Markus Baumknecht

I've been working a lot with Alex Harker's multiconvolve~ object recently (not for reverb; more artistically, for example by convolving two voices with each other) and one thing that I've wanted to control is the way in which some frequencies become overly resonant if they are contained within both the sample and the impulse response.

in order to solve this problem I've been trying to write an automatic bin amplitude scaling pfft~ patch. the idea was that I could attenuate frequencies differently according how loud they are, with more quiet bins hardly being rescaled at all while louder bins being attenuated more strongly. kind of like a dynamic EQ but finer and completely automatic.

the solution I've come up with is to use the scale~ object inside a pfft~ patch; my logic was to use it's fifth argument to determine the nature of the scaling curve so that frequency bins with a lower amplitude would be rescaled less strongly than frequency bins with a higher amplitude. in theory, with values below 1. (= logarithmic scaling) this should lead to more or less linear output for lower bin amplitudes whilst flattening the curve in louder regions.

the overall result puzzles me. when 'rescaling' a convolved input signal like this it seems to me that the result sounds less muffled and richer in higher-frequency content (which to me at least gives the perception of greater clarity).

however, louder frequency bins don't appear to be attenuated. additionally, the lower the value put as the fifth argument (e.g. 0.9 -> 0.5) , the more clearly audible are some artifacts that I cannot clearly identify (my guess was some kind of harmonic distortion ...?).

I was hoping anyone could shine a light on what I've been doing wrong or actually what I've been doing at all, because I don't actually understand what's going on.

if anyone knows whether I could achieve my goal more efficiently with other means, I'd also be very grateful.

convolution_unscaled.mp3

fft_filter_modified.maxpat

fft_filter_tester.maxpat

in order to solve this problem I've been trying to write an automatic bin amplitude scaling pfft~ patch.  the idea was that I could attenuate frequencies differently according how loud they are, with more quiet bins hardly being rescaled at all while louder bins being attenuated more strongly. kind of like a dynamic EQ but finer and completely automatic.

the overall result puzzles me. when 'rescaling' a convolved input signal like this it seems to me that the result sounds less muffled and richer in higher-frequency content (which to me at least gives the perception of greater clarity). 

however, louder frequency bins don't appear to be attenuated. additionally, the lower the value put as the fifth argument (e.g. 0.9 -> 0.5) , the more clearly audible are some artifacts that I cannot clearly identify (my guess was some kind of harmonic distortion ...?). 

I was hoping anyone could shine a light on what I've been doing wrong or actually what I've been doing at all, because I don't actually understand what's going on. 

hi everyone 
I've been working a lot with Alex Harker's multiconvolve~ object recently (not for reverb; more artistically, for example by co

pfft~ // automatic bin amplitude scaling using scale~