Ambi Encoder Stochastic

s3g Ambi Encoder Stochastic is a 64-voice stochastic synthesis instrument with first-through-seventh-order ACN/SN3D output. It can run freely, respond to MIDI, or combine both behaviors.

Each voice owns four evolving pressure-curve generators. Their breakpoint amplitudes and segment durations are controlled by reflected second-order random walks: one walk changes the step, and that step changes the audible curve. Segment durations set oscillator period directly, so pitch can leave its starting region and continue evolving.

Workflow

1. Insert s3g Ambi Encoder Stochastic on a 64-channel REAPER track, choose a factory preset or begin from FREE PERIOD, and select the ambisonic ORDER.
2. Follow it with an ambisonic decoder.
3. Choose FREE, MIDI, or BOTH.
4. Set the active voice count.
5. Choose a generator selection law and transition method.
6. Shape the random walks, time-fields, topology, AED projection, and optional field LISTEN mode.

Preset Library

The title-bar PRESET menu provides complete starting states. Choosing one resets the evolving generator bank under the new settings and briefly bridges the output for a continuous change.

User Presets

SAVE writes the current state as a portable .s3gstochastic file. LOAD restores one through the macOS file chooser. The file contains every synthesis, selection, time-field, envelope, topology, projection, listener, ambisonic-order, and output setting.

User presets are versioned JSON documents and can be moved between systems or kept with a project archive. Changing a control after loading or selecting a preset marks the current state as CUSTOM. REAPER project state stores the complete plugin state independently.

Parameter Surface

The SURF page turns as many as 24 captured or factory states into a two-dimensional performance space. Use ADD, DEL, and CAP in edit mode, assign a factory state from the selected-cell menu, then switch to PLAY and move the crosshair. CURVE changes the blend response, FOCUS changes how tightly the cursor favors nearby cells, and GLIDE sets the settling time for the effective cursor, toolbox controls, and topology geometry. The target diamond follows drag or automation immediately while the active crosshair and participating controls glide. POP opens SURF in an attached panel and returns the primary view to FIELD so both remain visible.

Continuous walk, time-field, envelope, topology, projection, and listener-amount controls interpolate by distance. Menus, counts, and listener mode follow the nearest cell, while the lanes represented by VOICES fade continuously into or out of both synthesis and the field drawing. OUT, ORDER, and the X/Y cursor remain live rather than jumping with a cell. SURFACE X and SURFACE Y are normalized, automatable CLAP parameters. The complete surface is saved in REAPER project state; portable .s3gstochastic files remain single states. See Parameter Surface for the shared workflow.

Second-Order Walks

The distribution menus provide UNIFORM, GAUSS, CAUCHY, LOGISTIC, ARCSINE, EXPON, and BINARY. The amplitude and timing walks can use different distributions.

Controls that become turbulent near their upper limit use a tapered GUI response. The lower and middle regions receive more slider travel, and the most extreme walk, density, spread, and topology values occupy the right edge. The taper applies only to slider travel; displayed values, user-preset values, and REAPER automation use the underlying parameter scale.

BASE, SPREAD, and DEV seed the initial periods. They are not pitch centers after initialization. LOW reflects the total polygon period before it becomes subaudio while preserving the relative duration of every segment. Generator identity and topology Z distribute voices upward from that floor. MIDI notes reseed a voice when assigned.

Generator Selection

A selection law chooses one of each voice's four generators at a waveform boundary. A choice persists across several cycles before the law is sampled again, giving each evolving pressure curve time to become audible. The history strip shows recent choices.

XFER lets nearby voices exchange generator tendencies and walk motion, and sets how strongly a new event can propagate to its nearest neighbors. MEM controls selection-law memory, generator persistence, and the stability of the slower macro state.

LINK moves to the selected generator, MERGE interlaces it with the previous generator, and VARY creates a topology-weighted variation between them. A short interpolation at each cycle boundary keeps the transition continuous without blurring the pressure curve for the full cycle.

Time-Fields

Free-running voices use renewal clocks: every active event is followed by a genuine rest before that voice can begin again. DENS sets the target activity, DUR sets the mean event duration, CONT increases duration variation and clustered propagation, and REST sets the minimum inactive time. The ADSR envelope smooths each change of state.

DENS is compensated for the active voice count: the same setting uses a lower per-voice target as the field grows from 8 toward 64 voices. Gentle activity regulation steers the ensemble toward that target without hard voice limits. This keeps the texture articulated instead of automatically becoming continuous; even at maximum density, REST remains in force.

Each onset can schedule delayed events at the two nearest topology points. Distance determines propagation time, while XFER and CONT set cascade strength. The network lines brighten as a cascade arrives, making clustered activity visible as well as audible.

Macro Scenes

The autonomous engine also moves through metastable macro scenes. A scene slowly changes four internal relationships together: activity, event duration, pressure-curve mutation, and cascade strength. MACRO sets the approximate time spent in each region. Higher MEM makes the changes smaller and more persistent; lower values allow wider movement between contrasting states.

In MIDI mode, note gates replace the autonomous time-fields. BOTH allows either source to open a voice.

Field Listener

LISTEN decodes the instrument's pre-output-gain ACN/SN3D field through eight cube-corner virtual microphones. Their smoothed directional energy charges a slow voice-local capture state rather than creating an audio return. LOCAL lets each voice read the pickup nearest its own direction, CROSS reads the opposing pickup, DIFFUSE gives every voice a stable weighted mixture of the complete aperture, and ROAMING moves independent read heads continuously around it. CAPTURE controls the depth of this negative-feedback viscosity, while OFF disables listener influence and leaves the generator autonomous.

As capture rises, new amplitude and duration accelerations are reduced, their existing second-order velocities lose momentum, generator evolution becomes less frequent, time-field clocks slow, and cascades are inhibited. The sounding configuration therefore settles into metastable stasis without lowering oscillator pitch or output level. Capture has a quick attack, a long release, and bounded minimum rates so the system can eventually thaw rather than lock permanently. It never biases a generator destination or renewal probability.

The LISTEN display shows the eight-ear auditory body, the route read by the selected voice, capture depth, and the effective mutation, evolution, field-clock, and cascade rates. Listening never writes to topology or projection, so enabling it cannot periodically translate or collapse the ensemble. The spatial field remains governed only by the Topology and Projection panels.

Topology and Projection

The main display is both a synthesis map and the source of each voice's AED placement. Points move in one persistent topology; there is no separate physics or spatial-motion layer.

• X biases pressure-curve mutation and generator weighting.
• Y changes event probability, field duration, and selection tendency.
• Z biases the segment-duration walk and therefore oscillator period.
• Distance from the center increases walk drive and transition variation.
• Nearest-neighbor links carry generator influence, walk motion, and delayed event cascades between voices.

SHAPE, ANIM, RATE, AMT, DEPTH, SCALE, COLL, and TWIST reshape and animate the shared field. AZIM, ELEV, and DIST place its center. FOLLOW sets how fully topology position and radius are projected into AED space.

Click a point to inspect its current and next generators. Drag blank field space to change the camera; use TOP, SIDE, 3/4, and the zoom buttons to restore useful views.

Pressure Display

The gray curve is the selected voice's current generator and the brighter curve is the next generator. Breakpoint markers belong to the next curve. A1 and T1 report the current primary-walk barriers for amplitude and duration.

Order and Routing

The output bus remains 64 channels wide. The selected order activates the corresponding ACN channels and zeroes all remaining outputs.