Evolving Convolution Field — User Guide
Transforms one Sound through a synthetic family of mutating convolution generations, then optionally adds adaptive spectral contouring and a stereo trail of recognisable source fragments that progressively erode into material memory.
What this does
Evolving Convolution Field is not a room simulator. It constructs a synthetic stereo impulse response as a sequence of delayed, mutated generations. Each generation can change its modal spectrum, glide behavior, survival, reversal state, stereo position, interchannel coherence, and internal broadband detail.
The selected Sound is convolved with this evolving IR field. The resulting wet branch can then be spectrally compared with the dry source, and a separate Material Memory branch can return short, recognisable fragments of the source as delayed descendants.
Signal flow
Quick start
- Select exactly one Sound.
- Choose a Character. The six characters define different modal, noisy, temporal, and spatial starting behaviors.
- Set Transformation amount and Wet mix.
- Use IR generations, Generation spacing, Generation decay, and Generational mutation to control the temporal field.
- Choose a trajectory, spatial mode, and survival shape.
- Set Adaptive contour if the wet field should follow the source spectrum or occupy what the source lacks.
- Set Material memory to introduce recognisable delayed fragments of the source. Set it to 0 for a convolution-only result.
IR generations
The IR is built as a family of generations rather than one static response. Every generation has its own onset, survival level, mutation state, pan position, frequency tendency, chirp tendency, reversal state, and coherence target.
Generation timing
Nominal generation offsets are derived from the requested spacing and a curvature parameter:
If the requested field would exceed the maximum allowed IR duration, the spacing is compressed to fit while preserving the generational order.
Survival shapes
| Mode | Behavior |
|---|---|
| Decay | Later generations progressively diminish. |
| Swell | Later generations become more prominent. |
| Arch | Energy rises toward the middle of the sequence and then falls. |
| Intermittent | Generations can disappear, producing gaps in the field. |
| Reverse bloom | Later generations are reversed so their energy swells toward their attacks. |
Spectral mutation
Each generation synthesizes an inharmonic modal population whose partial frequencies, decay times, onset spreads, amplitudes, and glides mutate as generational mutation increases.
The resonant partials use bounded, saturating frequency glides rather than unbounded chirps. This keeps trajectories inside the usable frequency region while allowing each generation to bend spectrally.
Broadband structure comes from two additional layers:
- Micro-impulse cloud — sparse broadband transients distributed through each generation.
- Dust descendant — decaying stochastic material whose left/right relationship follows the generation coherence model.
Stereo field
Stereo behavior is built into the IR rather than added as a final panner. A generation can move through the field while it rings, and individual partials can receive their own positions.
| Spatial mode | Behavior |
|---|---|
| Generation pan | All components broadly follow the generation trajectory. |
| Spectral prism | Partial frequency is mapped across the stereo field. |
| Dichotic comb | Alternating partial families are distributed toward opposite sides. |
| Partial scatter | Individual partials receive independent positions. |
Spatial width scales these differences. The coherence controls independently determine how similar, uncorrelated, or negatively related the two channels become across the generation sequence.
Adaptive contour
The adaptive contour operates on the convolution wet branch before Material Memory is added. It compares the current source spectrum with the wet spectrum frame by frame.
| Adaptive contour | Behavior |
|---|---|
| 0 | Off. |
| Positive | Retains wet spectral regions that the source currently shares and attenuates material the source has left behind. |
| Negative | Does the opposite: emphasizes wet material in regions that the current source occupies less strongly. |
Pitch class mode compares octave-equivalent chroma classes. Spectral mode uses a finer half-semitone spectral grid. The same real gain is applied to left and right, so the stereo geometry created by the IR remains intact.
Material Memory
Material Memory adds a second temporal layer after adaptive contouring. It does not reconvolve the source. Instead, surviving IR generations recall short fragments taken directly from locally strong regions of the original material.
Source-aware anchor streams
The source is divided into broad zones. Within each zone, the engine searches several candidate positions and selects a locally strong gesture. The same anchor is recalled by successive surviving generations, which creates a true descendant trail instead of unrelated grains.
Inherited generational behavior
Memory fragments inherit the generation plan: survival, delay, pan, reversal tendency, frequency tendency, and mutation. Later generations become shorter according to the erosion setting, while varispeed and source-position wander progressively loosen the identity of the original gesture.
Stereo Material Memory
Left and right recall related but non-identical descendants. Both channels remain tied to the same anchor and generation, but each side can differ in:
- source position inside the remembered gesture,
- micro-delay,
- varispeed,
- fragment duration,
- gain perturbation,
- and, at stronger divergence, reversal state.
