Keep the graph
The amplifier remains a network of stages, rails, loads, coupling paths, and observable operating points.
GreyboundOpen-source greybox guitar tone experiment.
Greybound
A public research base for guitar amp and pedal modeling, built around greybox thinking: analog circuits as a map, measurements as a check, and local learned laws that still report back into the circuit.
Alpha build distributed outside the App Store. macOS may ask for a manual open confirmation.
Open sourceexperiment-first project
Greyboxcircuit ideas plus measured behavior
Rustaudio model playground
Notestransparent research trail
Web alpha
The desktop UI, running in the browser.
The web target reuses the same iced interface, canvases, controls, and layout logic as the desktop app. No parallel HTML mockup, no duplicated Greybound application.
Open playground
Launch the browser playgroundDesktop alpha
A real app for testing the greybox chain.
Load the macOS alpha, route live guitar or a WAV file through the current rig, and record repeatable demo passes while changing amp, pedal, cab, doubler, and EQ controls.
Why it exists
Greybound is a shared workbench for greybox tone.
The project starts from a simple idea: guitar tone models should be understandable enough to change, measured enough to trust, and open enough for other builders to test their own instincts against the code.
Method
Not a captured amp. Not a frozen schematic.
Greybound treats tone as a living circuit state. The graph stays readable, the nonlinearity is learned locally, and every local choice must return useful electrical state to the rest of the model.
Learn locally
Tiny cells are allowed only where nonlinear behavior is too subtle for a clean equation.
Return state
The learned component reports voltage, current, bias, and headroom back into the shared circuit state.
Modeling map
White box, black box, grey box.
Greybound lives in the middle: more inspectable than pure machine learning, more organic than a purely linearized circuit solve, and more practical than a component-perfect simulator.
Equations first
White box
A model built from explicit equations. In practice it often captures the linear part of components well, while the messy nonlinear feel still needs careful choices.Pros
- Readable assumptions
- Stable linear behavior
- Good for design questions
Cons
- Can miss organic feel
- Can sound too ideal
- Nonlinear parts are hard
Examples first
Black box
A neural network trained from input and output examples. It can imitate what it has heard, but its internal reasoning is not a circuit you can inspect.Pros
- Can capture complex tone
- Fast once trained
- Works without schematics
Cons
- Replays existing sounds
- Hard to inspect
- Needs strong datasets
Circuit state plus micro-nets
Grey box
A circuit-level model where nonlinear components can use tiny neural networks, while voltage drops are reported back into the global circuit as shared state.Pros
- Keeps circuit context
- Learns nonlinear feel locally
- Auditable signal flow
Cons
- Requires careful coupling
- Needs component validation
- More design judgment
The subtle part
The model is not a capture. It is a circuit with learned local laws.
Many amp models choose one extreme: solve simplified equations and lose the messy feel, or train one large network that can only echo the sounds in its dataset. Greybound keeps the circuit as the organizing structure, then learns the small nonlinear behaviors that deserve it.
The circuit stays global.
Stages do not secretly mutate each other. They exchange explicit boundary data: voltage, impedance, DC offset, headroom, coupling style, level history, and latency.
Learning stays local.
Micro neural cells are used for nonlinear components, not for memorizing an entire amplifier. A learned cell has a small job and a visible electrical contract.
Voltage drops become shared state.
Current draw updates rails, bias, and recovery behavior. Those voltage changes are reported back into the circuit so the next stage feels the same stress.
Promotion needs evidence.
A cell can shadow the analytic path before replacing it. SPICE fixtures, held-out stimuli, render metrics, and listening notes decide whether it earns its place.
Signal chain
From string to speaker, every stage has a job.
- Guitar input
- Circuit clue
- Greybox model
- Reference render
- Listening notes
Greybox core
Not a black box, not a full circuit clone.
Greybound keeps the circuit graph visible, then lets local nonlinear cells improve the hard parts without swallowing the whole sound into one opaque model.Research lab
Experiments should leave a trail.
Model notes, diagrams, reference renders, and evaluation scripts live close to the code so every tone decision can be questioned, repeated, or improved.Sound design
Built for guitar tone as a living system.
The project starts with amp and pedal behavior, rig files, and listening tests, then grows only when the model earns its place in the chain.Research spine
A tone project with receipts.
Greybound keeps topology notes, model diagrams, calibration runs, render scripts, and comparison traces close to the code. The result is an experiment that can be tuned by ear and challenged by measurement.
What is inside
Built as an experiment others can open up.
- Top Boost-style greybox amp experiment
- Pedal model sketches for gain, time, filter, and dynamics blocks
- Circuit diagrams and model notes
- Offline render and reference comparison tooling
- Rust-first codebase for repeatable audio experiments