IR Analysis — User Guide

Measure the decay and clarity of a recorded impulse response, prepare it as a convolution kernel, reshape the measured space into creative IR material, and audition that space on another Sound — entirely inside Praat.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 2.1 (2026) License: MIT License Repo: Praat AudioTools

Contents:

What this does Measurement → sculpture Quick start Acoustic analysis Convolution-ready IR Early / late decomposition Creative IR sculpting Creative presets Audition by convolution Parameters Outputs Visualization Batch / headless use Notes & limitations

What this does

IR Analysis has two deliberately separated roles. First, it analyzes the selected Sound as a measured impulse response using decay and clarity measures. Second, if requested, it derives new Sound objects from that measurement: a trimmed convolution-ready IR, complementary early/late components, a creatively sculpted IR, and an audition result produced by convolving the IR with another open Sound.

The measurement is always completed before preparation or creative processing begins. The selected Sound itself is never modified by the creative stage, and all reported room-acoustic figures refer to the original analyzed response.

Core idea: the same recorded space can function as both measurement and compositional material. A response can be quantified, separated into early and late energy, reshaped, and then heard acting on another sound without leaving Praat.

Measured and creative results are not interchangeable. <source>_IRart is a derived artistic object. If it is analyzed again, the resulting figures describe that sculpture, not the original room.

Measurement → sculpture

SELECTED RESPONSE ↓ ACOUSTIC ANALYSIS start point → Lundeby truncation → Schroeder decay EDT / T20 / T30 / C50 / C80 / D50 / Ts ↓ OPTIONAL IR PREPARATION direct-sound alignment → measured tail cut → final fade ↓ OPTIONAL EARLY / LATE DECOMPOSITION ↓ OPTIONAL CREATIVE SCULPTING early/late balance → tail shaping → spectral tilt → late-tail reversal → pre-delay ↓ OPTIONAL AUDITION dry Sound × prepared or creative IR

The stages are intentionally one-way. Creative processing never feeds back into the measurement engine.

Quick start

Select exactly one Sound containing the impulse response or impulse-like room response to analyze.
Run IR_Analysis.praat. Channel 1 is analyzed by default.
Keep Lundeby (recommended) truncation unless you specifically want raw integration to the end of the Sound.
Choose Broadband + octave bands when comparative band information is useful; choose broadband only for a faster, simpler run.
Enable Prepare_convolution_ready_IR to create a trimmed kernel aligned to the direct sound by default.
Enable Creative_IR_processing to choose a creative starting point and then edit the six sculpting controls.
Enable Audition_on_another_Sound to hear the prepared or creative IR convolved with another Sound that is already open in Praat.

For audition: another Sound must already be open before the script starts. The first other open Sound becomes the default audition source, and its name remains editable in the Audition dialog.

Acoustic analysis

The broadband measurement follows an ISO 3382-1-style decay workflow implemented entirely with native Praat operations.

Measure / stage	Implementation
Start point	By default, the first sample reaching −20 dB relative to the peak. Advanced settings can instead use the peak sample or the start of the Sound.
Noise & truncation	Iterative Lundeby crosspoint estimation. If the crosspoint search fails but a noise estimate exists, the script falls back to the first smoothed point approximately 5 dB above the estimated noise.
Noise handling	Chu-style noise-energy subtraction in each decay block plus analytic late-tail compensation when a valid fitted slope is available.
Decay curve	Schroeder backward energy integration.
EDT	Least-squares decay fit over 0 to −10 dB, extrapolated to 60 dB.
T20	Least-squares fit over −5 to −25 dB.
T30	Least-squares fit over −5 to −35 dB.
Validity	A converged fit is refused when the available peak-to-noise range is insufficient: 20 dB for EDT, 25 dB for T20, and 35 dB for T30.
Curvature	`100 × (T30 / T20 − 1)`, reported only when both T20 and T30 are valid.
C50 / C80	Direct energy ratios around 50 ms and 80 ms boundaries relative to the detected start point.
D50	Percentage of total analyzed energy arriving within the first 50 ms.
Ts	Centre time from sample-domain integration of `(t − t0) × h²(t)`; it is not read from the block grid.

