We're new at this, so there's probably someone out there who long ago worked this out.
How do we do a snare or timp roll that doesn't have that "machine-gun" sound to it that so characterizes MIDI rolls? Is it a matter of going in and writing it in 64th notes, then offsetting them a tick or two here and there to create some sense of humanity?
Has anyone found a way to let GPO do this for them, so that they can just write the roll in the notation program (we use Sib v. 2)
For snare rolls, the most convincing ones would be the C# (not sure which octave) in the snare or orchestral percussion patch. You control the dynamic with the mod wheel and you can crecendo and decrecendo during the roll. For timpani rolls there are no pre-recorded keys, so I recomend you do what the link billp posted says.
The "after," of course, sounds considerably better. The main things for good-sounding rolls are:
1. Actually "play" them into the track using alternating left and right hands or using the new "playable" roll feature (be sure you have lowered the polyphony to under 16 to conserve CPU.) As with most simulation –related sequencing chores, perfectly spaced, quantized hits will always sound unnatural. Actually playing the part adds velocity and timing irregularities that are essential to the illusion.
2. Related to #1, make sure that velocities never max out for a repeated series of notes in a roll. A series of notes at a velocity of 127 will create a hideous result. A repeated velocity of 127 with no alternation of hands will be even worse. Save 127 for just the loudest part of the roll. If you look at the velocity data you should see very few identical values in a series. Left and right hand velocities should be somewhat different from one another during the roll. Think of it as one hand being naturally stronger than the other. If you play it in that will happen automatically.
3. Experiment with adding both VAR1 and VAR2 data to further increase the differentiation between individual hits.
...While it's true a live player strives for evenness, he/she never perfectly achieves it. Whereas, the computer musician must take steps to avoid perfect evenness since standard data entry methods make it all too easy to achieve perfect (and utterly unnatural) evenness.