Hi!

I am trying to build an FFT analyser so I can use it as visual feedback for an equaliser.
In order for this FFT graph to continuously average the amplitude values of each frequency, I wanted to be able to average this values as more data gets drawn in the graph.
The idea is to use a tabwrite of a certain block size, and then have several tabreads, with the block size of the initial one, divided by the number of tabreads, and then avergae the values outputted by the tabreads.
My biggest problem at the moment is to make the various tabreads read data from one value n to N/x; N being the block size, x the number of tabreads and n the value at which the previous tabread stops reading.
I have tried using [count] but it is very slow, since it depends on the metro, and I don't seem to make it work with [until] either...
Any suggestions?

Thanks for your time!
Filipe

If you can do the whole thing in the ~ signal domain, and keep it synced with [bang~], you might have better luck. Control domain objects like [metro] and [until] will not remain synced with signal, as cpu load varies.

I can't totally picture what you are trying to do (average x previous versions of the FFT?) but how about making a "leaky" FFT table integrator as in the attached patch. Every time you compute an FFT, compute a weighted sum of the new FFT and the old weighted sum. Basically it is a 1-pole low-pass filter on the the FFT value in each bin running at whatever metro rate you like.

Maybe this can be done at block rate without the tabread~ , but I got a "DSP loop" error when I tried that.

http://www.pdpatchrepo.info/hurleur/fft-leaky.pd

@emacpher said:

I can't totally picture what you are trying to do (average x previous versions of the FFT?) but how about making a "leaky" FFT table integrator as in the attached patch. Every time you compute an FFT, compute a weighted sum of the new FFT and the old weighted sum. Basically it is a 1-pole low-pass filter on the the FFT value in each bin running at whatever metro rate you like.

Maybe this can be done at block rate without the tabread~ , but I got a "DSP loop" error when I tried that.

All I was trying to do was to give an "averaged look" to my graphical representation of the FFT. So basically, instead of having an ever-changing graph, the frequency amplitudes will accumulate, and by the end of the track (for example), you can tell which frequencies/frequency bands are more present.

I think you solved this problem for me emacphe, so thank you ever so much. All I need to do now is adapt it to my application in order for it to work better with music and so on. Any suggestion on how to make a graphical representation logarithmic, instead of linear?

Thanks!

If you want the FFT values in dB, then change [sqrt~] to [expr~ 10 * log10($v1)].
If you want the frequency scale logarithmic, try http://puredata.hurleur.com/sujet-5282-array-logarithmic-scale

Great, thank you very much! I had seen the link you gave me before, but for some reason the representation of the lower frequencies isn't that accurate and user friendly. I'll try to modify it, in order to get a more accurate representation.

Thank you very much for your help once again.

Just discovered an error in my patch.
Please replace [tabread~ spectrum] with

[r tick]
|
[tabplay~ spectrum]

Otherwise the accumulated spectrum "signal" is not in sync with the new one.

Thank you very much. I find that when making that change, the curve is also a lot smoother and much more intuitive.

Thank you.