Yes - I am interested! If you have a way of doing this real-time, then it is facinating!
So - let me check.. You put a signal in (lets say a complex vocal monophonic waveform of 1kHz) and you can shift the pitch of this waveform using the position of ones hand reletive to the 'theremin' pitch antenna?
Note - A Vocoder will not do this.. it will only 'map' the harmonics of the input waveforms against each other. Also, feeding external audio into the ring modulator used to get the difference between the reference and pitch oscillators WILL NOT DO THIS! (I think this is what you have in mind) - It does produce an extremely interesting effect, but it does not shift the pitch of the input signal.. With no external input, the output would be the 'normal' theremin.. With an external audio input, the output would be complex.. I think it would be [T = Theremin pitch] [X = External waveform pitch + harmonics] Output = T+(T+X)+(ABS(T-X)) due to there being an output from reference and pitch oscillators = (REF+PITCH)which is high frequency and thrown away, and DIFFERENCE(REF,PITCH) which is the normal theremin audio output... This would modulate with the external signal giving Sum(T,X) and Difference(T,X).
If T and X were fed into a 2nd ring modulator (not involve the first modulator) the output would be Sum(T,X) and Difference(T,X) and T would dissapear from the audio output.
My advice to you is this - IF you really have a simple way to shift the pitch of complex waveforms (and talking about audio like polyphonic opera, which is indeed complex) then file a patent application on the idea before you talk about it! I have a patent (in last stages) on an "Analogue Function Duplicator" which will do what you describe, but only for pre-stored functions and waveforms, not on real-time input signals.
It is possible to do what you are describing by using a fast DSP - but there are still some difficulties .. like latency, particularly with fast changing 'reference pitch' signals.