasional hiccup. Anything beyond that, though, may or not be captured correctly depending on the situation and app you are using. Like a precocious and eager kindergartner, the app may initially dazzle you with its brilliance, but then smack into a wall, become confused, and start crying for help, and all bets are off as to what happens next. (Or in the case of using OMR apps, it may be you, the user, that will become confused and crying for help.)

Taking our analogy a bit further, all of the applications reviewed here — like a class of kindergartners — all can excel at certain things, but they each have their unique limitations and challenges. Once we progress beyond the simpler tasks, it was a mixed bag of results. Some applications were better than others, and you’ll find my specific conclusions within each section.

That’s not to say there isn’t promise here as the youngsters mature into adolescents. Ten years ago, I found OMR technology almost useless for just about everything I needed to do with it, but we’ve come a long way since then. That is no longer the case now, and the collective growth of this small industry is quite impressive.

We may finally be at an inflection point with OMR where more rapid change is on the horizon. OMR software used to assume the presence of a big, bulky scanner, a fairly powerful desktop computer, and the need for the OMR software to include lots of inbuilt editing options that resembled notation software.

That’s no longer the case. With the rise of machine learning (AI) and the ubiquity of mobile devices in nearly everyone’s pocket that can take high-quality scans, the paradigm has shifted — and with it, the goal of the technology itself. As the mechanics of recognition gets better every year, an OMR software’s value will start to become more about creating a holistic understanding of reading notation, as opposed to a strict digital representation of an image.

This means not only seeing there is a text field and figuring out the characters to display, but understanding what it is saying to do — the semantics. An example of this could be finding “accel.”, understanding it is tempo related, and trying to find a place where the music becomes steady again to create an end to the gradual increase in speed of the tempo. Another example would be seeing “3X” and understanding to change the playback instructions to repeat a section three times.

MusicXML

You may notice the screenshots below showing the initial import of MusicXML often don’t look that great. As a disclaimer for anyone not regularly moving notation around using MusicXML, this medium will always lose a lot of formatting data in the transfer process, especially since each notation application has its default way of interpreting and rendering certain elements. This is not to say that well formed MusicXML won’t make much of a difference — it very much matters.

Looking to the future again, an OMR application may try to detect music or text fonts, and the positioning of any element with more specificity. With the help of improved adoption from developers of the full capabilities of both the import and export formatting functionality found in MusicXML (or perhaps MNX in the future), in time we may see detailed engraving and positioning information transferred in the process as the norm and not the exception.

Testing content

I chose three different test files, all from the Classical or early Romantic periods of concert music. Naturally this testing set only scratches the surface of possibilities, but with six applications included in this overview, it was designed to limit the focus of this study primarily to music, and not to text like lyrics and chord symbols.

The test files:

1. Four pages of a Trumpet part from the first movement of Beethoven’s Symphony No. 3 (“Eroica”). It is relatively simple in terms of how complicated the elements of the notation are. It includes cue notes, multirests, repeats, and endings.
2. Three pages of a solo piano arrangement from Beethoven’s “Moonlight Sonata”.  Triplets are both labeled and unlabeled in the arrangement.
3. Five pages of a conductor’s score from Tchaikovsky’s “Nutcracker Suite” for Full Orchestra. The score is transposed, and several parts are condensed.

I tested the above pieces in two ways: “scanned” in via taking a picture with an iPhone 15 of the music in printed form; and as an import of PDF (originally created by an export of the music directly from Sibelius). The printed versions were printed using a standard HP home printer, not a professional grade printer. This limited the test to specific “clean” sources that could be easily controlled, and thus I did not test some of the many “real-world” scenarios one encounters, like scans of published editions, handwritten music, poor scans with pixelated or askew content, or other prints that may have been annotated or otherwise marked — elements that could well affect the results.

For desktop software options that did not offer taking a photo within the software, I imported a single PDF of the pages that I scanned using the iPhone using the app JotNot Pro which includes some modest processing of the image for optimizing contrast. For mobile apps supporting using the camera directly within the app, this option was used. Dedicated scanners will allow for better results, so this is an intentional effort to use a lower quality process with a phone camera, as it is the most widely available method of scanning a printed copy.

