Ken Murphy

I'm working on a project where I would like to analyze a chunk of audio of arbitrary length, and from that generate a smoothed-out "pad" composed of the average harmonic content of that chunk, which could be looped indefinitely, and not have any rhythmic component. Sounds that are louder and/or more sustained would have their harmonic content more strongly represented in the pad, and harmonic content of the the quiet/transient sounds would be be further down in the mix. Sort of the audio equivalent of a long-exposure photograph.

My math skills are middling, but have the gist of FFT/iFFT. It seems I should be able to convert the signal into the frequency domain, capture successive frames, and calculate the average for each bin over the desired period. Then I would need to convert the resulting averaged frame back into the time domain in a way where it can be played continuously and smoothly.

It seems like this should be possible with fft~/ifft~, or possibly pfft~. But I also stumbled on frameaverage~ (in Max7), which sounds like it might do just this. I'm still on Max5, so hoping for any clues on this before I make the plunge.

Any thoughts / suggestions? Am I on the right track?

I'm working on a project where I would like to analyze a chunk of audio of arbitrary length, and from that generate a smoothed-out "pad" composed of the average harmonic content of that chunk, which could be looped indefinitely, and not have any rhythmic component. Sounds that are louder and/or more sustained would have their harmonic content more strongly represented in the pad, and harmonic content of the the quiet/transient sounds would be  be further down in the mix. Sort of the audio equivalent of a long-exposure photograph.

generating-averaged-audio

yes, all that seems pretty reasonable, sounds like a plan. frameaverage~ is in max 6 too, is vectral~ available in max 5? That should do it too.

Otherwise, you could build your own frame accumulation/averageing using a delay that has a length of your frame size/buffersize.

hi!
yes, all that seems pretty reasonable, sounds like a plan. frameaverage~ is in max 6 too, is vectral~ available in max 5? That should do it too.
Otherwise, you could build your own frame accumulation/averageing using a delay that has a length of your frame size/buffersize.
cheers!


here's something I made a while back-- it's a pfft patch and a parent patch-- may be of use:

here's something I made a while back--  it's a pfft patch and a parent patch-- may be of use:


Woyteg, thanks for the pointers (and the vote of confidence that I'm on track)! Yes vectral~ is in Max5 so I'll investigate that.

Floating Point, thanks for that patch! I ran it on my trial version of Max7 and it's very close to what I have in mind. I suspected I wasn't the first to attempt such a thing... :)

Floating Point, thanks for that patch!  I ran it on my trial version of Max7 and it's very close to what I have in mind. I suspected I wasn't the first to attempt such a thing... :)

Unfortunately I can't help, but Im interested in your project!

the patch I posted doesn't need max 7 to work; it also works on max 6 if that's any help...

Another approach I'm trying is granular synthesis, with a very high grain density (e.g. dozens of grains playing simultaneously), selected randomly from the segment of audio I want to "average". Initial experiments seem to show some promise... with the high density, the transients are still audible but do get buried in the mix to some extent. And perhaps some transient content is a good thing as it helps capture the character of the audio.

(ps I'm surprised how many samples my 2011 mbp can play simultaneously without pinning the CPU!)

I recently updated a granular synth of mine, which can generate pretty good spectral blurring and unusual resonances (albeit in the time domain); you have control over grain shape (from reverse saw * sine -> saw * sine), grain location +/- N milliseconds, playback speed, and all/most of the usual parameters.

https://cycling74.com/project/granary-a-hybrid-granulation-engine/

And you are probably already familiar with these, by J F Charles

https://cycling74.com/tools/live-spectral-processing-patches-for-expo-74-nyc-2011/

Hi Ken
I recently updated a granular synth of mine, which can generate pretty good spectral blurring and unusual resonances (albeit in the time domain); you have control over grain shape (from reverse saw * sine -> saw * sine), grain location +/- N milliseconds, playback speed, and all/most of the usual parameters.
https://cycling74.com/project/granary-a-hybrid-granulation-engine/

And you are probably already familiar with these, by J F Charles
https://cycling74.com/tools/live-spectral-processing-patches-for-expo-74-nyc-2011/

Excellent, will give it a try! Thank you Brendan.

Generating "averaged" audio