Praat AudioTools · User Guide

GRM-Style Resonator v2.0

Turn short sounds and transients into sustained, frequency-controlled ringing with a bank of up to 64 independently tuned resonators. Shape harmonic or inharmonic structures, control their decay in seconds, and animate their frequencies over time.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 2.0 (2026) License: MIT License Repository: Praat AudioTools

Overview

GRM-Style Resonator v2.0 is an independently developed instrument for resonant sound transformation. It excites a parallel bank of tuned resonators with the selected Sound. Each resonator can continue oscillating after the excitation stops, producing an audible decay whose duration is specified in seconds rather than as a comb-filter feedback coefficient.

The resonator frequencies may form a harmonic series, an inharmonic collection, a logarithmically spaced bank, or a manually specified set. The bank can remain static or move continuously using sine modulation, smooth random drift, or sample-and-hold frequency changes.

Relationship to INA-GRM: “GRM-Style” refers to the aesthetic lineage of the classic GRM Tools Reson processor. This is not INA-GRM software or a reimplementation of its proprietary algorithm. The v2.0 architecture is an independent resonator bank implemented entirely in Praat.
What changed from v1.5.1? The older Hann band-pass plus integer-delay/feedback-comb architecture has been removed. v2.0 uses tuned, decaying resonators with frequency-independent T60 control, frequency movement, transient-aware excitation and automatic ringing tails. Settings named ringIterations, ringDecay, bandwidthHz and Resonator_mode no longer exist.

Quick start

  1. Select exactly one mono or stereo Sound in the Praat Objects window.
  2. Run GRM-Style_Resonator.praat.
  3. Start with Harmonic Organ for pitched sustained resonance, or Percussive Resonance for short attacks.
  4. Set Dry_wet to 1.0 to hear only the resonator bank. Reduce it if you want the source articulation in the mix.
  5. Use Automatic (longest T60) for Tail_mode, so the ringing can continue after the source ends.
  6. Leave Safety ceiling as the output level mode for attenuation-only overload protection.
  7. Run. The output is named <source>_GRMstyle_<preset>; the original Sound is retained.
A revealing first test: Use a short, broadband attack (a struck object, click or percussive gesture), followed by silence. With a long decay and a wet-only mix, you can hear the bank's pitches continuing after the excitation. Tonal source material can also excite the bank, but the response depends on how its spectral energy overlaps the selected frequencies.

Signal flow

Input Sound
Mono or stereo · original preserved
↓
Excitation
Whole source, transient-focused source, or a blend
↓
Parallel resonator bank
1–64 tuned oscillatory filters · independent T60, gain and stereo coefficients
↓
Optional frequency modulation
Changes resonator frequency while preserving decay and state continuity
↓
Stereo wet sum
Bank scaling · optional automatic/fixed tail
↓
Dry/Wet + output protection
Natural level, safety ceiling or peak normalize
↓
Output Sound + diagnostics

All processing is carried out within native Praat. No Python backend, external DSP library or pretrained model is used.

Presets

These presets define different resonator structures, rather than combinations of the previous filter/comb settings. The table lists effective values established by the v2.0 script.

PresetFrequency structureDecay / excitationMovement and space
CustomForm values; default 8 harmonic resonators from 110 Hz3.0 s base T60; tilt 0.7; Balanced; transient focus 0No modulation; Stereo Spread default
Harmonic Organ12 harmonics of 110 Hz4.0 s; tilt 0.5; Balanced; full-source excitationStatic; alternating stereo
Metallic Resonance6 inharmonic ratios × 220 Hz6.0 s; tilt 0.4; Unity impulse; transient focus 0.5; −6 dBStatic; hard left/right
Resonant Cloud32 log-spaced resonators from 300–2400 Hz; ±40 cents dispersion2.5 s; tilt 1.0; BalancedSmooth random drift, ±8 cents at 0.3 Hz; random pan
Moving Resonance10 inharmonic ratios × 180 Hz3.0 s; tilt 0.8; BalancedSmooth random drift, ±300 cents at 0.15 Hz; V-shape pan
Percussive Resonance8 inharmonic ratios × 330 Hz1.2 s; tilt 0.6; Unity impulse; transient focus 1.0; −6 dBStatic; hard left/right
Deep Resonance8 log-spaced resonators from 40–320 Hz12.0 s; tilt 0.5; Balanced; transient focus 0.3Sine LFO, ±15 cents at 0.07 Hz; alternating stereo
Preset override behavior: Presets set the bank, decay, excitation, pan and modulation parameters specified in the script (including gain profile and, in some cases, modulation depth/rate). Dry_wet, Tail_mode, Output_level_mode, Draw_visualization and Play_result remain user-controlled. The advanced settings for decay_list$, frequency spread/shift, tail caps and random seed also remain in force unless edited explicitly.