For each product and most examples, I summarized my findings into observations, a heuristic collection of my impressions of the results; and a conclusion about whether or not using the product on the given example would save the “average” user time by using it, as opposed to entering the music from scratch in music notation software.

Without further ado: The reviews!

---

Newzik

Newzik
Official web site
Price: $49.99/year or $149.99/lifetime (Premium version)
Platforms: iOS; Web

As a sheet music reader, Newzik is known as an alternative to forScore. But in 2021, Newzik launched LiveScores which converts the selected piece in your library to an interactive score that can be listened to and exported to MusicXML.

Subscribers to Newzik Premium get unlimited LiveScore conversions of scores in a user’s Newzik library.

Experience

Importing a PDF or snapping a picture into Newzik is easy, and connects to the core function of the Newzik’s reader functionality. Taking a picture of printed music reminds me a bit of scanning in a check when using remote deposit in certain banking apps, where as soon as Newzik has the image optimized to its liking, it takes the picture automatically for you. It’s a nice feature that makes it easier to take the picture and ensures the picture is not blurry in the process.

Newzik uses a machine learning approach to its recognition, which currently requires a lot of processing power. Results generally take a few minutes to get back.

Results

Symphony No. 3 – Eroica – Trumpet

Without requiring me to do anything, Newzik on its own understood that the part it reviewed was transposed and exported it as concert pitches while indicating that part was transposed in the MusicXML. The score imported in E-flat major, and when viewing as transposed view, displays up a step as seen in the PDF.

Observations:

• Struggled with cue notes, bringing in a mix of cue notes and grace notes, requiring a re-write.
• Multirests were handled with 100% accuracy.
• Recognized opening tempo well and creates usable MusicXML that converted perfectly.
• Recognized notes and articulations well.
• Missed fortissimo dynamic markings, rehearsal marks and endings within repeats.
• This example was over 650 measures long and it accounted for every measure perfectly.
• Some text anomalies showed up along the way that would need to be deleted.

Conclusion: Cleaning up this scan would ultimately save you a significant amount of time over entering it manually.

Moonlight Sonata

Newzik did well interpreting the missing triplet indications correctly. There were a few differences in the scanned version versus the PDF import, but not many. In this example, Newzik detected a certain note correctly from the scanned file, but not from the imported PDF (highlighted in green below).

Observations:

• Did well with inferring triplets, but got tripped up more when adding a second voice to a measure.
• When processing an imported PDF, it sometimes mistakenly added a note to voice 3, but this did not happen when the music was scanned.
• Processed cross-staff beaming on an implied triplet correctly.
• Notes were processed slightly more accurately in the scanned version.

Conclusion: Whether using a scanned version or a PDF import option, cleaning up the MusicXML imported from Newzik would save you significant time over manually entering it.

Nutcracker Suite

This example is the most challenging. The staff size is quite small in the printed and scanned version, and with multiple key signatures and many instruments, there is a lot to make sense of. Even so, surprisingly, Newzik processed the scanned version slightly more accurately than the imported PDF.

Observations:

• Measure 5 in the Bassoon and measure 4 in Violin 2 were handled better in the scanned version.
• Depending on the measure, Newzik struggled in both versions with different things in instances where the Clarinet, Horn and Trumpet parts have more than one note presented simultaneously.
• The scanned version was better at assigning the correct noteheads.
• Both versions completed all 25 measures with no major breakdowns.
• Correctly identified all transposing instruments and understanding the concert key of G Major. This information was transferred in the MusicXML data, correctly allowing playback and display to work perfectly.

Conclusion: The net result would be a savings of time cleaning up versus entering everything in again.

