I can see how the external control of this is tempting BUT
Each of the 24 bands has:-
An analysis filter and envelope follower giving a voltage out that represents the amplitude of that frequency band at that instant in time.
A reconstruction filter, allowing an identical band through (if any harmonics exist in this band!); this feeds into a VCA. This is normally controlled directly by that band's envelope follower so that a similar amount of carrier is allowed through to that seen by the analysis filter.
OK, so for patching, I see only a few tricks that are likely to be tried:-
1/ Scrambling - cross-connect the analysis channels to different reconstruction channels, giving weird results.
2/ Slew-limiting - making the reconstruction filters slow to respond, I guess this might sound a bit 'drunken' or blurred?
3/ Leaving the patching as designed and tee-ing in to all (or a selection) of patchpoints to play with unvoiced detector ideas - a seperate circuit that could, on detecting an unvoiced sound, would substitute the vocoder's output for a filtered noise source.
4/ (Depending on the smoothness of the overall filterbank profile) Using the reconstruction filters as a kind of 'String Filter' similar to the JH String Filter project, requiring all channels to be open all the time - maybe easiest just putting the same waveform in to both vocoder inputs and, possibly, killing every other channel to give a more 'peaky' response
As far as external control of the reconstruction filter bank is concerned, I personally can't imagine anything short of a computer being able to get close to the agility and speed required to give useful results.
OK, flame away
We still need to know the CV range for this thing!