A great idea! Damn!
So, let me get this correctly... You used multiple EQ3s to split the signal into individual frequency bands, and then slapped a compressor on all of those individual signals. Then you split another signal in the similar manner, and use each of those bands to feed the key inputs on the corresponding compressors, right? Edit: No! This would have a completely inverse effect
... I'm guessing that the Utility comes into play at that point. Yes.
you are very close, with your guess.
The sidechain input of compressor has a filter, which is used to asses the incoming vocal, so for example a band pass on the sideband at 1k will only make the compressor compress when a signal is present in the vocal at 1k.
Now, of course the synth is the carrier, so the compressor will turn down
the synth when the vocal goes loud (at 1k), which as you say , is the wrong way round. So how do we invert it? Like you guessed, A duplicate rack-chain in parallel, with a Utility set on phase invert. This cancels out similar signals and lets disimilar signals through. When the compressed synth goes quiet it is because A - B = 0 , IE both channels are the same, but if A is quieter due to the sidechain modulation then A- B = 0.14231234 (or whatever) and we hear the difference.
so I can extract positive values from each frequency band by bandpass, sidebands and phase inversion.
Well, you may notice that I have linked the demo Project (ALP) in the first post, so you can play with it yourself. It can be improved, I just did this late last night. I would suggest playing with the compression response times a little (remember that 'copy to siblings' is your friend here
It sounds a little different to the mp3, because I swapped the synth I was using for a Simpler and added a little pink noise to try and get some sibilants enunciated.