Tutorials Audio

Mixing Music Under Dialogue: Ducking and Sidechain Compression

By Infinite Creation · Intermediate · ~25 min

Some links on this page are Amazon affiliate links. If you buy through them, we earn a small commission at no extra cost to you.

A 1960s mid-century comic book style editorial illustration showing two audio engineers at a mixing desk, with a mixing desk, a studio microphone on a boom arm, an audio waveform display visible in the frame.

Overview

Background music fighting with dialogue for space is the single most common mix problem in podcast and video production. Sidechain compression, automatically lowering the music's volume whenever dialogue is present, is the professional solution, and it's available in every major DAW for free.

What You Need

A DAW with sidechain-capable compression (all major DAWs support this)
Separate dialogue and music tracks

Steps

Route dialogue as your sidechain trigger

Sidechain compression uses one track's signal (dialogue) to control processing on a different track (music). Confirm your DAW's routing so the dialogue track can act as an external trigger, usually via a send or a sidechain input selector on the compressor.

Add a compressor to your music track

Insert a compressor (most DAWs' stock compressor supports sidechain input) on the music track. This is the track that will actually get quieter.

Set the sidechain input

In the compressor's sidechain/external input settings, select your dialogue track (or its send bus) as the trigger source. The compressor will now react to dialogue level, not the music's own level.

Dial in threshold, ratio, attack, and release

Set the threshold so it only triggers on actual speech, ratio around 4:1-8:1, a fast attack (1-10ms) so the duck starts immediately with speech, and a release around 200-400ms so the music doesn't pump obviously back up between words.

Automate manually where sidechaining overreacts

For a handful of moments where automatic ducking sounds unnatural (a musical swell you want to protect, a brief pause you don't want the music jumping back up during), draw manual volume automation over the sidechained result.

Pro Tips

A release time that's too fast causes audible "pumping" (the music visibly breathing with every word). Too slow and the music stays ducked through short pauses. Start around 250ms and adjust by ear.
Duck by 6-10 dB as a starting point for most podcast/video mixes, enough to clear space for dialogue without music disappearing entirely.
If your DAW doesn't support sidechaining well, manual volume automation on the music track achieves the same result with more manual effort but total control.

Tested: what the release time does to a duck

We ran a sidechain duck through the sidechaincompress filter in FFmpeg 8.1.2 on 2026-09-23 and read the music's level every millisecond. A steady tone at -18 dBFS stood in for the music. Bursts of noise at -12 dBFS stood in for speech: four 300 ms words with 100 ms gaps, a one-second pause, then four more words. Threshold -22 dBFS, ratio 4:1 and attack 5 ms stayed fixed while the release changed.

Release	Music during a word	Music in a 100 ms gap between words	Music halfway through a 1 s pause	Back within 1 dB after speech ends
50 ms	-9.7 dB	0.0 dB	0.0 dB	32 ms
100 ms	-10.1 dB	-0.1 dB	0.0 dB	70 ms
250 ms	-10.5 dB	-5.4 dB	0.0 dB	190 ms
500 ms	-10.6 dB	-8.1 dB	0.0 dB	391 ms
1000 ms	-10.8 dB	-9.5 dB	-4.1 dB	797 ms

Every setting ducked the music by about 10 dB while a word played. The release decided what happened in between. At 50 and 100 ms the music came fully back up in every gap between words, which is the pumping step 4 warns about. At 250 ms, the middle of the range this page recommends, it still rose from -10.5 to -5.4 dB in each gap. At 500 ms it held within 2.6 dB of the full duck between words and still came fully back during the one-second pause. At 1000 ms the music was 4.1 dB down halfway through that pause, the other fault the Pro Tips describe.

Attack mattered less. Ten milliseconds into the first word the music was 10.7 dB down with a 1 ms attack, 10.0 dB with 5 ms, 8.4 dB with FFmpeg's default of 20 ms and 6.6 dB with 50 ms, so a slow attack lets the first syllable of each phrase land on music still near full level.

For the depth, the threshold did more than the ratio. Doubling the ratio from 4:1 to 8:1 deepened the duck from -10.5 to -12.2 dB. Moving the threshold from -20 to -24 dBFS took it from -9.0 to -12.0 dB. Set the depth with the threshold first.

Real speech has gaps of every length, and each compressor defines its release a little differently, so the same number will not behave identically in your DAW. Use the table for the direction: if the music breathes between words, lengthen the release until it stops.

The 500 ms setting, in FFmpeg, with the voice as the second input: ffmpeg -i music.wav -i voice.wav -filter_complex "[1:a]asplit[sc][v];[0:a][sc]sidechaincompress=threshold=0.079:ratio=4:attack=5:release=500[m];[m][v]amix=inputs=2:normalize=0" mix.wav

Validated. Measured 2026-09-23 with FFmpeg 8.1.2 on macOS, Apple silicon. Synthetic signals, a tone and noise bursts, not a real voice and music. Gain read over a 9 ms window. A threshold of -22 dBFS is 0.079 in FFmpeg's linear units.

Knowledge Base

Why Ducking Beats a Static Volume Reduction

Simply lowering music volume for an entire segment either leaves it too loud during dialogue or too quiet during instrumental moments. There's no single static level that works for both. Sidechain ducking responds dynamically, staying appropriately loud when there's no dialogue and automatically dropping the instant speech begins.

How Sidechain Compression Actually Works

A normal compressor reduces a track's own level based on its own signal. A sidechained compressor reduces one track's level based on a different track's signal entirely. The music compressor "listens" to the dialogue track and reacts to it, rather than to the music itself.

Getting a Natural-Sounding Duck

The difference between an obvious, distracting duck and an invisible one is almost entirely attack and release timing. Too fast a release creates audible pumping. Too slow leaves music oddly quiet during pauses. Small adjustments here matter more than the exact threshold or ratio values.

Where This Fits

Ducking solves one balance problem, and Mixing a Full Song covers balance across a whole mix.

FAQ

Q: Do I need a paid plugin for sidechain compression?
A: No. Every major DAW's stock compressor (Logic, Ableton, Pro Tools, Premiere's Essential Sound panel, REAPER's ReaComp) supports sidechain input at no extra cost.

Q: Should I sidechain sound effects too, not just music?
A: Often yes for continuous ambient beds (traffic, room tone) that would otherwise compete with dialogue the same way music does, short, punchy sound effects usually don't need it.

Vocabulary

Ducking. Automatically reducing one signal's level when another is present, most commonly lowering music while someone speaks. Keeps speech intelligible without anybody riding a fader manually.

Sidechain input. The signal a processor listens to when deciding what to do, separate from the signal it processes. Feeding speech into the sidechain of a compressor on music is what produces ducking.

Intelligibility. How readily speech can be understood, which depends on the gap between voice and background rather than on absolute level. A few decibels of separation matters more than making anything louder.

Music bed level. How far below speech the underlying music sits. Consistency across a programme matters more than the exact figure, because a bed that varies draws attention to itself.