Octave-band analysis

Optional octave bands use Praat's zero-phase Filter (pass Hann band). For each requested band, the script also measures the filter's own decay from a unit impulse. A band RT is accepted only when it is more than ten times longer than that self-decay. Bands more than 60 dB below the strongest measured band are treated as filter leakage and their decay/clarity values are refused.

Not IEC 61260 octave filtering. The per-band values are intended for comparative and compositional work, not standards-conformant octave-band reporting. The Hann filter is zero-phase/acausal and can move a very small amount of energy ahead of the impulse; this is negligible for RT but can bias per-band C50 slightly toward higher clarity.

Convolution-ready IR

When any downstream processing is requested, the script builds a prepared base from a copy of the measured response. The end point is derived from the measured truncation rather than from a fixed duration: the Lundeby crosspoint is backed off by the selected tail margin along the fitted decay.

Start alignment. By default, the direct sound is aligned to 0. Choose Preserve measured timing to retain the pre-arrival timing from the original Sound.
Measured tail cut. The kernel ends according to the measured decay/truncation logic.
Final fade. A short raised-cosine fade is applied only at the very end, leaving the preceding measured decay unchanged.
Peak protection. Generated IRs are never normalized upward. If a peak exceeds 0.99, the whole derived IR is attenuated by one scalar so the internal early/late energy relationship is preserved.

If Prepare_convolution_ready_IR is enabled, the prepared Sound is kept as <source>_IRprepared. When preparation is needed only internally for sculpting or audition, the temporary prepared base is removed at the end.

Early / late decomposition

The preparation dialog can split the prepared IR into two full-length complementary Sounds:

<source>_IRearly — direct sound and early reflections;
<source>_IRlate — late reverberant field.

The boundary is measured from the detected direct sound and can be 50 ms, 80 ms, or custom. A raised-cosine crossfade is applied around the boundary. The early and late windows are complementary, so the two components sum back to the prepared IR without a level bump or discontinuity.

Why 50 / 80 ms? These boundaries also connect directly to the script's C50 and C80 clarity measurements, turning an acoustic distinction into reusable sound material.

Creative IR sculpting

Creative processing works only on the prepared copy and produces <source>_IRart. The controls deliberately target properties that are meaningful for an impulse response rather than turning the tool into a general effects processor.

Control	Range	Effect
Early_gain_dB	−24 to +12 dB	Changes the weight of the direct/early component before the late field.
Late_gain_dB	−24 to +12 dB	Changes the overall prominence of the late reverberant component.
Tail_decay_shaping_dB_per_s	−30 to +30 dB/s	Applies an exponential amplitude envelope after the early/late split. The internal gain change is limited to +12 dB and −80 dB for safety.
Spectral_tilt_dB_per_octave	−6 to +6 dB/oct	Applies one broad spectral tilt around a 1 kHz pivot. Per-bin gain is limited to ±12 dB. Negative values darken the IR; positive values brighten it.
Reverse_late_tail	Off / On	Preserves the causal onset and early reflections while reversing only the late field. The reversed content is windowed after reversal, with a fresh complementary boundary fade and an end fade.
Additional_predelay_ms	0 to 2000 ms	Adds silence before the derived creative response.

Decay shaping is artistic. A positive or negative Tail_decay_shaping_dB_per_s changes the derived kernel's amplitude envelope; it does not claim that the measured room has a different reverberation time.

Creative presets

Presets are starting points, not hidden modes. Choosing one fills the six creative controls on the following dialog; every value remains visible and editable before processing.

