What Is Sidechain Ducking?
Sidechain ducking (audio ducking) automatically lowers one sound, usually music, while another sound, usually a voice, is playing, then brings it back up in the pauses. A compressor on the music listens to the voice through its sidechain (key) input instead of to the music itself, so the voice decides when the music dips and by how much.
It’s how podcasts, vlogs and ads keep background music full between lines and out of the way under them. Because a sidechain reacts to level, not language, any loud sound in the key (a laugh, a door, a crowd) dips the music too.
How Sidechain Ducking Works
The voice, or whatever should take priority. Only its level is used.
Threshold, ratio, attack and release turn that level into gain reduction.
The reduction is applied to the music, which dips and recovers.
In a music app you insert a compressor on the music track and route the narration into its sidechain input, which is the ducking setup Ableton’s Compressor reference describes. In a video pipeline, FFmpeg’s sidechaincompress filter takes the same two inputs: the stream to compress and the stream that controls it.
Threshold, Ratio, Attack, Release, Hold and Lookahead
| Setting | What it does | For voice over music |
|---|---|---|
| Threshold | The key level where ducking begins. | Set it so speech crosses it and room tone doesn’t. |
| Ratio | How hard the music is pushed down for each dB the key rises above the threshold. | 4:1 gives a moderate dip. At 10:1 and above, the music drops almost as many dB as the voice rises over the threshold. |
| Attack | How fast the dip happens once the key crosses the threshold. | Fast attack catches the first syllable; slower attack sounds smoother but lets the start of a word through. |
| Release | How fast the music returns after the key falls away. | Too short and the music pumps up between words; too long and it stays down through real pauses. A few hundred milliseconds is the usual zone. |
| Hold | Keeps the dip in place for a set time before the release starts. | Stops the music bobbing up in the tiny gaps between words. |
| Lookahead | Delays the signal a few milliseconds so the compressor reacts before the sound arrives. | Catches the first syllable without a harsh, very fast attack. |
| Knee | How gradually compression starts around the threshold. | A soft knee lets the dip ease in instead of switching on. |
Not every compressor has every control. Ableton’s Compressor offers sidechain routing and a lookahead of 0, 1 or 10 ms; FFmpeg’s sidechaincompress has threshold, ratio, attack, release, knee and makeup gain, but no hold or lookahead (both checked October 6, 2026). Apple documents Logic Pro’s version in its Compressor guide.
How Deep to Duck Music Under a Voice
| Depth | What you hear | Use it for |
|---|---|---|
| Light: 3–6 dB | Music stays present and gives way a little. | Music-led edits with occasional lines. |
| Medium: 8–12 dB | Music sits clearly behind the voice. | Most narration, tutorials and vlogs. Valmera’s step duck and voiceover duck use 12 dB. |
| Deep: 15 dB or more | Music nearly disappears under words and swells back in the pauses. | Dense talking, quiet voices, or tracks with vocals. |
Depth adds to the bed’s resting level: a bed that already sits low needs less dip, and a loud one needs more. The real test is a phone speaker. If any word is hard to follow there, duck deeper or lower the bed.
Work Out the Depth
Once the dip has settled, a hard-knee compressor reduces the music by the amount the key sits above the threshold times (1 − 1/ratio). With the key at −18 dBFS and the threshold at −30 dBFS, the key is 12 dB over. At 4:1, 3 dB of that excess remains, so the music drops 9 dB; at 12:1 it drops 11 dB. A quieter key produces less reduction at the same settings.
An illustrative, settled hard-knee model with no makeup gain. It ignores attack, release, knee smoothing and detector behavior. These sliders explain the arithmetic; they do not change Valmera settings or predict its rendered mix.
Modeled gain reduction:
12.0 dB above threshold × (1 − 1/4) = 9.0 dB of attenuation. 35.5% of the music's pre-duck amplitude remains; this is not a perceived-loudness percentage.
Put the key at or below the threshold, or the ratio at 1:1, and the reduction is zero. Real compressors add attack, release and knee on top of this arithmetic, so the dip builds and fades rather than switching.
Hear It
Two 12-second renders from Valmera’s renderer: a steady synthesized chord as the music, and a higher test tone as the key at 0:02–0:04 and 0:06–0:08. With smooth ducking on (B), the chord drops about 14 dB while the tone plays and returns to full level within a second after it stops. A is the same mix with ducking off.
Download the measured results (JSON) or the steps to reproduce them (README).
Sidechain vs. Keyframe Ducking
| Approach | What drives it | Strength | Watch for |
|---|---|---|---|
| Steady low bed | One fixed level | Never pumps. | The music stays low in the pauses too. |
| Keyframe or timed duck | Volume changes placed at the speech | Exact timing; ignores non-speech noise. | A snapshot: re-cut the dialogue and it has to be redone. |
| Sidechain compression | The live level of the key signal | Follows every edit and the voice’s dynamics automatically. | Any loud sound in the key dips the music. |
Automatic ducking doesn’t always mean a sidechain. Premiere Pro’s Auto Ducking, for example, writes volume keyframes onto the music clip that you can then edit by hand. In editors that offer both, “smooth” ducking usually means the sidechain kind, and “step” ducking means a fixed drop in level for as long as someone speaks.
How Valmera Ducks Music
When you add music, Valmera checks how much speech survives under it. If at least a second does, the track becomes a bed at −18 dB with smooth sidechain ducking keyed by the video’s own audio: it dips with the voice and swells back in the gaps. If nobody speaks under it, the music is the lead at −4 dB with no ducking, because there’s nothing to duck under.
| Smooth duck | Setting |
|---|---|
| Key (sidechain input) | The video’s own audio |
| Threshold | 0.03 linear, about −30 dBFS: speech, not room tone |
| Ratio | 12:1 |
| Attack | 180 ms |
| Release | 550 ms |
Step mode (“use step ducking”) is the older behavior: a fixed −12 dB dip during the transcript’s speech windows. It switches exactly at word boundaries and ignores non-speech sounds, but the jumps are audible. You control the bed level, ducking on or off, and smooth or step mode in plain English; the compressor’s individual settings aren’t exposed.
A voiceover works differently. When you add your narration, Valmera lowers the video’s original audio by 12 dB while it plays. The music’s sidechain listens to the video’s own audio rather than to the narration, so under a voiceover, set the music level directly (“music at −24 dB under the voiceover”). Every audio control is in the audio and music docs.
Frequently Asked Questions
Made with Valmera: real exports
Let Valmera Mix the Music Under Your Voice
Add a track and it sits under your speech with smooth ducking by default. Creating an account and uploading are free; editing requires a subscription.
Create an account →