Preset frequencies above Nyquist − 20 Hz, or at/below 20 Hz, are omitted and reported. The final resonator count can therefore be smaller than the requested number.

Parameters

Resonator bank and tuning

Form controlDefaultFunction
PresetCustomChoose a bank configuration; non-Custom presets override their respective form fields.
Number_of_resonators8Maximum number requested, from 1 to 64. Manual/ratio lists can provide fewer.
Frequency_layoutHarmonic seriesManual list, Harmonic series, Inharmonic ratios, or Equal logarithmic.
Base_frequency_Hz110Fundamental/reference for harmonic, inharmonic and logarithmic layouts.
Top_frequency_Hz1760Upper end of the Equal logarithmic layout.
Frequencies_or_ratios1 1.41 1.89 2.37 2.98 3.56Space-separated Hz values in Manual mode or ratios in Inharmonic mode; not used by harmonic/logarithmic layouts.
Dispersion_cents0Independent initial random detuning of each resonator, uniformly between ± the specified cents.
Manual-list reminder: The default Frequencies_or_ratios text contains ratios, not a usable manual list of musical frequencies. If you change the layout to Manual list (Hz), replace the text with real frequencies such as 220 330 440 660 880.

Decay, excitation and gain

Form controlDefaultFunction
Resonance_decay_s3.0 sBase T60: the time for the resonant amplitude to decrease by 60 dB after excitation stops (allowed 0.01–60 s).
Decay_tilt0.7Scales decay across log-frequency position. 1 = equal T60; below 1 gives shorter high-frequency decays; above 1 gives longer high-frequency decays.
Excitation_scalingBalancedBalanced (constant noise power), Unity peak (filter-like), or Unity impulse (strike). Affects how strongly the same input excites resonators of different decay times.
Transient_focus00 = whole source excites. 1 = emphasize onset regions while suppressing steady input; intermediate values blend both.
Resonance_gain_dB0 dBOverall linear gain applied to the resonant bank before stereo summing and Dry/Wet.
Stereo_panningStereo SpreadSets left/right output coefficients of each resonator (see Stereo architecture).

Movement, mixing and output

Form controlDefaultFunction
ModulationNoneStatic bank, Sine LFO, Smooth random drift or Sample-and-hold.
Modulation_depth_cents50Frequency excursion in cents; active for non-static modulation (0–2400 cents).
Modulation_rate_Hz0.2 HzModulation rate (above 0 and up to 20 Hz when modulation is enabled).
Dry_wet1.00 = exact dry bypass; 1 = wet resonator bank only; values between are linear dry/wet blends.
Tail_modeAutomaticAutomatic (longest T60), Fixed (advanced tail_s), or None (truncate).
Output_level_modeSafety ceilingNatural level, attenuation-only safety ceiling or peak normalization.
Draw_visualizationOnGenerate the analytical picture page.
Play_resultOnPlay the finished Sound, making a temporary scaled copy for playback only if the peak exceeds 1.

Advanced settings in the script

To keep the Praat form compact, the following settings are edited directly in the ADVANCED SETTINGS section of the script.

SettingDefaultMeaning
gain_Profile21 = Flat, 2 = Dampen highs, 3 = Boost highs. Frequency-relative weighting, separate from the main gain.
decay_list$EmptyOptional space-separated T60 values in seconds. Replaces base decay + tilt; if shorter than the bank, the last entry repeats.
freq_spread1Scales log-frequency intervals around the lowest initial resonator (1 = unchanged).
freq_shift_cents0Transposes the whole bank by a fixed cent offset.
tail_s5 sTail duration used by Fixed mode.
max_auto_tail_s30 sMaximum tail under Automatic mode, even when the longest T60 is higher.
random_seed00 = unseeded variation; nonzero makes initial dispersion, random movement and random panning reproducible.
finalPeak0.99Ceiling for Safety ceiling and Peak normalize.
transient_attack_s / transient_release_s0.05 / 0.15 sTransient detector time constants.

How the resonators work

Each resonator is implemented as a complex first-order resonator whose real output has a two-pole/one-zero response. The poles lie at a radius r and an angle set by the requested frequency. This permits continuous tuning without rounding the frequency to an integer delay length.

