Studio Dynamic Compressor — User Guide

RMS-based downward compression with separate attack and release, quadratic soft knee, phase-safe multichannel linking, optional sidechain filtering, and explicit output-level policies.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 1.3 (2026) License: MIT License Output: Sound object

Contents:

What this does Quick start Controls Presets Detector & sidechain Compression law Output policies Visualization Notes & limitations

What this does

The script applies one linked gain envelope to the selected Sound. It first builds an RMS-style detector from all input channels, optionally filters the sidechain, smooths the detected level with separate attack and release constants, converts the smoothed level to a gain-reduction curve, adds optional makeup gain, and multiplies that curve into every source channel.

Multichannel behavior: the detector is phase-safe. Each channel is optionally sidechain-filtered in its original signed form, squared, accumulated, averaged across channels, and square-rooted. The result is sqrt(mean(channel²)), so opposite-polarity channels do not cancel each other. The same final gain curve is then applied to every channel.

The processor is an offline RMS compressor. Even the two fastest presets are not sample-accurate peak limiters. The optional sidechain filters are zero-phase frequency-domain filters, which are useful offline but are acausal and can create a small amount of gain movement before a transient.

Quick start

Select exactly one Sound object. Mono, stereo, and wider multichannel Sounds are supported.
Run Compressor.praat.
Choose a preset or leave Custom.
Set threshold, ratio, knee, attack, release, and makeup behavior.
Enable Advanced settings only when detector, sidechain, output policy, statistics, or original-object handling needs to be changed.
Run the processor. The result remains selected and can optionally be played.

Default Custom behavior: threshold −20 dB, ratio 4:1, knee 6 dB, attack 10 ms, release 100 ms, Auto makeup on, 8 ms detector window, 2 ms detector update, no sidechain filter, and the scale-down-only output policy with target peak 0.99.

Controls

Main form

Control	Default	Implemented behavior
Preset	Custom	Selects one of six fixed starting configurations. Presets override several detector and dynamics fields; details are listed below.
Threshold_dB	−20.0	Detector level at which compression begins. No fixed GUI range is enforced.
Ratio	4.0	Downward-compression ratio. The script rejects values below 1:1.
Knee_dB	6.0	Width of the quadratic soft-knee region. 0 gives a hard knee; negative values are rejected.
Attack_ms	10	Smoothing time used when the detected level rises.
Release_ms	100	Smoothing time used when the detected level falls.
Makeup_gain_dB	0.0	Flat gain added after gain reduction. If Auto makeup is on, this value is replaced by the script's automatic estimate.
Auto_makeup	Yes	Computes makeup from threshold and ratio; it is not loudness matching or input/output RMS matching.
Advanced_settings	No	Opens the detector/output-policy dialog.
Draw_result	Yes	Draws transfer, waveform, metering, gain-reduction, and summary panels.
Play_result	Yes	Plays the final Sound after processing.

Advanced settings

Control	Default	Implemented behavior
Rms_window_ms	8.0	Controls the Praat Intensity analysis window through `pitch_floor = 3.2 / window_seconds`.
Detector_update_ms	2.0	Time step of the Intensity analysis.
Sidechain_filter	Off	Off; high-pass 80 Hz; high-pass 150 Hz; or low-pass 8 kHz. Filtering is detector-only.
Output_mode	Limiter (scale down only if needed)	Three final level policies are available. In this guide the middle mode is described more precisely as a peak ceiling; it is global scaling, not a dynamic limiter.
Scale_peak	0.99	Target used by the scale-down and normalize modes. Must be >0 and ≤1.
Show_stats	Yes	Reports output peak, output RMS, and maximum compression-only gain reduction.
Keep_original	Yes	If disabled, the original Sound is removed after successful processing.

Preset interaction: the advanced dialog appears before preset values are applied. Non-Custom presets therefore override their own RMS window, detector update, sidechain filter, makeup gain, and Auto makeup values. They do not override Output mode, Scale peak, Show stats, or Keep original.

Factory presets

Preset	Threshold	Ratio	Knee	A / R	RMS / update	Makeup	Sidechain	Suffix
Vocal Leveler	−24 dB	2.5:1	8 dB	15 / 150 ms	15 / 3 ms	+4 dB, Auto off	HP 150 Hz	`_Vocal`
Drum Punch	−18 dB	6:1	3 dB	1 / 50 ms	3 / 1 ms	+3 dB, Auto off	Off	`_Drum`
Mix Bus Glue	−14 dB	2:1	10 dB	30 / 200 ms	25 / 5 ms	+2 dB, Auto off	Off	`_Bus`
Fast Compressor RMS	−3 dB	20:1	0 dB	0.5 / 50 ms	2 / 0.5 ms	0 dB, Auto off	Off	`_FastComp`
Squash	−30 dB	10:1	6 dB	5 / 80 ms	5 / 1 ms	+8 dB, Auto off	Off	`_Squash`
NUKE Extreme RMS	−40 dB	100:1	0 dB	0.5 / 30 ms	2 / 0.5 ms	+15 dB, Auto off	Off	`_NUKE`

Custom uses suffix _Comp. Every preset preserves the source duration, sample rate, and channel count because compression is applied to a copy of the selected Sound.

