Physical Models

Most of the built-in synthesizer's engines start from a waveform and shape it. The acoustic-style engines start from an instrument instead — a string under a hammer, a bore under a reed, a metal tongue swinging through a slot, or a glottal source through formants. They compute each note from a mechanism rather than from a recording. Each note is a small simulation running at the sample rate, and what you hear is whatever that simulation radiates.

This page is about that family: what the model is actually computing, which acoustic quantity each voice is built around, and how far to trust each voice today.

Calibration status

Only the acoustic piano is calibrated against a reference recording. Every other physical model, including the harpsichord, modal and membrane models, is provisional: it is meant for data-free preview and as the fallback floor, and its sound can change across patch releases.

Which method for which job

Whether a physical model is the right choice at all — against a SoundFont, against sampling, against subtractive synthesis — belongs to Sound Sources. This page assumes you have already landed on one. For the shared patch layer these voices sit inside, see Built-in Synthesizer; for the terms, Synthesis Basics.

What a waveguide model is

A vibrating string or an air column carries waves that travel, reflect at each end and come back. A waveguide model is the shortest honest way to compute that: a delay line stands in for the travel, a filter stands in for the losses at the reflecting end, and the two are wired into a loop. Feed energy in, and the loop rings at a pitch set by nothing but its own length.

That is the whole trick, and it has two consequences worth knowing before you use one.

It needs no recorded data. Sampled grand pianos can be large because they store many recorded notes, velocity layers, and articulations. A model keeps a compact parameter set and computes notes and performance values within its modeled range. That is why these voices can be the floor under a MIDI file that arrives with no SoundFont at all.

Each voice costs compute. A sampler reads memory and advances a playback voice; a model solves its physics once per sample per voice, and each voice carries its own delay lines, loss filters and body resonators. Polyphony, not disk, is the budget you spend here. A dense arrangement of bowed strings is a heavier render than the same arrangement played from samples.

And it is a model, not a recording. It reproduces the mechanism, so it gets the behavior a sample cannot: a hard blow is brighter because the felt compresses further, not because a second sample was recorded. What it does not get for free is the specific voice of a specific instrument, which requires adjustment and calibration.

The loop and model-specific stages

A shared abstraction with model-specific stages

• the loop, shared by every voice
• exciter slot, swapped per voice
• output tap, outside the loop

Excitation, travel, loss, radiation. A delay line can represent propagation and a loss filter can represent decay, but each model may add its own topology, radiation, and control behavior. The bow, reed, lip, and jet therefore share an abstraction rather than one identical loop.

For a waveguide-style model, read it left to right: the exciter injects energy into the delay line, the wave travels, and at the far end it turns around and comes back through the loss filter to meet the exciter again. The length of that round trip sets the pitch in this abstraction, so these voices do not need a separate tuning oscillator. For a string or an open pipe one round trip is one period, so the loop holds the sample rate divided by the fundamental. A cylinder closed at one end, the clarinet's bore, is the exception: its open end sends the wave back inverted, so the wave needs two round trips to return to its starting shape. The reed engine's cylindrical bore therefore runs a loop of half the period with an inverting reflection, which is why it sounds an octave below an open pipe of the same length and favors odd harmonics. A loop too short to hold its pitch at the host rate (a high note at a low rate) runs internally at a rate of at least 48 kHz and is decimated back, so a top note sounds the same pitch at 8, 16, 24, 44.1 and 48 kHz. The radiation stage is a tap taken outside the loop, where the body or the bell turns the internal wave into something a microphone would hear.

Where a model exposes them, brightness and damping describe the spectral openness and energy loss of its propagation or radiation stages. Their exact response is model-specific. On a bowed or blown voice the exciter can replenish losses, so the note can sustain while those controls still change its color and decay.

The exciter is one important difference between instruments, but it is not the only one. A bow is stick-and-slip friction against the string. A reed is a valve that mouth pressure drives against the bore. A lip adds its own resonance, and a jet has its own transit behavior. The delay topology, losses, radiation, and body coupling also vary by engine.

The struck and plucked voices — piano, plucked-string, karplus-strong, harpsichord — use waveguide or delay-line structures with a brief exciter. After the hammer, pick, or quill transfers energy, the remaining state rings down. A control moved mid-note can therefore reach a sustained bowed voice while a struck voice may have no active exciter axis to receive it.

