Music OCR Review

Music OCR Review

I’m not a reviewer, I’m an engineer.  However, one of my ‘careers’ is a composer/artist.  As such, I want to make sure that the software I use is actually good.  Everybody defines good in a different way, but if I’m going to use some software, I don’t want to have to relearn it each time I open it up.

I’m not a musician, but I am a composer/arranger.  It takes me a long time to play a piece (not a lot of practice), so when I am tackling a project, I’m looking to start by inputting a midi file.  There are occasions where I cannot start with my own composition or a web-found midi file, but I do have the sheet music.  The next step is to look for a Music-scanning program.  There are a few around, but, in general, not a lot of reviews.  If you read the websites, every program is ‘award-winning’ and “has the best recognition”.  Which, of course, is marketing. 

I use Vuescan because I have a Canon 4200F scanner.  It’s over 20 years old and still works like a charm.  I tried a newer Canon scanner and immediately sent it back.  I do not like software that tells me what resolution I have to scan with.  Plus the USB was poorly implemented and required me to go through most of my USB ports before I found one where it would not quit halfway through the scan.  Then the scan had artifacts.

Since I’m beginning to do my Christmas album (starting again), I have a Christmas Fakebook with some 100-odd songs.  These are simple one instrument staffs (meant for singing).  These make a good start for arranging because you don’t get stuck with someone else’s idea of an arrangement.  I’m starting with Gesu Bambino.  It has one or two oddities that could catch an OCR off-guard.  There is one section of grace notes and the repeat begins after the first note.

At the end of this article, I will present a matrix with a 0-10 scale (10 best) for those Music OCR areas that are important to me.  Just to note, if I couldn’t test a function due to demo restrictions, that rates an automatic zero.

The test system:  I7-4770K, 3.5GHz (stock), Asus Maximus VII Hero Motherboard, Roland A-49 Midi controller, ALVA Nanoface Audio Interface, GeForce 750 Video card with 3 monitors, Canon Canoscan 4200f running with Vuescan.

General Observations

I believe that choirmasters, conductors and music teachers are used to putting up with all kinds of crap from software.  And, pretty much, that’s what they get.  The mildest thing I can say about all these Music OCR programs is that the interface is non-intuitive.  I know from my own day job that when you work on something day after day, it always seems like the best way to do things.  You have to have the ability to approach your interface from a naïve point of view.  Most users of these programs will use them for a week or two every quarter.  In that amount of time, it becomes a learning experience every time to get back to it.  All these programs have user interfaces that range from bad to horrible.  Not one has taken the steps to name things intuitively (as a whole, there are individual commands in each one that make sense) or make the workflow intuitive.  They all include reams of pages of a user quick-start.  My take is that if your ‘quick-start’ is 20 pages, you’re doing something wrong.

Sharpeye

Sharpeye installed easily and quickly.  I cringed when I saw the install program because it looks like Windows for Workgroups 3.1.  The installer is not newer than XP.  Like most of these programs, when started, you are presented with an interface that tells you nothing about the workflow.  Here the user interface was also XP vintage.  Then the fun began.  I clicked the button to scan and Vuescan opened up. 

When Vuescan opened up, I proceeded to try and scan an image.  It didn’t seem to work right.  Sharpeye just gave me rotating circles on both windows.  As some point I did manage to get an image into it, but it was screwed up.  The right hand 2 inches of the score were transposed to the left side.

The interface was completely non-intuitive.  I decided that I would have to scan an image and open it into Sharpeye to see how the engine worked.  Vuescan, opened on its own, refused to save an image.  Apparently Sharpeye had done a good job of sabotaging it, so I had to reboot.  Vuescan worked perfectly as long as Sharpeye was not open.

I scanned a tif image and opened it into Sharpeye.  At least Sharpeye has a button labeled ‘Read’.  As an image, I was using a page out of a Christmas fake book, the song being ‘Gesu Bambino’.  I would have to rate the read as excellent and I was also able to export a midi file that played perfectly.