Detector and sidechain

Phase-safe linked detector

For each source channel c: 1. optionally apply the selected zero-phase sidechain filter to the raw signed waveform 2. square the filtered channel 3. add it to a one-channel accumulator After all channels: sidechain(t) = sqrt( (Σ channel_c(t)²) / Nchannels )

This prevents anti-phase channels from cancelling the detector. A mono Sound and the same mono signal duplicated across two identical channels therefore drive the linked detector consistently.

Sidechain filters

Mode	Praat filter	Purpose / limitation
Off	None	Full-band detector.
Highpass 80 Hz	Pass Hann band: 80 Hz → Nyquist, 50 Hz smoothing	Reduces low-bass influence on gain reduction.
Highpass 150 Hz	Pass Hann band: 150 Hz → Nyquist, 50 Hz smoothing	Further reduces low-frequency influence; the Vocal preset uses this mode.
Lowpass 8 kHz	Pass Hann band: DC → 8 kHz, 50 Hz smoothing	Reduces high-frequency contribution to the detector. It is explicitly not a dedicated de-esser.

Praat's Hann-band frequency-domain filtering is zero-phase. That avoids phase shift in the detector path, but its symmetric impulse response is acausal, so gain movement can begin slightly before a transient. The filter also has a Nyquist validation check before processing.

RMS-style level conversion

The linked magnitude signal is passed to Praat's Intensity analysis with subtract mean = no. The script then applies a fixed calibration offset:

dBFS detector ≈ Praat Intensity dB − 93.9794

This follows Praat's pressure convention (reference pressure 2×10⁻⁵ Pa) and keeps a sidechain filter's level change intact. It is an approximation to a conventional rectangular-window RMS meter because Praat's Intensity analysis uses its own analysis window.

Attack and release

The detector is smoothed in the dB domain. Rising detector values use the attack constant; falling values use the release constant. For each direction:

N = round(time_constant / detector_frame_step) α = 1 − exp(−2.2 / max(1,N)) smoothed = previous + α × (current − previous)

Compression law

Gain reduction is computed from the smoothed detector level. With a hard knee (Knee_dB = 0), levels below threshold pass unchanged and levels above threshold follow the selected ratio.

Above threshold: GR_dB = −(level − threshold) × (1 − 1/ratio) gain = 10^(GR_dB / 20)

For a non-zero knee, the transition region extends from threshold − knee/2 to threshold + knee/2. Inside that region the script uses the standard quadratic interpolation:

over = level − threshold + knee/2 GR_dB = −[over² / (2 × knee)] × (1 − 1/ratio)

The visualization's transfer curve uses the same formula as the audio engine. That graph shows the static compression law only: it does not include makeup gain, sidechain timing, attack/release behavior, or final output scaling.

Auto makeup

Auto makeup is a fixed heuristic, not loudness matching:

headroom = min(abs(threshold_dB), 20) auto_makeup_dB = (headroom / 2) × (1 − 1/ratio)

When Auto makeup is enabled, this calculated value replaces the manual Makeup gain value for the run.

Output policies

Mode	Exact behavior	Effect on makeup gain
Preserve level	No final scaling. If the processed peak exceeds 0 dBFS, the Info window warns but the Sound is left unchanged.	Fully preserved.
Limiter (scale down only if needed)	If the final peak exceeds `Scale_peak`, Praat scales the entire file down so its peak equals the target. If it is already below the target, no scaling occurs.	Preserved unless the global peak ceiling is hit.
Normalize to target	For any non-silent result, Praat scales the entire file so its peak equals `Scale_peak`.	The absolute magnitude of makeup gain is cancelled by the subsequent normalization.

Terminology: the GUI retains the word “Limiter” for compatibility, but Output mode 2 is not a dynamic limiter. It is a whole-file, attenuation-only peak ceiling. The Fast Compressor RMS and NUKE presets are likewise RMS compressors, not sample-accurate peak limiters.

The default target is 0.99. The script rejects targets ≤0 or >1. A fully silent result is left silent in Normalize mode rather than being passed to Scale peak.

Visualization

When Draw result is enabled, the Picture window shows:

Transfer curve: input dB vs output dB, including threshold and knee region. This is the static compression curve before makeup and output scaling.
Waveform comparison: source in gray and final processed output in green. The amplitude range is shared and is chosen from both input and output extrema.
Metering: input/output peak and RMS plus maximum compression-only gain reduction and makeup value.
Gain-reduction timeline: compression gain in dB with makeup subtracted from the display, so 0 dB means no compression and negative values show reduction.
Summary: threshold, ratio, knee, attack, release, sidechain selection, makeup, gain reduction, peak, and RMS values.

The waveform and gain-reduction panels use the Sound's actual xmin/xmax; they do not assume the file starts at 0 seconds.

Notes and limitations

The detector is RMS-style and frame-based; it does not guarantee sample-peak containment.
Praat Intensity is used as the detector analysis engine, so Rms_window_ms controls it indirectly through the script's 3.2 / window mapping rather than creating a literal rectangular RMS window.
Sidechain filters affect detection only; the source audio itself is not EQ'd.
Zero-phase sidechain filtering is offline/acausal and may produce slight pre-transient gain movement.
All channels share one linked gain curve; there is no independent per-channel compression.
There is no dedicated look-ahead buffer, true-peak detector, brick-wall limiter, multiband split, external sidechain input, or dry/wet mix.
Threshold has no fixed form range. Ratio must be ≥1 and knee ≥0.