I’m trying to use Photoscore Ultimate. Just a horrible experience. Their website’s latest news is in 2019 when the last upgrade was released.
I was trying to convert PDFs into XML files to open in Dorico. I just wanted to see if anyone knows of a reliable PDF to XML converting software out there?
I’ve recently started using PlayScore 2 with good results. Depending on the quality of the PDF it’s been pretty reliable.
Despite their website looking awful and ancient, Neuratron apparently still exists. I recently managed to upgrade Photoscore after contacting them by mail.
It seems like Phtoscore isn’t the main focus of Neuratron anymore.
I guess it’s partly going to depend on the quality of the .pdf in the first place. I was asked by a quite well-known symphonist if I wanted to transcribe the neat but hand-written score from his piano concerto. I tried out all the top recommended OMR packages and decided it was a waste of time as too many corrections would be required in every case-- he was better off getting his usual copyist to enter everything manually (into Sibelius). It’s certainly an area where a really good newcomer would be very welcome.
This round-up A review of optical music recognition software - Scoring Notes is probably as good as any even if it’s not completely up to date.
I use the “pro” version of PlayScore2 and it’s generally OK, but has a 11 page limit, so workarounds are needed to scan larger scores. After reading the ScoringNotes article I downloaded the trial version of SmartScore64, but it won’t let you export an xml file as it’s the trial version! How are you supposed to evaluate this most critical aspect of the program? Needless to say, I won’t be buying it.
Every couple of years, I take a serious look at the state of the art in music recognition. I marvel at how unimpressive all the products are. Everywhere else in the music technology space, we see rapid progress and some truly amazing improvements in work flow and capabilities. But in the scanning space, we have a bunch of products that haven’t been touched for years.
I foolishly started another cycle of futility when I received an email saying one of these decades-old products (which will remain nameless) had an amazing new release that was so good, it justified a new name for the product line. I downloaded it and found it had errors on the first two documents that were handled better in the prior release. And before long, something in this software crashed my Windows system to where I couldn’t even revert to the prior release. I had to roll back to a Windows restore point. I reported this experience to the author. I was surprised to receive a response because he had never responded to similar pleas for help in the past. His response this time was quite nasty, ending with “take it or leave it.” That’s a pretty caustic reply from a person who has never shown much pride in that product, happily letting it do things like completely miss notes that are on the top staff line (A in BC, F in TC) consistently.
I have revisited 4 other products today. Same old s***. One product never recognizes any accidentals or key signatures. One has a trial that only allows 2 pages to be processed – total, and no XML demo. That had some ridiculous errors on a very simple score.
Other newer products are more mobile-based, which creates a real morass of work flow because I had to move things to Google Drive, but even then the Android app still wouldn’t open the PDFs, and the program provides no useful diagnostic info.
So it still seems like a wasteland. That’s ironic because I discovered one open source project and another commercial venture where the teams are building AI-based music recognition engines, but they would have to be embedded into an app to be useful to anybody. It seems like the pieces are all just sitting there waiting for a person or company that really understands the challenges and the opportunity. As I understand it, this is not unlike the strategy in the latest SpectraLayers project where the developer has utilized several different AI engines to accomplish different parts of the unmix challenge, and the results are nothing less than spooky amazing.
Why is so little happening on the music recognition front?
Thanks for this, @cparmerlee But please, can you edit and tell us which apps you tried, so we’re all fully aware of the pitfalls? There’s nothing to be lost with naming and shaming where appropriate.
And for example, the program(s) with no XML output for the trial version, someone who bought those products may have had a relatively good experience. If we knew which ones…
Scanscore was the program that did so poorly on accidentals and key signatures. It was a highly legible score. Indeed, I think the test file I used in that case was a PDF generated from Dorico – it wasn’t even run through a scanner, so there would have been no noise at all in that input. Maybe others have better luck or more patience than I have. There is no excuse for that kind of thing, IMHO.
The one that was a workflow nightmare (for me, considering my document scanner saves to a NAS directly accessible by my Windows machine that runs Dorico), was PlayScore 2. Frankly, I wasn’t in a very good mood by that time and I tried moving files from my NAS to an Android tablet using Google Drive and also by copying it to a MicroSD. I had no luck with either of those, so I never got a proper PDF into the program. I tried 3 or 4 pages of various music using the camera on the Android tablet and they all had more errors than I was willing to put up with (missing measures altogether, missing notes, misread multimeasure rests, etc.). Perhaps with a $1200 tablet and very careful attention to lighting, the program would do better, but that is not how it is advertised.
