Airwindows

kCathedral5

TL;DW: kCathedral5 lets you place yourself anywhere you like in the space.

kCathedral5 in Airwindows Consolidated under ‘Reverb’ (CLAP, AU, VST3, LV2)
kCathedral5.zip (613k) standalone(AU, VST2)

It’s high time I moved on to the many other sonic spaces I’ve promised to cover, but there’s reasons why this immediately follows kCathedral4.

Yes, I’d hit upon using a proper Householder matrix for the 3×3 early reflection matrix, yes the work on the genetic algorithm was giving me better sounding reverbs every day, yes I’d worked out ways to measure the results for peak energy over RMS in a way that exactly translated to ‘actually sounding far away rather than just a wash…

But two problems remained, only one of which I was aware of.

Firstly, early reflections were weird. They totally replaced pre-delay: the pre-delay included in kCathedral4 proved nearly useless as it was always too much delay, you couldn’t turn it down and sit things in the reverb. Turns out I had to invent a whole new way of doing these early reflections, still taking only values from the big reverb for tonal consistency, but instead using a spread of delays, always in order of bigness, and not only ‘the ones on the end’ but any set of delays in the reverb. I thought it’d be best to only use the longest delays, but the need for less predelay forced my hand.

At which point I learned that doing it let you move the sound source in the room, from right up next to the mic to back against the back wall. And every position sounded distinct and different, yet part of the space. And very low positions sounded like the instrument was right up against the wall, and adjusting that was just like moving drums out into the room and searching for that ideal spot, just like musicians and producers and engineers have done in real rooms back in the glory days of recording in studios…

And then it turned out I’d been doing undersampling wrong this WHOLE time (which showed up as one user observed a funny burst of energy over 30k when recording at 96k, and wondered what it was)

This was something I saw too. I’d run a test and at very high frequency there’d be this little ripple. And I experimented around and discovered: yes, and it comes from counting both the beginning sample and the end sample (of a downsampling Bezier calculation) rather than having them work as effectively the same sample. It involved a tricky little calculation, and then another for when the Bezier undersampling was continuous rather than stepped, and I ended up sorting it out.

Only to learn that the Bezier filtering used exactly the same fix. And the Bezier de-rezzing, which is about the same thing but has a bitcrush… And that every single reverb plugin I had going, including ones I was actively using in music, required this fix on every plugin for every build and every platform and every version…

I’ve been doing that. kCathedral5 shows you what is coming. I know I’m very interested in exploring the much smaller reverbs as people have asked for: kBeyond isn’t representative because it doesn’t have this fix and also doesn’t have proper Householder early reflections, or the ability to position at different places in the room. It does have the 6×6 matrix, though. Think of it as on the right track but more bright and tizzy than it has to be, and without nearly enough depth compared to what is possible. Sheesh, every time I name a plugin ‘kBestEverSeriouslyGuys’ it’s not a week before I find out something better to do…

Having done a whole bunch of revising and fixing, I’m pretty tired, so I’m going to step away and take a little time just to decompress until early or mid September. I’ve got all my firewood in and my fuel oil paid for, so I can gather my strength for the further sprint that will be ‘finishing up the component parts for, and the dedicated plugins of, ConsoleH’ which I’ve promised by 2026.

It seems like a great time to try to do something nice for people, so hang in there while I take a few weeks to breathe, and then it’s going to be the universe of hip-hop sound reinvention. We’ll probably see the ‘Donuts’ plugin. It’s NOT going to be ‘a clone of the thing Dilla used’ because I don’t do those, but all the same I think I can get the vibe.

Talk to you folks in a few weeks, and I hope you enjoy kCathedral5. :)

Airwindows Consolidated Download
Most recent VCV Rack Module
download 64 Bit Windows VSTs.zip
download Signed M1/Intel Mac AUs.dmg
download Signed M1/Intel Mac VSTs.dmg
download LinuxVSTs.zip
download LinuxARMVSTs.zip for the Pi
download Retro 32 Bit Windows VSTs.zip
download Retro PPC/32/64 Mac AUs.zip
download Retro PPC/32/64 Mac VSTs.zip
Mediafire Backup of all downloads
All this is free and open source under the MIT license, brought to you by my Patreon.

Date August 24, 2025
Author Chris
Comments 9 Comments

TapeHack

TL;DW: TapeHack is a new dimension in tape realism, abstracted to software.

TapeHack in Airwindows Consolidated under ‘Tape’ (CLAP, AU, VST3, LV2)
TapeHack.zip (491k) standalone(AU, VST2)

It started with a paper by Spencer Walters, on this subject: what’s the role of hysteresis in magnetic tape recording? We tend to assume that’s fine, we’re making it linear, apart from all the soft saturation of course. Even then, that’s not quite where it starts: it’s research on Dolby HX-Pro that got Spencer thinking. I’m not sure the Dolby people quite understood the implications of what they had, as that was an era of ‘remove all distortions’.

