audio-ducking
Installation
SKILL.md
Audio ducking
The voice carries the meaning, so the music works for it: lower and steady under speech, back up in the gaps that matter, never pumping on every breath. Do the work in the user's editor when they have one (volume keyframes or a sidechain compressor exist everywhere). Without one, use the ffmpeg recipes in references/ffmpeg-recipes.md.
Defaults
| Setting | Default | Why |
|---|---|---|
| Delivery target | YouTube and streaming: -14 LUFS integrated. Podcast: -16 LUFS. Broadcast (EBU R 128): -23 LUFS, within 0.5 LU. True peak at most -1 dBTP | One master cannot serve every destination. If the user named none, use -14 LUFS and say so. |
| Music under speech | 12 to 18 dB below the voice | Below 12 the music competes for the same ears; beyond 18 it disappears. |
| Duck depth | 10 dB (range 6 to 12) when speech starts | Deeper ducks are heard as the music switching off. |
| Method | Volume automation from speech timings | Predictable, editable, and it can start before the first syllable. |
| Automation timing | Ramp down over 150 ms, ending at the first syllable; hold 200 ms after the last word; ramp up over 500 ms | Clips no word onsets and does not snap back on a breath. |
| Short pauses | Stay ducked through any pause shorter than the whole envelope (about 0.85 s with the timings above) | Releasing between phrases is the "pumping" that sounds amateur. |
| Sidechain (no timings available) | Ratio 4, threshold about 13 dB below the voice's typical peak level, attack 30 to 80 ms, release 250 to 700 ms | Gives about 10 dB of duck; see the depth formula below. |