---

Soundslice

Soundslice
Official web site
Price: $5/month for a single user; plans for teachers and group licensing also available
Platforms: Web

You may know Soundslice as a web-based notation viewer, editor, and generally innovative application. It is particularly good at syncing notation to audio and video, and is considered one of the best practice tools for musicians. Soundslice added OMR functionality in late 2022 and has been refining the machine learning based recognition engine ever since.

You can create a “slice” from a photo or PDF with a subscription of Soundslice Plus which is $5 per month, or $50 per year, and allows you to scan 100 pages per month. You are able to try out the OMR functionality as part of the free plan as well, with access to scanning up to two pages a month.

Experience

As with Newzik’s LiveScores, the machine learning-based approach to OMR requires significant processing power, and generally takes 10-30 seconds per page to process. You receive an email when it finishes processing, or you can refresh the Overview page to see if it has finished processing.

Soundslice does not offer any kind of specific assistance in creating a scan, but allows you to simply use any picture or PDF you have as a starting point. For the scanned version that was tested, I used a PDF created using JotNot (as mentioned at the beginning of the article).

Soundslice uses a unique approach that does not require you to configure any settings for optimizing the scan. Instead, it asks you to clarify items in which its confidence is lower than the threshold for automatic conversion. This approach is simple to use, and helps improve the accuracy of the results over time.

After conversion, you can clean up your scan using the editor. For the purposes of this test, though, no cleanup was done using the editor before exporting to MusicXML.  You can see what features are currently supported by the PDF import functionality here along with guidance to resolving issues in the editor.

Results

Symphony No. 3 – Eroica – Trumpet

Simply put, Soundslice was the most accurate scan of any service on this example.

Observations:

• Soundslice recognized cue notes perfectly, but did not set them to cue notes when opening the MusicXML in other software.
• Multirests were handled with 100% accuracy, with a caveat: I corrected four multirests in the process of verification.
• Soundslice did not recognize the opening tempo.
• Soundslice did not determine on its own that this part, while written in F Major, is actually in Concert E♭ Major.  This can be achieved by manually updating using this process after the scan.
• Recognized notes, dynamics and articulations well.
• No hairpins or rehearsal marks were included in the recognition.
• The left repeat was placed in the correct place, but the right repeat endings were not in the correct place.
• This example was over 650 measures long, and it accounted for every measure perfectly.
• I was only able to find one error related to a wrong note or rhythm, which was found in both the scanned and imported PDF.

Conclusion: Using Soundslice on this piece would leave you very little cleanup work to do and would result in an enormous savings of time over manually entering it.

Moonlight Sonata

Soundslice was able to interpret the implied triplet indications tested in this example, but occasionally drops them as being triplets here and there.

Observations:

• Soundslice did well with inferring triplets, but occasionally processes them as normal eighth notes resulting in an overfilled measure.
• Processed notes well in cross-staff beaming, but you would need to clean up results to present it as seen in the original.

Conclusion: Using Soundslice with this score would save you significant time over manually entering everything.

Nutcracker Suite

Not surprisingly, the scanned version caused the most issues, as the staff is quite small once printed and then scanned using a phone.

Observations:

• The imported PDF version completed all 25 measures included in the sample.
• The imported PDF did an excellent job with identifying grace notes and slurs.
• The imported PDF did well displaying percussion notation.
• Scanned version has a number of problems that made it difficult to be used.
• Notes in the Contrabass are shown up an octave.
• Results were displayed with the right key signatures, but had some issues with transposing.
  • The results imported with the correct key signatures and transposition information, but were not spelled out just right in the MusicXML to work as smoothly as it should.
  • Because there is no transposition data in the MusicXML for each part, if you select transposed view in Sibelius, the results are updated again (as seen below) as it assumes it has been given concert pitches.

Conclusion:  The scanned version is not usable, but the imported PDF can be edited quickly in order to produce very usable results.  Using Soundslice would save you significant time over entering manually.