Gradually, Spencer started to figure out what he had, and Spencer wrote a paper, talking of magnetic flux and relative coercivity and hysteresis loops. I frankly don’t have the math he does, but I worked with analog tape growing up, and have a good idea of how it responds. And there was this set of diagrams, of a 1kHz wave magnetizing tape, and what you actually get out of it without and with tape bias…

It showed there was a dead spot, if you didn’t bias. Well, I knew that: I’d coded something like it into ToTape 7 and 8, letting you dial in a section of Type AB distortion if it was ‘underbiased’. But look again: that dead spot is still there, if you bias. The recording is sweeping back and forth across it at 100 kHz, but it doesn’t stop that dead spot, where magnetic tape’s coercivity hasn’t kicked in, from still existing. You’re just keeping it way busier than before, but a 100k wave still has to sweep through zero each cycle.

What if this dead spot was central to tape sound as we know it? What if there’s always a hint of it, a bit of antisaturation in there at the opposite dynamic extreme from TAEP PHAT, as shown in Spencer’s diagrams? More importantly, could I code something like that?

Sure could :)

TapeHack does a number of interesting things due to what I came up with. First, it’s very efficient: it’s a simpler, lower-CPU form of a sin() function, so it has the softclip nature. It’s made to have this ‘extra zone’ by manipulating how an approximation of sin() is made, so I can also do variations on it and asin() and combinations of them, for the purposes of future Console versions. Doing that tends to take away the saturation effects, so there would be additional layers of TapeHack for making ‘vintage consoles’ sound as we experienced them.

The differences in waveform can be described. Unlike sin(), TapeHack hits a soft saturation point and then takes the flat top and begins to extend it, while turning the ‘quiet part’ of the wave into more of a trapezoid shape that’s remarkably persistent in its slope. The steepening curve of the soft-clip gets sharper and sharper while these other factors stay relatively consistent.

As a result, TapeHack does ‘tape compression’ better than anything I’ve ever made, full stop. Turns out you can make sharp softclip corners, but we hear ‘brighter’ as a function of how much more steeply that center part of the waveform slopes. It’s heard as a shockingly believable ‘tape bark’ on things like drums, it’s heard plainly when you run a heavy guitar sound hot to ‘tape’ to make it sit more up-front, but just as importantly, it livens up a more quietly ‘recorded’ track by giving it that characteristic dynamism and optimal distribution of peak energy.

A side note: I’ve posted classic vinyl records and measured them, tracking peak vs. RMS energy, as that’s been my own parallel pursuit of what went on in those very sonic Seventies, when things just sounded better to me. A common factor is WAY more emphasis on peak energy and crest factor, with it persistently sitting much better dynamically than digital accuracy gives you. Well, this lines up with that observation exactly. Applying this hysteresis (or, like with TapeHack, just using the transfer function directly) produces both the peak distribution and the desired sound.

I’ll be using what I learned in TapeHack for many things. Hope you like it!

Airwindows Consolidated Download
Most recent VCV Rack Module
download 64 Bit Windows VSTs.zip
download Signed M1/Intel Mac AUs.dmg
download Signed M1/Intel Mac VSTs.dmg
download LinuxVSTs.zip
download LinuxARMVSTs.zip for the Pi
download Retro 32 Bit Windows VSTs.zip
download Retro PPC/32/64 Mac AUs.zip
download Retro PPC/32/64 Mac VSTs.zip
Mediafire Backup of all downloads
All this is free and open source under the MIT license, brought to you by my Patreon.

Date August 17, 2025
Author Chris
Comments 8 Comments

TakeCare

TL;DW: TakeCare is a lush chorus ensemble universe.

TakeCare in Airwindows Consolidated under ‘Effects’ (CLAP, AU, VST3, LV2)
TakeCare.zip (575k) standalone(AU, VST2)

The idea was to revisit plugins like Melt, with an eye to even more disconcerting sonic spaces like the works of The Caretaker. Seemed reasonable to do. I started off with my Melt plugin, to see what would happen.

Then, in working on Householder matrices, I found that using a Householder matrix for a 3×3 grid of delays (rather than simply using them as delays, which I’d done before, thinking ‘what difference could there possibly be?’) made a HUGE difference. This is what’s creating early reflections in my plugins post kBeyond (which it turned out I immediately went beyond).

So, if you can do that with a 3×3 grid of delays that produces 27 echoes per channel… what if those delays were instead a chorus ensemble machine?

This is TakeCare. Most of the work I did on it, was using synth pads in order to make giant illbient soundscapes, and indeed it’s quite good at that. You can get real seasick off TakeCare. I’ve also set it up so the ‘buffer’ control can be manipulated to intentionally create horrible noises: at full regen, it’ll even feed back at small buffer settings, and it’s made to go into distortion a little below full scale in case you want to get real crunchy. It’s also using an old Console algorithm internally, to expand the space inside the blast of noise.

