Granular Displacement — User Guide

Single-layer grain-based temporal displacement: the source plus a silent tail is divided into equal regions, each with its own random delayed recursive contribution.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 0.5.1 (2026) License: MIT License Repo: https://github.com/ShaiCohen-ops/Praat-plugin_AudioTools
Contents:

What this does

Granular Displacement appends a silent tail, divides the resulting Sound into equal temporal grains, and assigns a random delay and amplitude coefficient to each grain. Within each grain, delayed samples are added recursively to the evolving wet signal.

This is a single-layer grain map. It differs from Fractal Feedback, which repeats processing over a hierarchy of 2/4/8/... nested layers.

Important: the DSP is additive delayed feedback-like accumulation within each grain. It is not a crossfade formula of (1-a)×dry + a×delayed, and grains are not extracted and reassembled as separate Sound objects.

Quick start

  1. Select exactly one Sound.
  2. Run Granular_Displacement.praat.
  3. Choose a preset or Custom.
  4. Set tail duration, grain count, delay range, amplitude range and Wet/Dry mix.
  5. Click OK. Output: <source>_granular_<preset>.

Presets

The built-in presets override Tail duration, grain count, delay controls and amplitude range. Wet/Dry, Scale_peak, visualization and playback keep the form values.

PresetTailGrainsMin delayMax-delay factorAmplitude range
Subtle1.5 s53 ms0.200.15–0.60
Medium2.0 s85 ms0.250.20–0.80
Heavy2.5 s128 ms0.330.25–0.95
Extreme3.5 s1810 ms0.400.30–1.00
CustomUses the form values.

Parameters

ControlDefaultBehavior
Tail_duration_s2.0 sSilence appended before grain processing.
Number_of_grains8Requested grain count. Internally limited by total sample count and capped at 4096.
Delay_min_ms5 msConverted to samples, with a minimum of one sample.
Delay_max_factor0.25Nominal maximum delay = grain size in samples × factor. If minimum and maximum are reversed, the two effective bounds are internally swapped.
Amplitude_min / Amplitude_max0.2 / 0.8Custom bounds are ordered and both are capped at 1.0 before random values are generated.
Wet_dry_percent60%Clamped internally to 0–100%.
Scale_peak0.95Target peak normalization for every non-silent result/channel.

Grain construction

The source is first copied to a zero-based working Sound and followed by a silent tail with the same sample rate and channel count. The resulting total sample count is divided by the effective grain count:

grainSizeSamples = floor(totalSamples / effectiveGrains) grainDuration = grainSizeSamples / sampleRate nominal max delay = round(grainSizeSamples × Delay_max_factor)

The final grain absorbs any integer-division remainder samples. One delay and one amplitude are generated per grain before processing; these values are also used by the visualization for mono/multichannel processing and for the left channel of stereo processing.

Displacement processing

Within each grain, the wet working signal uses:

if col > delaySamples: self + amplitude × self[col - delaySamples] else: self

The delayed read is taken from the evolving working signal. Therefore contributions can include already modified past samples; the operation is not a simple static dry-plus-tap mix.

After all grains, Wet/Dry is applied against the extended dry Sound (source + silent tail), then the result is peak-normalized to Scale_peak.

No identity bypass: at 0% Wet, the result is still the dry source plus the requested silent tail and is target-normalized if non-silent.

Channel behavior

Mono and 3+ channels

All channels use the same grain boundaries, delay values and amplitudes. Each channel reads its own delayed samples. Channel count is preserved.

Stereo

The left channel uses the pre-generated grain map. The right channel generates a decorrelated map: its delay range is derived from approximately 1.1 × left minimum to 0.9 × left maximum (collapsed to a single value if needed), and its amplitude range is slightly reduced.

After Wet/Dry mixing, left and right channels are peak-normalized separately to Scale_peak before recombination. This can change their relative peak balance.

Visualization

Stereo: the grain map represents the left-channel assignments. The right channel uses separately randomized, decorrelated delay/amplitude values that are not drawn.

Output behavior