One thing to note is that Sharpeye did properly see my midi input and output devices.

Conclusion:  User and Scanner interfaces suck.  Music OCR is excellent.

Photoscore

Here again we have an XP vintage installer.  Visually the interface is more modern, but still cryptic.  It is worlds away from intuitive.  There is a ‘Scan Pages’ button, so I clicked on that.  Amazingly, it began to access my scanner and scanned a page.  Then it scanned the page again.  Then it scanned the page again.  I thought it would stop, but not a chance.  I escaped out of the scanning and immediately was presented with a pop-up that declared that Photoscore was now unstable and quickly save my work.  I decided to see how unstable and tried to get the program to read any of the 6 times it scanned the same page.  Each page presented an error ranging from ‘There’s nothing there’ to ‘Liszt Engine failed’.  I restarted Photoscore and, as soon as I clicked on one of the scanned images, I got the Unstable error again. 

Again, having to back-pedal, I restarted Photoscore and imported the tif I had used with Sharpeye.  It read the image in and I was presented with what looked like a photocopy of the score.  Not having a clue as to what to do next, I opened the Play drop-down and nothing happened.  I tried this a couple of times, then went in search for the preferences.  First, there is no tab for Midi.  You have to go to the Advanced Tab (since when is Midi ‘advanced’?) and there I found out that Photoscore had detected my controller as an output device.  Once changed to the Microsoft Synth, I got music.  The OCR was perfect.  Then I searched in vain for the output of a midi file.  The closest I found was ‘Save As’ but I couldn’t tell what I could save as, because I got the ‘limited functionality’ pop-up.

Conclusion:  Scanner interface is crap.  User interface is crap.  Music OCR is excellent.

The only note on this one is that while I know the reason for the limited functionality, I refuse to buy a pig-in-a-poke.  So if I can’t test the functionality I need, then I cannot recommend this software.

I further can’t recommend this software as it is produced by a German software company.  Every German software company I have encountered has the same attitude.  Our software is perfect and if you have a problem, it’s your problem.  So you can expect to be treated like you’re a dumbass and not worth their time.  So far none of them have proved me wrong.

Capella Scan & Play

Another German software company and another non-recommendation.  They don’t really have a demo version except that you can install the software and run it without a key.  The demo limits on this program are more draconian.  It will only play the first couple of notes from each line, so you have no idea whether the OCR worked or not.  Again, saving is disabled, so you have no idea how good they convert to midi.  On the plus side, it does work with Vuescan.

Conclusion:  Scanner interface is Good.  User interface is mediocre.  Music OCR is looks good visually, only half credit since you can’t hear it.

Not Recommended.

Smartscore X2

My first experience ever with Music OCR was with an earlier version of Smartscore.  At that time, the conversion (yes, we are converting from sheet music to digital format, that’s the proper word, not ‘Read’ or ‘Recognition’), was bad.  It missed several very clear notes.  That was done from a PDF.  I did not check the scanning feature.

Smartscore does the preliminary/final scan dance.  At least with Vuescan, it doesn’t always work.  I scanned an image and got the rotating circle for a long time.  To get an image, you have to wait a certain amount of time, then cancel.  Then your image appears.  You hope.  After the image is acquired, you hit the ‘Recog’ button—intuitive, isn’t it (it isn’t) and you are presented with a split screen with the scan above and the ‘Recog’ below.  One thing I noticed is that Smartscore is very random with its errors.  You can rescan the same image several times and each time it gives you different errors.  Well, let me put it this way, in the ‘Recog’ window, some of the measures will be pink, which I’m assuming at this point is a potential error.  If you don’t like where the pink is, just scan again and it will change.  You don’t even have to readjust the paper in the scanner!

Playback in Smartscore is uneven.  At times it seems that it revaluates the notes as something longer.  Although I do have to praise them that at least I got to hear the whole page.

