01 Analog modeled. Test enforced.

The first drum machine that can prove what it does at 96 kHz.

VoltageDrums is an analog-modeled drum machine built from the schematics up: every voice traceable to the circuit, every model tested against ground truth, every render reproducible to the bit. Written in Rust. Shipped as CLAP.

Interactive, with engine rendersExplore the 808 Methods, measurements, testsRead the proof

Circuit-traceable voice models
Verified at six sample rates, 44.1 to 192 kHz
Bit-identical renders, by construction
Calibrated against real hardware
Zero-allocation audio path, enforced by tests

VoltageDrums 808-style interface with voice controls and a performance monitor — **120.0 BPM · fixed pattern · deterministic animation** The lower panel visualizes incoming triggers. It is not a promise of an included v1 sequencer.

16synthesized voices

6verified sample rates

0audio-path allocations

∞dB self-null

1 msarea-correct trigger

02 The approach

Built from the schematic up.

Most drum machine plugins are tuned by ear until they sound close enough. VoltageDrums starts from the other end: the service manual, the schematic, and the published circuit analyses. Component values go in; behavior comes out.

The kick drum’s click is not a sample or a filter preset. It is the actual pulse-shaper network from the original circuit, four components and a diode, solved every sample with a closed-form Wright-omega equation. The tom’s pitch bend is the diode pair that conducts when the resonator swings hard. The snare’s rattle is a rectified noise burst through the same envelope topology the hardware used.

When the schematic says a resistor loads the network and pulls the kick from 45.7 Hz to 49.5 Hz, the model moves with it, because the loading is part of the model, not an error to be tuned away.

And when the documentation runs out, we measure. Where no source on the internet could tell us a voice’s true behavior, recordings of real units did: our models are fitted to hardware, and the fit is a test that runs on every build.

Take the kick circuit apart

Bass drum pulse shaper, bridged-T resonator and measured pitch trajectory — **808 kick, component by component.**Vector schematic plus deterministic engine measurement.

03 The circuit analysis

We read the service manuals so the models cannot lie.

Six details that change what you hear — and that a generic “analog” switch cannot reproduce.

CV 01

Accent is a voltage, not a volume knob.

Accent raises the trigger voltage feeding every voice. Downstream nonlinearities see a bigger pulse: attack, spectrum and pitch jump change, not just output level.

BUS 02

The hats and cymbal are relatives.

One bank of six square-wave oscillators feeds cowbell, cymbal and both hats. Closed hat terminates the open-hat envelope like the hardware’s ENVELOPE SHUT OFF block.

BD 03

A kick with three pitch mechanisms.

Attack pitch jump, retrigger injection and slow pitch sigh are three separate circuits. Each is modeled and tested separately.

909 04

TUNE does not change settled pitch.

Across two measured units, the fundamental stays within a couple of hertz while pitch-sweep time triples. The model follows the circuit, not the label.

SD 05

Folklore dies in the lab.

The famous “180 and 330 Hz” shell pair is not a hardware measurement. The schematic’s 1.47 capacitor ratio and recordings from two units agree.

NOISE 06

The hiss is a shared source.

The original avalanche transistor was selected batch by batch. Its calibrated level and shared distribution matter because every noise voice drinks from the same source.

808 signal path showing trigger and accent voltage, resonator voices, shared noise and metallic oscillator buses, hat choke and mix — **One trigger bus. Shared sources. A real choke.**Every line in this map corresponds to an implemented signal relationship.

04 The proof

Claims you can check.

Every method and toleranceOpen the technical dossier

Kick pitch trajectories overlaid at six sample rates

RATE MATRIX

Six sample rates, one specification.

Pitch, level and decay are conformance-tested at 44.1, 48, 88.2, 96, 176.4 and 192 kHz. Drift budgets are measured in cents.

Two deterministic renders and their difference at true digital silence

DETERMINISM

Bit-identical renders.

Same input, same seed, same bits — at any block size and across arbitrary event boundaries. Analog character never means hidden randomness.

Model kick decay anchors compared with measured hardware references

GROUND TRUTH

Calibrated against real hardware.

Knob sweep by knob sweep: fundamental, decay law, shell pairs, metallic frequencies, hat ranges and the attack pitch signature.

Level-normalized spectra demonstrating accent changes timbre rather than only gain

SOURCE AUDIT

The tests police the engineers.

The production audio path is audited for allocation, locks, panics and transcendental calls. A mutation test proves the guard still bites.

05 The engineering

Hardcore Rust. Boring on purpose.

The audio thread is where plugins go to misbehave: a stray allocation, a mutex, a denormal storm, and your session glitches. VoltageDrums’ audio path is allocation-free, lock-free and thread-free, and that is not a code-review promise.

audio_path.contractPASS

allocations          0
locks                0
threads              0
panics               0
transcendentals      0

design_math          f64
coefficients         checked f32
filter_state         f64
denormals            flushed
seed                 pinned
block_partition      bit exact

A tracking allocator counts every allocation across prepare and process in the test suite; the count must be zero. The inner loop calls no transcendental functions. Filter coefficients are designed in double precision at prepare time; knob motion runs through precomputed curve tables; nonlinear stages use closed-form solutions and bounded polynomial approximations with stated error limits.

The synthesis core is a dependency-free Rust crate. No frameworks in the signal path, no runtime surprises, every struct fixed-size and copyable. The plugin ships CLAP-first. Because voices are data, the same engine that produces the 16-voice kit scales to the machines that come next without a rewrite.

This is not a sampler. No sample import, no sound library, no content packs. Where a modeled circuit genuinely contains a ROM, we model the readout circuit, and the shipped content must meet the same provenance bar as everything else. The instrument synthesizes.