---

PlayScore 2

Playscore 2
Official web site
Price: $6.99/month or $49.99/year (Professional version)
Platforms: iOS, Android (Google), Windows

PlayScore 2 was first developed over a decade ago and has been refining its notation recognition ever since. This app is available on iOS, Android, and, more recently, for Windows users via the Microsoft Store. A PlayScore 2 Professional subscription allows you to process large scores and export MusicXML and is priced at $6.99 USD a month or $49.99 USD annually. The app has made significant updates since it was last reviewed by Scoring Notes in 2021.

Experience

PlayScore is the fastest of all the apps tested for conversion time: as soon as you take the picture, you can hear it play back. The app is very intuitive, and allows you to easily change tempos and make useful changes after conversion, like adjusting playback instrument sound and transposition.

PlayScore is popular among vocalists in choir settings. To this end, it has implemented an easy tool called “Split staves” to separate out voice 1 and voice 2; or, if there are two notes together in voice 1, it will play back the top note. This is helpful for vocalists wishing to hear their part, for example, in instances where a Soprano and Alto part share a staff.

While the quality of MusicXML generated from ScanScore 2 is generally quite good, it is intended to be used as a free standing application and not primarily a vessel to produce MusicXML for use elsewhere.

Results

Symphony No. 3 – Eroica – Trumpet

PlayScore will not attempt to automatically assign instruments when you scan in a piece, and will assume you are seeing concert pitches with piano unless you tell it otherwise. You can change the settings on a staff within PlayScore 2 to hear the part play back with the correct transposition and instrument sound, but that information will not make it into a MusicXML export.  If you are planning on cleaning up a scan for use in the notation software of your choice, don’t bother changing any settings in the app and just go right to exporting it to MusicXML.

Normal notes and dynamics are generally recognized accurately. But if you’re working with multirests, you’ll have some cleaning up to do in order to match up your measure numbers.

Observations:

• No title information came through.  You can however in the settings elect PlayScore to recognize text which defaults to off in order to maximize the speed of processing.
• Cued notes notes came through as normal notes.
• Cued notes were recognized better in the scanned version with the camera versus the imported PDF.
• Cued notes had more recognition issues than normal notes.
• Multirests at the beginning were not recognized by PlayScore 2.
• No tempo data came through.
• No instrument name or transposition data included by default.  You can update this manually in the Play and MIDI settings after import and it will play back correctly in app.  Any update to transposition does not translate to a MusicXML export however.
• Non-cued notes and dynamics were recognized well.
• All crescendos needed to be deleted.
• No repeats information captured.

Conclusion: Cleaning up this scan would save you some time over entering it manually.

Turning on “Auto transposition” in the settings does not fix the transposition of the MusicXML automatically as it is designed for use with a score that contains different instrument transpositions where it can identify the different key signatures and reconcile transpositions from that (as seen in the Nutcracker below).  In the case of a single staff using a transposing instrument, it must be handled post MusicXML import.

Moonlight Sonata

PlayScore 2 did well interpreting the implied triplet indications correctly — one of the main tests of this piece.

Observations:

• Did well with inferring triplets, but got tripped up more as you add in a second voice to a measure.
• The time signature was analyzed differently in the scanned and PDF imported versions.
• Processed cross-staff beaming on an implied triplet correctly.
• Processed all the notes with 100% accurately.

Conclusion: Using PlayScore 2 with this score would save you significant time over manually entering everything.

Nutcracker Suite

As already noted, this example certainly pushes the boundaries of any scanning application. The size of the staff once printed is quite small, and this adds to the complications with the scanned version. With the PDF import, PLayScore 2 did better, but there was a significant amount of cleanup required.

Observations:

• The imported PDF version completed all 25 measures included in the sample.
• No instrument and transposition information included in the MusicXML file.  Please note: In PlayScore 2, this situation has a setting that by turning it on fixes this however.  By toggling on “Auto transposition” the app fixed all transposing errors and labeled the MusicXML transpositions correctly when imported.
• Major issues were often found around measures that include grace notes.

Conclusion: When using the default settings in the app, the scanned version is not usable, but the imported PDF while requiring significant clean up, would save you some time over manually entering everything.