Preset	Early	Late	Decay shape	Tilt	Reverse	Pre-delay
Custom — keep current values	current	current	current	current	current	current
Natural — no change	0 dB	0 dB	0 dB/s	0 dB/oct	No	0 ms
Close / percussive	+3 dB	−8 dB	−8 dB/s	+1 dB/oct	No	0 ms
Distant bloom	−6 dB	+5 dB	+5 dB/s	−1 dB/oct	No	40 ms
Dark resonance	−3 dB	+6 dB	+3 dB/s	−3 dB/oct	No	20 ms
Reverse bloom	−4 dB	+5 dB	+2 dB/s	−1 dB/oct	Yes	30 ms

Preset philosophy

The names describe transformations of the recorded space, not simulated room types. A stairwell remains that recorded stairwell; the preset changes how its measured response is weighted, colored, shaped, or reversed.

Audition by convolution

Audition_on_another_Sound is a quick perceptual check, not a replacement for the library's dedicated convolution tools. It convolves another already-open Sound with either the prepared IR or, when available and requested, the creative IR.

The script finds the named dry Sound among the Sounds that were open when the run began.
If the dry Sound is multichannel, channel 1 is used for audition.
If the IR and dry Sound have different sampling frequencies, the IR audition copy is resampled to the dry Sound's rate.
Praat convolution is performed, followed by optional preview-only wet-level matching.
The wet result and dry Sound are mixed according to Wet_percent.
Final peak protection attenuates only when the audition output exceeds 0.99.

Wet-level matching

A clap-derived IR carries the arbitrary gain of the recording chain, so a raw wet percentage is not necessarily perceptually meaningful. With Match_wet_level_to_dry enabled (default), the wet convolution RMS is matched to the dry Sound RMS over the dry Sound's own duration before the wet/dry mix. The matching correction is limited to ±36 dB and affects only the audition preview; neither _IRprepared nor _IRart is changed.

The audition output is named <dry>_through_<IR>.

Parameters

Main dialog

Parameter	Default	Meaning
Channel	1	Channel to analyze. 0 explicitly sums all channels to mono; the script warns when that sum loses more than 1 dB relative to the channel mean.
Bands	Broadband + octave bands	Runs broadband only, or broadband plus nominal octave bands.
Truncation	Lundeby (recommended)	Measured truncation/crosspoint processing or raw integration to the end.
Prepare_convolution_ready_IR	Off	Keeps the prepared convolution kernel as an output Sound.
Creative_IR_processing	Off	Opens the preset and creative-sculpting stages and produces `_IRart`.
Audition_on_another_Sound	Off	Convolves another open Sound with the prepared or creative IR.
Draw_visualization	On	Draws the AudioTools analysis page; creative processing adds a measured/prepared vs sculpted comparison panel.
Play	Off	Plays the last Sound produced by the current run.
Advanced_settings	Off	Opens detailed analysis and table-output controls.

Advanced analysis defaults

Parameter	Default	Meaning
Lowest_band	125 Hz	Lowest nominal octave-band center.
Highest_band	8000 Hz	Highest requested nominal center, subject to Nyquist protection.
Block_length_ms	1.0 ms	Decay-curve energy block size, with a minimum of four samples.
Noise_subtraction_and_tail_compensation	On	Enables Chu subtraction and analytic tail compensation when supported by the Lundeby estimate.
Start_point	ISO 3382-1 (−20 dB re peak)	Direct-response start rule.
Tail_margin_dB	5 dB	How far above the estimated noise/truncation region the prepared IR should end along the fitted decay.
Results_table	Off	Creates broadband and, when available, octave-band Praat Table objects.

Preparation defaults

Parameter	Default	Meaning
Start_alignment	Align direct sound to zero	Removes the leading pre-arrival region from the prepared kernel, or preserves it when requested.
Final_fade_ms	5 ms	Raised-cosine fade at the prepared IR's final edge.
Early_late_split	80 ms	50 ms, 80 ms, or custom boundary measured from the detected direct sound.
Custom_split_ms	80 ms	Used only when Custom split is selected.
Crossfade_ms	4 ms	Complementary raised-cosine early/late transition.
Create_early_and_late_Sounds	Off	Keeps the two complementary decomposition Sounds.

