Wurl

Physically-modeled Wurlitzer 200A electric piano for Ableton Move, built for the Schwung open-source plugin framework.

Port of OpenWurli by hal0zer0 (GPL-3.0) to single-file C, with additional features for the Move platform.

Signal Chain

Every stage of the Wurlitzer 200A is modeled from first principles — no samples, no impulse responses.

Reed Oscillator (7 modal modes, Euler-Bernoulli beam theory)
  → Electrostatic Pickup (capacitive 1/(1-y) nonlinearity — the source of "bark")
  → Attack Noise (bandpassed burst on hammer strike)
  ──── voice sum ────
  → [2x Oversampling] ─────────────────────────────────────────────────
  │  → Preamp (HPF → gain × LDR tremolo modulation → asymmetric tanh)  │
  │  → Volume (squared audio taper)                                     │
  │  → Power Amp (Newton-Raphson Class AB with crossover distortion)    │
  │  → Speaker Cabinet (polynomial + HPF/LPF + thermal compression)     │
  │  → Soft limiter                                                     │
  → [Downsample] ──────────────────────────────────────────────────────
  → Darken (smooth one-pole LPF)
  → Spring Reverb (dispersive allpass chain)
  → Output (int16 stereo)

Controls

Knobs (8)

Knob	Function
Volume	Output level (squared audio taper)
Tremolo	Depth of 5.63 Hz tremolo (timbral, not just volume)
Attack	Reed onset speed: 0% = slow ring-up, 100% = fast/punchy
Decay	Note sustain: 0% = short/percussive, 100% = long sustain
Brightness	Upper harmonic content: 0% = very dark, 50% = neutral, 100% = bright
Darken	Smooth low-pass filter: 0% = bypass, 100% = very dark (800 Hz)
Bark	Pickup nonlinearity: clean bell → aggressive growl
Reverb	Spring reverb macro (controls mix + decay + damping together)

Menu Parameters (jog wheel)

Parameter	Function
Speaker	Cabinet character: 0% = flat bypass, 100% = authentic 200A
Tune	Detune ±100 cents
Preset	Cycle through 10 built-in presets (applied in real time)

Presets

#	Name	Character
0	Classic 200A	Neutral starting point, no reverb
1	Dreamy Keys	Slow trem, warm, darkened, lush reverb
2	Barky Soul	High bark, punchy, bright, authentic speaker
3	Surf Spring	Deep trem + heavy spring reverb
4	Dark Ballad	Dark, mellow, long sustain, gentle reverb
5	Percussive Clav	Instant attack, very short, bright, dry
6	Warm Pad	Slow attack, very long sustain, big reverb
7	Lo-Fi Tape	Very dark, full speaker, slight detune
8	Bright Bell	Clean (no speaker), bright harmonics, reverb
9	Gospel Growl	Maximum bark + speaker, aggressive

What's Ported from OpenWurli

The core physical model is faithfully ported from OpenWurli:

• Modal reed oscillator — 7-mode quadrature oscillator with Ornstein-Uhlenbeck frequency jitter, cosine onset ramp, and progressive damper
• Per-note physics — tip mass ratio, eigenvalue interpolation, reed compliance, spatial coupling (Simpson integration of mode shapes), velocity S-curve
• Electrostatic pickup — time-varying RC circuit with bilinear discretization, C(y) = C0/(1-y) nonlinearity
• Attack noise — bandpassed burst on hammer strike
• MLP v2 corrections — 2→8→8→11 neural network (195 weights) runs once at note-on, correcting H2-H6 frequencies, decay rates, and displacement scale. Trained on real Wurlitzer recordings.
• Power amplifier — behavioral Newton-Raphson Class AB model with crossover distortion
• Speaker cabinet — Hammerstein polynomial waveshaper (BL asymmetry + Kms hardening) with tanh excursion limiting, thermal voice coil compression, HPF/LPF
• CdS LDR tremolo model — asymmetric attack/release envelope, power-law resistance curve
• Per-note variation — deterministic FNV hash detuning, mode amplitude offsets, register trim
• Voice stealing — 5ms linear crossfade on steal (releasing voices first, then oldest)
• NaN guards — on voice sum and final output, with filter state reset

What's Different from OpenWurli

Feature	OpenWurli (Rust)	Wurl (C port)
Preamp	DK method 12-node MNA circuit solver (2750 lines generated)	Simplified 2-stage CE: HPF → gain → asymmetric tanh → LPF
Tremolo oscillator	Melange-generated Twin-T circuit (2100 lines)	Behavioral sine LFO (same CdS LDR model)
Oversampling	2x on preamp only	2x on entire output chain (preamp + power amp + speaker)
Polyphony	64 voices	16 voices (Move CPU budget)
Parameters	4 (Volume, Tremolo, Speaker, MLP on/off)	11 (8 knobs + 3 menu), plus 10 presets
MIDI timing	Sample-accurate event splitting	Block-rate (128 samples)
Spring reverb	None	Dispersive allpass model (Parker 2011)
Darken filter	None	One-pole LPF (800 Hz → 20 kHz)
Preset system	None	10 built-in presets with real-time switching

What's Not Ported

• DK method preamp — the melange-generated 12-node Modified Nodal Analysis circuit solver is 2750 lines of Rust with matrix constants. Too complex to port; replaced with a behavioral approximation.
• Twin-T tremolo oscillator — the melange-generated 7-node circuit model is 2100 lines. Replaced with sine LFO (the original's legacy-tremolo feature flag).
• Melange power amp circuit — the 20-node 7-BJT generated circuit option. The behavioral NR model IS ported.

Install

1. Download wurl-module.tar.gz from the latest release
2. Open the Schwung Installer desktop app
3. Click Install custom module and select the downloaded .tar.gz
4. Restart your Move — Wurl appears under Sound Generators

Building from source

./scripts/build.sh      # Docker ARM64 cross-compile
./scripts/install.sh    # SCP to move.local

Requires Docker Desktop or an aarch64-linux-gnu-gcc cross-compiler.

Credits

• hal0zer0 — OpenWurli: original Rust implementation of the Wurlitzer 200A physical model (GPL-3.0)
• OldBassMan (Freesound) — Wurlitzer 200A recordings used for OpenWurli's MLP training data
• Robbert van der Helm — nih-plug framework (original plugin host)
• Julian Parker — "Efficient Dispersion Generation Structures for Spring Reverb Emulation" (2011) — spring reverb algorithm reference
• Charles Vestal — Schwung plugin framework for Ableton Move

License

GPL-3.0 — inherits from OpenWurli.

See LICENSE for full text.