Pro tip: If you toggle on “Lyrics and text” and “Auto transposition” in the settings, the results on the PDF import of this piece are quite good and it requires very little cleanup.  All the transpositions are handled perfectly in this case and it finds text not seen before as well now saving you significant time.

---

Sheet Music Scanner

Sheet Music Scanner
Official web site
Android: $4.49 (one-time purchase, but switching to a subscription model soon)
iOS: $4.99/month or $22.99/year (with free 7 Day Trial or $17.99/year with no trial)
Platforms: iOS, Android (Google)

Sheet Music Scanner is an iOS and Android app that allows you to scan with your phone, import photos, or import other files to process.  Sheet Music Scanner is available for $4.99 a month or $22.99 per year.

Experience

The app features a clean design and is easy to use. It is meant to provide an intuitive experience that does not overwhelm you with options and allows you to use it without any prior experience. Like all of the other OMR applications, in general the results will not be good if the image is blurry or difficult to work with, so there is a short help section mostly dedicated to troubleshooting issues that may occur with problematic images.

Sheet Music Scanner mentions transparently that it does not yet support symbols like grace notes, dynamics, and other features in their App Store description which is validated in this testing.

The scans happen quite fast — generally within about 10 seconds for a larger piece. The goal is to allow you to hear the playback of what you’re looking at, with the option to easily move it into your software of choice.

Results

Symphony No. 3 – Eroica – Trumpet

Sheet Music Scanner did well with the scanned version, and slightly better in its interpretation of the PDF import. While no transposition is inferred from it being a B♭ Trumpet, it will export the MusicXML as a transposing instrument if you manually edit the instrument selection.

Sheet Music Scanner did well fully determining the repeat structure, showing left and right repeats in the correct places, as well as where the first and second endings were. The presence of multirests appeared to affect the measures around it. At times, the music would suddenly go from three beats per measure (the correct number) to one, as seen here:

Observations:

• Title information on the MusicXML was tied to the file name, versus what is shown on the scan itself.
• Cued notes came through as normal notes.
• Multirests were not recognized by Sheet Music Scanner.
• Tempo was not recognized by Sheet Music Scanner, and only reflected what the tempo defaulted to in the app unless manually updated.
• No instrument name or transposition data was included.
• No dynamics were recognized.
• Information about repeats were captured well.
• Occasional anomalies with incomplete measures.

Conclusion: Cleaning up this scan would save you some time over entering it manually.

Moonlight Sonata

Sheet Music Scanner was able to interpret the implied triplet indications tested in this example in both the scanned and imported PDF examples, but had trouble with cross-staff beaming and accidentals in the scanned version when with more than one voice.

Observations:

• Did very well with inferring triplets.
• Did not process cross-staff beaming as written, and lost implied triplets in the process. Caused instability in the results in the measure and sometimes the measure immediately following.
• Scanned version struggled with interpreting accidentals in multiple voice settings, correctly generally only handling one or the other.
• Interpreted notes using mid-measure clef changes correctly, but did not include the clefs themselves in the MusicXML when found mid-measure.

Conclusion: Using Sheet Music Scanner with this score would save you significant time over manually entering everything.

Nutcracker Suite

When scanned in using the app, Sheet Music Scanner gave an honest assessment that the resolution was too small to work with. This is a very reasonable and fair conclusion, since the staff size is quite small. However, the PDF import worked well.

Although there is plenty of cleanup work to do here, with a few instrument changes and transpositions, this scan is actually quite usable.

Observations:

• The imported PDF version completed all 25 measures included in the sample.
• Sheet Music Scanner struggled to identify grace notes and slurs.
• Did not identify any articulations.
• Results were displayed with the right key signatures, but not the right instruments or transpositions.
• Other than with measures containing grace notes, the recognition of notes was good.
• About 10-20% of the measures would need to be re-entered in again from scratch.
• The MusicXML import was a little bumpy coming into Sibelius. Sibelius complained it was not valid MusicXML, but pushed through and opened.

Conclusion: Cleaning up this scan would save you some time over entering it manually.