Static resonator: z[n] = r · exp(jω) · z[n−1] + x[n] y[n] = g · Re(z[n]) ω = 2π f / fs r = 10^(−3 / (T60 · fs)) Impulse response: h[n] = g · r^n · cos(ωn) n ≥ 0

After the input stops, the resonator continues at its tuned frequency, with an exponentially decreasing envelope. A T60 of 3 seconds means an approximately 60 dB decrease in amplitude over 3 seconds of unforced ringing, independent of center frequency.

The native Praat implementation uses a heterodyned state recursion and compiled Formula operations. It maintains the resonant state through frequency changes rather than restarting the filter every time the tuning moves.

Approximate resonance bandwidth and Q: B ≈ ln(1000) / (π · T60) ≈ 2.199 / T60 Hz Q ≈ f / B ≈ f · T60 / 2.199

These are narrowband/high-Q relationships. For very broad resonances the measured magnitude-response maximum need not fall exactly at the pole frequency. The script therefore reports the response-peak offset in cents separately from the tuned ringing frequency. Q is derived from T60 rather than independently adjustable.

Why this is different from v1.5.1: A comb filter reinforces a periodic series of frequencies and, with short delays and moderate feedback, often sounds mainly like a spectral coloration. Here the resonators have individually controlled oscillation frequencies and seconds-long decays, so a short excitation can produce a sustained pitched response.

Excitation and decay

Three excitation scaling modes

ModeImplemented gain factor before bank/profile gainPractical effect
Balanced (constant noise power)sqrt(2 × (1 − r²))Balances long and short resonators for ideal broadband/noise excitation.
Unity peak (filter-like)2 × (1 − r)Approximately equalizes narrowband steady-state gain near each center; can sound comparatively restrained.
Unity impulse (strike)1Preserves the initial impulse coefficient, letting long decays produce strong sustained ringing; useful for bells and impacts.

Each resonator is additionally weighted by the bank gain profile and the user gain. The entire bank is scaled by 1 / sqrt(number of active resonators). This helps control overall level without forcing every resonator to have equal output energy.

Transient focus

When Transient_focus is above zero, a linked amplitude detector compares a fast peak-following envelope with a slower envelope. A gate favors rapid rises and attenuates sustained input in proportion to the selected focus amount. The same gate drives both source channels, preserving their timing and relative excitation.

0 excites continuously; 1 emphasizes attacks; values in between retain some sustained excitation. This is a transient-sensitive amplitude gate, not a classifier or pitch tracker.

Individual decay times

If decay_list$ is empty, decay varies with the resonators' position in log frequency:

T60(k) = Resonance_decay_s × Decay_tilt ^ frequency_position(k) frequency_position ∈ [0, 1], from lowest to highest center

A custom decay_list$ overrides the tilt. All effective T60 values are limited to 0.01–60 seconds.

Frequency movement

Modulation changes the resonators' actual oscillation frequencies, not merely the loudness of fixed frequency bands. The oscillator phase is integrated from the instantaneous frequency, and its resonant state remains continuous throughout the movement.

ModulationBehavior
NoneFixed tuning throughout the output.
Sine LFOSmooth periodic frequency movement; different resonators receive distributed initial LFO phases.
Smooth random driftPer-resonator random control values connected by raised-cosine interpolation.
Sample-and-holdPer-resonator frequency targets jump at the control rate, while the resonant state and phase remain continuous.

Modulation_depth_cents is the maximum positive/negative cent excursion applied before clipping the instantaneous frequency to the valid 20 Hz … Nyquist − 20 Hz range. Modulation_rate_Hz controls the LFO or random-control update rate.

Initial Dispersion_cents, freq_spread and freq_shift_cents modify the starting bank layout; modulation subsequently moves those selected frequencies over time.

Stereo architecture

The wet output is a two-channel stereo sum. Each resonator has a left and right coefficient; these coefficients distribute its contribution across the wet bus.

Pan modeCoefficients
All Center (dual mono)Left 1.0 / Right 1.0
Stereo Spread (Alternate)Alternates 1.0/0.3 and 0.3/1.0
Wide Spread (Hard L/R)Alternates 1.0/0 and 0/1.0
Left Heavy / Right Heavy1.0/0.3 or 0.3/1.0 for all resonators
V Shape (Outside In)Sweeps coefficient emphasis across the ordered resonator list
RandomIndependent random per-resonator right coefficient between 0 and 1; left is 1 minus right
Important with stereo sources: Both input channels excite separate resonator states, and the left/right gain coefficients then scale the corresponding existing channels. Material present only in the source left channel is not moved into the source right channel by selecting a rightward pan. For a mono source, each resonator is sent to both wet channels according to its coefficients. More than two source channels are not supported.

