Vintage Glue Compressor — User Guide

Channel-linked RMS-style compression with a standard soft knee, causal attack/release, optional nonlinear saturation, dry/wet mixing, and explicit output-level control.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 1.3.1 (2026) License: MIT License Repo: https://github.com/ShaiCohen-ops/Praat-plugin_AudioTools

Contents:

What this does Quick start Compression model Saturation models Presets Parameters Output & visualization

What this does

Vintage Glue Compressor is a feed-forward dynamics processor with a shared power-based sidechain. The selected Sound is measured at the audio sample rate, converted to a causal RMS-style level estimate in dBFS, smoothed with independent attack and release time constants, and passed through a hard- or soft-knee compressor curve. The resulting gain is applied uniformly to every channel. An optional nonlinear stage can then add Tube, Tape, Transistor, or FET-style coloration before dry/wet mixing and the final output-level stage.

Channel linking: the detector squares each channel first and then averages channel power. It does not downmix amplitudes before detection, so anti-phase stereo material cannot cancel in the sidechain. One gain curve is then applied to all channels, preserving inter-channel gain relationships.

Signal flow:

Input Sound
→ per-channel square → mean channel power
→ causal RMS window → amplitude → absolute dBFS
→ causal attack/release smoothing
→ soft-knee gain computer → makeup gain
→ gain applied to all channels
→ optional saturation and Harmonics mix
→ Dry/Wet mix with the original
→ optional safety ceiling / peak normalization
→ output Sound

About the “vintage” names: the presets are practical style presets, not circuit-level emulations of specific hardware. Likewise, Tube, Tape, Transistor, and FET are simplified nonlinear transfer functions chosen for different symmetric/asymmetric saturation characters.

Quick start

Select exactly one Sound in Praat.
Run Vintage_Glue_Compressor.praat.
Choose Custom or one of the six named presets.
For Custom, set the RMS threshold, ratio, knee, attack/release, saturation type, Drive, Harmonics mix, makeup, and Dry/Wet mix.
Enable Advanced_settings only when you need detector-window, saturation-oversampling, output-level, statistics, or source-retention controls.
Click OK. The processed Sound remains selected; playback is automatic when Play_result is on.

Useful starting point: the default Custom settings are threshold −15 dBFS RMS, 4:1, 6 dB knee, 10 ms attack, 100 ms release, saturation Off, Auto makeup On, and 100% wet. The default advanced output policy is an attenuate-only peak ceiling at 0.99.

Compression model

1. Power-linked RMS detector

For a multichannel Sound, each channel is squared before the channels are combined. The detector therefore operates on mean channel power:

p[n] = mean_c( x_c[n]^2 )

A causal one-pole smoother then estimates average power. With Rms_window_ms = W:

α = exp(−1 / (Fs · W))
p̄[n] = α p̄[n−1] + (1−α) p[n]
L[n] = 20 log10( sqrt(max(p̄[n],0)) + 10⁻¹⁰ )

L[n] is an absolute dBFS level estimate; the threshold does not move with the peak or crest factor of the loaded file.

2. Attack and release

The dB level is smoothed causally at the audio rate. A rising level uses Attack_ms; a falling level uses Release_ms. These time constants are separate from Rms_window_ms, so detector averaging and compressor ballistics are independently controllable.

3. Hard / soft knee

Below the knee the gain is unity. Above the knee, the output slope is 1/Ratio. When Knee_dB > 0, the transition uses the standard quadratic soft-knee form:

GRdB = −(1 − 1/R) · (L − T + K/2)² / (2K) inside the knee

For Knee_dB = 0, the compressor becomes a hard-knee design.

4. Makeup

Makeup_Gain_dB is folded directly into the gain curve. With Auto_makeup enabled, the script uses a simple heuristic: half of the gain reduction that a signal 10 dB above threshold would receive. This is a convenience estimate, not loudness matching.

Saturation models

The nonlinear stage comes after compression and makeup. Drive is converted to drive_amt = 1 + 3·Drive. The selected nonlinear signal is blended with the clean compressed signal using Harmonics_mix.

Mode	Implemented transfer	Character
Off (Clean Digital)	No nonlinear stage	Compression only.
Tube	`tanh(x·d) / tanh(d)`	Symmetric soft saturation. Because the transfer is odd-symmetric, a symmetric input primarily produces odd-order nonlinear products.
Tape	Asymmetric tanh: stronger positive than negative drive	Asymmetry introduces a mixture of even- and odd-order products.
Transistor	Bounded normalized cubic soft clip	Symmetric, monotonic, C1 transition into ±1 clipping.
FET	Asymmetric tanh using `x + 0.1·x·\|x\|`	Punchier asymmetric saturation.

Drive = 0 is not a saturation bypass when a nonlinear model is selected. It gives the minimum model drive (d = 1). For truly clean processing choose Off (Clean Digital), or set Harmonics_mix = 0.

Harmonics mix

y = Harmonics_mix · y_sat + (1 − Harmonics_mix) · y_compressed

Harmonics_mix = 0 is the clean compressed signal; 1 is the fully nonlinear signal.

Oversampling

The Advanced dialog offers Off (source rate), 2x, or 4x saturation oversampling. When enabled, only the nonlinear stage is processed at the higher rate and then resampled back. With oversampling Off, high-order components above Nyquist can alias into the audible band, especially at high drive.

