Click an image for a slideshow of every plugin, each shown with its full controls both hidden and open.
Audio plugins
Built to a professional standard, with published research behind the sound and safeguards built in. Each has a few main knobs and a set of presets for working fast, with the full controls one click away for fine-tuning later.
Made for ourselves and our friends, partly as an experiment and partly to get some sounds and textures we were trying to capture. Built with AI assistance.
macOS: AU · VST3 · Standalone, Apple Silicon & Intel Windows: VST3 · Standalone
PC 01 · Variable-mu limiter
comp
A 670-style variable-mu tube limiter, modelled on the original 1950s circuit.
Three main knobs: Squash sets how hard it compresses, Time how quickly it lets go, and Gain the output level. The ratio rises with level and a Knee control sets how gently it starts, while the tubes add more harmonics the harder they work, so it can compress heavily without sounding obvious.
How it works
- Circuit model after Peter Raffensperger's wave digital filter study of the classic variable-mu limiter.Raffensperger 2012
- Push-pull pair of 6386 remote-cutoff triodes, using the paper's equation fitted to the tube's datasheet curves, pre-solved into a table so it costs two lookups per sample.Raffensperger 2012
- Feedback sidechain with the original's crossover deadband (the Knee control), diode bridge and six-position RC time-constant network. Positions 5 and 6 release according to the material, and Time sweeps smoothly between positions.Raffensperger 2012
- Runs oversampled to at least 176.4 kHz, so the feedback loop stays tight.
- Linked, Dual or Mid/Side stereo, a drive control for more tube colour at the same level, and a sidechain low cut that affects only the detector.
Safeguards
- Silent tracks cost almost nothing: after 50 ms of digital silence the tube stage sleeps, while the time-constant network keeps running, so gain recovers through a gap exactly as it would have.
- Plus the suite-wide safeguards.
PC 02 · Tape machine
tape
A studio tape deck or a cassette four-track.
Choose a studio deck at 30, 15, 7.5 or 3.75 ips, or a cassette four-track with narrow heads, a narrower band, more hiss and an unsteady transport. Crosstalk and dropouts are there when wanted. Age, Wobble and Heat adjust the whole machine at once. Bounce records the result to tape again, once or twice, adding more wobble, hiss and loss each time.
How it works
- Speed-domain transport: a commanded speed passes through motor inertia, then picks up periodic wow with Ornstein–Uhlenbeck drift, three-harmonic capstan flutter and a slight mains-frequency speed ripple. All of it is integrated into tape position, so pitch is always the derivative of timing, as on a real machine.Uhlenbeck & Ornstein 1930
- Record stage with magnetic saturation and pre-emphasis, so highs saturate and compress first, then de-emphasis restores the balance. Slower speeds use more pre-emphasis, so loud highs soften sooner. 4× oversampled.
- Head losses from a published physical model: spacing loss and gap loss for the chosen speed as a linear-phase filter, plus head bump and a low-end limit. Calibrated against classic studio-deck specifications, then exaggerated a little so the speeds are easy to tell apart.Chowdhury 2019
- Noise is on the tape, not added after it: pink hiss with replay-EQ lift and a slowly wandering level, plus modulation noise that rides on the music and goes silent when it does. Speed, wear and dropouts shape the noise as they shape the music.
- Dropouts are shared by both channels, with smooth raised-cosine edges and high-frequency loss, so the stereo image stays put.
Safeguards
- Reports its exact latency (transport delay plus head-loss filter, per Bounce pass), so it stays aligned with the rest of your session.
- Plus the suite-wide safeguards.
PC 03 · Double tracker
twin
Double tracking with two tape machines.
Two tape machines each play a copy of your track, slightly early or late and slightly out of tune, panned either side of the original. A simulated operator nudges each machine's speed by hand, so the copies drift and catch up irregularly instead of sweeping evenly like a chorus.
How it works
- Operator model: random holds, nudges and pushes with cosine easing feed a position controller, which drives the transport through motor inertia. The two machines have different motors and random seeds, so they never move together.
- Each machine is a valve record stage (biased tanh), a Lagrange-interpolated tape delay and a valve replay stage, 4× oversampled, followed by the same physical head-loss model as Tape.Laakso et al. 1996Chowdhury 2019
- The double is centred up to 40 ms late, or up to 18 ms ahead of the source using a 20 ms lookahead, and the operator moves it up to ±20 ms around that.
