Random DurationTier Multichannel Generator - User Guide

Creates multiple independent random-walk DurationTiers, resynthesizes the source once per tier with Praat Manipulation, and assembles the variants as a multichannel Sound.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 0.5.1 (2026) License: MIT License Repo: Praat AudioTools

Contents:

What this does Quick start Presets Random walk Parameters Processing pipeline Visualization Notes & limitations

What this does

The script takes exactly one selected Sound and generates N independent time-warped variants. Each channel receives its own DurationTier. The tier begins at factor 1.0, follows a bounded Gaussian random walk through evenly spaced interior control points, and returns to 1.0 at the end. Each tier is inserted into a Praat Manipulation object and rendered with Get resynthesis (overlap-add). The variants are then combined into one multichannel Sound.

Interpretation of the factor: 1.0 is the unity-duration reference. Values below or above 1.0 locally contract or expand time according to the DurationTier used by Praat. The script changes duration; it does not construct a granular cloud or independently repitch the channels.

1 or more generated output channels.
Six named presets plus Custom.
Three boundary policies: Clamp, Reflect, and Reject.
Independent stochastic walk for every generated channel; no user seed is exposed.
Optional retention of the generated DurationTier objects.
Five-part visualization showing tiers, factor distribution, resulting durations, output waveform, and summary statistics.

Quick start

Select exactly one Sound object.
Run Random_DurationTier_Multichannel_Generator.praat.
Choose Custom or one of the six named presets.
Choose the boundary policy and, if needed, adjust the pitch-analysis and output controls.
Run the script. The result is named from Output_stem: for example dur8_2ch or dur8_8ch.

Preset semantics: a named preset overwrites only Number_of_channels, Control_points, Variability, Min_factor, and Max_factor. Boundary handling, pitch analysis, naming, peak target, tier retention, visualization, and playback remain the values chosen in the form.

Presets

Preset	Channels	Control points	Variability	Factor range
Custom	form value	form value	form value	form value
Subtle Variations	4	8	0.20	0.80-1.25
Standard Multi-texture	8	10	0.40	0.50-2.00
Extreme Time-stretch	8	15	0.60	0.25-4.00
Dense Polyrhythm	12	20	0.50	0.40-2.50
Minimal Duo	2	6	0.25	0.70-1.50
Chaotic Cluster	16	25	0.70	0.20-5.00

The form defaults to Custom with 8 channels, 10 control points, variability 0.40, and a 0.50-2.00 range. Therefore the default numerical settings happen to match Standard Multi-texture, but the selected preset remains Custom.

Random-walk construction

For N = Control_points: stepSigma = Variability / sqrt(N) state starts at 1.0 for each interior point: step ~ Gaussian(0, stepSigma) update state using the selected boundary policy DurationTier anchors: (tmin, 1.0) N evenly spaced interior points (tmax, 1.0)

Variability is a walk parameter, not a requested standard deviation of the finished tier. The scaling makes the unconstrained N-step random-walk variance comparable across different point counts. Boundaries, the unity anchors, and the fact that the visualization pools all interior stages change the observed distribution.

Boundary handling

Mode	Implementation	Consequence
Clamp	Take one Gaussian step, then clip the state to Min/Max.	Large walks can accumulate samples exactly at the boundaries. This is the default and preserves the earlier v0.3 behavior.
Reflect	Reflect an overshoot back across a boundary; repeat up to 10 reflections, then clamp if still outside.	Avoids the direct edge clipping of Clamp for ordinary overshoots.
Reject	Re-roll an out-of-range step, up to 20 attempts. If none is accepted, take one additional step and clamp it.	Keeps accepted ordinary steps inside the range without reflection.

v0.5.1 validation: Number_of_channels >= 1, Control_points >= 1, Min_factor < Max_factor, and the factor range must include 1.0 because the start and end anchors are fixed at unity.