Memory stereo divergence controls how much the channels disagree. It is multiplied by Spatial width, so a width of 0 remains mono-compatible.
Characters
The Character presets define the starting DNA of the IR field and also different Material Memory behaviors.
| Character | General tendency |
|---|---|
| Glass Bloom | Longer, clearer resonances and relatively sparse, recognisable memory fragments with moderate stereo divergence. |
| Metallic Fracture | Shorter metallic structures, stronger mutation, more fractured memory behavior, and wide left/right disagreement. |
| Spectral Dust | Short modal activity, stronger broadband dust, smaller fragments, and a highly dispersed stereo memory field. |
| Alien Resonator | Lower and longer resonances with slower, larger source memories and moderate stereo divergence. |
| Panoramic Orbit | Strong trajectory motion and a spatially active memory trail designed to move through the field. |
| Unstable Matter | High mutation, unstable survival, aggressive erosion, pitch mutation, reversal, and maximum stereo memory divergence. |
Main controls
| Control | Default | Role |
|---|---|---|
| Character | Glass Bloom | Selects the core timbral and memory behavior. |
| Transformation amount | 0.92 | Controls the strength of the synthetic IR components. |
| Wet mix | 0.72 | Final wet/dry blend. |
| Spatial width | 0.88 | Scales stereo motion, spectral spread, coherence departure, and Material Memory divergence. |
| Adaptive contour | 0.5 | −1…1 spectral relation between wet field and current source; 0 disables it. |
| Material memory | 0.34 | Level of the direct source-fragment memory branch; 0 disables it. |
| IR generations | 8 | Number of generations in the evolving IR family. |
| Generation spacing | 32 ms | Base temporal separation between generations. |
| Generation decay | 0.78 | Base survival behavior used by the selected survival shape. |
| Generational mutation | 0.68 | Amount by which later generations diverge from the initial generation. |
| Trajectory | Spiral Orbit | Controls generational movement and related spectral tendencies. |
| Spatial mode | Generation pan | Determines how partials are distributed inside each generation. |
| Survival shape | Decay | Controls how generation energy survives through the sequence. |
| Random seed | 0 | 0 gives a fresh stochastic realization; a positive integer makes a take reproducible. |
Advanced controls
| Group | Controls | Purpose |
|---|---|---|
| Generation timing | Spacing curvature, Max IR duration | Shapes inter-generation timing and limits total IR length. |
| Spectral / spatial field | Spectral spread, Coherence start/end, Reverse probability, Motion depth, Dry pan | Controls partial distribution, interchannel relation, reversal, and motion. |
| Output | Keep kernels, Normalize output | Optionally retains the L/R IR fields and protects the final output from peaks above the target range. |
| Adaptive contour | Contour mode, Memory ms, Lookahead ms | Sets the spectral comparison domain and temporal memory of the contour process. |
| Material Memory | Fragment ms, Fragments per generation, Erosion, Pitch mutation, Reverse probability, Delay scale, Jitter ms, Stereo divergence | Controls the temporal identity, degradation, timing, and channel disagreement of recalled source fragments. |
Visualization
The Picture-window report uses the AudioTools house style and shows the process rather than only the final waveform.
- Source / Output — direct waveform comparison.
- Adaptive contour strip — when active, displays spectral attenuation over time.
- Generational evolution — survival and mutation across generations, with reversed generations marked.
- Partial field / trajectory / coherence — partial positions, generation motion, and measured left/right coherence.
- Generated IR L/R — both synthetic impulse-response channels.
- Summary strip — generation, mutation, spatial, contour, memory, and output diagnostics.
Outputs
The main result is a stereo Sound named from the source and selected Character. The original Sound is not modified.
When Keep kernels is enabled, the generated left and right IR fields are retained for inspection. The Info window reports the effective spacing, IR duration, generation survival/reversal, adaptive-contour behavior, Material Memory activity, output correlation, and mono fold-down behavior.
Notes & limitations
- Input handling: multichannel input is intentionally collapsed to a mono processing source; the tool then creates a new stereo image.
- Not a room model: IR generations are artistic compositional objects, not estimates of physical reflections.
- Material Memory is source-derived: its fragments preserve recognisable source identity and therefore depend strongly on the character of the selected Sound.
- Strong negative coherence: can create impressive width but may lose energy in mono fold-down; the report measures this.
- Adaptive contour is frame based: fast harmonic changes can be smoothed by its analysis window and memory.
- Output normalization: the optional final protection is down-only; it does not boost a quiet result to a fixed peak.
Citation
Cohen, S. (2026). Praat AudioTools — Evolving Convolution Field.