If it is similar to vocoder, yes, it is tremendously confusing and needs a better interface (or begin to make use of 1.1 "send" feature?), but to do this with vocoder you must send both "instruments" to a dedicated FX channel (AudioFileProcessor for voice and something like tripleosc for midi).
Next, pan voice to one side and midi to the other side and try it out. If it doesn't work, invert your panning.
Note, I've never tried this with an AutoTune, but vocoder works well (vocoder sounds like draft punk, rather than AutoTune which sounds like T. Payne)
Hope this helps. Please provide feedback on whether or not it works!