As we all know, the Pro 3 wavetables have 16 "reference waveforms." But each waveform is represented four times inside the system exclusive data, and the waveform among these that the Pro 3 actually uses is based on which key is being played:
(1) One 1024-sample waveform
(2) One 512-sample waveform (begins around F2)
(3) Two copies of a 256-sample waveform (begins around F3)
(4) Eight copies of a 128-sample waveform (begins around F5)
I say "begins around" because there seems to be sort of a morphing from one waveform to another around that split point. Also, I'm just using my ears, so you might perceive the split point to be slightly different.
There are three wavetables attached to this post. Each one starts with a sawtooth waveform in the 1024-sample data, and then changes to a square for the 512-sample (sq2, #35), 256-sample (sq3, #34), and 128-sample (sq2, #33) data. If you load these and play up the keyboard, you can hear where the changes happen. Note that the data is identical for all 16 reference slots in these proofs-of-concept.
This adds an interesting dimension to Pro 3 wavetables. Acoustic musical instruments change their harmonic content depending on pitch, and Pro 3 can imitate this, to some extent.
For me, I'm going back to the lab on how I'm designing my Pro 3 wavetable library, as I want it to be able to take advantage of the Pro 3's 64-waveform reality, while keeping things simple for normal 16-waveform generation.
I'm going to work on a 64-waveform wavetable that uses an acoustic piano at various pitches, and then I'll have a better idea of whether this is worth the time in practice.