Phase Shaper — User Guide

Creative convolution engine with 12 synthetic impulse-response generators for smearing, resonance, rhythmic tails, noise fields and stylized glitch textures.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 0.6 (2026) License: MIT License Repo: https://github.com/ShaiCohen-ops/Praat-plugin_AudioTools

Contents:

What this does Convolution & impulse responses Quick start 12 IR modes Intensity presets Controls Processing pipeline Channels, randomness & level Visualization Technological & compositional context

What this does

Phase Shaper is a creative convolution processor. For each mode it synthesizes a mono impulse response (IR) — a chirp, gated noise, inharmonic resonator, glitch pattern, low rumble, long noise smear and so on — then convolves that IR with the source.

The historical name is not literal: this script does not directly edit FFT phase. Convolution applies the complete transfer function of the synthetic IR, so both spectral magnitude and phase of the result are affected indirectly.

What are convolution and an impulse response?

An impulse response is the signal that describes how a linear time-invariant system responds to an impulse. Convolution combines a source x with an IR h to produce a new signal:

y(t) = x(t) * h(t)

Room convolution normally uses an IR measured from a real space. Phase Shaper instead synthesizes the IR from scratch. The IR is therefore better understood as a compositional kernel: its timing and spectrum are imposed on the source as filtering, smearing, resonance or rhythmic structure.

Quick start

Select exactly one Sound.
Run Phase Shaper.praat.
Choose one of the 12 modes.
Start with Medium; presets override both Intensity and Wet/Dry.
Leave Trim_to_original off when you want to hear the full convolution tail.
For mono sources, keep Stereo_output on if you want the built-in 15 ms pseudo-stereo version.

12 IR modes

Mode	IR construction
Hyper-Dispersion	`3 × Intensity` s linear chirp from 50 Hz to 80% of Nyquist, with linear decay.
Quantum Rain	`2 × Intensity` s Gaussian noise gated by a sinusoid at `15 × Intensity` Hz, under exponential decay.
Fractal Zap	`0.5 × Intensity` s accelerating/FM sinusoidal texture; modulation rate `500 × Intensity` Hz; quadratic fade.
Reverse Black Hole	`1.5 × Intensity` s Gaussian noise with an exponential swell toward the end.
Alien Resonator	`1.5 × Intensity` s inharmonic four-sine chord at base `400 × Intensity` Hz with ratios approximately 1, √2, √5 and √10, exponentially decaying.
Cyber Glitch	`0.4 × Intensity` s two-level square-like chirp, then linearly faded.
Bouncing Ball	`1.2 × Intensity` s thresholded cubic-phase sinusoid creating increasingly dense impulse-like events. This is an acceleration gesture, not a gravitational bounce model.
Deep Space	`4 × Intensity` s low-passed Gaussian noise. Cutoff is `max(50, 150/Intensity)` Hz, with a curved decay.
Spectral Freeze	`6 × Intensity` s low-level Gaussian noise with very slow decay. “Freeze” is descriptive: the IR is long, not mathematically infinite.
Demon Growl	`1.2 × Intensity` s Gaussian noise amplitude-modulated at `25 × Intensity` Hz and one-third of that rate, with exponential decay.
Shattered Glass	`0.25 × Intensity` s bright stochastic high-frequency texture around `min(12 kHz, 0.9 × Nyquist)`, with very rapid decay.
Tape Deterioration	`2 × Intensity` s tonal IR around 180 Hz, frequency-modulated by slow “wow” and faster “flutter”, then exponentially decayed. It is a stylized convolution texture rather than a tape-machine simulation.

Noise-based modes have their measured DC mean removed when non-zero. Every generated IR is then target-normalized to peak 0.99 before convolution.

Intensity presets

Preset	Intensity	Wet/Dry
Custom	Form value	Form value
Subtle	0.5	50%
Medium	1.0	75%
Heavy	1.5	90%
Extreme	2.5	100%

The selected mode is not changed by a preset. Trim, stereo output, scale peak and display/play switches also remain independent.

Controls

Control	Default	Meaning
Intensity	1.0	Mode-specific master parameter controlling IR duration and, depending on mode, density, rates or aggression.
Wet_dry_percent	100	Linear dry/wet blend after convolution; internally clamped to 0…100.
Trim_to_original	Off	Off keeps the convolution tail. On extracts exactly the original duration from the wet result before mixing.
Stereo_output	On	For native multichannel input, preserves its channels. For mono input, creates the script's 15 ms pseudo-stereo delay. Off explicitly collapses multichannel input to mono.
Scale_peak	0.95	Final target peak, valid range >0…1.

Processing pipeline

Prepare a wet source and dry copy. With Stereo_output on, multichannel relationships are preserved; with it off, multichannel input is converted to mono.
Synthesize the selected mono IR at the source sample rate.
Remove DC in flagged stochastic IRs, then peak-normalize the IR to 0.99.
Convolve the working source with the mono IR using Praat's Convolve: "sum", "zero".
If requested, trim the wet signal to original duration.
When dry contribution is non-zero, zero-pad the dry signal to the wet duration if necessary and mix wet × convolution + dry × source.
For a mono source with Stereo_output on, keep the left channel exact and build the right channel as a 15 ms delayed copy.
Target-normalize the complete output to Scale_peak.

Channels, randomness, duration and level

Native multichannel: preserved when Stereo_output is on; the same mono IR is convolved with every channel.
Stereo_output off: multichannel sources become mono.
Mono + Stereo_output on: output becomes pseudo-stereo with a 15 ms delay only in the right channel.
Duration: with Trim off, approximately source duration + IR duration (minus one sample in ordinary discrete convolution). With Trim on, original duration.
Randomness: Quantum Rain, Reverse Black Hole, Deep Space, Spectral Freeze, Demon Growl and Shattered Glass use unseeded Gaussian randomness; repeated runs can therefore differ.
0% wet: the mixed signal becomes dry, but the final Scale peak still target-normalizes it. It is therefore not an amplitude-identical bypass.
Final level: every non-silent result is target-normalized to Scale_peak.

Output name: source name + mode suffix + preset, for example <source>_hyper_medium.

Visualization

The figure uses the actual IR used for processing, not a regenerated copy. It shows source waveform, IR waveform, IR spectrum, output waveform and a summary. Source and output waveforms share one amplitude scale. The IR spectrum is capped at the smaller of 8 kHz and source Nyquist.

Technological and compositional context

Convolution is commonly associated with measured room impulse responses, but an IR can also be designed. In that role, convolution becomes a general timbral-compositional operation: a sparse IR behaves like a delay/filter structure, a resonant IR imposes spectral peaks, a long noisy IR creates diffuse smearing, and a pitched IR transfers strong tonal structure to the source.

Phase Shaper belongs to this creative-convolution tradition. Its mode names — “Black Hole,” “Fractal,” “Freeze,” “Tape” — are musical metaphors for different synthetic kernels, not claims of physical modeling. The historical script name likewise survives as an identity: the processor changes phase only as a consequence of convolution with the IR, not through explicit phase-bin editing.