(so to say)
spectral smoother takes the spectrum of a signal and smoothes it out -it literally lowpasses the Fourier transform. It's not very interesting on synths, but on recorded stuff it (heavily) lifts the noise that's buried down (maybe adding some distortion due to the FFT windowing, I'm short on maths there). I use it mainly on voices, to get weird doubling growls or to make traces/bursts of noise that seem to "come from" the base sound.
There are two parameters, each controlled through a spline curve -width and height. Width is the horizontal spread of the smoothing across the spectrum (for any given frequency, 100% width means a bandwidth of smoothing>36.8% equal to the frequency itself). Height controls the mix between the original and smoothed spectra, changing the level of the spectrum valleys with respect to its peaks. Negative height makes the peaks and valleys more instead of less pronounced (the opposite of smoothing); in this case, increasing width makes the peaks more isolated.
The spectral display is just indicative -it shows the smoothed spectrum as sent to the iFFT, which then does its best.
(Thanks to Gabriel Mulzer, Martijn Deluxe and Jan Brähler for the FFT implementation.)
* % is now called width, mix is called height
* Range for width controls set to 0..100 (instead of -100..100)
* Mixing (height) got its own curve; both curves can be scaled
* Geometric (instead of arithmetic) mixing, to allow the use of negative values
* RMS (instead of abs avg) smoothing filter, to avoid a large difference in level between positive and negative height -output level for positive height is now higher than before
* Fixed mode and max settings removed -current version corresponds to relative mode, 100% max of the previous one
* First breakpoint removed
* GUI and automation IDs reorganized
* Many small changes in implementation