Auto ducking: how to lower music under speech automatically
Auto ducking lowers music while someone speaks and lifts it between lines. How deep and how fast to duck, and how to stop it pumping between words.
hopeengineer · Founder of Sharpy and Threadify18 October 2026 · 7 min read
The short answer
Auto ducking turns background music down automatically whenever someone speaks and brings it back up when they stop, so the voice stays clear without you drawing every dip by hand. Duck by about 6 dB, ramp over about a quarter of a second, and hold the music down through the short pauses inside a sentence so it does not pump between words. Sharpy's duck follows the voice track itself, looks ahead so the music is already down when a line starts, and takes how long to hold from the speaker's own pauses, all on your Mac.
What is auto ducking?
Auto ducking is music that lowers itself under a voice. The editor sets it once, and the music level follows the speech for the whole piece: down while anyone talks, up in the gaps between lines. It replaces a job editors used to do by hand, drawing a dip in the music's level line under every sentence.
Audio ducking is not the same as setting the music quiet. A bed that sits low all the way through sounds timid in the gaps and still fights the voice in a busy passage. A duck lets the music fill the pauses and step back for the words.
How does ducking in video editing work?
It listens to one track and moves another. The classic tool is sidechain ducking: a compressor on the music track whose trigger, the sidechain, is fed from the voice. When the voice rises past a threshold, the music is turned down by a set amount; when the voice stops, the music comes back. Five settings shape it:
| Setting | What it does | A good starting point |
|---|---|---|
| Threshold | How loud the voice must be to trigger the duck | Just under the voice's quiet words, over the room noise |
| Depth | How far the music goes down | 6 dB, Sharpy's default |
| Attack | How fast it goes down | About a quarter of a second, started early so it is down before the first word |
| Hold | How long it stays down after the voice stops | Long enough to cover the pauses inside a sentence |
| Release | How fast it comes back up | About a quarter of a second |
Sharpy's defaults are a 6 dB duck with 250-millisecond ramps. The hold is the setting most tools get wrong.
Why does simple ducking pump between words?
Because the hold is too short for the way people talk. Speech is full of short gaps: between words, at commas, for breath. A duck that lets go in each of them brings the music up and down several times a sentence, and the bed lurches in and out under the voice. Mixers call it pumping, and it sounds like a mistake.
The fix is to hold the music down through the pauses inside a sentence and let it up only between sentences. But no single hold suits everyone: across Sharpy's test recordings, the pause after a sentence ran from 110 milliseconds on a fast-paced reel to 480 to 780 milliseconds on four others. A hold set for one speaker pumps on another.
How do I make music quieter when someone talks?
Give the music a duck keyed to the voice. In Sharpy there are three ways, from quickest to most precise:
- One click in the window. Drop the music in and use the window's bed ride button. It writes a 6 dB ride under the lines and back up between them, and the step's name says what it followed, for example "rides 6 dB under the words on the picture, in 1 place(s)". When the edit carries its words, the ride follows them to the word; when it does not, it follows the other sound playing underneath.
- A live duck from the agent. Ask for the music to duck under the voice track (
edit_trackwithduck) and give a threshold the voice sits over. The Mixer strip then names what the track ducks under. This duck follows the voice track's actual sound, so re-cutting the voice needs no rewriting. - By hand on the level line. Every sound clip carries a level line: drag a dot to ride it, and ⌘-click to add or remove a point. The line is drawn so that −6 dB sits halfway down, and the level moves in decibels between points, so a two-point dip sounds like one movement.
A duck under no track, under itself, or under a track that is itself ducked is refused, so you cannot build a loop by accident.
What does Sharpy's duck do differently?
It is timed to the speaker rather than to a fixed number:
- It looks ahead. The duck starts early by its attack time, so the music is fully down as a line begins, instead of the first word arriving over full music.
- It holds by the speaker's own pauses. Leave the hold empty and Sharpy measures it from the voice track's words, from that speaker's pauses at commas. The music stays down through a sentence and comes up between sentences, not between words.
- It stays with the edit. Because it listens to the voice track, cut, move or retime the voice and the duck follows.
On a test recording the hold found was 250 milliseconds. While the speaker talked, the bed sat at a median of −6.00 dB, within half a decibel in 369 of the 400 fifty-millisecond stretches with speech, and it stayed down through all 16 of the speaker's gaps. Audio mixing for video puts the duck in the context of the whole mix.
How much should music duck under a voice?
Start at 6 dB and trust your ears on a phone speaker. The duck is a dip on top of a bed that already sits well under the voice; it is not the whole difference between them. If words still get lost, lower the whole bed rather than deepen the duck, which only makes the music's return more obvious.
For voice-led audio there is a firm reference: the W3C's accessibility guidelines (WCAG success criterion 1.4.7, level AAA) ask that background sound in a recording made mostly of speech sit at least 20 dB below the speech. It is written for audio-only recordings such as podcasts, not video, but it is a good bar for any piece where every word matters.
How do other editors on the Mac handle ducking?
Most offer a version of it. Apple's iMovie has a "Lower volume of other clips" checkbox in a clip's volume controls, with a slider for how far the other clips go down while that clip plays. Final Cut Pro's guide describes ducking by marking a stretch with the Range Selection tool and lowering the level inside it, which adds the keyframes for you. Both lower the music over a range you choose; neither follows the speaker's pauses.
For a version in another language, ask Sharpy for undipped stems as well: each ducked part is written again at a steady level, the bed's dipped stem sitting 6 dB under its undipped one while the voice speaks, so a mixer can duck it again under a new voice.
Frequently asked questions
01What is auto ducking?
It is music that turns itself down while someone speaks and back up when they stop, keyed to the voice track, so you do not draw each dip by hand.
02What is sidechain ducking?
A compressor on the music whose trigger comes from another track, usually the voice. The voice sets when the music goes down; the compressor's settings set how far and how fast.
03Why does my music go up and down between words?
The hold is too short, so the duck lets go in every small gap. Lengthen it until the music stays down through a sentence, or let Sharpy measure it from the speaker's pauses.
04Does ducking work with more than one speaker?
Yes. The window's bed ride follows the edit's words, whoever says them, and a duck keyed to a voice track follows whoever speaks on that track.
05Can I duck sound effects too?
Yes, the same way: duck the effects track under the voice track. Ambience usually needs less ducking than music, because it carries no tune to compete with the words.
Sources
Checked on 18 October 2026.
Written by
hopeengineer
Software architect and founder of Sharpy and Threadify. Builds scalable systems and AI tools, and writes here about how Sharpy understands and edits video.
Sharpy opens on 1 November 2026.
The world's first context-aware agentic video editor, on your Mac. One plan, unlimited edits.