Terms used in the tables below

• Bore — the air column inside a wind instrument's tube; its shape (cylinder or cone) decides which resonances it supports.
• Embouchure — how a player's lips are set against a mouthpiece; on brass it tunes the lip's own resonance.
• Helmholtz motion — the steady motion of a bowed string: a single kink travels around the string, and once per period the string sticks to the bow, then slips back.
• t60 — the time a string's vibration takes to fall by 60 dB, the same measure a room's RT60 applies to its reverberation.
• 8′ and 4′ — organ and harpsichord registration, named after pipe length in feet: an 8′ stop sounds at the written pitch, a 4′ stop an octave higher.

The acoustic models

Engine mode	What it models	The quantity it is built around	Identifier
piano	Felt hammer, coupled unison strings, soundboard	Where the hammer strikes	piano.strike_position
pipe-organ	Flue pipe drawing on a shared wind chest	Wind pressure at the mouth	pipe_organ.breath
bowed-string	Rosined bow gripping and slipping on a string	Contact point and bow force	bowed_string.bow_position
reed	Cane reed valving a cylindrical or conical bore	Blowing pressure and reed stiffness	reed.breath_pressure
brass	Lip valve on a flared bore	Blowing pressure and embouchure	brass.lip_tension
flute	Air jet across an edge, driving an open pipe	Jet transit time against the bore period	flute.jet_ratio
plucked-string	String grazing a curved bridge	Pluck point and bridge buzz	plucked_string.pick_position
vocal	Glottal source through five vowel formants	Which vowel	vocal.vowel
free-reed	Metal tongue swinging through a slot	Bellows pressure and tongue stiffness	free_reed.breath_pressure
harpsichord	Quill-plucked string choirs with a short rear segment	Pluck position and rear coupling	harpsichord.pluck_8a

Most entries above use waveguides in the sense the figure describes. vocal is a source-filter model with no loop, and free-reed is a driven tongue with no coupled air column. harpsichord uses separate delay lines for its string choirs and a short segment behind the bridge. These voices belong together because they are solved from a mechanism rather than played from a recording.

The identifiers below are the engine's own field names. They are how each model is voiced, and they are what the named presets set — and a patch overrides them one by one through engineParams, spelled in lowerCamelCase; see Reaching them from code.

The Struct default column is the value the voice starts from before any preset is applied. synthEngineParamInfo(mode) reports something else as default: the value of the engine's base preset (the violin for bowed-string), which differs from the struct default for most voiced fields. Ranges are the same in both.

piano — hammer, unison strings, soundboard

Field	Struct default	Range	What it is
piano.strike_position	0.085	0 – 0.5	Where the hammer meets the string, as a fraction of its length. It notches the partials that have a node there — the default is about a twelfth of the length; real grands strike between roughly 1/7 and 1/12 depending on register.
piano.hammer_exponent	2.5	1.5 – 4	The felt's compression law, force against compression. It sets how much shorter and brighter a hard blow is than a soft one.
piano.hammer_contact_ms	1.2	0.2 – 10	How long the felt stays in contact at A4, mezzo-forte. A hammer that stays on the string for T seconds cannot push energy much above about 1/T: 1.2 ms caps the blow near 1 kHz, and a harder, shorter contact reaches higher.
piano.strings	3	1 – 3	Coupled unison strings per note.
piano.detune_cents	1.6	0 – 50	Micro-detune across the unison. The strings fall out of step, and the prompt sound gives way to the long aftersound.
piano.dispersion	1.0	0 – 1	Scale on the keyboard-graded stiffness stretch; 0 is a perfectly harmonic string.

Two things this model does that a knob-driven plucked string cannot. Inharmonicity follows a U-shaped curve with its minimum around C2, rather than climbing monotonically toward the treble: the wound bass strings turn it back upward, so the very bottom of the keyboard is more inharmonic than the notes just above it. Longitudinal string modes are modeled — a struck string stretches along its length as well as across it, and the tension change radiates the inharmonic growl that fills a low note's attack. Without them the bass is felt more than heard.

pipe-organ — a flue pipe on a shared wind chest

Field	Struct default	Range	What it is
pipe_organ.breath	0.35	0 – 1	Mouth pressure driving the jet. It colors the tone rather than setting the level — the pipe self-oscillates across the whole band, and loudness rides the amp envelope.
pipe_organ.brightness	0.5	0 – 1	How brightly the open end reflects the upper partials: principal at the top, stopped flute at the bottom.
pipe_organ.stopped	false	—	A covered pipe. Fundamental-dominant, octave pump muted, reflection darkened.
pipe_organ.rank_count	0	0 – 8	How many ranks sound together. 0 is a single implicit 8′ built from the two fields above.
pipe_organ.wind_sag	0.0	0 – 1	How far the shared wind pressure drops as more pipes draw on it, so a full chord sinks slightly and then recovers.
pipe_organ.tremulant_rate_hz	0.0	0 – 12	Undulation of the wind supply, heard as pitch and level together. 0 is off.