Audition defaults

Parameter	Default	Meaning
Dry_Sound_name	First other open Sound	Name of the Sound to audition through the IR.
Wet_percent	100%	Wet/dry mix. The script accepts values above 0 and up to 100.
Use_creative_IR_if_available	On	Uses `_IRart` for audition when it exists; otherwise uses the prepared IR.
Match_wet_level_to_dry	On	Preview-only RMS matching before wet/dry mixing.

Input & output behavior

Output	When created	Meaning
`<source>_IRprepared`	Prepare_convolution_ready_IR enabled	Direct-sound-aligned or timing-preserved, measured-tail-trimmed IR with final fade and attenuation-only peak protection.
`<source>_IRearly`	Create_early_and_late_Sounds enabled	Full-length early component with complementary boundary crossfade.
`<source>_IRlate`	Create_early_and_late_Sounds enabled	Full-length late component complementary to `_IRearly`.
`<source>_IRart`	Creative_IR_processing enabled	Artistically derived IR after visible/editable sculpting controls.
`<dry>_through_<IR>`	Audition_on_another_Sound enabled and a valid dry Sound is found	Convolution audition with optional wet-level matching and wet/dry mix.
`<source>_IRresults`	Results_table enabled	One-row broadband Table containing metadata, noise/truncation information, EDT/T20/T30 validity, curvature, clarity, definition, and Ts.
`<source>_IRbands`	Results table + octave bands	Per-band Table containing T30, T20, EDT, C50, relative band level, validity flags, and measured filter self-decay.

The source Sound is preserved. Internal work objects are removed at the end, while every requested user-facing output remains in the Objects window.

Visualization

With Draw_visualization enabled, the script draws the AudioTools analysis page with:

Energy-time curve with estimated noise floor and truncation point;
Schroeder decay curve with accepted T20 and T30 fit lines;
Octave-band reverberation-time panel comparing accepted T30 and EDT values, or a clear message when bands were not requested;
Summary strip with T30, T20, EDT, peak-to-noise range, C50, C80, D50, Ts, noise floor, truncation, curvature, and analyzed-channel information.

When a creative IR exists, the page expands with a fifth panel: Prepared vs sculpted IR (dB). The prepared envelope is drawn in grey and the creative envelope in orange against the same reference, so the effect of gain balance, decay shaping, reversal, spectral color, or pre-delay can be inspected visually.

Batch / headless use

Analysis, octave bands, report, results tables, and visualization can run headless across the Praat versions documented in the script. The preparation, creative, and audition stages use beginPause dialogs and therefore require interactive use in newer headless builds.

Praat 6.4.63+ headless behavior: the script documents that beginPause can terminate a headless run silently. For batch analysis, leave Prepare_convolution_ready_IR, Creative_IR_processing, and Audition_on_another_Sound unticked. To automate those stages, set their defaults directly in the script and remove the corresponding pause calls.

Notes & limitations

ISO-style, not a universal standards claim. The broadband start, decay-fit ranges, curvature logic, and noise-clearance checks follow the ISO 3382-1-style procedure documented in the script. The octave-band implementation is explicitly non-IEC.
Channel choice matters. Real multichannel IR channels are separate measurements. Channel 1 is the default; channel 0 is an explicit sum. If that sum cancels energy, the script warns that a comb-filtered sum should not be treated as the room decay.
Prepared IRs are not normalized upward. Peak protection only attenuates derived Sounds when necessary.
Spectral tilt is deliberately broad. It is a single tilt around 1 kHz, not a multiband equalizer.
Reverse processing affects the late field only. The onset/early structure stays forward, and the late content is reversed before new boundary/end windows are applied.
Audition is intentionally basic. Multichannel dry audition uses channel 1, and the feature exists to hear the measured/derived space quickly rather than to replace a dedicated convolution processor.
No external runtime. The implementation uses native Praat only: no Python, subprocesses, or temporary external processing files.