And then it turns out that if you keep it more calm, less regen, very small buffers and the depth not too extreme… it’s a gorgeous chorus ensemble on something as revealing as a vintage Fender Rhodes. It generates stereo width, supplies a lovely pad of lushness around the instrument, and doesn’t have to be weird at all. This could come in handy on things like synth pads, voices, who knows what? Think ‘lush stereo chorus ensemble’ and dial it in to whatever your needs are. The range on this is pretty nuts, because it was originally designed for radical uses but cleans up really well.

I hope you like it :)

Airwindows Consolidated Download
Most recent VCV Rack Module
download 64 Bit Windows VSTs.zip
download Signed M1/Intel Mac AUs.dmg
download Signed M1/Intel Mac VSTs.dmg
download LinuxVSTs.zip
download LinuxARMVSTs.zip for the Pi
download Retro 32 Bit Windows VSTs.zip
download Retro PPC/32/64 Mac AUs.zip
download Retro PPC/32/64 Mac VSTs.zip
Mediafire Backup of all downloads
All this is free and open source under the MIT license, brought to you by my Patreon.

Date August 10, 2025
Author Chris
Comments 6 Comments

Silken

TL;DW: Silken is a high frequency boost that gives ambience and texture.

Silken in Airwindows Consolidated under ‘Filter’ (CLAP, AU, VST3, LV2)
Silken.zip (543k) standalone(AU, VST2)

Here’s one where the plugin is probably better than my ability to demo it. Silken is a kind of high frequency boost, based on PrimeFIR used in ‘prime mode’ but backwards, so rather than being a lowpass with lots of ambient leakage, it’s a method of subtracting such a brickwall mostly linear phase lowpass with different leakage.

So that’s a lot. You can simply listen to it and see whether it is able to do ‘silky high frequency boosts such as you might use for lead vocals’ or you can bear with me as I try to explain HOW it does that thing. Because, even though stuff’s kinda hectic around here and my video wasn’t good, the plugin I made really brings a useful texture. If I could sing better I’d be all over this demonstrating how great it is, and even so it might help out. I feel it might click with people so it behooves me to explain how it’s real.

So, you can run a brickwall filter, using a ‘window’ (one of the controls) to determine just how steep the filter’s gonna be. It’s an algorithm called a sinc filter, and the wider the window, the steeper your brickwall can be. It’s a phase-linear filter, so it has pre-echo and it has latency. Silken does not compensate for this latency, which depends on how wide the window is: it’s a slightly unusual arrangement because it’s not completely symmetrical in an effort to cut down on the latency.

So far so good. It’s like a shelf for boosting highs and cutting lows. But then, bring in what PrimeFIR does. That lets you make the filter only from prime numbered samples, and not every sample. What happens when you make a sparse filter like that? I’ve made multiple plugins that use this TYPE of effect: BrightAmbience, and in fact my classic plugin Iron Oxide. There’s plenty of experience in using this type of effect… at least with me :)

PrimeFIR can make a ‘lowpass’ out of only prime-numbered samples (part of the filter, not just counting every sample in your audio) and it lets through a sort of ‘haze’ around the filtered sound, as audio across the whole window bleeds through. What Silken does is different in two ways. First, it’s subtracting the filtered part, to make it a highpass. Second, it’s constructing the filter out of only NON-prime samples, this time. So what’s happening is, it’s more effective at being a highpass than PrimeFIR is at being a lowpass, but the stuff that leaks through is still out of the prime numbers because those are now the ones NOT being subtracted.

You get a highpass where, the harder you push it, the more of an ‘aura of silky ambience’ you get around the highs. It will sort of diffuse super high frequency transient information, like a diffusion filter does for visual information. The result is flattering in exactly the same way a camera’s diffusion filter is. It should work fine on even the most high quality sources, but it should be an absolute lifesaver on the kind of nasty mic (like certain lavaliers!) that puts out distressing hyperfocussed bright transients. Now, you can diffuse that and change the texture of it, not just turn the brightness you’ve got up and down.

Hope you like it :)

Airwindows Consolidated Download
Most recent VCV Rack Module
download 64 Bit Windows VSTs.zip
download Signed M1/Intel Mac AUs.dmg
download Signed M1/Intel Mac VSTs.dmg
download LinuxVSTs.zip
download LinuxARMVSTs.zip for the Pi
download Retro 32 Bit Windows VSTs.zip
download Retro PPC/32/64 Mac AUs.zip
download Retro PPC/32/64 Mac VSTs.zip
Mediafire Backup of all downloads
All this is free and open source under the MIT license, brought to you by my Patreon.

Date August 3, 2025
Author Chris
Comments 2 Comments

Chris

kCathedral5

TapeHack

TakeCare

Silken

human-made bespoke digital audio

Kinds Of Things

The Last Year

Patreon Promo Club