However, I see that PlayScore has a Windows version, which I did not try. Perhaps that would allow a more successful workflow. Maybe I will try that tomorrow.
I posted this last week, about SmartScore.
… though the thread took an unexpected turn. 
Generally, I find that ScanScore’s results require so much editing (in its own app before XML export), that you might as well just type it all in. But if anyone wants me to run an image for them, I’ll happily test it.
As said, OMR is something that AI would be ripe for.
I recommend trying app called Newzik. You can export a musicXML from there and clean it in Dorico.
I like that the ”cleaning up” part is in Dorico and not within the program itself - it is much faster that way. And if the export is rubbish, you notice it very fast
I think Newzik is actually AI-based!
A quick test of a piece (via Newzik’s website) seems to show better results than SmartScore. I can’t download the XML without paying a subscription, but at least it hasn’t done this:
Of course, Newzik wants to sell you the cloud subscription stuff, rather than just OMR; and it’s not for macOS: only iOS.
disappointing but not surprising to see this area is still a wasteland. I too am a bit surprised that no-one seems interested in investing in what surely could be a lucrative area if done better.
You can also use Newzik in a web browser, so I guess with any OS. I have the lifetime subscription, as I use it mainly as a score-reading app.
It syncs automatically all of your scores to a cloud, which is very useful. I prefer it to forscore because you can include all of the parts to each piece - this is very useful in a session if you would like to quickly see how an individual part looks like.
A while ago I also tried Newzik (and a few others), and was not happy with the results, you can better type in as someone mentioned.
I avoid using OMR software, but if I “must” I use Audiveris, which is a free one, written in Java so you have to have that too. You can also cleanup in the app itself before exporting. It certainly is not perfect (far from) but that goes for all the other applications, but you do not have to pay for this one…
I did a brief session with PlayScore 2 for Windows. Here are my impressions:
As far as I can tell, there is no way to correct these errors or even see them within Playscore. I had to open the XML in Dorico to see what is happening.
TOTALLY WORTHLESS APP. Don’t waste your time.
Even if it worked perfectly, I don’t think I would pay $80/year for a subscription.
Now I have to figure out how to cancel my subscription before they charge my credit card.
Here is the original part:
Here is what the Playscore XML looks like when you open in Dorico.
Interesting. I’ve never fed PlayScore 2 anything in a handwritten style font, but I’ve had usable results from it (or at least from the iPhone/iPad version) with IMSLP scans in the past.
Yes, it’s different to Photoscore/SmartScore etc. in that you can’t see the results within the app itself, only by taking MusicXML elsewhere, but I’d argue that’s a pro rather than a con, in that that’s one less program to learn - I’m much more comfortable editing in a program I use for notation on a daily basis.
(And actually, though it doesn’t have the means to show the results, it does have the means to play back the results, which, if you’re working with anything in more than one voice is a useful guide to whether your XML is going to be worth editing.)
With that in mind, though I’ve always had to fiddle with it to get good results, or do a little editing of the resulting XML, I’ve found PlayScore 2 + Dorico to be a much quicker workflow for me than Photoscore + editing directly in Photoscore.
Try fiddling with the advanced settings: the error correction and sampling ones, and hopefully you’ll find it’s not a “totally worthless app”.
Or at least don’t put other people off on the basis of your one and only test page.
If it fails so badly on such a simple test, I think everybody deserves to know that. Perhaps others have better luck, but there was absolutely nothing unusual about my test.
People are free to ignore my example. It won’t hurt my feelings. And if somebody has better results that justify an $80/year subscription, I would be happy to see those results. I don’t trust anything the company is saying about the product, based on my hands-on experience.
This is the nature of AI. People are excited about the possibilities of fully autonomous cars, but the reality is there aren’t any out there that can handle the range of conditions we encounter as humans every day. AI systems are only as good as the training set. It is entirely possible that PlayScore was trained with data such that it works better for your files than mine. But my test was about as basic as it gets.
==========================
I left a review on the Microsoft store. The developer replied to that with this comment:
PlayScore does not support handwritten style music. Please see description.
The email from Microsoft had links inviting me to update my review, but none of the links worked. I would note:
All of that aside, the biggest problem is not the failure to recognize sharps. The biggest objection I have is that once it makes that error, it becomes an avalanche of other errors, such as splitting bars and changing clefs, where none of that is in the original sheet. And you can’t find out what has happened until you import the XML into Dorico.
