Lol. Great analogy.fineincrements wrote: Sun Apr 19, 2026 5:19 pmYeah i know why you'd think it'd work that way. The thing is the FFT processing pipeline is one giant complex system of several buffers and sequential operations, and it gives a single latency of x samples that have to be reported to the host. that latency affects things like dry/wet phase alignment, which is completely outside the FFT pipeline. Having one pipeline is complex enough. trying to run 2 in parallel would be like "I want the left half of my car to be a ferarri and the right half to be a monster truck"kraster wrote: Sun Apr 19, 2026 11:58 amJust spitballing but would it be possible to design it so when lower FFT settings are used that it only processes the higher frequency end of the spectrum.fineincrements wrote: Sun Apr 19, 2026 5:56 amThank you so much for the compliment and suggestion! My only hesitation to do a FFT switch is that I first tried this at FFT 1024, and the low end artifacts were so bad. When you have these extremely sharp angled wavetable shapes running through the low end, and a low resolution, the discontinuity between bins causes a light but horrible distortion. I really wanted it to work at 1024 and spent gobs of time adding in gaussian smoothing and slew rate limiting and whatever I could come up with that might tame these artifacts and sadly nothing worked. at least nothing that preserved the integrity of the wavetable shape a user might want to draw in. I sort of decided I had no choice but to move to 2048 and the artifacts were handled. There is still a tiny bit of smoothing in the low end but it doesn't have to be much at 2048.Ah_Dziz wrote: Sat Apr 18, 2026 7:43 pm Having a user adjustable fft size would be a great move. Keep your default as whatever you like most but the change in sound with different sized bins is huge and a useful tool. It also allows for much lower latency at the expense of frequency resolution.
This is the best implementation of this idea I've seen in a plugin.
So as you go to smaller fft less bass is being processed and the affected and unaffected bass signal are recombined at the output. Not too dissimilar to the bass preservation mode that’s already but more implicitly tied to FFT size.
I find that in practice the areas where wavefield often has the most audible effect is in the Mid range.
As I said just spitballing!
Personally the latency doesn't bother me since I'm not using it in realtime.
It also seems like way too much work for edge cases with inherently dubious results.
, even if I'm not overly fond of the results.
