Blur a sound into a wash
Spectral blur works on the spectrum rather than the waveform. A short time Fourier transform splits the signal into 1025 frequency bands four times per window, each band is smeared across time and across its neighbours, and the waveform is rebuilt from what is left. Pitch survives, rhythm mostly does not.
The three controls do different jobs and fail differently, which is why they are separate.
Blur
Blur is a one pole filter on each band’s level, with a time constant from 15 ms to 4 seconds. A band that suddenly gets loud takes that long to reach full level, and just as long to fall again once the source has stopped.
Rising and falling at the same rate is what makes this a blur rather than a gate or a hold. At 40% a drum hit turns into a swell. At 100% a chord goes on standing for several seconds after the file has moved on to something else.
Spread
Spread is a triangular kernel applied across the bands, up to a quarter of the spectrum wide. A single note stops being a partial at one frequency and becomes a band of energy around it, so pitch turns into colour. Sung vowels lose their formants, and cymbals turn into a flat hiss.
Because spreading a partial over hundreds of bands would drop the level, the spectrum’s energy is restored after the kernel runs. Spread stays a texture control rather than a volume control.
Scatter
Scatter adds a fixed offset to each band’s phase, different in each channel. No energy moves in frequency at all: the spectrum analyser reads the same before and after. What changes is when each frequency arrives, which spreads a transient over the window and widens the image because the two channels are dispersed differently.
Spread carries Scatter with it whether or not the control is up, because a spread spectrum rebuilt with phases that agree is a click rather than a wash.
Latency and level
The window is 2048 samples, about 43 ms at 48 kHz, and the tool reports that as its latency so the runtime can align the dry path and trim the export. In a chain, everything downstream is aligned to the same figure.
The wet path is matched to the level of the input over 1.5 seconds. Without it the same visible settings would differ by up to 15 dB, since coherence between overlapping frames is what sets the output level once energy has been moved between bands.
Where it fits
Late in a chain, after anything that shapes dynamics, since a blurred signal has almost no dynamics left for a compressor to read. Mix around 40 to 60% keeps the source in front of its own wash; at 100% the source is gone and what remains is a pad built out of it. A filter afterwards is worth having, because Spread pushes energy into the top of the spectrum on bright material.