- The copies are loudness-matched to the source over about a second. Flip phase-inverts them for close-delay phasing; Double Only drops the source so you can print the doubles or use Twin on a return.
Safeguards
- Low Latency switch removes the lookahead entirely for tracking through it, keeping the double behind the source at all times.
- Plus the suite-wide safeguards.
PC 04 · Small-speaker drive
wreck
Distortion and small speakers, for lo-fi sounds.
Lo-fi distortion for vocals or anything else: filters, a drive stage, a small speaker model and a resonant mid boost. The Wreck knob turns all of it up together, hotter, thinner, darker and more nasal, from clean at 0 %, through the preset at 50 %, to completely broken at 100 %.
How it works
- Three drive characters: tube (biased tanh), transistor (a hard p-norm knee with the crossover grit of a class-B output stage) and starved fuzz (asymmetric two-transistor clipping, with gating and sputter when the bias is starved).
- 4× oversampled with linear-phase half-band filters, and the dry path delayed to match, so Mix stays phase-aligned.
- Four speaker models (practice amp, phone, PA horn, transistor radio), each designed from band limits, cone and cabinet resonances and breakup ripple, turned into a minimum-phase impulse response with the real-cepstrum method, plus early cabinet reflections.Smith, SASP
- Filters and the resonant mid are zero-delay-feedback (trapezoidal) state-variable filters, smooth even under fast automation.ZavalishinSimper 2013
Safeguards
- Speaker models are level-normalised in the 300 Hz–3 kHz band, and switching speakers crossfades over about 30 ms on zero-latency convolution, so changing it mid-song doesn't click or jump in level.
- Plus the suite-wide safeguards.
PC 05 · Room leakage
bleed
Simulated mic spill between tracks with distance, angle and room reflections.
Bands used to record in one room, often a garage or a basement, all playing at once. Every mic picked up a little of everything else, and that's a big part of why they sounded like a band. Bleed adds it back. On each track it acts as that track's mic, also hearing up to eight other tracks, each at its own distance and angle, with the room's reflections. Parts recorded separately come back sounding like one take. A Space switch also offers the leakage of a console (bright crosstalk) or of neighbouring tracks on tape (dull crosstalk).
How it works
- Per source: distance sets the delay and inverse-distance level, angle sets the off-axis high-frequency loss, and a simple room model adds early reflections. Polarity, mute and solo per row.
- Space: Room uses distance, angle and reflections. Console leaks highs through a 1.5 kHz high-pass, like a console's stray capacitance; Tape leaks lows through a 1.2 kHz low-pass, like crosstalk between adjacent tape tracks. Switching fades the bleed out and back in.
- Tracks find each other over a lock-free shared-memory bus inside your DAW's process, so AU and VST3 instances hear each other too.
- Reads run exactly one block behind, so tracks your DAW processes in parallel never race, and in Room that lag is subtracted from the distance delay.
Safeguards
- Only dry audio is broadcast, so a mic never hears its own output: no feedback loops.
- A deleted track is never mixed in by mistake, even when another track takes over its place on the bus.
- Exact Timing reports the bus lag as latency, so even very close mics line up to the sample after delay compensation. It is always on in Console and Tape.
- If your DAW sandboxes plugins or blocks shared memory, Bleed passes audio through unchanged and says why.
- Plus the suite-wide safeguards.
PC 06 · Echo chamber
chamber
A 1960s studio echo chamber: a loudspeaker and two mics in a tiled room.
Reverb from a simulated echo chamber: the track plays through a loudspeaker in a hard, tiled room and two mics pick it up. Chamber sets the mix, Decay the reverb time and Pre-Delay the gap before it starts. Low and high cut filter what goes into the room. Era chooses the amp and speaker: a 1960s valve amp with a paper-cone speaker, a 1970s solid-state amp with a studio monitor, or Now, with no amp or speaker at all. 100 % Wet sets it up for a return track.
