In this guide
Audio ducking lowers one sound while another is active—for example, music under a narrator—and brings it back afterward. Use separate tracks where possible, then check the timing and intelligibility throughout the finished mix.
Audio ducking temporarily lowers one sound when another sound is active. A common example is background music getting quieter while a narrator speaks, then returning between sentences. The voice acts as the control; the music is the sound being lowered. Ducking can be automatic or created with manual volume keyframes.
It is a mixing technique, not automatic source separation. If voice and music are already combined in one audio file, turning that file down lowers both. Look for the original separate tracks first; otherwise, an estimated separation may provide a useful starting point, with limitations described below.
Choose a ducking method
| Situation | Starting method | Check |
|---|---|---|
| A few spoken lines over music | Manually adjust music volume with keyframes or an envelope. | Music fades before speech and returns without a distracting jump. |
| A long narration with separate music | Try an automatic ducking effect controlled by the narration. | Soft words and short pauses trigger appropriate changes. |
| Live commentary or a stream | Use a suitable sidechain or ducking processor in the live mixer. | Test routing, delay and level during actual speech. |
| Voice and music already mixed together | Find original tracks or test source separation before independently changing the music. | Words, music bleed and timing; estimates can lose detail. |
What do duck amount, threshold and fades mean?
The duck amount controls how far the music level falls. A threshold or sensitivity setting determines when the control sound triggers a change. Fade-down and fade-up times control how quickly music falls and returns. A very short fade or frequent triggering can make the music pump; a slow return can leave it too quiet for the next cue.
Start with a modest dip and listen to the whole sentence. Quiet speech, breaths and pauses can trigger automation differently from a loud presenter. There is no universal setting: the music, voice and delivery environment determine how much reduction is helpful.
Duck music under speech in Audacity
Audacity's Auto Duck manual describes a specific track arrangement. Keep an unchanged project or source copy; labels may vary by version.
- Arrange separate tracks. Place the music track above the narration track and align them. The first unselected track beneath the selected audio acts as the control.
- Select the music to change. Select the music track or intended region, leaving the narration control track unselected.
- Open Auto Duck. Choose Effect → Volume and Compression → Auto Duck. Adjust the duck amount, threshold and fades, then apply the effect to the selected music.
- Review in context. Play the opening word, short pauses, quiet phrases and the return to music. Undo and adjust if the music pumps or hides speech. The manual notes that applying effects to time-stretched clips can render the speed change.
Automatically duck music in Premiere
With separate dialogue and music clips, Adobe's Premiere ducking instructions use the Essential Sound panel. Tag the voice as Dialogue and the music as Music, enable Ducking on the music, select Dialogue under Duck Against, then choose Generate Keyframes.
Listen and adjust the generated keyframes where words begin, pauses are short or music must return at a particular cue. Adobe notes that generating keyframes again overwrites manual changes to those generated keyframes. An ending fade is a separate edit; the Premiere fade guide covers clip-edge fades.
What if voice and music are already combined?
First ask for the source project or independently recorded dialogue and music. If only a finished recording exists, test the voice isolator on a representative passage. Describe the spoken voice and compare Isolated sound with the original; the remaining mix may be a candidate for the background. The outputs are estimates: music can remain in the voice, speech can remain in the background, and adding both together may not reproduce the original perfectly.
Saturalabs accepts supported MP3, WAV, MP4 and MOV uploads up to 50 MiB. AI WAV downloads require an account and sufficient credits, displayed before download. It does not automatically duck a timeline or render a new video. Import the chosen audio into an editor, preserve the original start position and check transitions and sync using the audio replacement guide.
Check the final mix
Listen at a comfortable level on headphones and another playback system. Check that the first and last words remain clear, music does not jump between phrases, and the ending returns at the intended time. Play the exported file itself; a successful preview is not proof that the final render has the same routing or levels.