PDF to Musescore file converter unsuccessful

• Nov 16, 2017 - 18:20

I used the converter a number of times several months ago with 100% success.
I've tried using it again, with a number of different PDF files, in the last few days and either the conversion is unsuccessful, or the output files are corrupted.
Has anything changed recently?
This utility is fantastic when it works, but I can no longer get it to work properly. Help!


Comments

Welcome.
The system is experimental ...
Can you make an example?: describe (and attach if you can) a file that was converted and one unsuccessful (Dimensions, number of instruments, number of pages etc.)
Someone can give you some answer...

In reply to by ebennyt

From my experience with OMR:
1. Shaashuei_Hannukka.pdf
Very clean looking, truly horizontal staves and vertical barlines, consistent system widths, no lyrics, no
fingering numbers, no chord symbols/diagrams, no ossias, segnos, codas. (This score would be easy to manually enter into MuseScore were it only a few measures.)

  1. Lee_Celloschule_OP_30_11.pdf
    Good image contrast. More complex with fingering numbers, staccato dots, slurs. The quarter rest could be weird looking (compared to the previous score) but would be deduced, if not recognized by the OMR as such.

  2. CCF11112017_00005.pdf
    Image is skewed (look at right/left margins) - therefore the horizontal/vertical reference is not accurate (can be tweaked in the software).
    Three 'songs' on a single page, also containing fingering numbers, slurs. This would be a good candidate to submit as three separate images.

Regards.

In reply to by Jojo-Schmitz

Good point...
Score 1 is also a candidate for an app like PDFtoMusic Pro, which requires that particular type of pdf file - rather than a 'scanned image', or 'picture', of a score.

Also, in many instances, a pdf like score 1 (the kind with graphical elements) can be modified using a pdf editor to make Audiveris' job easier.
For example, text (e.g. lyrics), images (e.g. fretboard diagrams, ossia), chord symbols, and title pages (containing no important score elements) can be deleted; and then, the modified score can be loaded (or even re-scanned) into Audiveris so that the OMR can quickly identify - and focus on - the notes, measures, staves. This cuts down on having to tweak the (plethora of) recognition parameters from their default settings to get a good output from the OMR.

Regards.

MuseScore's PDF file converter service runs the Audiveris OMR (optical music recognition) application. You can try it yourself.
From their site:
Audiveris is a very fast OMR, but with limited recognition capabilities and requiring good quality scores. Its implementation is based on an iteration mechanism like that: build glyph, assign shape, check validity. If the validity check fails, we simply try another iteration. The main difficulty is of course that doing so we have no way to know if we have actually reached the best interpretation.
See:
https://github.com/Audiveris/audiveris/wiki

OMR presents a very complex dilemma: How to translate the various glyphs, lines, symbols, text of notated music into a machine readable form that can then be delivered to a notation editing/playback app. I have had varying results using Audiveris (before the MuseScore converter service was available). Most issues dealt with the quality of the pdf - if either scanned, or created from a scorewriter program.
See:
https://musescore.org/en/node/261949#comment-806203
In the above link, have a look at the first (scanned) pdf image. The top of the page was not perfectly aligned to the top of the scanner glass. This makes it more difficult for any OMR to discern horizontal staff lines, vertical measure lines, note stems, beams, etc. Sometimes re-aligning the page and re-scanning works (sometimes also changing dpi settings).

Audiveris, over time, has developed an intricate user interface that allows a user to tweak dozens of 'default' parameters, so strict horizontal/vertical detection settings can be adjusted for better (or worse) results. Similar problems occur with noteheads - e.g. quarter note, or ink blob on half note? Whole note, or half note with barely visible stem?
Such that... when one finally obtains an acceptable OMR score... he could have saved time by initially entering the score into MuseScore!

So...
There are lots of musical tools out there, for all sorts of purposes. To choose what works best for you often requires some hands-on experimentation.

Regards.

In reply to by Jm6stringer

It's not clear to me what you are telling me. As I said, the first example was successful, the second resulted in a corrupted .mscz file and the third didn't work at all. (The last two examples were taken from .pdf files in the IMSLP library).
Is there anything I can do to get these into .mscz format?

In reply to by ebennyt

Hello ebennyt...

As Thomas wrote:
When the online convertor fails, there is much we can do from our end.
But, I agree it should read (as Jojo suggests):
When the online convertor fails, there is not much we can do from our end.
Which means:
The MuseScore online pdf converter service is a one shot deal. It either works, or it doesn't.

There is, however, much you can do:
1. Transcribe the score yourself into MuseScore (no problem if, for example, the pdf is 32 bars of single staff music.)
2. Use another OMR tool and see if that one works any better on first try.

My point is that whatever OMR you use, results will vary depending on the quality of information contained in the pdf. Some pdfs are exported from music notation software. Some pdfs are created from scanned images (essentially a photograph). These differences can make it easy (or difficult) for any OMR to "get it right" on the first pass.
Here's a simple example: An image could appear tilted on a pdf's page. That is, the original was placed on the scanner glass and not aligned to the top and sides. A human looking at the scanned results can instantly recognize that fact. A machine without eyes and intellect needs to be told "what to look for". So in this case the parameters of the OMR need to be manually adjusted to recognize the skewed page. Thus horizontal and vertical recognition must be tweaked in the software, so that the OMR has a greater chance of recognizing things like horizontal staff lines, vertical note stems, etc.

Another example: A scanned score is a scan of a previously scanned score. This will result in problems with the sharpness, or resolution, of the image. (Pdfs generated from scorewriting software generally don't have this problem, but IMSLP's pdfs are notorious for this. Just look at how old some the scanned editions are - original or otherwise!)

So... transcribing the score yourself into MuseScore always works, and using another OMR may work with just a single pass at conversion. Most likely though you will have to attempt an initial conversion , then adjust some parameters; perform another pass, adjust some other parameters, rinse and repeat until you get an acceptable result.

I have used Audiveris in the past, and I don't mean the online converter. The app produces initially what it thinks is its best interpretation, but then allows the user to fine tune the recognition parameters to produce successively better results. For a small score, if not recognized initially (or after one or two tweaks), one's time may be better spent entering the notation directly into MuseScore. For a large score, it may be worthwhile to tweak all the relevant OMR parameters to finally obtain something worthwhile. Sometimes it's just getting the staves recognized, or fixing a few dotted notes - e.g. where the dots are barely visible in the original pdf scan. Other times, entering the score directly into MuseScore could be quicker.

I think Audiveris is an amazing attempt to perform a Herculean task. (At the time I was using it, they were even talking about hand-written score recognition!) The learning curve - especially considering all the adjustable parameters - is steep.
Give it a try! It's free, and you may only need one or two tweaks beyond the default settings to get acceptable results for any given score.

Regards.

P.S. MuseScore is great, too! :-)

Do you still have an unanswered question? Please log in first to post your question.