Skip to content

Lozenge

The Lozenge editor, a tall narrow panel in dusty rose, sage and cream: large Protect and Size dials stacked down the centre line reading 0 and 15.0 ms, a smaller Sens dial and a Listen button below them, and a Removed strip counting repairs along the foot, over a flat cartoon drawing of a mint lozenge spinning inside two big whoosh ribbons that run off the top and bottom of the panel.
ProtectProtects the voice from audible cutsSizeSelect clicks ranging from 0.5 to 15msSensHow quiet peaks are detectedListenPlay only what is being taken outRemovedVisual: live repairs so you can keep track of what is being removed

Get Lozenge — Free

Free, three knobs

Three knobs - Protect, Size, Sens - and 7 presets to start from. Three knobs do the same work as many expensive declickers.

One editing pass fewer

De-clicking is normally done by hand or with expensive specialized software - Lozenge does the same work with one render through it, or live while you record.

Your consonants survive

What puts people off a declicker is it eating their Ts. Lozenge tells them apart automatically.

How Lozenge works

01

Find the click

On paper a mouth click and a hard consonant’s release are the same event: 1.27 milliseconds against 1.38, both around 25 dB under the voice. Separating them means listening closely to what comes after it.

02

Protect, on a knob

How far the audio around a click has to fall, relative to the click itself, before anything is removed. Turned up, less is removed; turned down, it catches more clicks while reaching more into running speech.

03

Gap refilling

What comes out is chosen to minimise the prediction error of the speech either side, so the hole gets the voice that was going to be there.

04

Listen

One button plays what is being removed instead of what is left. Ticks mean it is working; speech means turn Protect up. What it plays and what it keeps add back up to exactly the input.

Numbers

your voice untouched

Specification

Repair
Autoregressive interpolation, order 32, banded solve
Detection
Band-limited transient, gated on what follows it
Latency
70.3 ms, and it does not move when Size moves
Range
Size 0.5 to 15 ms; Sens 0 is a true bypass
Presets
7
Formats
VST3, CLAP, AU, LV2
Systems
macOS (Apple Silicon and Intel), Windows x64, Linux x86_64

Measured

Sens at 0, against the input
0 samples differ
Held vowel, 3.4 s of steady voicing
0 samples differ
Same passage, any buffer size
the same result
Speech at default settings
32.3 dB SNR
Samples changed while the voice is loud
0.015% of the file
Samples changed in the gaps
0.45% of the file
Synthesised stop release
0 samples differ
Sustained fricative
0 samples differ
Glottal-style pulse train
0 removals per second
Injected clicks, held-out readers, 20 files
55% found, 2 false fires a minute
Kept plus what Listen plays
the input, exactly
CPU, stereo at 44.1 kHz
0.37% of one core, 270× realtime

Presets

Voice - default · Careful - keep every T · Sung vocal · Audiobook - close mic · Watery clicks · Listen to what goes · Off - straight through

Get it

Lozenge — Free

More plugins

They are built the same way

The Excursion editor: ten emitters on the left, a cone-pressure diagram and playback level in the centre, the signal path on the right, and a telemetry row reading low corner, gain reduction and level match.

Excursion

Plays your track through a speaker too small for it.

€39/VST3 · CLAP · AU

The Lossy editor: encoder, spectrum and image, and transport panels above a band-allocation strip showing which frequencies are kept in stereo, flattened to mono, or discarded.

Lossy

Audio compression you can starve until it breaks.

€29/VST3 · CLAP · AU

The Inkey editor: a root keybed, scale and Scala import on the left with Auto Fit reporting D# Minor at minus three semitones, a pitch map of cents off key against frequency on the right, and correction, detection and output controls below.

Inkey

Retunes a sample into your key while it plays, chords intact.

€39/VST3 · CLAP · AU

The Negative editor: the Inside Out preset at Amount 100, a sixteen-band mirror with input level above the centre line and the reflected output below it, and Speed, Depth, Floor, Mix and Output along the bottom.

Negative

Inverts a mix's dynamics band by band, so what sat between the hits becomes the loudest thing in it.

Free/VST3 · CLAP · AU · LV2

The Downstream editor: a decibel axis with the master at −9.2 LUFS, the −14 target below it, a −4.8 dB arrow between the two and 16.1 dB of peak left marked above; the Delivered knob at 100, and the measured integrated loudness, true peak, range and peak headroom listed beside it.

Downstream

Applies the loudness a delivery target implies, so you hear what the listener gets instead of a number.

Free/VST3 · CLAP · AU · LV2

Lozenge is free and unlicensed: it asks for no address, creates no account, and reports nothing to any server. Audio is processed on your machine and never leaves it. It is an audio repair tool: nothing about it is a medical device, a diagnosis, or advice about anyone’s voice or health. No product, company or service is named, matched to, or compared with anywhere in this plugin. The repair method is described in published literature from 1986 and implemented here from that description rather than from anyone’s source.