Ducking

Ducking means lowering the level of one sound, usually music, while another sound, usually a voice, is playing, and raising it again when the voice stops. The music "ducks" under the words, so speech stays clear without switching the music off.

In short video it is the difference between a voice-over that fights its music bed and one that sits on top of it.

Two ways to duck

• Volume automation. A lane, or envelope, on the music track: it ramps down just before the voice starts, stays low while it talks and comes back up after. Every move is drawn on purpose, so the result is predictable and easy to change.
• Sidechain compression. A compressor on the music that reacts to the level of the voice. It ducks automatically, but set too hard it makes the music pump in and out.

Many editors have an automatic ducking switch that does one of the two for you.

Starting levels

In HyperFrames, data-volume sets a level where 1.0 is the file as it is:

In an editor that shows decibels, start the music about 14 dB below the voice. These are starting points: listen on a phone speaker and on headphones, then measure the final mix, which should land at about -14 LUFS with the true peak below -1 dBTP.

Ducking in a HyperFrames video

To duck exactly, first find where the voice speaks. FFmpeg lists the silences:

ffmpeg -i voice.wav -af silencedetect=noise=-35dB:d=0.4 -f null -

The gaps between the silences are the speech. A volume lane on the music, or on a group that holds several copies of a loop, then dips under each spoken stretch and rises again in the longer pauses.

In Motion Cheat Code

The mcc-sfx-pass skill does this as part of a sound pass: it finds the speech, draws the lane under it, adds fades and measures the loudness of the render. The music tracks also come as stems, so you can drop only the drums under a voice instead of turning the whole track down.