Audio mixing for video: levels, ducking, busses and stems

Q: Can I mix with the agent instead of the window?

Yes. edittrack sets faders, rides, busses, sends and ducks, adjustclip sets a clip's level and sound chain, and readsound reads the meter. The window takes in the agent's changes as it makes them.

How do you mix audio for a video?

Start with the voice: it is levelled first, and music and effects are set against it. A good order of work:

Clean each voice. Cleaning changes how loud a clip is, so do it before you set levels. Removing background noise covers this step.
Level the dialogue. Every speaker and every clip at the same loudness.
Fill the gaps. Room tone under every hole a cut leaves.
Bring in music and effects. Under the voice, with the music ducked while people speak.
Group and balance. Dialogue, music and effects on their own busses.
Measure and deliver. Read the whole piece on a loudness meter, then export at your platform's target.

Sharpy's window plays the same mix the render writes, so you never judge a mix you are not going to ship.

What does each control in the audio mixer do?

The Mixer tab has a strip for every sound track, one for every bus and a master strip.

Control	What it does	How to use it
Fader	Sets a track's level, from silence to +10 dB, with 0 dB three quarters of the way up	Double-click for 0 dB, or type a value
Balance	Turns the far side down and leaves the near side as it is, so a centred track keeps its level	Used instead of pan, because every sound track is stereo
Level line	Rides a clip's level down, drawn across every sound clip	Drag a dot; ⌘-click adds a point or removes one
Track ride	Points of level fixed on the timeline, applied after the fader	Draw it, or record fader moves in Read, Write, Latch or Touch mode
Monitor level	How loud the edit plays in your room	Kept on your Mac, never written into a render

A fader move is heard while you drag and committed as one step when you let go. The level line only rides down: making a clip louder is the gain control's job. A track ride stays put when clips move; a clip's level line moves with the clip.

How do you level dialogue so every clip sounds the same?

Measure the speech in each clip, set one gain that brings it to a common loudness, then ride gently inside the clip. Ask the agent to level the dialogue (adjust_clip) and Sharpy brings the clip's speech to −23 LUFS, the reference level in EBU R128, with a gentle ride where there is speech so the room in the pauses is not lifted too. It reports what it did, for example "speech at −34.6 LUFS, +11.6 dB to −23 LUFS".

In a test, two lines recorded 17.5 dB apart both came out at −23.0. Sharpy moves a clip by at most 12 dB either way; one further out is reported as a recording to fix. Levelling is for consistency between clips; on export the whole mix is set to its destination's target, as what LUFS for YouTube explains.

Each clip can also carry a sound chain: click and plosive repairs, hum removal, a gate, EQ, a compressor, a de-esser, pitch, reverb, width and pan. Give values in plain units (Hz, dB, milliseconds), or ask Sharpy to find the hum and clicks for you.

What is audio ducking?

Audio ducking is turning music down while someone speaks and back up in the pauses, so the voice stays clear. Done crudely, it pumps: the music lurches up in every gap between words.

A man's hand rides a fader down a few millimetres on a small mixing desk while a woman speaks into a microphone across the room. — Ducking is a fader ridden down under each line and back up between them; timed to the speaker's own pauses, it never pumps.

Sharpy's duck follows the voice track's actual sound and looks ahead, so the music is fully down as a line begins. It takes how long to hold from the speaker's own pauses at commas, so the bed stays down through a sentence and comes up between sentences, not between words. The default is 6 dB down with quarter-second ramps. On a test recording by Sharpy's maker, the bed sat at a median −6.00 dB while he spoke and stayed down through all 16 of his gaps.

In the window, the bed ride button writes a 6 dB ride under the lines and names the step, for example "rides 6 dB under the words on the picture, in 1 place(s)".
From the agent, ask for the music to duck under the voice track (edit_track). This duck listens to the voice, so re-cutting the voice needs no rewriting.

Music for your video covers choosing the bed.

What are busses and sends for?

A bus is a strip with no clips of its own: tracks pass through it and share one fader, one balance and one sound chain. Make one with + Bus in the Mixer and choose it from a track's output menu. The usual layout is dialogue, music and effects: three faders balance the piece, and one compressor treats every voice alike.

A send feeds a copy of a track to a bus at its own level, beside where the track normally goes. Send several voices to one reverb and each keeps its dry sound. A pre-fader send ignores the track's fader; a post-fader send follows it.

What are stems in video editing?

Stems are the separate parts of a finished mix, usually dialogue, music and effects, each written as its own file so someone else can rebalance them. Ask the agent to render with stems and Sharpy writes each track or bus that feeds the master as a 24-bit WAV beside the render, the full length of the piece.

Played together at unity gain, the stems are the mix, even where the master limiter is working: on a real 20-second recording with a music bed, the master minus the sum of its stems left at most −138.5 dBFS. For a version in another language, ask for undipped stems too, each ducked part at a steady level, so a mixer can duck it again under a new voice.

An EDL, FCPXML or OpenTimelineIO file cannot carry a sound chain, so when you hand an edit on, Sharpy names every chain, bus and send that stays behind. What is an edit decision list explains those formats.

What should you check before you export the mix?

Room tone under every gap. Clear, Leaving the Gap takes a stretch out and leaves the hole; ask the agent to fill it with room tone. Sharpy lifts the room from the recording's quietest stretches, never repeating one within 20 seconds: a filled 1.5-second hole measured −46.4 dBFS against the room's own floor of −46.7.
Clean joins. In the inspector, choose a cut, a crossfade or constant gain for a sound clip's join, and its length. A crossfade holds constant power, for two different sounds. Every end of a clip's sound is faded over 3 ms, so no join clicks.
The meter. Under the Mixer's strips sits an EBU Mode meter as EBU Tech 3341 defines it: momentary, short-term and integrated loudness, loudness range and true peak. Reset it and play the whole piece for a true integrated reading.
Mono. Sharpy sums the mix to mono and warns where wide stereo would cancel out on a phone speaker.
The target. Choose the destination and loudness when you export; export the finished video goes through the settings.

Frequently asked questions

Do I need to mix a video with only one voice?

Yes, lightly. Level the voice, fill the gaps with room tone and set the loudness on export: those three steps are most of what makes a video sound finished.

How loud should background music be under a voice?

Low enough that every word is clear on a phone speaker, and no lower. Sharpy's duck takes a bed 6 dB down under each line by default; trust your ears on top of that.

Can I mix with the agent instead of the window?

Yes. edit_track sets faders, rides, busses, sends and ducks, adjust_clip sets a clip's level and sound chain, and read_sound reads the meter. The window takes in the agent's changes as it makes them.

Does my sound leave my Mac?

No. Every measurement, clean-up and render runs on your Mac.

Sources

Checked on 18 October 2026.