Does such a thing exist? I am fairly new to pd and am having a very hard time knowing where to look for externals. I am detecting pitch using [helmholtz~] and would like to do FFT on the audio stream with frames the size of the pitch that [hh~] outputs

Thanks!

@vobb "set" messages to [block~]........ block~-help.pd
From Pd-Extended doc.
Only power of 2 if you need audio in/out from the sub-patch.
Other values only If it doesn't communicate with the outside world.

This will probably be useful info....... Pd timing https://forum.pdpatchrepo.info/topic/7436/vline-metro-and-tabwrite-timing....... and problems with Fourier windowing @katjav ....... https://www.katjaas.nl/home/home.html
David.

@vobb said:

Hello and thanks for taking the time. FFT is not restricted to power-of-two but it runs most efficiently that way, unless this SE thread is completely off it's O(N log N) vs O(N^2))

Good link, thanks, I knew that the optimal FFT recursively divides the frame in half, but hadn't thought about other small-integer factors. Actually I can't quite visualize how it works except for the optimal case, but I'll trust that the FFT library knows what it's doing.

It's very likely, though, that the pitch will be a non-integer number of samples. And syncing the block size to a changing pitch is likely to be... amusing.

Anyway hope you get some useful data from it.

hjh

@whale-av Thanks, that's very useful. I am on vanilla and the docs for [block~] was a bit anemic. @katjav's site and posts have already been invaluable! I am targeting frame size = wavelength so that I do not have to worry too much about boundary effects. Unless my limited experience with FFT is wrong, the best FFT frame size will be the one that is exactly the wavelength of the monophonic sound I'm analyzing, hence the combination of her [helmholtz~] component with FFT. I may still end up writing this in some sort of external such that I have full control over the frame, upsampling etc.

I have not yet tried windowing with overlap and such, but somehow I doubt that it will be as good as a perfectly fitting rectangular window since my frame sizes are so small. If my initial plan does not work out, then I might try this.

@vobb If you can tolerate a little latency, maybe you could turn the problem around and instead pitch shift the signal down so that its period is the same as the next larger power-of-2 FFT frame?
pitch shifting to match FFT f0.pd

@jameslo I am very latency sensitive since this is a realtime application hooked up to a guitar. I will investigate your patch and see if the latency added is acceptable, thanks!

@vobb Here's a version with only the latency incurred by the FFT window, see what you think. I added the ability to change window sizes on the fly and did the pitch shifting in a delay line instead.
pitch shifting to match FFT f0 v3.pd

(I revised this post, too many mistakes in the last one)

@jameslo very kind, I will try this and report back results.

@jameslo Have not yet hooked it up to the rest of my patch but just doing some prelimimary testing it looks very promising. At the very least this will allow me to develop and test the rest of my ideas, then if I absolutely need the lowest possible latency I will write an [fftvar~] external when I'm done with everything else.

Massive thanks! I'm finally realizing a project I've been thinking about for years

@vobb Did you ever finish this? If so, can you share some sound?