Equal Power Crossfading of Reverb

[sadhu]

I was looking at the code in this post and adapted it to what I need. Which is mixing the dry signal and a completely wet reverb and effects chain. But the final post at the end mentions equal power crossfading to avoid a volume drop in the middle of the dial:

Convolution Dry/Mix

I don’t believe there’s one included in the DSP library, but writing your own should be trivial. Just create a secondary buffer that gets allocated to the proper buffer size in your prepare method and make a copy to it as the first step in your process method, then apply a gain to both streams based on the normalized position of the dry/wet knob, and finally mix them together into the output buffer using addition. Here’s an implementation I did for an old project using the dsp:: API. The projec…

After some reading online I think I basically need to do this (I am assuming the signals are de-correlated). I suck at math and just want to know if I am on the right track I sounds like it is working to my ears.

wetMix.setGainLinear(sqrt(fxKnob)); dryMix.setGainLinear(sqrt(1.0-fxKnob));

Am I on the right track? Also, if the signal was correlated would I use a more linear sweep with 0.5 in the middle? What would be an example of a more correlated signal? A chorus? Sorry for all the questions I just want to understand this better so I don’t need to ask about it in the future.

[martinrobinson-2]

sqrt(0.5) gives you the right -3dB midpoint but, from what I recall trying that, you get nasty steps at the extremities in the signal that is going to towards silence when close to 0 or 1

[martinrobinson-2]

Algorithm to pan audio

Something like that using sin() and cos() with ±45° as the limits for the angles.

[martinrobinson-2]

(I know that linked question is about panning, but the principle is the same for crossfading.)

[sadhu]

Thanks for the response! That panning math using the sin and cos looks simple enough I may try to implement that. I am not really noticing the issues with the edges of the Square Root pan though so maybe my use case is fine to just use that. I may just code up both and see which one I intuitively like better. I assume the SQRT uses less cycles so if I can get away with that I maybe just keep it.

[martinrobinson-2]

OK if it works then that’s good. This graph should help illustrate what I mean:

[sadhu]

I coded this, does it look pretty much correct? It is sounding fine to me but I want to be sure. In this case the “fxKnob” sweeps from -45 to +45

const double degrad = M_PI / 180; const double sqrt2d2 = sqrt(2)/2.0;

double fxAngle = degrad * fxKnob; // convert degree to rad

wetMix.setGainLinear(sqrt2d2 * (cos(fxAngle) + sin(fxAngle)) );
dryMix.setGainLinear(sqrt2d2 * (cos(fxAngle) - sin(fxAngle)) );


[martinrobinson-2]

45° is pi/4

So assuming your fxKnob is in the range [-1,1] then just multiply by that for your fxAngle

[sadhu]

Yeah I guess I wouldn’t want the "mix"parameter exposed to the user as going from -45 to 45. I switched it to the -1 to +1 system and it sounds great. Thank you

[martinrobinson-2]

Ah yes sorry, I see now. I missed the bit where you said the fxKnob was ±45!

[refusesoftware]

In Juce 6 you can get a bunch of these crossfading options by using dsp::DryWetMixer and choosing a DryWetMixingRule.

[sadhu]

Oh that is interesting, would the equivalent of the above algorithm be the “balanced” rule then? I finished the project I needed this for but I am sure I will have to implement the same thing at some point in the future.