You could try to look for rms amplitude changes, if these are large enough then you get a trigger. I would think that would work ok for drums. Generally speaking I would think finding the onset of a sound with a fast attack would be easier than with a slow attack.
What are the problems you are finding?
Oeyvind should also have something to say about this, he has a lot of experience with these kinds of applications.
Prof. Victor Lazzarini
Maynooth University
Ireland