The wind supply is shared across the ranks rather than per note, which is what makes wind_sag mean anything: it is a property of the chest, and it only shows up when several pipes pull on it at once.

bowed-string — stick and slip

Field	Struct default	Range	What it is
bowed_string.bow_position	0.13	0.02 – 0.5	Contact point as a fraction of the string length from the bridge. Small is sul ponticello — near the bridge, bright and edgy. 0.13 is the natural playing point. Moving past it toward the fingerboard (sul tasto) does not soften this model's tone; on the violin preset it brightens again.
bowed_string.bow_force	0.5	0 – 1	Downward pressure. It widens the sticking region of the friction curve: low is a light, whistly bow that barely captures the string, high is firm and slightly rough.
bowed_string.bow_speed	0.5	0 – 1	How fast the bow is drawn — the dynamic level.
bowed_string.brightness	0.5	0 – 1	Bridge reflection openness.
bowed_string.damping	0.4	0 – 1	Bridge loop loss. The bow replenishes it either way, so the note sustains.
bowed_string.rosin	0.0	0 – 1	Texture of the rosined hair, as a small deterministic velocity noise on the bow.

Bow position and bow force are the pair a player actually controls, and they are not interchangeable: position decides which partials the string emphasises, force decides how cleanly it locks into the Helmholtz motion that makes a bowed note a bowed note. A violin-family body resonator sits on the radiation tap; it is a measured violin corpus, and bowed_string.corpus_scale can only place its lowest mode, because a cello is not a scaled violin.

The demo below starts from the violin preset's own bow and moves one of the pair at a time.

SYNTH · PRESETIDLE

Bow force and bow position

Scroll to load

A violin A4 rendered by the bowed-string model, starting from the preset’s own bow. Lower the bow force and the note thins out and turns airy, with more noise around the pitch; raise it and the tone goes rough and louder. Move the bow position down toward the bridge and the note brightens sharply. The position slider stops at the preset’s natural contact point: moving past it toward the fingerboard does not darken this model’s tone.

Bow force

0.55

Bow position (from bridge)

0.12

reed — a valve the breath opens

Field	Struct default	Range	What it is
reed.breath_pressure	0.6	0 – 1	Steady mouth pressure — the dynamic level. The reed only speaks above a threshold, so very low values approach silence.
reed.reed_stiffness	0.5	0 – 1	Slope of the reed table: soft and dark through hard, bright and buzzy.
reed.reed_opening	0.5	0 – 1	How far the reed tip sits open at rest. Low pinches shut sooner and the tone turns nasal.
reed.conical	false	—	false is a cylinder closed at the reed — a clarinet, where odd harmonics dominate. true is a cone, approximated as an open pipe with the full series: saxophone, oboe, bassoon.
reed.brightness	0.5	0 – 1	Bell reflection openness.
reed.chiff	0.4	0 – 1	The brief bright noise as the reed starts to speak, so the note articulates rather than swelling in.

conical changes which resonant modes the bore supports, rather than applying a post-EQ curve. That changes the instrument's resonance and harmonic balance at its source.

brass — the lip is the valve

Field	Struct default	Range	What it is
brass.breath_pressure	0.7	0 – 1	Steady mouth pressure driving the lip valve.
brass.lip_tension	0.5	0 – 1	Embouchure: where the lip resonance sits relative to the bore. Tighter puts it a touch above, the outward-striking point that gives brass its edge; looser is a broader, more slotted center.
brass.lip_damping	0.5	0 – 1	Width of the lip resonance. It sets the resonator's Q from 30 at 0 down to 8 at 1, held constant across the range so the resonance stays proportionally as narrow on a low note as on a high one. Low is a sharp, buzzing lip; high is a broad, rounded one.
brass.conical	false	—	Cylindrical body (trumpet, trombone) against conical (horn, tuba, cornet, flugelhorn). Both radiate the full series; the flag colors the bell reflection.
brass.brightness	0.5	0 – 1	Bell reflection openness.
brass.damping	0.4	0 – 1	Bore loop loss.

Unlike a reed, a brass lip has a resonance of its own, and the note is where the lip and the bore agree. That is why lip_tension is an intonation control as much as a timbre one, and why the same fingering can slot several notes.

flute — a jet across an edge

Field	Struct default	Range	What it is
flute.jet_ratio	0.5	0.1 – 0.9	The jet's convection time as a fraction of the bore period. Around 0.5 the jet drives the fundamental; smaller ratios drive the octave and above, so this is the physical seat of the register.
flute.breath_pressure	0.55	0 – 1	Mouth pressure — dynamic level and jet drive.
flute.jet_reflection	0.5	0 – 1	How strongly the bore's returning pressure deflects the jet at the flue.
flute.end_reflection	0.5	0 – 1	How strongly the open end sends the wave back down the pipe.
flute.damping	0.35	0 – 1	Bore loss. Low is a flute; high is an ocarina or a blown bottle, which will not overblow.
flute.breath_noise	0.15	0 – 1	Jet turbulence — a flute's signature texture. High for a shakuhachi, low for a tin whistle.

