YW. Frankly, it was a combination of meticulous and toss together. I use a pair of python scripts I wrote that do a lot of the busy work (cutting & labeling the samples, and then automatic keyboard mapping of the resulting sample files to produce the .sfz file.) The hard part was that my pitch detection algorithm only worked for about 2/3 of the samples (which is better than I expected, given how strange steel drums are.) So I had to manually review all the files and rename many of them by hand based on the correct pitch. That also gave me a chance to delete the clinkers.
For folks who fiddle with it, please give me feedback on the overall dynamic range. Using the default loudness for velocity curve made the highest velocities not nearly loud enough compared to the quietest notes. The default is based on (V/127)^2, or the square of the velocity as a fraction. I used the 4th power instead of the square and it got a lot better but I'm interested in feedback on that.
Also, there's some fun background noises in some samples. Extra credit for whoever finds any samples with a crow. A damn crow kept cawing while I was trying to record! So far I haven't noticed that in the tails of notes while playing but I'm sure it's in there somewhere.
There were also a few samples where a truck passed by but I think I deleted them all.
I'll also appreciate feedback about whether the lowest and highest velocity layers are useful. I have my doubts.
If I get a chance to sample again, I plan to focus on consistency of the timbre of each note rather than on loudness or how hard I strike. That'll be time-consuming, but should deliver better results. I expect to toss out over half of the samples as I'll be trying to hit the target sweet spot for that velocity layer. There's a fairly narrow force range where (for the first octave notes) the first harmonic kicks in -- that is, when the strike activates the higher octave pad in addition to the main one being struck. In the 2nd velocity layer you can hear most of the notes don't activate that area but some begin to. In the middle layer it's very clearly active for all notes. What I hope to get is a layer below without it, a layer between where it's starting to emerge -- which will be the tricky one -- and a third layer that's like my current middle layer but more consistent, with the 1st harmonic very clear.
Thank goodness the instrument has only two octaves.
One thing I've learned is to learn from the first attempt, rather than trying to get it all right the first time. Sort of like, when you first get a DAW, don't start out trying to record your magnum opus. Record a few simple tunes first and learn!