computer-aided input of printed scores
I think it would be valuable for Muse Score to read a score in pdf
and generate a corresponding score in Muse score. I spent a working lifetime
in computer-aided pattern recognition and think that I have something to
contribute here. The key term here is computer-aided NOT automatic. The user
can supply key parameters such as sp, the distance between staves and
systems and other quantities used by Muse score. The pattern recognition
program would use these to generate a blank score which it would 'subtract'
from the PDF score. The note in the resulting image would then be represented
by a solid or open ellipse with the equator deleted, a relatively easy object
to recognize. Optical character recognition can be used to deal with the
lyrics although I suspect that the major work in reading a score will be in
the notes. Computers are much faster now than they were in the majority of my
working lifetime so that our program could evaluate several possibilities
essentially in parallel to find the most appropriate one. Please let me know
what you think of this suggestion.
Comments
MuseScore does have an 'Import PDF' function that works together with https://musescore.com/import
It is pretty experimental though, so don't get your hope set too high
In reply to MuseScore does have an … by Jojo-Schmitz
I note what you say. I am not asking to use it; I am applying to become one of its developers.
In reply to I note what you say. I am… by Keith Paton
Then you should be talking to Hervé Bitteur (creator of Audiveris, which is used under the covers)
In reply to Then you should be talking… by Jojo-Schmitz
Actually, I'd suggest talking to Liang Chen, who has worked on an in-house version of this more recently: #110306: OMR in musescore
I was actually coming here to suggest this. This is a really good idea sine there are so many weird idiosyncrasies that are hard to code but easy to identify if you're actively sitting there just answering questions that a parser isn't sure about.
My original thought was something like a background image that was behind the input layer but that would be rather messy. Having it determine where the barlines are and checking for things like that and then having you input things it isn't sure of would be great.