Here again, saving is locked out, so I did not get a chance to find out how well the generated midi works outside of Smartscore.  This earns Smartscore an instant ‘Not Recommended’.

Audiveris

The only thing this program has going for it, is that it is freeware.  The conversion is pretty good, although at this point I can’t tell as it has ceased to work.  Audiveris requires Java to run, so if you are concerned about security, stay away.  Another negative is that Audiveris only exports MusicXML files.  Since Sonar Platinum cannot import them, it’s useless to me.  And I don’t feel like tacking on another conversion program just to get to where I want.  As far as I know, MusicXML is a very niche format only used by Music OCR programs.  And don’t tell me it’s ‘universal’---it’s not.  Only a handful of programs use it and while the idea is good, since it includes dynamics as well as notes, if you can play with a DAW, you’re not universal.

OpenOMR

This is a GNU license program and was written back in 2006 for a dissertation.  It requires Java and the start is a 200 character command line entry, so I just didn’t bother.

Gamera

You may see this touted as a Music OCR.  It isn’t.  It’s a toolkit for removing staff lines.

Vivaldi Scan

This software is touted as ‘easy to use’.  We’ll see.  The install did not go well, it halted telling me that an uninstall or an install had not been completed.  If that is the case, then Smartscore did not remove itself properly.

Reboot.

Same error on install.  I noted that the version of Vivaldi Scan was 2003a, so the installer is way out of date.  Chalk up a big fat zero for this one.

Tried an alternate download site, same version.  The developer’s website leads to a forbidden page.  This one is no longer maintained.

Midi Connections Scan

Again a demo version.  Note that on the install, they forgot to translate the German into English.  See what I mean about German software companies.  It did install and then refused to work.  It would not work with Vuescan.  It complained about needing 400 to 600 dpi scans in B&W, but then defaults to 255 colors.  Although it says that it needs a tif file, it refused to load a tif with the ‘detailed’ error of ‘Not Loaded!’.  I refuse to play these types of games.  Avoid this one.

Just for completeness, the user interface is Windows 3.1 and while I don’t care if the program happens to work, here it just adds aggravation.

PDFtoMusic Pro

Did not install this one as it does not work on scans.  It will extract music information from a music PDF.  While I can scan to PDF, it’s still an image and will not work with this program.

Box Score:

Program	Install	User Intf	Workflow	Scanner	Vuescan	OCR	Playback	Midi Output	Notes
Sharpeye	8	7	4	0	0	8	9	9	Unable to scan
Photoscore	9	5	4	0	5	9	9	0	Unstable, unable to scan
Capella Scan & Play	9	5	6	5	8	5	0	0	Scans but unable to evaluate due to demo limits
Smartscore X2	9	7	5	3	5	9	5	0	Uneven playback
Audiveris	8	5	3	5	5	7	5	0	Only MusicXML output, scanning problems
Vivaldi Scan	0	0	0	0	0	0	0	0	Did not install
Midi Connections Scan	9	3	0	0	0	0	0	0	Refused to work at all.

So the bottom line is that I can’t recommend anything.  Every program has problems.  Vivaldi Scan will not work unless you roll back to Windows 2000.  Photoscore is right out due to stability problems.  Midi Connections Scan is right out as well.  Audiveris may work for you, but with no midi out, it’s useless to me.  If I could test the midi output on Capella & Smartscore, they may work for me.

As it is, it looks like my best best is Sharpeye, but I would have to scan first then import it into the program.  Less than ideal, but workable.

That said, I do invite any of these developers to change my mind.  I’m willing to admit I’m wrong when I’m wrong.  But no, I won’t buy a new scanner just to work with your multi-hundred dollar program.  If you care to ask, even the updated Canon drivers for the 4200 no longer work.  Something changed in Win7 that broke the driver.

Remember that I didn’t do anything but try your demo programs on my system.  The system I would use them on.  I’m not a reviewer, I’m not a software hacker.  I’m just someone who needs a damn program that works.  Prove it to me!