The defining quantity for the flute is a ratio of two times: the jet's travel time across the mouth divided by the bore's round-trip time. That ratio helps determine which register the pipe supports.

plucked-string — the bridge the string grazes

Field	Struct default	Range	What it is
plucked_string.pick_position	0.2	0 – 0.5	Plucking point as a fraction of the period; the excitation comb notches the harmonics with a node there. 0 is no comb.
plucked_string.buzz	0.0	0 – 1	How hard the curved bridge limits and sprays the returning wave into the upper partials. 0 is a cleanly terminated harp or koto; higher is the shimmering rattle of a sitar's jawari or a shamisen's sawari.
plucked_string.brightness	0.7	0 – 1	Loop-lowpass openness — how slowly the upper harmonics decay relative to the fundamental.
plucked_string.decay_s	4.0	0.05 – 60	String t60 at A4.
plucked_string.decay_stretch	0.5	0 – 1	How much longer the low strings ring: t60 scales by two to the power of stretch times octaves below A4.
plucked_string.release_damp_s	0.12	0.01 – 10	Damped t60 applied at note-off — the finger or palm coming back.

buzz is the whole reason this engine exists beside the plain plucked loop. A curved bridge does not terminate the string at a point; the string keeps grazing it, and each graze throws energy back up the series, so the note shimmers for its entire ring instead of decaying smoothly.

vocal — a source and five formants

Field	Struct default	Range	What it is
vocal.vowel	0	0 – 4	/a/, /e/, /i/, /o/, /u/. Selects the formant table.
vocal.brightness	0.5	0 – 1	Tilts the glottal source and opens the upper formants.
vocal.breath_noise	0.1	0 – 1	Aspiration mixed into the source — pressed at 0, breathy above it.
vocal.vibrato_rate_hz	5.5	0.1 – 12	A singer's own vibrato, read at note-on.
vocal.vibrato_depth	0.3	0 – 1	Vibrato amount. 0 skips the LFO entirely.

This one is not a waveguide. It is a source-filter model: a glottal source — a sawtooth shaped by a one-pole tilt, plus aspiration — feeding a bank of five resonant bandpasses tuned to a sung vowel. There is no loop, so nothing here reflects, and the signal path is strictly feed-forward. One consequence is worth knowing: the source is a PolyBLEP band-limited sawtooth, so high notes do not alias at the oscillator.

The demo below runs a pad through a separate vowel filter, to show what formant bands do to a spectrum; it is not the vocal engine itself.

A/B PROCESS · VOWELIDLE

Vowel filter — shaping formant bands

Scroll to load

Choose a, i, u, e or o and compare the pad with the filtered signal. The real vowel filter combines three resonant bands with a direct path, changing the spectral shape without transposing the notes. Both spectra share one reference scale, and neither A/B signal receives extra peak or loudness matching. This is a timbre effect, not speech synthesis; output level can differ between vowels.

Compare

Vowel

free-reed — a tongue through a slot

Field	Struct default	Range	What it is
free_reed.breath_pressure	0.7	0 – 1	Bellows pressure — the dynamic level and the drive.
free_reed.reed_stiffness	0.5	0 – 1	Asymmetry of the tongue nonlinearity: softer is rounder, stiffer is buzzier.
free_reed.detune	0.3	0 – 1	Musette beating between the pair of tongues on a note. 0 collapses to a single tongue.
free_reed.brightness	0.6	0 – 1	How much high content the reed plate radiates: mellow reed organ at the bottom, buzzy harmonica at the top.
free_reed.slot_duty	0.0	0 – 0.9	The fraction of the period the tongue's first pass holds the slot open. It places the null the measured references carry near the seventh partial. 0 uses the simpler shaped-saw source.
free_reed.radiation	1.0	0 – 1	How much of a free monopole (a small source that pushes air out equally in every direction) the plate is: 0 radiates the flow, 1 its derivative. A harmonium measures at 1, a cupped harmonica at 0.

This one is not a waveguide either. A free reed's pitch comes from the tongue itself, not from a coupled air column, so there is no delay line and no round trip — which is exactly why an accordion plays the same pitch whatever the bellows are doing, while a flute jumps an octave when you blow harder.

