ZDF vocoder for serious vocoding
"we have daft punk at home"
to be frank, i've never been terribly floored by the sound of many software vocoders. have always had a suspicion that using the computationally expensive and state of the art ZDF reaktor filters might go some of the way closer to getting a 'classic vocoder' sound. after some experimentation, the curious combination of the 4 pole NLB ladder (to 2p bandpass breakout) for the modulator, and butterworth bandpass for the carrier, seemed to give a surprisingly good result. OK, the next thing to do was figure out a heuristic approximation of the voiced/unvoiced detection that nicer vocoders have, with a burst of noise exciting the carrier for fricatives and plosives, but is inaudible or mostly inaudible at all other times: check. this has a side effect of also working nicely for transients when vocoding things like drums. this noise channel is also slowly gated by the carrier.
the number of voices corresponds to the number of bands. it is set to 32, but it can be set to fewer (if you're going for a more 'authentic' sound), or more (if you're a dork). dialing in the number of voices and hearing the quality of the vocoder change the instant you let up the mouse should not be missed*
for maximum effect, the carrier signal aught be driven with something like an unfiltered saw wave, a pro tip is to mix in some unfiltered impulse train (to taste) as well, since the best thing to have for a carrier signal is something with a flat frequency response or close to it, to avoid a 'hat on a hat' or 'transfer function on a transfer function' effect on the sound. conversely, some filtering of this carrier signal could also be desirable, since the unvoiced-envelope gated noise can excite the high frequencies on its own, and providing separation between the lower vowels driven by an oscillator and higher frequency speech components driven by noise could be argued a more realistic approximation of human speech
version 1.0 is mostly de-jenkified. the resonance for the modulator's filters is at a constant .94, so setting the number of bands too low *might* cause high frequency response to suffer, seeing as there is no consideration of adjusting for bandwidth. this is because it seemed to sound the best this way, so a future update may or may not address this. also a B panel with expanded controls, mostly for fine tuning the parameters of the unvoiced envelope and the bandwidth of the noise, could be a possibility. the display will no longer will correctly when the number of voices is changed, but other than that, this thing just works, so don't @ me