---

SmartScore Pro 64 NE

SmartScore Pro 64 NE
Official web site
Price: $399, or $199 (perpetual license) with proof of purchase of Finale or Dorico
Platforms: macOS, Windows

SmartScore Pro 64 NE is a desktop application available for Mac and Windows that contains both a music editor component and an OMR component. It is available for a one-time purchase of $399, or $199 with proof of purchase of Finale or Dorico.

Experience

The NE in SmartScore 64 NE stands for “New Edition”, and represents an overhaul of its recognition system introduced since the last review on Scoring Notes in 2021. You should approach a purchase using SmartScore as an investment in music notation software where you will want to learn your way around in order to get best results.

When you first open the program, you are prompted with a Getting Started section and are encouraged to learn the keyboard shortcuts and explore the detailed documentation when needed.

If you are using a scanner, SmartScore helps you walk the process, and helps you optimize the image you are working with before getting into the recognition portion.

Once your file has been converted, you are given a system report and then able to compare the conversion with the original file.

From here, you are encouraged to use the editor tools provided to make updates to the processed file, or you can export it to MusicXML for cleanup elsewhere. The editor is quite robust, offers you a lot of options catered to the cleanup process, and is well documented.

While adjustment of import settings and in-app editing functionality were not part of this review, you can find out more about SmartScore’s great editing tools in the previous Scoring Notes article on this topic as well.

Results

Symphony No. 3 – Eroica – Trumpet

SmartScore was able to recognize the pitches on the staff well, but struggled with many other aspects.

Observations:

• Title information comes is as scanned, versus using a file name, which is nice.
• SmartScore struggled working with cued notes, and many measures were overfilled.
• Multirests were sometimes recognized correctly, and other time not.
• Tempo text was recognized, but not the metronome mark portion.
• SmartScore did not interpret the transposition of a trumpet automatically. This can be done manually before export.
• Assigned to Guitar in the MusicXML file which displays this way in the import.
• Repeat information is not captured.
• Anomalies with incomplete measures.
• Slightly better results with the PDF import versus scanned version.
• Text was often duplicated.

Conclusion: Cleaning up this scan would probably not save you time over entering it manually.

Moonlight Sonata

SmartScore was not able to interpret the implied triplet indications tested in this example in both the scanned and imported PDF examples. This caused alignment and playback issues throughout the piece. On the positive side, SmartScore can interpret cross-staff beaming, and accidentals across multiple voices were handled well.

Observations:

• SmartScore did not infer triplets.
• Cross-staff beaming is supported.
• Accidentals were handled well.
• Interpreted and displayed mid-measure clef changes correctly.
• Consistent results across the scanned and imported PDF versions.
• Good note and accidental recognition.

Conclusion: Because of the implied triplets not being recognized, using SmartScore with this score would not save you time over manually entering everything.

It should be noted that if you start fixing missing triplets using the editor within SmartScore, the system will make it easy for you to fix them all suggesting “Do you want to apply this tuplet to similar note groupings?”.  As mentioned earlier, learning what tools are available to you within an application can help you get the most of using it and this fix would improve the results before export significantly.

Nutcracker Suite

When using the scanned version in SmartScore, the smaller staff size and lower resolution understandably led to issues, but the imported PDF worked quite well.

While there is still plenty of cleanup work to be done, the notes were interpreted accurately when not combined with grace notes, and allows for a solid starting point.

Observations:

• The imported PDF version completed all 25 measures included in the sample.
• SmartScore struggled to identify grace notes and renders surrounding notes unusable.
• Slurs were identified well.
• Identified some but not all staccato markings.
• Results were displayed with the right key signatures, but not the right instruments or transpositions.
• Other than measures containing grace notes, the recognition of notes was good.
• About 20-30% of the measures would need to be re-entered in again from scratch.
• The MusicXML import was a little bumpy coming into Sibelius. Sibelius complained it was not valid MusicXML, but pushed through and opened.

