I've been working on adding a MIDI simulator into my roll editing software, and
it would be interesting to be able to recover the printed dynamic marks that
many 88-note rolls have. So far my simulation for 88-note rolls is very basic, a
flat level with a fixed boost for themodist-style accents when they occur. If
some dynamic-line data is available it would be possible to derive something
that scales the theme and accompaniment proportionately. Even just a fixed theme
gives a much more musical result, so simplistic dynamic scaling is likely to be
pretty effective.
Given that I'd want to process the dynamic line in that manner, I'd prefer to
have the information raw rather than already processed into MIDI levels. If the
printed line can be located down the length of the roll, how about converting it
into something like a MIDI controller which downstream software can interpret as
it sees fit?
I don't think that it needs to be too complex to interpret these lines once the
image has been recognised. If we assume that every roll exploits the full
dynamic range (which is probably a fair starting assumption), the controller
could be linearly scaled with value zero as the leftmost position located for
the dynamic line and the maximum value for the rightmost position. Every such
line I've seen works from left (soft) to right (loud) so such scaling would work
automatically for all types of roll both in terms of its scale width and
position. Users could then impose min/max MIDI values to suit, and adjust the
linearity of the MIDI response between the extremes.
I suspect that for decent results there would have to be some capability to
fiddle with the lines to deal with printing errors, particularly minor
positioning errors along the roll that give dynamic shifts at the wrong time.
Thinking about my roll editor, I don't think it would be too hard to write code
to edit the controller values on screen when superimposed on the roll image. The
program allows immidiate playback of what's on screen so processing a roll up to
fairly decent standard shouldn't take too long. Not something to do in bulk,
perhaps, but useful for selected things such as the high-quality hand-played
classical 88-note rolls from Hupfeld.
Julian