Hype aside, nobody builds a filter bank like this any more

Nobody asked what it cost

Albiswerk Zürich built the tube version in the fifties and this germanium one in the sixties. As far as I can tell they were made to order, which makes sense, because the studios signing for one were state-owned. Public money, and no need to design anything down to a price.

You can hear where that went. The filters are passive and inductor-based, and every channel carries its own transformer. Thirty-six channels, thirty-six transformers. The topology is the one you might know from a Pultec: filter passively first, then make the level back up with a line amp, here a germanium one.

Then there are the faders. A fader on this thing is not a slider running along a track. It's a small mechanical lift, and pulling it drags a wire that turns a mechanism inside the unit to set that band's gain. You feel the parts moving with you. Somebody decided the mechanism deserved the same care as the circuit, a very Swiss way to think, and all that machinery ends up in the sound.

A filter built today, modular or a bank by Moog or Buchla, has to land at a price somebody will pay, and every choice inside it bends toward that number. This one got inductors, transformers and the best components available on all thirty-six channels because there was no reason it shouldn't. It also weighs enough to break my back whenever I even consider moving it.

Three years of looking

I spent three years looking for one I could touch. Once a month I went through the forums, the auction listings, the obscure local sites that only post in their own language, then a plain Google search in case something had surfaced somewhere I wasn't watching.

Two museums in Germany hold these units. I wrote to both, naively thinking they might let one mangle some music for a change. Neither replied. I get it. Now that I have one, I wouldn't let go of mine either.

And of all places, I found one two hours away from where I live

Then I bought some gear from a man who runs a vintage studio here in Poland, and at the back of his control room sat a pair of them, wired up for mastering. Three years of watching auctions across Europe, and they were a drive away. He wasn't keen on selling. I was so close, though!

I called him once a month after that and told him, gently each time, that I'd love to take one off his hands. Six months of this. And then he said yes.

He brought them over while I was still renovating the building. They stood there for a couple of months before I had a room finished enough to hook them up. When I did, they mostly worked, which is about as much as anyone can ask of a sixty-year-old instrument.

Two units, one spectrum

The 502/74 is the main unit and covers 94 Hz to 6250 Hz. The 502/72 takes the two ends the main unit leaves alone: 31.8 Hz to 78.5 Hz at the bottom, 7.5 kHz to 16.5 kHz at the top. Run them together and the whole audible spectrum is split across thirty-six bands. The BBC Radiophonic Workshop could only afford one of the two, and took the 502/74.

Every band only cuts. The top of the travel is 0 dB, the filter doing nothing (in theory), and the bottom takes that slice of the spectrum away entirely. There is no boost anywhere on the unit.

And the filters sit in series, bands overlapping. Drop two sliders side by side and leave the one between them up: the two you dropped pull the middle one down with them.

So what looks like the vintage hardware equivalent of a Pro-Q doesn't work like a Pro-Q at all. Razor-sharp cuts are not its forte. Shape is. You draw with it.

How the studios actually played it

Photographs from GRM and WDR show these units mid-session: bands arranged into waves, cliffs, irregular geometry. No single-band notches, no surgical work of any kind. Long curves running across the spectrum, or a whole region taken out with a run of bands pulled down together. Shapes you'd draw with your whole hand.

I rebuilt those settings as presets inside Smugi. Click the plugin name and they're waiting, in the positions the engineers left them.

Warsaw, and the 'three at a time' idea

The Polish Radio Experimental Studio never had an Albiswerk. They worked with an RFT bank, built in East Germany and a considerably simpler machine.

There's a documentary from the era where you can watch Eugeniusz Rudnik working it. He doesn't reach for one fader. He lays his hand across three, moves them together, and reacts to the change in real time, hunting for a response he hasn't heard yet. He isn't setting a filter. He's playing one.

That's where Smugi's front panel comes from. Three faders move as one, which turns thirty-six bands into twelve you can get your hands around.

None of the bands went anywhere. All thirty-six still run underneath, still overlapping, still pulling on each other as they do in the units. The only thing that changed is what you touch. Producers are running Smugi live now with the faders mapped to a MIDI controller, which is the best endorsement of that choice I could ask for.

Sampled, not modelled

Sampling the pair took a couple of weeks, and it's a deep sample. The path is the one I use for everything: DA converter into the units, straight back into an AD converter, with the conversion kept as clean as I can get it so the only thing colouring what you hear is the Albiswerk.

Modelled plugins have always sounded dead to me next to the hardware they emulate. My best guess why: the frequency response is the easy half. You can measure a curve and match it. Phase and transient response are the rest, and they're what your ears read when something sounds like a device instead of a picture of one.

So what Smugi has in it is the real interaction between those bands, phase and transients included. You pay for it twice: once in development, because it takes ages to get right, and again in CPU, because a sampled model costs more than an algorithmic one. If I have to choose between the sound and the session's buffer size, I'll take the sound every day.

What I added, and what I removed

Every plugin I make starts with a design brief I write for myself, always the bare minimum to capture and test the concept. Then I spend around two months using it. Real sessions, different material, no plan. Everything in Smugi beyond the original units came from something I noticed in those two months.

Mute is the plainest, and the one I enjoy most: drop a band, bring it back, by clicking its label. It sounds like nothing until you're performing rather than mixing, and then it's the control your hand goes to.

Movement puts an envelope follower on every band, tuned so each one answers to what's passing through it. Feed it something rhythmic and the bands breathe in time with the material. Push it further and the source comes apart into pieces of itself. Since it reacts to the input, the input gain is worth experimenting with here too.

Feedback sends a band back into its own input.

The macros move all of it at once. They're the fastest way to play the plugin: perform the whole bank as a single filter, low-pass, high-pass or tilt, with thirty-six interacting bands doing the work out of sight, so it bends in ways an ordinary filter won't. And there's a step sequencer for when you want that movement strict rather than played.

I also built pan per band. On paper it's the obvious feature. In practice it never gave me anything musical back, so it went out. Adding features is easy. Pile them up and you get an interface nobody wants to explore.

What it's best for

I made Smugi to get somewhere else with a sound. To put something in and have it come out as a texture with no obvious relationship to what went in.

It turned out more flexible than that. There's a mastering studio running it on the master. Francesco Donadello told me he uses it on the orchestral bus for A-rate film scores. So the question was never what it's made for. It's what you make with it.

What I like most is the thought of somebody loading one of the WDR or GRM presets, putting a hand across three faders, and going somewhere none of those engineers got to. That's what the device was built for, and I'm proud it can do that job again, for people who were never going to get within a hundred miles of one. It's being played, instead of sitting in a museum.

The trial runs on your own sources, free, with no time limit. Load a preset and take it somewhere new.