Two more plucked models

karplus-strong — the plain plucked loop

Field	Struct default	Range	What it is
ks.pick_position	0.18	0 – 0.5	Plucking point as a fraction of the period.
ks.decay_s	3.0	0.05 – 60	String t60 at A4.
ks.decay_stretch	0.5	0 – 1	How much longer the low strings ring.
ks.mute_harmonic	0.0	0 – 16	Hand mute, quoted as a multiple of the note's own fundamental rather than as a fixed frequency — a palm puts the break at the same harmonic whatever the pitch. About 2 is a dead thump, 6 a light palm, 0 open.
ks.exc_brightness	0.85	0 – 1	Excitation lowpass openness at full velocity.
ks.release_damp_s	0.08	0.01 – 10	Damped t60 at note-off.

This is the cleanly terminated plucked string, and it is what a great many GM programs fall back to — the guitars, the basses and the harp. Compared with plucked-string the difference is the bridge: here it reflects and that is all, so the note decays smoothly rather than shimmering. Several richer behaviors live in this engine and default to off, including fret slap, a second polarization and bridge coupling between the two planes.

harpsichord — a quill, and three string choirs

Field	Struct default	Range	What it is
harpsichord.pluck_8a	0.14	0 – 0.5	Plucking point of the first 8′ choir, as a fraction of the string.
harpsichord.pluck_8b	0.22	0 – 0.5	The second 8′ choir, plucked at a different point — which is what makes the two registrations sound different rather than merely louder.
harpsichord.pluck_4	0.11	0 – 0.5	The 4′ octave choir.
harpsichord.plectrum_edge	0.8	0 – 1	Sharpness of the quill's release.
harpsichord.velocity_range_db	5.0	0 – 24	The whole dynamic range key speed buys. A real instrument gives three to six decibels, and not even monotonically — past a certain speed the plectrum slips off sooner and the note gets quieter.
harpsichord.rear_coupling	0.35	0 – 1	How much of the short undamped segment behind the bridge reaches the output. This is where the instrument's inharmonic shimmer comes from; the speaking partials stay harmonic to within a couple of cents.

Registration uses separate string choirs: two 8′ unisons and a 4′ octave are three independent delay lines at three periods, not one string with a mix control.

Trim the gain before you A/B it

Raw output levels are not matched across engines, so a harpsichord can sit substantially higher or lower than the voice beside it. Lower the patch gain before putting a harpsichord next to another engine, and audition the pair at a matched level.

Hear the contrast

The demo below plays one note through each of the sixteen engines that sound without a sample bank, the waveguide models on this page and the non-waveguide ones beside them. Listen to the onset above all — that is where the models and the waveform part company hardest. Subtractive starts loud and gets carved; the plucked loops start with a burst and immediately begin losing energy; the bowed string takes a few periods to grip before the tone locks in; the wind voices ramp their pressure and speak with a chiff. The cutoff control acts on the shared filter, so it has the same meaning across these voices. Raw levels are not matched between engines, so match gain before judging tone.

SYNTH · SUBTRACTIVEIDLE

One note, sixteen engines — waveform against physical model

Scroll to load

The same A3 through every synth engine that sounds without a sample bank. Subtractive, FM and additive build the tone from oscillators: one carved by a filter, operators modulating each other, a sum of partials. The rest solve a model of a vibrating thing — a plucked or struck string, a bow gripping and slipping, a ringing bar, a column of air driven by breath or a reed. Percussion maps the note to a drum-kit piece rather than a pitch. Only subtractive has an oscillator to choose, so that control disappears in the other modes. Only the piano model is calibrated against a reference recording; the other physical models are approximations, and levels differ between engines. Press play and listen for how each one begins, which is where the models and the waveform part company most.

Engine

Cutoff

2200 Hz

Where the family ends

Of the non-waveguide engines in the demo above, three sit close enough to this family to need a word. They are documented on Built-in Synthesizer rather than here. modal is a bank of tuned resonators struck at once — the right model for a bar or a bell, where the modes are known and none of them travels. percussion models the vibration modes of an ideal circular drumhead (not whole-number multiples of one pitch) with filtered noise and, for shakers and scrapers, a stochastic particle source; it is a modal model with a noise bed, not a loop. additive is additive synthesis: it sums drawbar partials rather than modeling a physical object.

