Temporal Warping — User Guide

Align, stretch, or compress speech and musical phrases in time using dynamic temporal mapping functions.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 1.0 (2025) License: MIT License

Contents:

What this does Quick start Warping model Script options Applications Examples Troubleshooting FAQ

What this does

The Temporal Warping script performs non-linear time mapping of Sound, Pitch, Intensity, or TextGrid objects in Praat. You can define warp functions to accelerate, decelerate, or synchronize two signals (like dynamic time warping, but customizable). It supports both analytic mappings and control-point-based interpolation curves.

Design: This documentation maintains your signature clean visual style for the plugin set (same spacing, typography, and components).

Quick start

Load a Sound or Pitch object in Praat.
Run Run script… and select Temporal_Warping.praat.
Choose Warping mode: manual (control points) or from anchor file.
Define mapping pairs: original time → warped time.
Apply to desired object(s). The script resamples or interpolates as needed.

Tip: You can also export the warping curve as a TableOfReal for reuse in other scripts.

Warping model

Temporal warping is defined by a monotonic function w(t) mapping original time t to warped time. This allows temporal stretching, compression, or non-linear alignment.

Mapping function

w(t) = Σᵢ Bᵢ(t) * Δtᵢ, where Bᵢ are basis splines or linear segments between control points.

The derivative dw/dt indicates local time dilation or contraction.

Type	Description	Use case
Linear	Piecewise-linear mapping between defined anchor points.	Simple tempo alignment.
Spline	Smooth continuous curve preserving monotonicity.	Gradual accelerando/ritardando.
Dynamic	Derived from time alignment of two signals (DTW).	Automatic synchronization.

Script options

Inputs

Primary object: Sound, Pitch, or Intensity
Reference object (optional): target for alignment
Warp source: manual | anchor file | DTW | formula

Control

Interpolation: linear | spline
Resampling rate: Hz
Anchor smoothing: frames
Clamp boundaries: yes/no

Output

Warped Sound or corresponding Pitch/TextGrid
Warp function Table: time→time mapping
Warping ratio plot (optional)

Applications

Speech alignment: Match timing of two utterances (e.g., mimicry studies).
Performance synchronization: Align sung and spoken versions.
Tempo normalization: Stretch/compress recordings to a common time grid.
Expressive timing analysis: Visualize local acceleration/retardation curves.

Examples

1) Manual warp

Define a few anchor pairs manually.

Anchors:
0.00 → 0.00
0.50 → 0.60
1.00 → 1.20
Mode: linear

2) From anchor file

Load a text file with time pairs.

# Anchor file (t_original t_target)
0.0 0.0
1.0 0.9
2.0 2.2

3) Dynamic time warping

Automatically compute mapping between two sounds with DTW, then apply to one of them.

Warp source: DTW
Reference: target Sound
Output: Warped Sound

Troubleshooting

Warp jumps backward — Ensure anchors are sorted and monotonic.
Aliasing artifacts — Increase resampling rate or apply spline smoothing.
Boundary errors — Enable clamp boundaries or extend anchor coverage.
Misalignment — Check reference object duration; DTW requires similar scale.

FAQ

Can I reverse time?

Yes, but only with explicit negative mapping; typical warps require monotonic increase of time.

Can I apply the same warp to multiple objects?

Yes. Once generated, the Warp function Table can be re-used on other objects with identical duration.

Is this real-time?

No, processing is offline but efficient for up to 10-minute signals.