Soft clipping or compression in reverbs

[Marvinh]

https://en.wikipedia.org/wiki/Mu-law_algorithm

This might work as a good saturator

If you are modeling old reverbs

This is juce convolution normalized not sure if it’s the input IR normalized or this is ran on the output of the algorithm .

Code: Select all

static float calculateNormalisationFactor (float sumSquaredMagnitude)
{
    if (sumSquaredMagnitude < 1e-8f)
        return 1.0f;

    return 0.125f / std::sqrt (sumSquaredMagnitude);
}

[mystran]

https://en.wikipedia.org/wiki/Mu-law_algorithm

μLaw is designed to be used as non-uniform quantization for narrow bitwidths (eg. 8 bits), to store on storage and/or transmission costs.

It's not really suitable for processing (well, obviously you can use any non-linear function for distortion, but beyond that).

[kernaudioio]

worth distinguishing clip-in-the-feedback-loop vs clip-at-the-output. in the loop a soft clipper changes the effective feedback gain dynamically, which couples to t60 and gives you that pumping character valhalla room has. at the output it is just coloration, never touches the tail shape. for an arbitrary-decay architecture (fsv-dvn style) the clipper wants to live on the pulse gain g(m) not on the summed tail, otherwise your t60 target and your peak-limit fight each other at long length settings. depends entirely on whether you are going for character or transparent bus hygiene.