Leveller

The leveller evens out a speaker who moves on and off the microphone. It turns each phrase up or down toward a target, working from the whole clip ahead of time, so a phrase starts at its own level and doesn’t catch up after it begins. To tame sudden loud moments use the Compressor, and to bring a whole clip to a delivery loudness use Normalize.

Where: Clip tab ▸ Sound ▸ Leveller
Needs: The clip's loudness and its speech, both measured on your Mac with no key and no upload.
Agent tools: set_audio

Ride the level

Select the clip and open the Sound sub-tab of the Clip tab. The Leveller section starts closed, so open it.
Tick Ride the level. It turns on at the defaults.
Read Status. If it says Aims at and a level, the clip is being levelled. If it says anything else, it’s waiting or it can’t work on this clip.
Play the clip, and adjust the controls if a phrase is still too quiet or too loud.

The section’s header shows the target, such as “0 dB”, or “off”. Reset appears there once the leveller has settings, and turns it off. Unticking Ride the level also throws away its settings, so ticking it again starts from the defaults.

How it levels

The leveller looks at the clip’s speech and splits it into phrases. A pause in the speech of at least half a second ends a phrase, and a shorter gap stays inside it.

In a phrase, it measures the level the way a loudness meter does, over about three seconds of speech at a time. It never reaches across a pause into the next phrase.
In a pause, it holds the gain it had and doesn’t follow the room. In the last quarter of a second of the pause it moves to the next phrase’s level.
A phrase with under a second of speech takes the level of the last longer phrase before it, or, when there’s none, the first one after it. A short “yes” doesn’t get a level of its own.

The controls

Control	What it does	Range	Default	Quick picks
Target	Where it aims.	−20 to 10 dB	0 dB	−6, −3, 0, 3
Max Boost	The most it turns the clip up.	0 to 24 dB	12 dB	6, 12, 18
Max Cut	The most it turns the clip down.	0 to 24 dB	12 dB	3, 6, 12
Response	How smoothly it follows the voice within a phrase.	0.5 to 10 s	2 s	1, 2, 4 s

A number typed outside the range goes to the nearest end of it.

Target

The target is in dB from the clip’s own loudness, not an absolute level. At 0 dB each phrase aims at the clip’s loudness, and at −3 dB it aims 3 dB under that. Because it follows the clip’s loudness, a trim that changes what plays moves the level it aims at.

Max Boost

Caps how far a quiet phrase is turned up. Lower it if a faint background voice or a distant mumble is being pulled up too far. At 0 dB the leveller never turns anything up.

Max Cut

Caps how far a loud phrase is turned down. At 0 dB the leveller never turns anything down.

Response

Changes in a phrase’s level shorter than this are smoothed out, before and after alike. A longer Response follows the voice more slowly, and a shorter one follows it more closely.

Status and Loudness readouts

Status says where the leveller is aiming, what it’s waiting for, or why it isn’t levelling.

Status says	What it means
“Aims at −41.7 LUFS”	It’s levelling, toward that level: the clip’s loudness plus the Target.
“Waits for its loudness”	The clip’s sound analysis hasn’t finished.
“Waits for its speech”	Speech in the clip hasn’t been found yet.
“Not levelling: it plays no sound in this timeline”	The clip has nothing to measure here.
“Not levelling: it plays for under 0.4 s”	The clip is too short to measure.
“Not levelling: it is too quiet to measure”	The clip has no measurable loudness.
“Not levelling: it has no phrase with 1 s of speech”	No phrase is long enough to measure.
“Not levelling: speech detection is unavailable”	Speech detection isn’t available, and the media has no saved speech.
“off”	Ride the level isn’t ticked.
“—”	Several clips are selected, so select one to read it, or the clip is a nest, where the leveller has no effect.

Loudness shows the clip’s loudness, which the target is measured from, such as “−41.7 LUFS”. LUFS is the unit of loudness. It reads “measuring…” while that’s being worked out, and “—” with several clips selected or when there’s no reading. It’s measured over the part of the clip that plays, before the clip’s volume, so moving the Volume slider doesn’t change it.

Change the leveller over time

With an action selected, the controls write the action’s end. Target, Max Boost and Max Cut move smoothly across the action. Response changes at the action’s end, and so does turning the leveller on or off.

Good to know

The leveller works from the speech found by the clip’s sound analysis, or the speech saved for Duck by Speech.
Until the clip’s loudness and speech are both measured, it plays without levelling. An export waits for them for up to five minutes, then stops with a message naming the clip. Wait for Status to show the clip’s loudness and export again.
A clip it can’t level, for one of the “Not levelling” reasons above, plays and exports without it.
It works the same in the preview and in an export, from wherever playback starts.
It runs after the Gate and before the De-esser and the Compressor, which then get the levelled voice.
Normalize in Loudness mode measures through it.
Enhance Voice turns the leveller on at its defaults when the speech’s loudness varies a lot, and off when it doesn’t.
On a nest clip the leveller has no effect. Open the nest with ⌘↓ and set it on the clips inside.

With an agent

set_audio sets the same controls through its leveller argument. {} turns the leveller on at its defaults, and null turns it off. A value outside its range is an error that changes nothing. See Tools: clip sound.