Conclusion: Cleaning up this scan would save you some time over entering it manually.

---

ScanScore Professional 3

ScanScore 3
Official web site
Price: $79 for a one-year license (Professional version)
Platforms: macOS, Windows

ScanScore is a desktop application available for Mac and Windows that contains both a music editor component and an OMR component. A one-year license for ScanScore Professional costs $79.

Experience

ScanScore can work with your home scanner as a starting point, but can import images and PDFs as starting place for the conversion.  There is also a mobile app that can allow you to use your camera as a starting point for the scan. I was able to get it set up and connected to my license, but unfortunately received an error message every time I tried to use the app.

The desktop application itself is easy to use. Getting a PDF imported and finding where to export it to MusicXML was intuitive. This review is not exploring the editor portion of the application, but more can be found in the previous Scoring Notes article on this topic.

Results

Symphony No. 3 – Eroica – Trumpet

In this example, there was a disconnect in what I saw within ScanScore and the MusicXML export.  The MusicXML spells out time signatures not visible in the application when comparing the initial results within ScanScore.  For example the 1/2 time signature is presented in the MusicXML, but not seen within ScanScore before export.

Observations:

• Title information was not included in MusicXML.
• Cued notes were interpreted as grace notes at all times.
• Multirests were not recognized.
• Tempo text and metronome mark were not correctly recognized.
• Did not interpret the transposition of a trumpet automatically and reflect it in the MusicXML.
• Assigned to Piano in the MusicXML file which displays this way in the import.
• Some repeat information was captured, but not fully.
• Lots of anomalies with incomplete measures when cue notes were used.
• Good results when there were no other complications around the notes.
• Captured articulations well.

Conclusion: Cleaning up this scan would save you some time over entering it manually. It would be best to start a new score and copy/paste into it versus trying to fix the existing problems.

Moonlight Sonata

Recognition from the scanned version was generally not as accurate as from the PDF import. In all cases, ScanScore was not able to interpret the implied triplets. Even when triplets were spelled out, the software had trouble recognizing those correctly, usually adding “3” as text under the middle note of regular eighth notes instead of grouping the three together as an actual triplet.

In the screenshot below from the imported PDF, in the first measure, you can see that only the second group of triplets are interpreted correctly. On the original PDF, the triplets are shown in the first meaure.

Because of the triplet issue, the alignment between the treble and bass clef fell apart. This could be a factor in the reason ScanScore did not determine that there was a single piano on a grand staff, and instead listed it as two different single staff instruments in the MusicXML export.

Observations:

• ScanScore could not infer triplets.
• Cross-staff beaming was not interpreted correctly.
• Accidentals were handled well.
• Interpreted and displayed mid-measure clef changes correctly.
• ScanScore was not able to create a single-staff piano part on a grand staff.
• Correctly identified the written “Adagio” as 66 bpm.

Conclusion: Because of issues with triplets — implied and not implied — not being recognized, using ScanScore with this score would not save you time over manually entering everything. It did well recognizing notes, and I believe a simpler piece, or one without implied triplets, would have had good results.

Nutcracker Suite

The smaller staff size and lower resolution in the scanned version understandably lead to it being unusable, but the imported PDF had better results. This piece also includes triplets, of which none were successfully recognized.

However, one bright spot in the recognition from ScanScore was its ability to interpret grace notes. It did a great job creating usable results in measures that included grace notes.

Observations:

• The imported PDF version completed all 25 measures included in the sample.
• ScanScore did a great job with single grace notes, as well as with multiple grace notes.
• Slurs and articulations were identified well.
• Results were displayed with the correct key signatures, but as concert pitches instead of transposed.
• Triplets not being recognized is a constant problem in this particular conversion.

Conclusion: Cleaning up this scan would save you some time over entering it manually. The usability of the measures with grace notes offsets the issues with the triplets. If you are a fast editor in the software of your choice, you can quickly turn much of this conversion to something usable.

---

Files for review

If you would like to see the original PDFs and exported XML files from each application to explore further, you can download them all here.