absolutely manually.
I do a lot of similar stuff, it’s actually fun to decipher these manuscripts.
Thanks for your opinion! Of course it’s great fun. Some of these are real puzzles, especially if there are e.g. holes or mistakes 
This has been my experience exactly. THis software consistently fails to recognize notes on the top line of the 5-line staff (F in TC, A in BC). It rarely misses with any other notes. Virtually 100% of the misses are on that one line. I have tried to call this to the author’s attention several times over the years, only to be attacked as a person who “just doesn’t get it.”
It also does a lousy job of recognizing multimeasure rests. If those two things worked most of the time, then I might recommend the product. As it is, I have learned enough of the built-in editor tricks to be marginally faster using the program (on simple parts only, not scores) then entering entirely from scratch.
My current workflow for converting big band scores is to use Soundslice to convert the rhythm section parts. They come through nearly 100%, except for DS, segno, coda, and those are easy to add in Dorico. That gets my score set up quickly with the right number of measures and all the repeat sections and double bars. I would use Soundslice for everything except that: 1) the cloud process is quite slow, and 2) that *&$#$%# subscription. I can only do 100 pages of music per month.
Once I have the rhythm section parts set up, I use SmartScore on each of the wind parts. I fix what I can do quickly in SmartScore (there are almost always dozens of errors on each part). I just ignore the multimeasure rests and deal with that in Dorico. But I do make sure that all the measures have the right number of beats before exporting XML.
Once in Dorico, it becomes a somewhat tedious process of working through each player, bushing sections of music to the right because SmartScore missed the mm rests.
In a tyical big band chart of, say, 120 measures, I can usually get all of the above done in about 2 hours. But then it is a process of fuxing dynamics, slurs LV, ties, accidentals that Smartscore got wrong. I know many people feel they are more productive just entering everything from scratch. It is a close call for me. You would think that the technology would be better by now.
Musitek is preparing a Mac Silicon/M1 update for SmartScore 64 NE at this moment.. It will be made available as a free update for current Mac users later this wee from the Musitek site here: MUSITEK - Music Scanning Software
The manuscript image posted is not a job for SmartScore or any other OMR application. Anyone interested should simply test the demo version of any or all available programs for themselves.
NOTE for users testing or using SmartScore to Dorico MusicXML functionality: Some XML conversion issues still exist (specifically system and page formatting). These issues do not appear when the same SmartScore XML output files are imported into Finale or MuseScore or Sibelius.
The scanning ability in SmartScore 64 Pro is very good. Good copy and you get a good scan most times 100%. Poor copy and you get poorer results. SmartScore 64 Pro is not very good if it has a lot of instruments or is a very long chart.
The editing part is terrible. I just do the scanning in SmartScore 64 Pro then export XML to Finale, now to Dorico, for the editing.
Thank you, Janus! I had no idea that was possible. Does your message appear on the forum, or should I correct it? Although I read other’s posts, I rarely post to this or any other forum and don’t really know the rules. I just wanted to plug PlayScore 2, since I didn’t see it mentioned in this thread and it is very accurate in my use case. Photoscore has been my go-to for years, and far better than the other products I tried until recently. A shame they haven’t kept with the times. Maybe, given access to Claude, they will reconsider and provide a new product with a perpetual license. Or not.
Tim
I have NEVER gotten anywhere close to 100%. Never. And most of my music is in good condition recently (within 10 years) printed from Finale. As I mentioned, the notes on the top line of the staff are missed about 20% of the time.
We don’t see anything like what you’re describing from our SmartScore 64 users.
Send us an email and attach one or two of the PDF or TIF files you had trouble with. We will get right back to you with clarification.
tech@musitek.com
I can’t help much but like I said good copy and good results. I even do it from a PDF and all is good. I have Brother all in one large format printer/scanner and I set resolution around 400.
Gladly. I can supply any number but let’s start with this one:
Notice that notes were missed in mm 79, 81, and 89. They were all top line Fs, and the scanned input is not bad at all. As mentioned, almost all of the misses are notes on that top line. In addition, it missed the 9-bar mm entirely and read the 8-bar mm as a 2. Ironically, it did seem to get the 15-bar rest despite a pencil marking.
I will email you the source PDF so you can see how it performs for you. My machine is Win 10 using the latest version of SmartScore I can find on your site.
I think it is safe to say that there are no programs that can deal with a manuscript like that.
But there should be. Fundamentally, what we call artificial “intelligence” has no intelligence whatsoever. No reasoning. No ability for original thought. But it is extraordinary at discerning patterns, far better than humans can do in many cases.
The basic objective should be, if a human can see (or hear) what is going on, then it should be possible to train AI for that task. There are perhaps 15 measures in your example that are so damaged that a person would have to rely on music theory and period knowledge to make an educated guess at what the original music was. But the rest of it can be read directly with some difficulty. We should be expecting the software to reach this point sooner rather than later.