The percussion engine does carry one pair of fields worth stating here, because a kit tuned against an older build will need retrimming:

Field	Struct default	Range	What it is
percussion.wire_threshold	0.1	0 – 1	How far the head must swing before the snare wires make contact, as a fraction of the swing a full-velocity strike reaches on that piece. 0 is always in contact and 1 is never, so the whole interval means the same thing on every kit piece. Strike velocity enters the same comparison, which is the nonlinearity a strainer actually is: a soft hit can stay under the threshold for its whole length.
percussion.wire_decay_ms	0.0	0 – 2000	Damping of the wire bed in its own right, so the rattle can outlive the head that started it — which is why a snare's tail is broadband after its head tone has gone. At 0 the rattle is a plain gate, which is what every piece without a strainer uses.

More engine keys

Beyond the voicing fields above, each engine takes further engineParams keys, all listed by synthEngineParamInfo(engineMode). Ranges and units are the engine's; the Default column is the base preset's value. A key whose feature is off at its default leaves the sound unchanged until you raise it, and a gated key reaches the voice only while the key named in its activeWhen holds.

velocityExponent (0 – 4) is the note-on level curve (velocity / 127)^exponent. It is a key on the bowed-string, reed, brass, flute, plucked-string, karplus-strong, free-reed, vocal, modal, subtractive and fm engines.