Parameters

Parameter	Default	Behavior
Preset	Custom	Named presets overwrite only channels, control points, variability, and factor range.
Number_of_channels	8	Number of independent DurationTiers/resyntheses and intended output channels. Must be at least 1.
Control_points	10	Number of evenly spaced random-walk points between the two unity anchors. Must be at least 1.
Variability	0.40	Sets Gaussian step sigma as Variability/sqrt(Control_points).
Min_factor / Max_factor	0.50 / 2.00	Bounds for interior states. Min must be lower than Max and the interval must contain 1.0.
Boundary_handling	Clamp	Clamp, Reflect, or Reject. This remains user-selectable with named presets.
Pitch_floor / Pitch_ceiling	75 / 600 Hz	Passed to `To Manipulation`. v0.5.1 requires ceiling > floor.
Time_step	0.01 s	Passed directly to `To Manipulation`.
Name_prefix	DurRand_	Names the generated DurationTiers (`DurRand_1`, etc.). Relevant after completion only when tiers are kept.
Output_stem	dur8	Names temporary variants and the final multichannel result.
Scale_peak	0.99	Final `Scale peak` target applied to the combined result. It can attenuate or amplify.
Keep_duration_tiers	Off	Keeps the N generated DurationTier objects instead of deleting them after visualization.
Draw_visualization	On	Draws the 8x8 summary before tier cleanup.
Play_result	On	Plays the final result after processing.

Processing pipeline

Input: require exactly one selected Sound; use its original time domain for all DurationTiers.
Preset: if a named preset is selected, overwrite the five preset-controlled duration parameters.
Validation: check channels, control points, factor range/unity anchors, and pitch floor/ceiling.
Tier generation: create one independent random walk per output channel.
Resynthesis: for every tier, run To Manipulation: Time_step, Pitch_floor, Pitch_ceiling, replace its DurationTier, and call Get resynthesis (overlap-add).
Assembly: one variant is copied as mono; two are combined to stereo; 3+ variants are combined iteratively into the multichannel result.
Normalization: apply Scale peak: Scale_peak once to the assembled result.
Visualization and cleanup: draw while the tiers still exist; then remove the temporary variant Sounds and, unless requested, the DurationTiers.

Output duration: each DurationTier can produce a different resynthesized duration. Panel C records those durations individually; the final combined Sound reports its own resulting duration after assembly.

Peak scaling: because Scale peak is unconditional, Scale_peak=0.99 is not an attenuation-only ceiling. A result whose peak is below 0.99 is amplified to the target.

Visualization

v0.5.1 uses the five-part v0.5 visualization and draws it before cleanup, so the DurationTier curves are available even when Keep_duration_tiers is off.

Panel	What it shows
A - Random-walk duration tiers	All N DurationTier curves overlaid, with unity plus Min/Max reference lines.
B - Factor distribution	24-bin histogram of interior factor samples only; the fixed 1.0 endpoint anchors are excluded.
C - Per-channel duration	Actual duration of every resynthesized variant, with the original source duration as a reference line.
D - Output waveform	Mono output, or channels 1 and 2 overlaid when the result has multiple channels. Channels 3+ are not drawn in this waveform panel.
E - Summary	Preset, source, channels, point count, boundary mode, pitch range, Variability parameter, interior-factor SD, mean factor, output duration, final peak, and total factor-sample count.

Interior-factor SD is the population-style standard deviation computed from all interior factor samples across all generated tiers. It excludes the two fixed unity anchors per tier. It is descriptive and is not expected to equal the Variability parameter.

Factor samples is channels × (control points + 2), including the two unity anchors in each tier.

Notes & limitations

The random generator is not seeded by a user parameter, so repeated runs can produce different tiers.
The script uses Praat Manipulation/overlap-add; quality therefore depends on whether the selected pitch-analysis settings are appropriate for the source.
The code does not promise that the observed factor distribution is uniform. Boundary policy materially affects that distribution, especially with wide ranges or high variability.
More control points do not simply mean "more variability": step sigma is reduced by sqrt(N) as N increases.
The six named presets do not overwrite Boundary handling, pitch-analysis controls, naming, Scale_peak, or visualization/playback options.
The final peak normalization changes absolute level across runs; compare temporal behavior rather than untreated loudness if evaluating different random walks.