Dialogue–Music Ducking in Film Mixing: Carving Space Automatically with Sidechain Compression

On a series mix, after getting the note 'the music is beautiful but I can't hear the line' from the director three episodes in a row, I stopped fighting with fader automation. Dropping the music 3 dB on every line and bringing it back after each sentence meant hundreds of automation points in a 45-minute episode — and every time the director changed the cut, all of them drifted. The fix was a technique broadcast radio has used for decades: ducking. The music bus runs through a compressor that 'listens' to the dialogue bus; when dialogue starts, the music pulls back on its own, and when it stops, the music returns. Tuned properly, the audience doesn't hear an effect — they hear a well-balanced mix.

The Sidechain Logic: Who Is the Compressor Listening To?

A normal compressor looks at the level of the signal passing through it and clamps that same signal once it crosses the threshold. A sidechain (Pro Tools calls it 'Key Input') breaks that link: the compressor still clamps the music, but it takes its trigger from a different signal — in our case the dialogue bus. So no matter how loud the music is, if there's no dialogue the compressor does nothing. When dialogue arrives, the music ducks in proportion to the dialogue level. It's the 'turn the music down' automation move, made intelligent and self-adjusting to the cut.

Setting It Up in Pro Tools

• ―Make sure every dialogue track sums into a single Aux (DX Bus). You'll key the ducking off this bus, not off individual tracks.
• ―Add a send on the DX Bus (e.g. Bus 31–32, named 'DX Key'). Set it pre-fader so the trigger stays constant even if you later ride the dialogue fader down. Send level at 0 dB.
• ―Insert a compressor on the music bus (MX Bus) — the stock Avid Dyn3 Compressor/Limiter works, but a plug-in with sidechain filtering like Pro Compressor or FabFilter Pro-C 2 gives cleaner results.
• ―From the compressor's Key Input menu, pick the 'DX Key' bus. Make sure the compressor's 'External' or 'Side-Chain' switch is engaged; in Dyn3 that's the key icon at the top of the plug-in window.
• ―Loop a scene with dialogue and watch the gain-reduction meter: the needle should move on each line and sit at zero in the silences.

The Settings: A Recipe for Staying Invisible

For ducking to be a balance tool rather than a musical effect (the EDM pump), the settings need to be gentle. My starting point: ratio between 2:1 and 3:1 — never above 4:1. Set the threshold so that even the quietest normal line triggers 1–2 dB of reduction, and shouted lines reach 4–6 dB. Attack 20–50 ms: too fast and the music 'hiccups' on every syllable, too slow and the first word of a sentence gets buried. Release is the most critical parameter: somewhere around 400–800 ms stops the music from leaping back during the natural breath between two sentences. If the plug-in has a Hold control, 150–250 ms of hold removes flutter on the short pauses inside a sentence.

Keep the knee soft. A hard knee creates an audible 'step' in the music on every line that hovers near the threshold; a soft knee makes the transition gradual so the ear can't track it. Leave make-up gain at zero — the goal of ducking isn't to preserve overall level, it's to open space during dialogue; adding make-up gain just makes the music louder in the quiet stretches for no reason.

The Sidechain Filter: Cleaning Up the Trigger

The dialogue bus carries more than voice: breaths, clothing rustle, whatever room hum survived from set, and sometimes low-frequency 'thumps' from the edit. All of these can trigger the compressor, so the music flinches even when nobody is speaking. The fix is a band-pass filter on the key signal. Pro Compressor and Pro-C 2 have built-in sidechain EQ for this; if you're on Dyn3, route the 'DX Key' bus to an Aux and put an EQ III in front of it. My setting: cut below 200 Hz and above 6 kHz. This band keeps the intelligibility region of speech (roughly 1–4 kHz) while removing low rumble and sibilance as triggers. The filter changes only what the compressor 'hears' — it doesn't touch the dialogue itself.

When to Duck, When to Automate

Ducking is a safety net, not an artistic decision. In scenes where the music swells dramatically and the line is deliberately buried inside it — a farewell, a montage — I bypass the compressor and write automation by hand, because there the answer to 'how many dB' is emotional, not technical. Ducking works best on dialogue scenes with continuous underscore, documentary narration and episodic work: situations where the music is always 'there' but must never step forward. In practice I use both: ducking keeps the rough balance automatically, and I only touch the music bus fader by hand at the emotional peaks of a scene. That turns hundreds of automation points into ten or fifteen deliberate moves per episode.

Three Traps to Avoid

• ―Making the music's return audible. If release is too short, the music swells like a breath after every sentence — viewers notice and find it distracting. When in doubt, lengthen release; even a full second sounds natural in most scenes.
• ―Ducking the effects too. If you try to duck the SFX bus alongside the music, sync sounds like a door slam lose their weight whenever they coincide with a line, and the scene feels fake. For SFX, use EQ and manual automation, not ducking.
• ―Forgetting the delivery format. If you deliver stems (DX/MX/FX separately), the music stem will carry the ducking baked in — the recombined stems sound identical to the mix, but the music stem on its own will look wavy. Telling the client up front prevents the 'the music stem is broken' email.

The test for ducking is simple: if someone watching the scene doesn't say 'the music keeps dipping', it's working. You should see the compressor move but never hear it. Once a mixer has this set up right, the next project starts from the same template, the threshold shifts 2 dB for the new dialogue level, and the time that used to go into dialogue–music balance goes where it matters — the scenes that actually need a human decision.