@polyplexmescalia Most people won't notice the difference between aliasing and the weird smearing of low quality mp3s, so downsampling might be enough for your needs? Perhaps fake the smearing with an allpass filter/phaser on top of the downsampling?

@willblackhurst What are you using to make the screen capture? I never had an issue using ffmpeg and jack_rec. A bit of speculation, it is how PipeWire and Jack are setup, your screen capture app does not speak Jack and PipeWire is trying to sync the realtime audio from Jack with the non-realtime audio of what ever screen capture app you are using. Pure conjecture, I have not played with PipeWire enough to say for certain, and the time I have spent with it was quite awhile ago. Does your screen capture app speak jack or what ever you are using for audio in pd?

When I did this I believe I did something like this:

When you are done with your capture just run killall on ffmpeg and jack_rec and then use ffmpeg to combine the audio and video. Audio and video were not quite synced but they were close enough for the majority of cases and you really had to look for it to see they were out of sync, not difficult to fix if needed.

A little ridiculous, had to jump through some hoops to get that symbol back into a float. I used the -u flag of [fudiparse] so I could just use a simple [list split] in an earlier version but [fudiparse] and [fudiformat] should drop in now. There is some external which would make the symbol to float conversion easier but I can never remember which one it is and I have been trying to figure out the logic of why some float symbols will not convert with [float] and they weirdly are not even consistent regarding which hoops you have to jump through, first time I had to form a list and put it into a [text].

What about implementing your own fixed or floating point encoding for sysex? Break the float into two ints at the decimal, first first bit of the message is the length of the first int, reconstruct after the fudiparse?

Edit: Suppose a simple integer based fp implementation would not actually work here, did not think that through.

@Vnms nope, no python knowledge required, you just run the script with the output of the web app as args and it gives you what you need to stick into [biquad~].

@Vnms I have used https://github.com/drmartinjaros/biquadgraphpd for calculating this stuff, makes it simple and might help you find your flaw and should work better than ChatGPT.

@gentleclockdivider manual or help files or here on the forum, too long ago to recall. Seb gives the full answer above, I put it in pd terms since most (including myself) find pd far easier to understand than pd's dense C.

Regarding Max's join, make it, just an abstraction away. [list store] mostly does that and a great deal more.

@gentleclockdivider Yeah, will fix that error. The first inlet in the case of [pack] is always a list, a float or symbol is just a list of one element list. Internally [pack] works like this:

So left inlet will always add the appropriate selector even if it is a single element.

@gentleclockdivider First inlet of [pack] and many objects treats everything on the first inlet as a list which it unpacks and fans out to the other inlets in order, a symbol or float is just a list of one element in this case.

This is standard behavior for vanilla objects. [pak] is not vanilla and many external objects go against vanillas standard behavior.

@FFW try this, open ffw-test.pd and see if it expands on mouse over and shrinks when the mouse leaves. Restart pd before opening it just in case you have one of those weird things going on caused by a confused instance of pd. This should work,
Archive.zip
Edit: if it doesn't work, what version of pd are you running?