Sound Modeler

What is Sound Modeler?

Don't ask AI to write your music. Ask it to help invent the instrument you've never heard before. Sound Modeler turns a description — 'glass resonating underwater with a slowly breathing analog texture' — into a chain you can inspect and change at every link: intent, acoustic properties, synthesis strategy, DSP graph, parameters, patch, playable instrument. Every sound is stored as Sound DNA (eleven perceptual dimensions and a material) and a Sound Genome (the complete recipe), so 'make it darker' becomes an exact parameter diff with the dimension that moved it, two sounds can be morphed or bred with a named section kept verbatim, and the Sonic Microscope can trace a resonance back to the node that caused it. A model is consulted once, at design time; everything after that is arithmetic, which is what keeps model latency and cost out of the real-time audio path.

Problems it solves

• Turn a description into a playable, editable instrument
• Refine a sound in words and see the exact parameters that moved
• Morph and breed sounds, keeping the parts you name
• Trace a harshness or resonance to the node causing it
• Map velocity, aftertouch and MPE the way a real instrument responds
• Model a recording into a playable body, or invent a bell made of liquid crystal
• Take the instrument to Ableton Live, VST3/AU or a hardware synth

Key capabilities

Sound Modelling

Turn a described sound into an editable synthesis system — DNA, genome and DSP graph — rather than a rendered file.

Sound Morphing & Breeding

Morph continuously between two sounds, or breed them and keep a named section of one parent verbatim.

Graph Tracing

Ask what is causing a resonance, a harshness or a transient and get the nodes and parameters responsible.

Evidence Trail

Every figure carries the record it came from, so a reader can check the answer instead of trusting it.

How it is built

Sound Canvas

The sound at the centre; SOURCE, SHAPE, MOTION, SPACE, CHARACTER and OUTPUT around it, expanding only when needed.

Sound DNA and Genome

Eleven perceptual dimensions and a material, compiled deterministically into the complete recipe.

Hybrid engine strategy

Virtual analog, wavetable, FM, additive, granular, noise and modal resonance chosen per sound, with a reason for each.

Sonic Microscope

Predicted spectrum, transients, stereo field and modulation, traceable back through the graph.

Playable in the browser

A Web Audio engine plays the recipe from the keys, the computer keyboard or a MIDI controller — with pitch bend, mod wheel and pressure.

Physical bodies and impossible materials

Strings, membranes, bars, bells, plates, glass and tubes with physics-derived mode tables — and materials that do not exist, blended by density, stiffness and damping.

Sound-to-Synth

Record or upload a sound you are authorised to use; its transients and partials are measured into a model that plays. The audio is not kept.

Exports that build

Max for Live patchers, a JUCE project that compiles to VST3/AU/Standalone (the AU passes auval), and Standard MIDI Files for hardware.

How it works

How the work is structured, and what each step's output rests on.

1. 1

   Describe the sound

   AI-inferred

   Words become Sound DNA with the evidence for every dimension that moved.

2. 2

   Choose the engines

   Recommendation

   The strategy names each engine and why it is in the mix.

3. 3

   Compile the genome

   Observed

   DNA and strategy become the complete recipe, deterministically.

4. 4

   Play, refine, trace

   Observed

   Every instruction returns the exact parameters that changed; the microscope traces what you hear.

Example use cases

Each example has its own page you can read and share.

Ask Sound Modeler

Illustrative examples. Results depend on your own data.

• “Create a warm evolving synth that sounds like glass resonating underwater, with a slowly breathing analog texture and expressive MPE control.”
• “Make the high frequencies sound crystalline.”
• “Add another oscillator an octave down, but only bring it in when velocity exceeds 90.”
• “Remove this harshness without reducing brightness.”
• “70% of the pad, 30% of the bell, but keep the bell's movement.”

Questions

Does it generate audio files?

No. It produces an editable synthesis system — the recipe — and a browser engine plays it. That is the point: a recipe can be read, changed, morphed and bred; a rendered file cannot.

Which engines are real today?

Virtual analog, wavetable, FM, additive, granular, noise, modal resonance, stereo space, physical modeling (string, membrane, bar, bell, plate, glass, tube) and sample modeling (a body measured from a recording) are shipped. Spectral processing is approximated by additive partials. Neural modeling — embeddings and a learned timbre model — works when the Model Fabric serves audio embeddings and says so when it does not.

Can I use it as a plugin or in Ableton?

Export a JUCE project and build it: VST3, AU and Standalone from one recipe (the generated AU passes Apple's auval). For Ableton Live, export Max for Live patchers and Save As .amxd inside Max. Hardware gets a Standard MIDI File of the MIDI-defined Sound Controllers; vendor SysEx dumps are on the roadmap and are not fabricated.

Can I share an instrument with someone?

Yes — a link to listen, a link to remix (they fork a copy with the origin recorded), or named collaborators who edit the same sound with every revision attributed.