Polyphase allpasses update

That. The jumps at peaks make the system nonlinear: the peak envelope of a sum is not necessarily the sum of the separate peak envelopes. This means we cannot describe the filter as having a fixed frequency response, like, x dB down at each frequency for any input. There's probably some way to describe it, but not the usual, LTI way. In any case, the main feature of the release phase is a certain response in time, an exponential decay. Because this filter in particular doesn't produce purely exponential sections (the output is continuously mixed with the input), I could care a bit about its frequency response, but as it resists analysis (?) and the zdf version doesn't seem to help with the top end, it makes more sense to use a simple crossfade.

btw, I revised the release coefficient formula and replaced the file here, didn't want to make more noise about it 😁

Not sure, but I don't think so. I mean, zero-delay feedback is a way to describe an implementation of a digital filter converted from an analog model using the bilinear transform (which turns integration into summation using the trapezoidal rule, hence aka trapezoidal integration), as compared to other, "naive" implementations which are just wrong. So I guess it depends on the analog model -there may be filters for which the coeff of x[-1] is zero.

I think italics (with single asterisks) and bold (with double) still work.

Well, I wouldn't say they need to be, given how many compressors use a coupled arrangement. I just tried all the versions proposed in the Giannoulis paper. The core envelope follower is what they call a branching, level-corrected, release-to-signal detector (formula 16). It's not the most basic one, that would be (12). I found that all branching forms scale the whole signal -like, their release trajectories are entirely below the actual signal envelope, not just locally at peaks. Here yellow is average (50 ms), red is 1/50, green 10/50, magenta 30/50.

The branching ones are like intermediate steps from pure peak (zero attack) to pure average (attack = release). The decoupled ones keep the decay level, they just progressively smooth the peaks. None of them is "wrong", but I think the decoupled ones do more proper envelope extraction.

tbh, I don't feel quite at home with this thing, I've been using Discourse in other places for some time and it's been much more usable. I also don't like that they've completely removed the possibility of deleting comments -some days ago I commented, then realized I misunderstood the whole thing, then found out I couldn't delete it, then replaced it with ten dots to fill the minimum length requirement -.-

The JUCE forum, the SuperCollider forum, the Csound forum, for example.

I'm a bit reticent with Discord because I have a number of gamers around, and it tends to be like let's be all together at the bar all day (?). I may give up eventually though 😁

It's probably better on mobile than the old forum, but Discourse forums are good on mobile too. They're also rather slow, mainly because they have this Facebook-like infinite scrolling thing, but they're more... I don't know, usable. Like, here we can't make partial mentions, the "quote" thing doesn't count as a mention and doesn't keep a reference to the original comment. You load a file here, and there's no handle to it. If you want to replace it, you have to delete it with backspace. There are no options for the size of images. Notifications don't go away after you see them, you have to actually click them. It's all kind of... bare.