This is probably an economic problem and a legal (i.e. copyright) problem more than a technology problem. There is not a huge market for this kind of thing, and if it evolved too far, there will be intense resistance from the big holders of copyrights, because effective, efficient music recognition would make a fundamental change in our relationship with music. For the composer of written music, it would become the equivalent of people using sampling technology 20 years ago.
I doubt, for example, that companies like Yamaha (which owns Steinberg) would want to see this move very quickly.
Great news. It does look like it’s been released now.
I would like to thank you for engaging with the community here, many of whom already are, or could be soon, your users!
Better OMR is the goal for all of us, and dialogue between developers and users is ultimately beneficial to all.
Hello
You sent a screenshot. That doesn’t tell us enough. Please send just the PDF or the TIF file. We’ll process it on our end and get back to you. On the face of it, your issue seems to be multi-measure rests. You post suggest SmartScore missed something like 20% of the music. It would be interesting to see an example of that as well.
Thanks
Musitek Administration
805-646-8051
Tech Support:
admin@musitek.com
I emailed the PDF. Did you not reecive it? Here it is:
Alto 1.pdf (348.7 KB)
That’s not what @cparmerlee wrote
That 20% is not a percentage of the music, but a percentage of the notes on the top line
Yes. And it is probably less than 20%. Probably more like 5% of the top line notes that are missed. The point is that it rarely misses any other notes unless there are pencil marks on the score.
It misses other things that make the program marginally useful. For me, missing so many of the multimeasure rests makes it pretty tedious. It is really painful to repair that in the SS editor. So my workflow is to just try to eliminate the cases where there are too many or too few beats n a measure, then fix everything else in Dorico. On my last three conversion projects, I think I spent about 3 hours on each piece getting it more-or-less fully intact in Dorico. From there I might spend another 5-15 hours improving the arrangement and markings, but that’s unrelated to the scanning technology. If the scanning technology worked better, I think I could complete that first phase in 90 minutes instead of 3 hours. I’d pay good money for that.
As comparison, I opened the Alto 1.pdf with PlayScore 2 and got this .xml file:
Speak Low.xml (65.5 KB)
The direct import into Dorico (on iPad) displays the content of that .xml file:
So basically there are no multi-rests, some unwanted fingerings, two pencil marks misinterpreted. I consider this as an excellent interpretation by PlayScore 2. In Dorico I corrected the small mistakes, which literally took me 4 minutes. After changing the treble staff to Alto Saxophone I did a transpose from C down to Eb and got this very nice result:
We received your image file and noticed pages were skewed during scanning. SmartScore has a “Deskew” tool in the Image Editor prior to recognition. We straightened your page and there were fewer mistakes. We cleaned it up (for us, just a couple of minutes since we’re familiar with the tools). You should have received the edited image and SmartScore files by now. Here’s the first page and XML file:
Alto 1.xml (101.7 KB)
I ran that same Alto-1 file (the original scan) through SoundSlice. Here are the results.
Alto-1 direct from Soundslice.pdf (21.3 KB)
Alto-1 from Soundslice.dorico (906.8 KB)
Alto-1 from Soundslice.xml (73.2 KB)
I did no “editing” at all, either in SoundSlice or in Dorico to get to the dorico file you see here. However, this is a slow process.
-
Upload the scanned image to their website.
-
Wait for it to process on their site – might be a minute, could be longer.
-
Review the AI findings – that is to say, the AI process presents fragments of uncertainty with Soundslice’s best guess. In other words, customers are paying to train the AI for Soundslice. In my case, there were about 10 issues presented and Soundslice was correct in all but 1. The wrong one was the 15-bar multimeasure rest, which it took to be 1. I simply typed 15 into the box and that was all.
-
Download the XML and then send into Dorico as usual.
So this is a time-consuming workflow. I assume they can get more compute resources to speed up their website, and they should be able to implement a better batch process, which would cut the time way down. But the main outcome is that got almost everything right.
- It does not recognize coda, DS, Segno markings.
- It got all the notes, triplets, ties, slurs, and articulations. I have never seen any other program come close on that.
- It mostly ignored pencil markings.
- This chart had no dynamic markings, which was one reason why I converted it to Dorico
- It got all the multimeasure rests correct (with the proviso that it did ask me about the 15 and I had to type in the 15. The total number of measures is off by one because somebody had scribbled out the barline before that 15-bar rest.
- There is a pair of repeat marks between 9 and 120. They were both marked over in pencil. Soundslice recognized the opening repeat but not the closing repeat. I wouldn’t expect any program to deal with this because the original ink is mostly obscured.
Other than the subscription model, and the tedious workflow processing remotely, it was nearly perfect on this particular example. There is an odd thing I don’t understand. If you look at the PDF that came out of Soundslice, you can see it clearly recognized all the multimeasure rests. But when I brought this XML into Dorico, those measures of rest are missing. But strangely, the measure numbers have been adjusted. For example, the Dorico file has music through measure 16. Measures 17-20 are rest, but they don’t appear in Dorico. Yet, the next measure that does appear has a new time sig mark and resumes at measure 21, as it should.
Is there some XML import option in Dorico I have to set in order to get the rest measures to appear?