[LAD] Leslie and convolution

[Robin Gareus]

On 07/26/2012 07:10 PM, Florian Paul Schmidt wrote:
> On 07/26/2012 07:04 PM, Julien Claassen wrote:
>> Hello Flo!
>>   OK, I'm with you so far. So I suppose, that it is possible to
>> process the IR of a time-variant system. The question is: Is it done
>> in any library already optimised for the audio domain?
>>   Thanks and best wishes
>>            Julien
>>
> 
> I sense a misunderstanding. With a single constant kernel you can only
> model time-invariant systems. A Leslie is NOT time-invariant. I.e. the
> response changes with time (the speaker rotates). So what you could do
> as a workaround would be to get IRs from the Leslie at various
> positions. Then convolve your input signal with each of them and mix the
> outputs together with the mixing coefficients being a periodic function
> of time.. You could probably get away with calculating only a few of the
> convolutions if your mixing coefficients are sparse, i.e. you only
> crossfade between two consecutive (in time) output signals.. (keep in
> mind the window length of the convolution kernel, though)..
> 
> It's of course a tradeoff - You need quite a few IRs to make the
> transitions smooth.. But then it's just a matter of hacking some code
> around an existing convolution engine..

You're pretty much describing
  https://github.com/pantherb/setBfree/blob/master/b_whirl/whirl.c