Saturation intensity index: when statistics are enabled, the script reports a setting-derived index based only on saturation type, Drive, and Harmonics mix. It is an arbitrary comparison scale and not a THD measurement.

Presets

The named presets replace the compressor and saturation character settings shown below. They do not change Dry/Wet mix, saturation oversampling, output-level policy, ceiling, statistics, or Keep original.

Preset	Threshold	Ratio / Knee	RMS window	Attack / Release	Saturation	Drive / Mix	Makeup
Vocal Opto	−24 dBFS	3:1 / 12 dB	20 ms	10 / 500 ms	Tube	0.25 / 0.40	+4 dB
Drum VCA	−18 dBFS	8:1 / 3 dB	3 ms	1 / 50 ms	Transistor	0.40 / 0.50	+4 dB
Mix Bus Glue	−12 dBFS	2:1 / 10 dB	15 ms	30 / 200 ms	Tube	0.15 / 0.30	+2 dB
Tape Squeeze	−15 dBFS	4:1 / 8 dB	10 ms	5 / 100 ms	Tape	0.60 / 0.70	+3 dB
Tube Warmth	−20 dBFS	2.5:1 / 15 dB	20 ms	20 / 300 ms	Tube	0.35 / 0.50	+3 dB
FET Punch	−20 dBFS	12:1 / 0 dB	2 ms	0.5 / 50 ms	FET	0.50 / 0.60	+6 dB

All named presets switch Auto_makeup off and use the fixed makeup value in the table.

Parameters

Main dialog

Parameter	Default	Meaning
Preset	Custom	Custom or one of six named style presets.
Rms_threshold_dBFS	−15.0	Absolute RMS-sidechain threshold in dBFS.
Ratio	4.0	Compression ratio; must be at least 1.
Knee_dB	6.0	Soft-knee width; 0 = hard knee.
Attack_ms	10	Causal attack time constant applied to the dB detector.
Release_ms	100	Causal release time constant applied to the dB detector.
Saturation_type	Off	Clean, Tube, Tape, Transistor, or FET transfer.
Drive	0.3	Nonlinearity drive control. Must be ≥ 0; it is not restricted to 0–1.
Harmonics_mix	0.5	Blend between clean compressed and saturated signal, 0–1.
Makeup_Gain_dB	0.0	Constant gain added to the compressor gain curve.
Auto_makeup	On	Replaces manual makeup with the script's heuristic in Custom mode.
Dry_wet_mix	1.0	0 = original dry Sound, 1 = fully processed path.
Advanced_settings	Off	Opens the second dialog.
Draw_result	On	Draws transfer curves, gain reduction, output waveform, and summary.
Play_result	On	Plays the result after processing.

Advanced dialog

Parameter	Default	Meaning
Rms_window_ms	10 ms	Causal one-pole power averaging time constant, independent of attack/release. Named presets replace this value with their own detector window.
Saturation_oversampling	Off	Source rate, 2x, or 4x for the nonlinear stage.
Output_level_mode	Safety ceiling	None; attenuate-only ceiling; or always peak-normalize to Ceiling_peak.
Ceiling_peak	0.99	Linear sample-peak target, >0 and ≤1.
Show_stats	On	Reports input/output peak and RMS, gain reduction, crest factor, and level-stage details.
Keep_original	On	If Off, removes the source Sound after successful processing.

Very short detector windows: if Rms_window_ms · Fs / 1000 < 4, the script warns that the detector is following individual cycles rather than a stable level estimate. This can turn the gain path into waveform-rate modulation rather than conventional compression.

Output, level policy, and visualization

Dry/Wet

Dry/Wet mixing occurs after compression and saturation:

y_mix = Dry_wet_mix · y_processed + (1 − Dry_wet_mix) · x_original

The output-level stage is applied after this mix. Therefore 0% wet is not necessarily a bit-identical bypass if the selected output-level mode subsequently attenuates or normalizes the file.

Output level modes

Mode	Behavior
None	Leaves the processed sample level unchanged. Peaks above 1 can clip when exported to integer PCM.
Safety ceiling (default)	If the absolute sample peak exceeds Ceiling_peak, attenuates to that peak; otherwise does nothing. Quiet material is never boosted.
Peak normalize	For any non-silent result, applies one constant gain so the absolute sample peak equals Ceiling_peak. This can either attenuate or amplify and therefore changes the absolute level implied by makeup and Dry/Wet.

This is sample-peak management using Praat's absolute extremum, not true-peak limiting.

Output name

The named presets append _Opto, _VCA, _Glue, _Tape, _Tube, or _FET. Custom uses the original object name with no added suffix; if Keep_original remains on, both objects can therefore share the same visible name in the Objects list.

Visualization

With Draw_result on, the Picture window shows:

Saturation transfer curve: selected nonlinear character with the Harmonics mix blended into the plotted curve.
Compression transfer curve: threshold, knee region, ratio, and unity reference in dBFS.
Gain reduction over time: compressor gain reduction before makeup, over the Sound's actual time domain.
Output waveform: mono directly; for multichannel Sounds the first two channels are drawn while all channels remain in the processed output.
Summary: preset, input levels, threshold, maximum gain reduction, makeup, saturation settings, output levels, crest factor, and output-level action.