visual spectral filter

your mapping from "fft-bin-to-midi" looks correct, except for one very important error-- you are assuming that bins can be remapped from linear to log scale-- they can't-- not in the sense you have here, because bins, as discrete "objects", or structures are fixed in a linear context; they lose their meaning in this remapping process.
for example, the first two bins represent a whole octave range (of midi), while the last two bins represent 1/256 of an octave range.
To get around this you need to have something like a single row of about 2048 cells ( 8*256, 256 cells per octave) and interpolate values in the lower octaves to fill in the gaps (or just leave them blank).
You could do this by having a separate 256*1 matrix for each octave with a mapping like this:

bin octave
1 -> 1 (cells 0-255)
2,3 -> 2 (cells 256-511)
4,5,6,7 -> 3 (cells 512- 767)
8,9,10,11,12,13,14,15 ->4 (cells 768- whatever)

etc

then glue the matrices together to form a single row
the interpolation would be easy as matrices can be set up to interpolate automatically

another less heuristic way is to do the maths to map bin number to cell-number, with or without interpolation, and use the setcell message to set correct value for each cell in your midi-row, but either way is ok probably because theoretically fft information can't be remapped like this without loss or distortion of information

I was wondering how I could get around the log-linear problem. Probing wth [capture~] I saw exactly what you just explained (one giant bin at the beginning etc), only I didnt know how to fix it. I will try your solution and post the results here in case anyone else is interested.
Thank you!

So I tried your solution

Something doesn't work between the [jit.catch~] and the [jit.glue]

before I got your response I worked out a solution on my own and I was wondering if you could take a look at it and see if it does something similar to what you described here it is

let me know your thoughts

sounds very interesting, what exactly is the patch doing?

the patch is generating a single row matrix of magnitudes with a midi scale distribution. If you feed an audio signal to it you can use the matrix as a filter for other matrixes. I work with audio-visual patterns, they look like the midi piano roll notation, so I use the visual filter to only select the frequencies of the pattern that are being played. I will post some examples next week on this page
https://michelezaccagnini.com/electronicinteractive/

I looked at your first patch and fixed it for you (not perfectly, not sure whats happening with the last octave) before looking at the second one-- the second looks more efficient and simpler-- not sure why you divide the signal when it first comes out of the fft object-- anyway the first one has some good pedagogical value...

Thanks Floating Point, I divide the fft signal to normalize the magnitudes (I learned this form this tutorial https://cycling74.com/tutorials/advanced-max-ffts-part-3). I tried the fixed patch but I am getting errors from the pwindow. I see your point about splitting the matrix now...

oh yes, of course to normalise, I watched that tutorial last week!
the pwindow errors are just because max is in overdrive-- normally you need to take max out of overdrive when using jitter
btw if you want one-semitone "resolution" change the dimension of the interpolating matrix to 12

hey I realized, for the version I looked at, you don't need all that gating etc,, so here's a simplified /optimized version