Invert spectra
Hello
I've been playing around with FFT a little bit, but I can't figure out a way to totally invert the spectra. I mean, I want the whole harmonic relationship to be inverted without changing every harmonic's intensity, but I want the low pitches changed to be high and viceversa.
I googled this: http://stackoverflow.com/questions/3520557/how-to-spectrum-inverse-a-sampled-audio-signal , but I wouldn't know how to do it in PD.
Any ideas?
Thanks.
i'd guess, you will probably hear nothing. the upper half of the bins above the nyquist frequency will be a couple of zeros. if you mirror them to the lower half, you can imagine, what you would hear.
I've already tried blockmirror~ but is not what I want 'cause it takes a block and reads the bins backwards. Maybe a graphic will explain better what I intend.
Anyway I really didn't thought something you wrote (toxonic), "but since fft converts the signal from the time domain into the frequency domain". I'll pay attention to that and try to figure out something.
here's what i did earlier. i don't really understand FFT at all very much, so i know it is wrong. (whatever i did, it sounds pretty useless, anyway)
my guess, is that it would be better to keep the phase of the original signal, and just invert the frequency. there's a bit of FFT maths involved to do that though:
http://eamusic.dartmouth.edu/~book/MATCpages/chap.3/chap3.pops/3.4.pop.mag.html
i have no clue about fft too, but i guess you should be able to specify the amount of bins to be reversed.
i didn't know any object to do that, so i converted the signal of real and imaginary part to a list of floats with [pack~], splitted the list at a variable point via [zl slice] and reversed the lower part with [zl rev]. after this i repacked the whole stuff with
and reconverted it via [unpack~].
i'm pretty sure that this ain't what you wanted, but it sounds pretty funny, changing the "block size" of bins to be inverted.
wow man, you're amazing! it works perfectly!
You guys keep on saying that you don't understand FFT, but trust me when I say that you know way more than I do. I'll have to figure out how to implement it in Pure Data some day. Anyway, I don't really know how your patch works, toxonic, but it certainly sounds very interesting. I tried running some audio through it and got some really cool almost metallic sounds coming out. It could be useful in some contexts.
toxonic:
did you ever connect that FFT shift subpatch that is hidden in your patch? i just tried that and it works really well.
i can see a really neat school of fx coming out of this:
as far as i know, the following can all be pretty easily applied:
inverse
shift
all sort of EQ effects
feedback
then, some that i just brainstormed, but haven't tried yet:
reduce the resolution of each value output by the FFT
use lfo's (or envelopes, or whatever) to modulate FFT outputs. i already tried this a little bit with one of my DIY objects, and it works well.
rather than just an inverse, maybe try some random scrambling on a selected range of the FFT to see what happens.
again, might be even more interesting if the range gets modulated with an LFO or something.
however, the bug 'AHA' moment i just had was that it will be a million times more computationally efficient to include ALL of these things within one FFT analysis, and have switches to turn them on and off as desired.
you know, it's going in and out of the FFT domain that uses most of the cpu resources, so while you're in there, you may as well get as much done as possible.
i'll have a go with this in the coming week and see where i can get. any help would be excellent!
toxonic:
did you ever connect that FFT shift subpatch that is hidden in your patch? i just tried that and it works really well.
i can see a really neat school of fx coming out of this:
as far as i know, the following can all be pretty easily applied:
inverse
shift
all sort of EQ effects
feedback
then, some that i just brainstormed, but haven't tried yet:
reduce the resolution of each value output by the FFT
use lfo's (or envelopes, or whatever) to modulate FFT outputs. i already tried this a little bit with one of my DIY objects, and it works well.
rather than just an inverse, maybe try some random scrambling on a selected range of the FFT to see what happens.
again, might be even more interesting if the range gets modulated with an LFO or something.
however, the bug 'AHA' moment i just had was that it will be a million times more computationally efficient to include ALL of these things within one FFT analysis, and have switches to turn them on and off as desired.
you know, it's going in and out of the FFT domain that uses most of the cpu resources, so while you're in there, you may as well get as much done as possible.
i'll have a go with this in the coming week and see where i can get. any help or more ideas would be totally excellently appreciated!
yes, i built that subpatch after i found, that you can apply list objects on audio this way. it just rotates the bins order. there are a lost of nice list objects that can be applied that way. the pack~/unpack~ objects are very useful. this way you can also build audio busses with unroute/route, and many other applications.
please post more of what you know can be done to FFT bins (or any bins for that matter)
Oops! Looks like something went wrong!