Tail, Dry/Wet and output behavior

Resonant tails

Tail modeEffective behavior
Automatic (default)Appends silence equal to the longest effective T60, capped by max_auto_tail_s (30 s by default).
FixedAppends the duration specified by the advanced tail_s (5 s by default).
NoneNo extra time is appended. Ringing can be cut off when the source ends.

If the chosen tail is shorter than the longest T60, the script applies a 20 ms safety fade (or less for a very short output) to the wet tail end to avoid a sudden cut. This is not a substitute for allowing the resonance to decay naturally.

Dry/Wet

Dry_wet = 0 returns an exact copy of the original input samples, with the original channel count and without an appended tail; output-level processing is skipped. For Dry_wet > 0, the processed output is stereo. A mono source is converted to stereo for blending; a stereo source retains its original left/right dry channels.

output = (1 − Dry_wet) × dry + Dry_wet × resonator_wet

Output level

ModeBehavior
Natural levelNo scaling. Strong or overlapping resonances may create peaks above 1.0.
Safety ceiling (default)Attenuates the entire mix only when its peak exceeds finalPeak (0.99 by default).
Peak normalizeScales any non-silent output to the selected peak, even when amplification is needed. This can obscure the perceptual effect of changing gain.

Praat retains the source's original start time (xmin) and sample rate. The final Sound object is named <source>_GRMstyle_<presetName>. If Natural level produces peaks above 1.0, the Sound is retained unchanged and the Info window warns about integer-PCM clipping; playback uses a temporarily peak-scaled copy.

Visualization

The v2.0 visualization is an approximately 8-inch diagnostic page designed to show the resulting resonant behavior, not the removed v1.5.1 band-pass architecture.

1. Output waveform and tail

Output signal with a shaded extension after the source ends and a vertical marker at the source end. Its scale follows the output so a quiet ringing tail is not hidden by a strong input transient.

2. Output spectrogram and trajectories

Measured output spectrogram with red curves showing the actual resonator-frequency trajectories sampled from the phase used by the processor. The source end is marked.

3. Measured wet spectrum

Measured spectrum of the wet bus (blue), with intended static center frequencies (red vertical lines), on a logarithmic frequency axis. Under modulation the red vertical markers are the starting centers, not the complete moving paths.

4. Configured versus measured T60

Bars show the assigned decay time for each resonator; dots show decay estimated from its internal state envelope when a usable post-source tail is available.

5. Configured gain versus output level

Relative configured gain profile (line) compared with the measured per-resonator output RMS (bars). Each series is normalized to its own maximum for shape comparison.

6. Summary

Effective resonator count, frequency range, decay, excitation scaling, modulation, stereo mode, wet/out peaks, output protection, tail and duration.

Measurements versus settings: The spectrogram, wet spectrum, individual output RMS and state-derived T60 estimates are computed from generated sound or resonator states. Intended center frequencies, configured gain and configured T60 are design settings. A measured T60 can be unavailable when the output does not include enough post-source samples, or when there is insufficient state energy for reliable estimation.

The Info window additionally reports each resonator's center frequency, T60 set/measured, pole radius, Q, approximate bandwidth, response-peak deviation in cents, and output RMS.

Practical examples

A struck object becomes a resonant instrument

Start with Percussive Resonance or Metallic Resonance. Use Unity impulse, a wet-only mix and Automatic tail. Short attacks excite tuned ringing rather than remaining short filtered gestures.

Build a sustained harmonic field

Choose Harmonic Organ and adjust Resonance_decay_s in Custom mode if needed. Reduce Transient_focus to allow both attacks and sustained source material to excite the bank.

Introduce slow movement

Try Moving Resonance or use Sine LFO with a low rate and modest cent depth. The bank frequencies move while the resonant states continue evolving.

Deep, slowly decaying structures

Start from Deep Resonance, with a long Automatic tail. Low resonators can keep oscillating long after the source has stopped.

In non-Custom presets, changing a form value that the preset overwrites will not change that preset's effective setting. To design a custom bank, select Custom and edit the relevant controls directly.

Limits and technical notes