[pulseaudio-discuss] Best place/way to incorporate a convolver into the audio stream

[Jason Newton]

On 08/18/2009 07:35 AM, Neil Wilson wrote:
> Thanks,
> 
> My original thoughts were to take something like Brute-Fir and adapt
> it with a pulseaudio IO module. Since you can already prod BruteFir
> with a CLI it seemed like a good way to get something going that could
> have a GUI grafted onto it afterwards.
> 
> Did you look at this approach and reject it for some reason?
> 
> 
For FIRs >> ~100 samples, FFT based convolution is usually significantly
faster.  CLI can still be done with dbus - I already provide a way of
shoving in coefficients straight to the filter, I was just motivated to
do an equalizer and not a full on FIR sink.  All of my design decisions
beyond pure dsp were influenced by Lennart's ladspa module as it was the
closest preexisting thing and even then, it's a tricky thing because of
how much data one operates on at a time and how much time it takes to
process (<<1 s, but ~ enough to agitate pulseaudio on some border
condition environments).

I am interested in making it more of a fir-filter-sink but what type of
filters/kernels would you be interested in using?  I haven't really
thought of anything aside from echo that could be useful (equalizer
covers all of the variations of bandpass filters).

Jason