Engine	Key	Range	Default	What it does
reed	bellRadiationHz	0 – 8000	0	Bell radiation corner on the output side only, so the loop and its tuning stay as voiced. 0 is off.
brass, flute, free-reed	onsetBreathOvershootDb	0 – 8	brass 6.89, others 0	How far the mouth pressure (the bellows pressure, for free reed) stands above its held level at velocity 127 as the note speaks, before it settles. 0 is off.
	onsetBreathSettleMs	1 – 500	brass 134, others 40	Time the overshoot takes to settle onto the held level (about 95%). Gated on the overshoot.
	onsetBreathOvershootVelDb	-3 – 6	brass 1.35, others 0	Velocity dependence of the overshoot. Gated on the overshoot.
free-reed	onsetBreathOvershootKeyDb	-6 – 6	0	Key dependence of the overshoot. Gated on the overshoot.
bowed-string	hyperbolicFriction	on / off	false	Switches the junction to a hyperbolic friction law solved against the string's characteristic line, with the bow axes read in newtons and metres per second.
	stringTensionN	1 – 500 N	60	String tension; with stringLengthM it sets the characteristic impedance. Gated on hyperbolicFriction.
	stringLengthM	0.1 – 2 m	0.325	String length. Gated on hyperbolicFriction.
	forceRatio	0 – 12	0	Bow force as a multiple of the impedance times the bow speed, the operating point Schelleng's window is drawn in. 0 is off. Gated on hyperbolicFriction.
	bowWidthM	0 – 0.05 m	0	Bow-hair contact width along the string. 0 is a point bow.
	bowPositionFf	0 – 0.5	0	Bow position at velocity 127: a louder stroke moves toward the bridge, and bowPosition becomes the pianissimo point. 0 is off.
	bowPositionSpread	0 – 0.25	0	Half-width of a per-note spread of the contact point, so the harmonic the bow nulls changes from note to note.
	sectionSpreadCents	0 – 50 cents	0	Section vibrato depth: several players, each with an independent vibrato. 0 is off.
	sectionRateHz	0.5 – 12 Hz	5	Section vibrato rate. Gated on sectionSpreadCents.
	stringLossHarmonic	0 – 1000	0	Corner of a round-trip loss fixed in harmonic index rather than in Hz, quoted at A3. 0 is off.
	stringLossPerOctave	0 – 4	0	How that corner moves per octave. Gated on stringLossHarmonic.
	corpusSeries	0 – 4	0	The family member whose corpus the bridge signal passes through in series instead of the parallel bank; 0 keeps the parallel bank. Gated on body: 'violin' and bodyMix > 0.
flute	jetDrive	on / off	false	Replaces the cubic jet table with a flow-split source at the labium whose convection delay shortens as the breath rises.
	jetOffset	0 – 2	0	Jet centreline offset from the labium, in jet half-widths: the asymmetry that voices the even harmonics. Gated on jetDrive.
	jetReceptivityWindow	0 – 0.7	0	Half-width of the averaging window with which the jet reads the returning flow. Gated on jetDrive.
	jetCycleMean	on / off	false	Zero-mean jet deflection. Gated on jetDrive.
	jetDeflectionLimit	0 – 8	0	Limit on the swing at the labium, in jet half-widths; 0 is unbounded. Gated on jetDrive and jetCycleMean.
	jetCentringPerOctave	0 – 4	0	Share of the jet's rest offset removed per octave above the first register, so overblown registers lose their even harmonics.
	vortexLoss	0 – 400	0	Vortex loss at the mouth, which bounds the oscillation when the jet does not.
	latticeCutoffHz	0 – 20000 Hz	0	Tone-hole lattice cutoff, the same at every note. 0 is off.
	breathRadiation	0 – 1	0	Direct radiation of the mouth turbulence, which carries the broadband floor under the fundamental.
	loopPhaseServo	on / off	false	Keeps the nonlinear source's phase from pulling the pitch. Gated on jetDrive being off.
free-reed	musetteFlat	on / off	false	Adds a third tongue as far flat as the musette partner is sharp, so the tremolo straddles the in-tune tongue. Gated on detune.
	musettePartnerLevel	0 – 1	1	Level of each musette partner against the in-tune tongue. Gated on detune.
	musetteTaper	0 – 1	0	How the musette offset follows the compass: 0 holds the cents, 1 holds the beat rate in Hz. Gated on detune.
	slotSwing	0 – 16	0	Tongue swing at A4 in plate thicknesses, scaled with 1/f0. Gated on slotDuty.
	radiationKeytrack	0 – 1	0	Share of the monopole's +6 dB per octave kept across the compass. Gated on slotDuty and radiation.
karplus-strong	sympathetic	on / off	true	The open-string halo that rings in sympathy.
	haloDampS	0 – 10 s	0	Damped t60 of the halo once every key on the part and the sustain pedal are up. It changes the sound only below the halo bank's own ring time; at or above it the halo stays undamped. Gated on sympathetic.
	bodyAirHz	0 – 400 Hz	0	Air-mode frequency of a guitar body; 0 keeps the bank's 100 Hz. Gated on body: 'guitar' and bodyMix > 0.
	bodyAirT60S	0 – 2 s	0.1	Air-mode t60 of a guitar body. Same gate.
	bodyKnock	0 – 4	1.6	The pluck striking the body's air mode, as a ratio to the pluck's own fundamental. 0 is no strike. Same gate.
karplus-strong, plucked-string	pickPositionPerOctave	0 – 4	0	Octaves the pick position rises per octave of pitch, quoted at A2 and capped at the string's midpoint. Gated on pickPosition.
	excBrightnessPerOctave	0 – 4	0	Octaves the excitation lowpass corner rises per octave of pitch; 0 keeps the corner in Hz.
plucked-string	hfDecayS	0 – 60 s	0	t60 of the partials at 4 kHz; the loss filter is solved from this and decayS. 0 is off.
	midDecayS	0 – 60 s	0	Optional t60 at 1 kHz. Gated on hfDecayS.
	inharmonicityB	0 – 0.01	0	Inharmonicity coefficient B, the same at every note. 0 is a harmonic string.
	pickNoise	0 – 1	0	Pick or plectrum contact click at note-on.
harpsichord	rippleLossShare	0 – 0.4	0	Depth of a loss comb as a share of the loop's loss per traversal; the fundamental keeps its t60.
	rippleRatio	0.05 – 0.95	0.5	The comb's length as a share of the loop period. Gated on rippleLossShare.
	polCoupling	0 – 1	0	Coupling to a second string plane, heard as a beat and a two-stage decay. 0 skips the second plane.
	polDetuneCents	-50 – 50 cents	1	Second plane's mistuning. Gated on polCoupling.
	polT60Ratio	0.05 – 2	0.55	Second plane's t60 as a share of the first's. Gated on polCoupling.
modal	malletMs	0 – 50 ms	0	Mallet contact time at full velocity. 0 is off.
	malletVelExp	0 – 1	0	How far the contact shortens as the strike hardens: malletMs * velocity^-exponent. Gated on malletMs.
	contact	0 – 16	0	Level of the mallet's direct contact radiation, a pulse outside the envelope. 0 is off.
	contactMs	0.02 – 20 ms	0.5	Contact time of that pulse. Gated on contact.
pipe-organ	pipeDetuneCents	0 – 30 cents	4	Per-pipe tuning error at and below C4, fixed per rank and note; the slow beating between ranks it leaves is the chorus.
	loopPhaseServo	on / off	true	Keeps each rank's jet phase from pulling its pitch.
	windRegulated	on / off	true	Regulated wind: the sag is only the dip at a rise in demand, and a steady draw holds pitch and level.
vocal	sectionSpreadCents	0 – 50 cents	0	Section singers' vibrato depth, applied outside the source-filter chain. 0 is off.
	sectionRateHz	0.5 – 12 Hz	5	Section vibrato rate. Gated on sectionSpreadCents.

synthEngineParamInfo(...).unit is one of none, cents, Hz, ratio, ms, s, count, m, dB, N (newtons, for stringTensionN), mm or dBPerOct.

Reaching them from code

The deep fields above are settable from every binding through engineParams. The tables spell them with the engine's internal names; the key a patch takes is the section's field in lowerCamelCase (bowForce, ranks2.level; harpsichord.pluck_8a is pluck8a), and synthEngineParamInfo(engineMode) lists the exact keys with their range, unit and default.

