Intensity-Driven Temporal Walk

One recorded sound, two evolving paths through its material. Praat's measured Intensity contour controls the temporal fragmentation of the right channel, while the left maintains a constant segment clock.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Library: Praat AudioTools Category: CDP Version: 0.2 (2026) Script license: MIT CDP engines: LGPL-2.1-or-later

The musical idea: intensity reorganizes time

Intensity-Driven Temporal Walk treats the dynamic behavior of a recording as compositional information. Praat extracts an IntensityTier from the selected Sound, maps its decibel values to a sequence of segment durations, and passes that changing sequence to the Composers Desktop Project's Drunken Walk processor.

The instrument produces two related but independent temporal recompositions. Both move through the same recorded material according to the same forward-moving locus curve and the same CDP settings. The left channel uses one fixed segment duration; the right channel uses an evolving duration derived from the sound's own Intensity. Loud passages call for shorter segments, while quiet passages call for longer segments.

The recording's dynamics do not merely change the volume of the result. They change how the source material is traversed, cut, and recomposed in time.

Unlike Frozen Canon or Spectral Anticipation, this instrument does not sustain accumulating harmonic voices. It creates a simultaneous two-channel encounter between a stable temporal clock and an analysis-driven, changing clock.

How the instrument works

The complete process runs offline from one selected Praat Sound. An Intensity or IntensityTier object does not need to be prepared or selected. The script performs all measurements and constructs the required CDP breakpoint files internally. It requires no Python.

Analysis stage

From Sound to IntensityTier

Praat runs To Intensity, followed by Down to IntensityTier. The analysis is normally sampled at a 10 ms time step; the CDP clock control is thinned to approximately 50 ms spacing. Very low values are floored at 60 dB below the loudest measured point to avoid digital silence dominating the mapping.

Transformation stage

CDP Drunken Walk

The embedded extend drunk 1 process concatenates sections selected by a constrained, randomized movement through the source. The shared locus trajectory runs forward through the recording. The clock parameter controls the nominal duration of the selected pieces.

Two temporal voices in one stereo Sound

Channel 1 · Left

Constant Walk

The segment clock remains fixed throughout the output. At the default setting, CDP uses a nominal duration of 150 ms per segment.

Channel 2 · Right

Intensity-Driven Walk

The same process receives a clock breakpoint curve derived from Praat. With default settings, the clock varies between 70 and 300 ms depending on the source intensity.

Illustrative stereo organization (not measured data) One Sound Internal Intensity analysis CDP Drunken Walk Constant clock = 150 ms CDP Drunken Walk Intensity-driven clock Stereo Sound L = Constant R = Intensity-Driven Same input, locus, ambitus, step, splice, and seed; only the Clock differs.
The schematic shows the routing principle. It is not a waveform or an exact representation of a particular run. The script's own Praat Picture displays the measured controls and the rendered audio.

There is no panning or cross-channel summation. The Constant result is copied to the left channel, and the Intensity-Driven result to the right. The original recording is not mixed into either channel. A single attenuation factor is applied to both channels only when the combined peak exceeds 0.95; their relative levels are not normalized independently.

How Intensity becomes a CDP Clock

The internal IntensityTier expresses sound intensity in decibels. The script measures the usable minimum and maximum values and linearly maps them to a segment duration. The mapping is inverted: loudness corresponds to shorter segments.

E(t) = (I(t) - Imin) / (Imax - Imin)
Clock(t) = Clockmax - (Clockmax - Clockmin) × E(t)
300 msQuiet source region → long segment
185 msIntermediate intensity → intermediate segment
70 msLoud source region → short segment

The three figures above describe the default Custom mapping; the intermediate value is illustrative. The left channel always uses its separate fixed 150 ms clock, regardless of source intensity.

Breakpoints in output time

CDP interprets the clock breakpoint timestamps on the output timeline, not the original source timeline. The script converts each source-relative analysis time using the requested output-duration factor:

Output breakpoint time = Source-relative analysis time × Output_duration_factor
Clock value = mapped segment duration in seconds

A second breakpoint file controls the common locus: its timestamps are in output seconds, but its values specify positions in the original source. Both CDP passes read the same locus file. This distinction is essential to understanding how the source is navigated and recomposed.

The intensity-derived clock is not an amplitude envelope and is not applied as a gain multiplier. If the measured Intensity range is effectively flat, the driven branch falls back to the Constant clock, clipped to the configured minimum/maximum range if necessary.

