Many of us chose to defeat the purpose of pdf extractors by making sure that any pdf images we distribute are renderings of .tiff images. Scan at a high enough resolution and the eye has a hard time telling the difference but pdf extractors will not find any font information to import.
Since I have never allowed my work to be posted on the web, it’s annoying when I find it there. My only consolation is that, if it’s to be stolen, I’ve made it more difficult to do so. Not impossible, of course.
Piracy isn’t the issue but I’ll indulge it. In your case, PF Slow, one would only have to appeal to Photoscore or Smartscore (which you own, and surely use) instead of PDFtoMusic… It doesn’t make it more difficult, actually it doesn’t make any difference.
Recently. I tried the demo version of the followings OMR applications:
PDF to Music Pro
I used some pieces downloaded from IMSLP:
Madama Butterfly - a full score (scanned)
Madama Butterfly - a vocal score (scanned)
as well as other scanned PDFs as follows:
A 50-page vocal solo score without bar with Korean lyrics (a scan of a printed PDF created using a Notation Software)
We are the champions (a scan of a printed PDF created using a Notation Software)
I think all of them are not reliable… We are the champions was relatively OK, but about 50% was wrong.
Lots of people talk about AI and deep learning. However, the development status of OMR seems to be lagged behind.
The results of these apps are highly sensitive to different factors, among which picture quality (I can safely talk about Photoscore ultimate, not the others)
My experience with pdf output by notation software is that, except with piano pieces that have different voices per staff and cross-staff notes, the results are very good. For a Mozart symphony, it’s perfectly possible to have only 10 rhythmical problems to solve !
With scanned material, the results can be good to unusable. It depends on the scan quality and the content of the music. I’ve just worked on three pages from Pagliacci, it was almost perfect (10 errors) so very fast to correct and use.
The main problem of those apps is that they are a niche market, so nobody is going to invest the amount of money necessary to make it an equal to text recognition software (and it is way more complicated than text). I don’t expect to actually see a high performing app in that field in the next five years
The path … Photoscore Ultimate → music.xml → Sibelius → music.xml → Dorico
does also work well.
Neuratron (the developers of Photoscore) might have had some links to the old Sibelius team (at least that’s my guess).
I updated to thre latest Photoscore Ultimate version. So far I couldn’t find a real improvement, but I haven’t done much testing yet. I use Photoscore a lot to transpose single parts. This works quite well, but multi rests always had to be corrected. Therefore I hope the announced “Multi-rest recognition improvement” will speed things up.
I just looked at one XML file and found out, that Photoscore is still writing MusicXML 2.0 files. Version 3.0 and 3.1 are out now for quite some time …
The update is basically an adaption to 64bit Systemsoftware.
In macOS this is version 15, in Windows I don’t know which System is 64bit only.
Because most of my infrastructure still relies on 32bit drivers or software, I can not upgrade my system os at this time…
and therefore don’t need to update PhotoScore unless I just want to spend money.
Or unless there is more development happening in PhotoScore…
I updated to the latest Photoscore (primarily because I thought there could be some nice updates in it) without being under 10.15 The graphical update feels like a major one, because Photoscore Ultimate, before version 9.x.x, felt like a XXth century app. It’s now really more pleasant to use — and since the whole point of it is largely based on graphical stuff, it does make a difference. But, as I stated before, the other improvements are less amazing. Nevertheless, it’s a tool I still use and will do, every time I feel it can make me earn some time. But (that’s for any Neuratron’s dev who might be reading this) I do hope they get involved with people from Dorico and learn from their work (that’s what real musicians wait for, updates that do matter).
I frequently use Photoscore Ultimate to interpret scanned or downloaded pdfs of hymns and lead sheets. I find the music to be very accurate, but not so much the text. I usually select all text and lyrics, delete then re-enter by hand in Dorico. It’s enough of a time saver to make it worth purchasing.
The interface, like their website is less than elegant - sort of a late 1990’s gui, upgrades are so long in coming I sometimes wonder if anyone is at home at Neuratron. Yet it seems to be the best game in town of the software I tested when deciding which one to use a few years ago. I think MarcLarcher hit the nail on the head about this being a relatively small niche market verses a pretty complex problem to tackle.
I’m not excited about the upgrade as described in the OP, but might consider upgrading anyway to encourage the company to keep moving forward - as many of us invested in Dorico while it was fledgling. Maybe Steinberg will buy Neuratron and offer it as a companion product to Dorico.
I think this is a mistake that is often made in marketing. There have been numerous products in this category and not much revenue. A bigger company is tempted to look at this and conclude, “See. That proves there is no market.”
I would suggest the situation is actually quite different. There is a limited market for a product that is extremely tedious to use and certain to only be 98% accurate or something like that. Imagine a database system that returned the wrong results even 1% of the time. That would be a very limited market, wouldn’t it?
Maybe a better example is Google Maps. A few years ago, the results were so poor that I wouldn’t bother to use it most of the time. But today it is considerably better and people use it heavily.
I suspect we are seeing a similar thing with notation software in general. When Finale was the only real game in town, a person had to be very highly motivated to use it. But now there are better, easier to use notation products. In addition, the prominent DAWs incorporate some level of notation, so far more people are using computers to generate notation. I can’t remember the last time I went to a rehearsal and somebody pulled up a new chart that was written by hand. (Actually I can. I have a friend who will go to his grave arranging by hand. But he’s the only one I know still doing that.)
My point is that I believe there is a HUGE market if somebody can develop software that works reliably and accurately. Of course, such a program could become a real threat to copyright owners, so there could be some legal resistance to that.c
I don’t mean to minimize the scope of this work. Any of these pattern recognition processes are complex and error prone. When we use pattern recognition to identify faces, it is no big deal to miss 1 in 100. But if you do that with music, that may be enough to make the program practically useless.
kt-va, if you go into the Preferences>Reading of Photoscore, you can disable text. This way the application might work a bit faster as it focuses on just the music - and you don’t need to delete anything later.
I have actually deselected everything in Preferences>Reading.
Marc and Leo,
Photoscore is probably the best notation OCR app on the market…
I hope they will be interested in working with Dorico users, but since Neuratron is subdivision of AVID… They are more oriented to Sibelius users…
I wish you success in case you are going to contact them!
I’m not sure how independent is Neuratron… since their products are so linked with the Avid’s products. Even the editing functions in Photoscore are taken from Sibelius. The product can also be purchased on the AVID’s web site.