How it works
- Early reflections from image sources: the speaker mirrored in the walls of a rectangular room, each echo delayed and attenuated by its distance to each mic, then low-passed for the walls' absorption.Allen & Berkley 1979
- The tail is a 16-line feedback delay network with an allpass inside each line, mixed through a scattering matrix, so every echo splits into many on each pass and the tail is as dense as noise within about 30 ms.Schlecht & Välimäki 2020
- Instead of wobbling the delay times, the mixing matrix rotates slowly. Resonances never settle into a metallic ring, and nothing goes out of tune: a real chamber doesn't move.Schlecht & Habets 2015
- Each line has a ten-band octave EQ, fitted by least squares to the decay settings, so the measured decay time follows the setting within about 5 % in every octave from 125 Hz to 8 kHz.Prawda et al. 2019Bristow-Johnson
- Era runs all three chains at once, level-matched within 1 dB, so switching crossfades instead of clicking. The 1960s valve stage adds even harmonics that grow as the send gets louder.
Safeguards
- Tells your DAW how long its tail is, from the current decay and pre-delay, so the reverb plays out after the audio stops.
- Presets and Random never change 100 % Wet, so a return track stays a return track.
- Plus the suite-wide safeguards.
Built to fail quietly
- Fault guard on every output. If the DSP ever produces an invalid number, the block is muted and the plugin resets itself, so it goes quiet instead of screaming and then recovers. Anything past +18 dBFS is clamped. Normal audio passes through bit for bit.
- Gain, mix and level controls ramp over 20–50 ms, so preset changes and automation never jolt. Bypass crossfades and stays time-aligned.
- Real-time safe: no memory allocation, locks or file access on the audio thread.
- Stress-tested before release: every parameter at minimum, at maximum, and jumping to random values every 64 samples, with input 12 dB hot and a full-scale noise burst, and checked for memory allocation at several sample rates and buffer sizes. Validated with pluginval at strictness 10 (AU and VST3) and auval.
Sources
The published research these plugins are built on. Each one is credited where it is used above.
- Peter Raffensperger, “Toward a Wave Digital Filter Model of the Fairchild 670 Limiter”, Proc. 15th Int. Conference on Digital Audio Effects (DAFx-12), York, UK, 2012. dafx.de
- Jatin Chowdhury, “Real-Time Physical Modelling for Analog Tape Machines”, Proc. 22nd Int. Conference on Digital Audio Effects (DAFx-19), Birmingham, UK, 2019. Its head-loss equations follow Jay Kadis, The Science of Sound Recording (Focal Press, 2012). dafx.de
- G. E. Uhlenbeck and L. S. Ornstein, “On the Theory of the Brownian Motion”, Physical Review 36, 823, 1930. doi.org
- T. I. Laakso, V. Välimäki, M. Karjalainen and U. K. Laine, “Splitting the Unit Delay: Tools for Fractional Delay Filter Design”, IEEE Signal Processing Magazine 13(1), 1996. doi.org
- Vadim Zavalishin, The Art of VA Filter Design (rev. 2.1, 2018). dsprelated.com
- Andrew Simper, “Linear Trapezoidal Integrated State Variable Filter With Low Noise Optimisation” (technical paper, 2013, rev. 2016). cytomic.com
- Julius O. Smith III, “Minimum-Phase and Causal Cepstra”, in Spectral Audio Signal Processing, CCRMA, Stanford University. ccrma.stanford.edu
- J. B. Allen and D. A. Berkley, “Image Method for Efficiently Simulating Small-Room Acoustics”, Journal of the Acoustical Society of America 65(4), 943–950, 1979. doi.org
- S. J. Schlecht and E. A. P. Habets, “Time-Varying Feedback Matrices in Feedback Delay Networks and Their Application in Artificial Reverberation”, Journal of the Acoustical Society of America 138(3), 2015. cris.fau.de
- S. J. Schlecht and V. Välimäki, “Scattering in Feedback Delay Networks”, IEEE/ACM Transactions on Audio, Speech, and Language Processing 28, 2020. arxiv.org
- K. Prawda, S. J. Schlecht and V. Välimäki, “Improved Reverberation Time Control for Feedback Delay Networks”, Proc. 22nd Int. Conference on Digital Audio Effects (DAFx-19), Birmingham, UK, 2019. dafx.de
- Robert Bristow-Johnson, “Cookbook Formulae for Audio Equalizer Biquad Filter Coefficients”, published as a W3C Working Group Note. w3.org
Made with AI assistance
These plugins, and this site, were built with the help of large language models, including Anthropic's Claude. They helped find and read the published research, write and review the DSP and plugin code, and draft the words on this page.