Getting started

Installation: Place Intensity_Driven_Temporal_Walk.praat in plugin_AudioTools/CDP/, alongside the CDP Process Explorer. It uses the bundled internal/cdp_engine.praat, the process and parameter registries, and the available bundled CDP binaries. On supported Windows installations, no separate SoundLoom, CDP installation, or Python is necessary.
  1. Select exactly one Sound in Praat. The recording must be at least 2.5 seconds long. Mono and multichannel input are accepted; a temporary mono working copy is made for processing.
  2. Run Intensity_Driven_Temporal_Walk.praat. Leave Custom selected for the reference configuration or choose another preset.
  3. Render the result. The script performs Intensity analysis, creates the CDP control curves, runs two Drunken Walk processes, and assembles one stereo Sound.
  4. Listen by channel. Left is a fixed-clock walk; right is a clock driven by the source intensity. Stereo playback combines these complementary readings.
  5. Inspect the Picture window. With visualization enabled, the graphs show the measured Intensity, the actual breakpoint control, the shared locus, and the rendered stereo waveform.
No extra Praat objects are required: Unlike the original proof of concept, the instrument calculates its IntensityTier internally. Intermediate analysis objects and the two separate CDP result Sounds are removed after stereo assembly; the selected original is preserved.

Six compositional presets

Presets control duration, fixed and intensity-driven segment-clock ranges, and the spatial range of source traversal (ambitus and step). They do not introduce new DSP algorithms. Every preset preserves the comparison between a Constant left channel and an Intensity-Driven right channel.

PresetOutput factorConstant ClockDriven Clock rangeAmbitusStep
Custom1.5×150 ms70–300 ms0.30 s0.08 s
Gentle Temporal Drift1.25×220 ms150–350 ms0.15 s0.04 s
Balanced Walk1.5×150 ms70–300 ms0.30 s0.08 s
Fragmented Gesture1.5×100 ms40–180 ms0.40 s0.12 s
Wide Temporal Excursions1.75×180 ms60–400 ms0.75 s0.20 s
Extended Temporal Landscape2.5×250 ms100–500 ms0.35 s0.06 s

Gentle Temporal Drift

Longer pieces and smaller source excursions encourage a more continuous, less fragmented listening experience.

Balanced Walk

The working proof-of-concept configuration. A useful baseline for comparing a constant temporal clock with analysis-driven articulation.

Fragmented Gesture

Short pieces, larger movement steps, and faster potential clock variation favor sharper temporal recomposition.

Wide Temporal Excursions

A larger ambitus allows the source navigation to wander farther around the common forward locus.

Extended Temporal Landscape

A longer output and gentler stepping create a more extended environment for internal temporal motion.

Custom

Uses the numeric settings from the form rather than predefined preset values. The defaults are the baseline reference configuration.

Musical descriptions indicate intended behavior, not guaranteed sonic outcomes for all recordings. Presets do not override the entered Seed, Splice, or Intensity analysis settings. CDP may produce slightly different durations from the requested output factor; the stereo renderer safely pads the shorter branch without trimming the longer one.

Parameters and output

ControlBehavior and limits
PresetCustom plus five named settings. Presets override the duration factor, Constant Clock, driven Clock range, ambitus, and step.
Output_duration_factorNominal duration multiplier relative to the source. Default 1.5; valid interval 1–4. Output requests above 180 seconds are rejected.
Intensity_pitch_floor_HzPraat Intensity analysis pitch floor. Default 100 Hz.
Intensity_time_step_sIntensity analysis time step. Default 0.01 s; limited to 0.05 s. Clock breakpoint data are thinned to about 0.05 s source-time spacing.
Minimum_clock_sShortest driven segment, reached at the strongest measured intensity. Default 0.07 s. Must exceed the CDP/splice-derived lower limit.
Maximum_clock_sLongest driven segment, reached at the weakest measured intensity. Default 0.30 s; must exceed Minimum Clock.
Constant_clock_sFixed left-channel segment duration. Default 0.15 s.
Ambitus_sSource-navigation range around the locus; default 0.30 s. Limited to half of the source duration when necessary.
Step_sStep size of the Drunken Walk in source-time units. Default 0.08 s.
Splice_msCrossfade/splice duration between fragments. Default 12 ms; valid 1–50 ms; segment Clock values must exceed twice the splice duration.
SeedShared positive integer random seed for both CDP passes. Default 42, for repeatable runs on the same engine.
Draw_visualizationDraw the analysis-and-output Picture after rendering, without changing the audio. Enabled by default.
Play_resultPlay the finished stereo Sound after processing. Enabled by default.

