Spectral Smearing Reverb — User Guide

Filter-bank spectral dispersion: low-frequency bands arrive later than high-frequency bands, producing a diffuse time-smeared tail without convolution.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 0.3.3 (2026) License: MIT License Repo: https://github.com/ShaiCohen-ops/Praat-plugin_AudioTools
Contents:

What this does

Spectral Smearing Reverb is a frequency-dependent dispersion effect. It splits the source into a logarithmically spaced Hann filter bank, delays each band by a different amount, weights the bands with a Lorentzian spectral response, and sums them into a wet field. Low-frequency bands arrive later than high-frequency bands, so attacks and spectral components are spread across time rather than repeated as conventional echoes.

This is not convolution reverb. The wet path is built directly from filtered copies of the source. The term reverb describes the resulting diffuse tail; there is no separate impulse response or convolution stage.

The processor supports mono and stereo sources. Stereo channels are processed independently with slightly different delays, spectral weighting and slow amplitude modulation. The dry stereo image is preserved in the wet/dry mix.

Quick start

  1. Select exactly one mono or stereo Sound object.
  2. Run Spectral_Smearing_Reverb.praat.
  3. Choose Subtle, Medium, Heavy, Extreme, or Custom.
  4. Use Time_stretch to control the maximum low-frequency delay and Frequency_bands to control spectral resolution.
  5. Use Peak_frequency_Hz and Response_width_Hz to shape which bands contribute most strongly.
  6. Set Wet_dry_percent, then run the script.

Presets

PresetTailBandsTime stretchBase ampPeak / widthFade
Subtle Smear1.0 s120.40 s0.20600 / 800 Hz0.8 s
Medium Smear1.5 s200.60 s0.35600 / 800 Hz1.2 s
Heavy Smear2.5 s300.85 s0.50600 / 800 Hz1.8 s
Extreme Smear4.0 s451.20 s0.65600 / 800 Hz2.5 s
CustomUses the values entered in the form.
Presets overwrite the smearing, frequency-response and fade parameters shown above. Wet_dry_percent, Draw_visualization and Play_result remain user-controlled.

Parameters

ParameterDefaultMeaning
Tail_duration_s1.5Requested output tail. The effective tail is at least Time_stretch, so the longest low-frequency band is not truncated.
Frequency_bands20Number of logarithmically spaced bands; internally limited to 2–64.
Time_stretch0.6 sLongest left-channel band delay, occurring at the lowest band centre.
Base_amplitude0.35Overall scale for the wet-band amplitudes before spectral weighting and random variation.
Peak_frequency_Hz600 HzCentre of the Lorentzian weighting; limited to 95% of Nyquist when necessary.
Response_width_Hz800 HzControls how broadly the Lorentzian weighting spreads around the peak.
Wet_dry_percent60%Linear wet/dry blend, clamped to 0–100%.
Fadeout_duration_s1.2 sCosine fade applied to the wet tail only; limited to the effective tail duration.
Draw_visualizationyesDraw the AudioTools analysis/summary figure after processing.
Play_resultyesPlay the resulting Sound when processing finishes.

Processing model

Band centres: logarithmic from about 80 Hz to 0.90 × Nyquist Left delay(f) = Time_stretch / sqrt(f / baseFreq) Right delay(f) = 0.92 × Time_stretch / sqrt(f / baseFreq) Lorentzian weight(f) = 1 / (1 + ((f - peak) / width)^2)

The first filter-bank band is low-pass, the last is high-pass, and interior bands use Hann pass-band filtering with smoothed boundaries. Each band receives a random amplitude factor generated once per run, then a slow positive cosine modulation. The right channel uses slightly shifted response and modulation values for decorrelation.

Randomness: no random seed is set. Re-running the same settings can therefore change the wet-band amplitude pattern. The visualization shows the dispersion law and Lorentzian response, not the per-run random amplitude multipliers.

Channels, duration and level

Visualization

The display contains the dry waveform and the full output waveform on the complete output time span, plus two analysis panels:

Output

The result is named <source>_smear_<preset>, where the preset label is Custom, Subtle, Medium, Heavy or Extreme.