Fractal Feedback Reverb — User Guide

Recursive whole-signal delay accumulation with an exponential memory-depth cycle, secondary cosine-modulated delays and an appended reverb tail.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 0.4.1 (2026) License: MIT License Repo: https://github.com/ShaiCohen-ops/Praat-plugin_AudioTools

Contents:

What this does Quick start Presets Parameters Delay pattern Recursive processing Channel behavior Visualization Output behavior

What this does

Fractal Feedback Reverb extends the source with a silent tail, then repeatedly adds delayed copies of the evolving signal. Primary delays follow an exponential cycle controlled by Seed_delay_s, Chaos_factor, and Memory_depth. Every iteration also adds a shorter secondary delay whose amplitude is cosine-modulated.

This creates a dense, recursively accumulated delay texture. The term fractal here refers to the repeating exponential delay pattern. The script is not a conventional feedback delay network (FDN) with a matrix of interacting delay lines.

Different from Fractal Feedback: this tool processes the entire signal iteratively and appends a tail. Fractal Feedback instead uses nested temporal segmentation over the original duration.

Quick start

Select exactly one mono or stereo Sound.
Run Fractal_Feedback_Reverb.praat.
Choose a built-in preset or Custom.
Set tail duration, delay-cycle parameters, amplitudes, modulation frequency and Wet/Dry.
Click OK. The output is <source>_fractalReverb_<preset>.

Presets

Built-in presets override all tail, chaos and amplitude controls listed below. Wet/Dry and the two output switches keep the values currently shown in the form.

Preset	Tail	Iterations	Seed	Chaos	Depth	Amp base	Secondary	Mod Hz
Subtle	1.5 s	20	0.06 s	1.5	3	0.12	0.65	25
Medium	2.0 s	32	0.08 s	1.8	4	0.18	0.75	30
Heavy	2.8 s	48	0.10 s	2.1	5	0.24	0.85	38
Extreme	4.0 s	70	0.12 s	2.5	6	0.30	0.95	45
Custom	Uses the form values.

Parameters

Control	Default	Meaning
Tail_duration_s	2.0 s	Silence appended before processing; output duration becomes source duration + tail.
Iterations	32	Number of primary + secondary recursive additions.
Seed_delay_s	0.08 s	Base primary delay.
Chaos_factor	1.8	Exponential multiplier inside the memory cycle.
Memory_depth	4	Number of exponent positions before the delay pattern resets.
Amplitude_base	0.18	Starting scale for the per-iteration random primary coefficient.
Secondary_factor	0.75	Scales the secondary delayed contribution relative to the primary contribution.
Modulation_freq_Hz	30 Hz	Cosine amplitude modulation applied to the secondary contribution.
Wet_dry_percent	60%	Clamped internally to 0–100%.

Delay pattern

Primary delay(i) = Seed_delay × Chaos_factor^((i-1) mod Memory_depth) Secondary delay(i) = Primary delay(i) / Chaos_factor Example: seed 0.08 s, chaos 1.8, depth 4 80 ms → 144 ms → 259.2 ms → 466.6 ms → 80 ms → ...

The primary/secondary delay times themselves are deterministic for a given preset or Custom setting. Randomness enters through the per-iteration amplitude coefficient.

Recursive processing

For each iteration i, the left/mono path calculates:

amp1 = Amplitude_base × (1 - i/Iterations) × randomUniform(0.6, 1.4) amp2 = amp1 × Secondary_factor self ← self + amp1 × self(x - primaryDelay) self ← self + amp2 × self(x - secondaryDelay) × cos(2π × x × Modulation_freq_Hz)

Each formula operates on the evolving working signal, so later passes build on the result of earlier passes. At the final iteration (i = Iterations), the linear amplitude term reaches zero.

Wet/Dry is applied against the extended dry signal (source followed by silence). The output is then peak-normalized to 0.95.

No identity bypass: 0% Wet still returns the dry source plus the appended silent tail, and the non-silent result is normalized to a 0.95 peak.

Channel behavior

Mono

Mono input remains mono. One recursive path is processed and normalized to 0.95.

Stereo

The two channels are processed independently. The right channel uses Chaos_factor + 0.02, Seed_delay + 0.002 s, Modulation_freq - 2 Hz, a different random-amplitude range, and Secondary_factor - 0.03. Left and right are then normalized separately to 0.95 before being combined back to stereo.

3+ channels: the current script creates a one-channel tail for any input that is not stereo, then concatenates it with the source. Consequently the implementation is intended for mono or stereo input; 3+ channel input is not safely supported.

Visualization

Dry waveform.
Reverb waveform drawn only over 0 to the original source duration; the appended tail is not shown in this waveform panel.
Delay pattern for the first 20 iterations (or fewer): nominal left/mono primary and secondary delays.
Memory-depth wrap markers at cycle resets.
Parameter line and summary strip.

The delay plot does not show random amplitude coefficients and does not show the right-channel offset parameters used during stereo processing.

Output behavior

Duration: source duration + Tail_duration_s.
Sample rate: preserved.
Channels: mono remains mono; stereo remains stereo.
Normalization: fixed target peak 0.95. Stereo channels are normalized separately before recombination.
Randomness: random amplitude coefficient on every iteration; no user seed, so repeated runs can differ.
Time origin: processing concatenates a new extended Sound and does not explicitly restore a non-zero source xmin.