I’ve now heard examples of the sunvoc vocal filter, and it sounds like just an usual strong complex iir type filter, applying a frequency response of a vowel to the input sound. In some examples I’ve seen/heard the result also sounds very “synthetic”, this stuff is dependant on the input signal that’s filtered - if it’s very close the the base sound (and accompanying hiss/noise) the vocal chords produce, such filters sound very close to a real vocal, while very synthetic base sounds will mostly be modulated to a more “robot” type result.
How was the example posted above here actually generated? In sunvox, with what kind of synth/fx chain? It sounds very natural at least in the beginning.
Spectral stuff? Well I could imaging (not with the sunvox filter, but with the other free vst) that there can be some work done on the input signal before the filter to modulate it automatically into something closer to vocal chords pulses. Real spectral manipulation via an fft in high resolution sounds like not suitable for realtime audio, the latency would be too big. I’ve sometimes had the impression that certain advanced (commercial) vowel filters could have some fancy method of interresonating or whatever, creating the mentioned “dirt” by themselves, even from very clean signals that would sound very synthetic with a straight filter curve. I could also imagine those more advanced filters just strip/filter the input signal to the base sound frequencies, and re-add vocal chord/tract like harmonics by themselves in a way or another. Also it’s not just the formants, the way the base frequencies sound can also help with the voice impression.
Using a vocoder for these type of sounds - vocoders apply a rough frequency curve, but the usually peaks won’t move, just scale roughly, so the sound would only be close to natural if the vocoder had very many bands, or formant tracking.
Another oldschool way to create choir type stuff is using a wavetable of looped wavecycles of vowels. But this could sound too static, after all it’s the motion/modulation that creates the stronges suggestion of listening to something close to a real voice. Also when repiching vocal wavecycles, the formant peaks will move with the note frequency, natural formants stay roughly in their positions independent of the pitch that’s sung/spoken.
And yes, vocal tract is not just a few bandpass filters with the right freqs. On my todo-list is experimenting with a mic, praat, and the 10 band parametric eq in renoise to try to model vocal filter curves. I’m curious what kind of results this could provide.
Very short answer to the original question: get a good vowel filter vst, and learn to use it right. Or some synth sporting these kind of filters.