Re: [chuck-users] Convolution Reverb (Posed by: Julien Saint-Martin)

If the impulse response is short enough, or real-time isn’t important, then the FIR chugin can be used. FFT-based convolution is possible using the FFT/iFFT UANAs, but latency and real-time are issues. To deal with that, It would be possible to factor into a nice systolic topology using shorter FFTs and delay lines to line everything up. I don’t know of a specific chugin or example other than those. Someone should write one :-)) Prc Sent from my iPhone

Thank you Perry!
I want to use large IR so the FIR method is not possible.
I found the file attached from your Kadenze course (thanks also for that,
it was really cool!! ).
But as you write in it, it is not real time processing.
Regarding the "nice systolic topology using shorter FFTs and delay lines to
line everything up" method.
I searched the internet to understand. Please tell me if my understanding
is correct ?
- The IR must be divided into shorter chunks.
- Each chunk is associated with a FFT/IFFT Uana (kind of processing unit).
- Each processing_unit (except the first) is preceded by a delay line of
size : chunk_size * chunk_index (starting from 0).
- On processing, each audio packet (chunk_size) is sent to processing_unit
0 and all delay lines.
- Then when all processing_units are done, we sum every audio outputs.
Do you think I can use the method in your file attached to build a
processing_unit ?
Best Regards,
Julien
Le ven. 1 mars 2019 à 18:09, Perry Cook
participants (2)
-
Julien Saint-Martin
-
Perry Cook