So there are four things you can actually do.

Pick a preset, and get its physical voicing whole. Every physical-model preset ships with its deep fields already set, and selecting it by name is how you reach them. Read the catalog from the runtime rather than hardcoding it.

Browser / Node

const names = synthPresetNames();
// 'violin', 'cello', 'clarinet', 'alto-sax', 'trumpet', 'french-horn',
// 'concert-flute', 'shakuhachi', 'koto', 'sitar', 'accordion', 'harmonica',
// 'choir-aah', 'acoustic-piano', 'church-organ', 'harpsichord', ...

const patch = synthPresetPatch('cello');
// The wrapper sections only — patch.engineMode is 'bowed-string', and the
// bow's own voicing is not returned; override it with engineParams.

const audio = project.bounceWithSynthInstrument(
  { preset: 'cello', cutoffHz: 3600, ampAttackMs: 40 },
  { totalFrames: 48000, numChannels: 2 },
);

Python

names = sonare.synth_preset_names()
patch = sonare.synth_preset_patch("clarinet")

audio = project.bounce_with_synth_instrument(
    sonare.SynthPatch(preset="clarinet", cutoff_hz=3600.0),
    total_frames=48000,
    num_channels=2,
)

Override the engine's own fields. Put them in engineParams on a patch (engine_params in Python). Absent keys keep the preset's value. An unknown key, a key of another engine, a non-finite or fractional-for-integer value, or a value outside the field's range is refused by name. The values are calibrated per preset, so change one or two at a time and listen.

Browser / Node

const audio = project.bounceWithSynthInstrument(
  { preset: 'cello', engineParams: { bowForce: 0.7, bowPosition: 0.1 } },
  { totalFrames: 48000, numChannels: 2 },
);
const keys = synthEngineParamInfo('bowed-string'); // name, min, max, default, unit

Python

audio = project.bounce_with_synth_instrument(
    sonare.SynthPatch(preset="cello", engine_params={"bowForce": 0.7, "bowPosition": 0.1}),
    total_frames=48000,
    num_channels=2,
)
keys = sonare.synth_engine_param_info("bowed-string")   # name, min, max, default, unit

The keys stay lowerCamelCase in Python. A few bowed-string fields are gated: stribeck reaches the voice only while elastoPlastic is on, and corpusScale / corpusTiltHz only with body: 'violin' and bodyMix > 0. Those conditions are the entry's activeWhen (the same camelCase key in Python's SynthParamInfo); see Gated fields.

Override the shared layer on top of it. Anything in the wrapper — filter cutoff and resonance, both envelopes, the LFOs, glide, body mix, stereo spread, gain — applies over whichever engine the preset selected. The percussion engine is the exception: of these it honors only gain, busDrive and polyphony. The same fields are resolvable as continuous automation targets through resolveInstrumentAutomationId. What you cannot automate that way is structure: preset, engine mode, waveform, filter model, polyphony and the mod routings are refused rather than smoothed. The amp and filter envelope times are explained on Envelopes and Modulation. A patch gain must be finite and within [0, 4]; a value outside that range is refused by name.

Play the exciter live, through the excitation axes. This is the route that moves the physics during a note. Besides the static engineParams fields, the engine defines four abstract axes — force, position, brightness and morph — that a controller gesture or a mod-matrix route can move mid-note, and each engine declares which of them it reads.

Engine mode	Axes it reads
bowed-string	force, position
pipe-organ, reed, brass, flute, free-reed	force, brightness
vocal	brightness
additive	morph, at patch level rather than mid-note
piano, plucked-string, karplus-strong, harpsichord, percussion, subtractive, fm, modal, sample	none

The struck and plucked voices declare no mid-note excitation axes: their exciters are one-shot, so a live excitation route has no target after the note starts.

Two ways to drive the axes. A controller profile maps a device gesture to an axis — the default gm profile sends CC2 (breath) to force and CC74 to brightness or position, whichever the engine reads — and you select or extend it with setControllerProfile and bindController on the realtime engine. A mod-matrix routing does the same thing from inside the patch: route the breath, aftertouch, expressionCc or pitchBend source to the excitationForce, excitationPosition, excitationBrightness or spectrumMorph destination, and it travels with the patch and needs no rebinding.

One more thing to know before you audition: a bare engine mode voices that engine's base preset (fm the e-piano, modal the bell, percussion the GM drum kit). Only sample stays silent until you bind a bank. See Built-in Synthesizer for the details.

Related: Built-in Synthesizer · Sound Sources · GM and GS Fallback Bank