Fixed CDP settings

For both channels: Clock randomness = 0 and Overlap = 0. The locus advances linearly from approximately 0.65 s after the source begins toward approximately 0.65 s before the source ends, using the same breakpoint file for each pass. The two CDP branches share the same Seed, Ambitus, Step, and Splice.

Result

<sourceName>_IntensityTemporalWalk_Stereo

One Praat Sound containing exactly two channels, beginning at output time zero. Left = Constant Walk; right = Intensity-Driven Walk. The original recording remains in the Objects window and is not included in the rendered mix.

Reading the Praat Picture visualization

With Draw_visualization enabled, the script produces a measured 8 × 8-inch Praat Picture composed of four panels plus a summary strip. The graphs are derived from the analysis and the actual output, not from the conceptual illustrations on this web page.

Panel A

Source and Intensity

The source mono waveform is drawn together with its measured IntensityTier curve, using separate amplitude and dB scales. The graph also reports the actual analysis range and number of Tier points.

Panel B

Segment Clock

The blue horizontal line gives the Constant Clock of the left channel. The orange curve shows the clock breakpoints sent to CDP for the right channel. The shared axis is in output seconds, and Clock values are displayed in milliseconds.

Panel C

Source navigation

Both passes follow the same locus trajectory from the beginning toward the end of the source. The surrounding band represents the configured ambitus. Output time is compared directly with the source-position axis.

Panel D

Rendered stereo channels

The actual output waveforms are displayed separately: LEFT – CONSTANT WALK and RIGHT – INTENSITY-DRIVEN WALK. Both use a common amplitude reference so relative levels remain visible.

The summary strip reports the input duration and channel count, Intensity range, preset, breakpoint count, segment-clock settings, CDP process, output duration, and final peak level. Enabling or disabling the Picture does not alter the rendered Sound.

Technical implementation

The instrument uses the embedded CDP process drunk (extend drunk 1) from the Extend & Loop category. Praat handles the analysis, breakpoint construction, CDP dispatch, object management, and stereo assembly.

Two CDP passes, one controlled difference

Both executions share the source WAV, output duration, locus breakpoint file, ambitus, step, splice, clock randomness, overlap, and deterministic seed. The Constant branch passes a numeric clock; the Intensity-Driven branch passes the generated breakpoint-file path.

Registry-aware integration

The script uses internal/cdp_engine.praat to prepare the CDP arguments, confirms the expected argument-template positions for locus and clock, and then substitutes the breakpoint-file paths. It does not change the shared CDP engine or require a separate CDP installation when a supported bundled engine is present.

The audio input is downmixed to mono internally if it has more than one channel. Both CDP outputs are imported as mono Sounds; the final stereo Sound copies the corresponding sample arrays into separate channels. Any shorter branch is padded with silence to match the longer branch. A single attenuation-only peak check preserves interchannel balance.

Repeatability: both processes receive the same nonzero random Seed. Because the Clock values differ, this does not imply the two channels will choose exactly the same sequence of source segments. It makes each processing configuration reproducible, rather than forcing identical stochastic decisions.

Input constraints include a 2.5-second minimum duration, an upper limit of 180 seconds for the requested output, valid positive clocks, and splice-aware clock bounds. The script generates temporary breakpoint files in the AudioTools CDP working directory and removes its generated files after the processing passes.

Listening and limitations

A dialogue, not a stereo effect preset

Listen to each channel separately first. The fixed left-channel clock is a control condition for the Intensity-Driven right channel. Heard together, their differing temporal articulations may create an unstable, shifting stereo relationship.

Dynamics drive time, not gain

The IntensityTier does not amplify or attenuate the result directly. It controls the nominal duration of the source fragments chosen by CDP. Stronger passages can yield denser segmentation, while quieter passages can preserve longer gestures.

Source-dependent results

The amount of contrast depends on the original recording's dynamic range, articulation, and spectral characteristics. A recording with nearly constant Intensity can produce similar or even identical clock settings in both channels.

No direct transposition

This Drunken Walk configuration does not pitch-shift its fragments. Changes in perceived pitch organization arise from reordering and repetition of pieces of the source, not from a transposition algorithm.

Compatibility and verification: This documentation describes Intensity_Driven_Temporal_Walk.praat v0.2 as supplied. The bundled CDP programs must be compatible with the running operating system. Actual sonic behavior depends on the source recording and CDP engine; the examples and diagrams here are explanatory, not measurements from a rendered file.