Sorry - this can't be done. Only the volume of the performance can be changed at the set list level.

I guess if this is true - then you could modify the performance in liveset to add a dummy PART which stays at a constant volume (like an FM-X PART) and use the envelope follow "trick" to set the destination as pitch on all the other parts.

If you setup a user curve (I setup Uni - since I was only adjusting upwards in pitch) - then the ratio of +50 works well. On the user curve, the output value of "1" means raise by 1 semi-tone. "2" is 2 semi-tones. Etc. So setting your key-to-volume relationship is easy. If you do not want very low volumes to react - then you can use only 4 ranges at the top for 4 different keys. Left-to-right on the user curve represents the input envelope follower value which is the sensed volume of your dummy output FM-X part.

Keep in mind that the envelope follower itself has some parameters to basically set the gain and hysteresis to the change in "energy" - or you can think of this as volume since we've picked a fairly otherwise "uniform" voice that should track linear with the part volume.

I've checked this and it works - changing the volume from the setlist does change the key.

There are some limitations - I tried also bumping the volume back up of the active PARTs so when the SetList volume was turned down, the parts themselves would turn up. This works, but you can only get as high as the common volume (set by the setlist volume) - so once the volume gets down too low - the parts are all too low to be playable. So the trick would be to make the FM-X voice react to master volume changes very severely at the top end of the volume range (say between 100-127) to enable some headroom and "squish" all of your key changes on your user curve to the input=100-127 range. Maybe 3 keys at 0-108 = 1st key, 109-118 = 2nd key, 119-127 = 3rd key. This places lots of burden on making the FM-X voice also stable since the envelope follower needs some margin and 9 values isn't much margin to throw at it.

Apologies for the complexity - but there is at least a fairly reliable way to get at least two keys worth of difference by adjusting the volume in a live set if that's all that is available.

Last - pitch shifting isn't like transpose. It's like having a big (or small) pitch bend. So it's not the best tool for the application if there was a way to do a transpose instead. I say this because a bent note doesn't sound like the unbent note (of equivalent pitch) since the bent note is a pitch shifted version of a lower (or higher, wherever the bend started) note and the note at the final resting spot may be even a different sample altogether. Just play a pitch-bend adjusted note +2 (or more) and play the same pitched note. If bending from middle C +2, bend up to the D sound (+2) then play the standard D. Pick different notes and different (maybe wider) bends and you'll see how different in character the notes are. It's fine that it is this way - just wanted to explain how pitch shift has its drawbacks.

Pitch shift is worst at the edges of a sample range since that's already where the programmer decided they needed another sample for the next note (and "stretching" would sound unnatural) -- place the welcome mat for unnatural.

Side note:

Doing this, since I wasn't using lanes/motion control - I couldn't control how the input parameter (Part 2 Envelope Follower) could map to the control curve. When I was trying to compensate the volume, I wanted low inputs to produce high outputs - or a reverse standard "5" curve (triangle with highest output at input=0 and lowest output at input=127).

This was brought up in another topic - and there are ways to deal with this under motion sequence since there is the pulse available to adjust. However, with more direct destinations without pulses available - we have to use user curves (linear, not step in this case) to make a reverse curve.

I know I can set a reverse standard curve - but the problem is offset. The offset is fixed at 0 (unipolar) so I start with zero and get smaller (negative values) instead of starting at 127 and get smaller. User curves already provide the ability to do this for simple curves with straight lines - but it may be nice to enable an additional parameter of offset so instead of a user curve (for this reverse volume example) - I could use the negative ratio standard curve with a +127 offset (or the current positive limit of unipolar).

As mentioned, the simple case of standard curve param1=5 is covered with user curves - but some of the other more exotic curves cannot be duplicated in user curves due to the lack of inflection points (just 8).

Using standard curves, the dogleg gave me something closest to a reverse ramp if I setup the parameters correctly. Input values of 0-70 or so (eyeballing) were very high with a ramp down to 0. Rather than do a user curve - this is actually how I tested the volume "bump".