The mix with speech as the anchor

Mix the words first and build everything else around them [official guidance]. Speech is the anchor of the mix: everything else is balanced to it.

1. Dialogue and voiceover first, consistent in level from line to line.

2. Clean each speech take first: trim the voice engine's tail click, add a high-pass at about 80 Hz, and match the room tone across cuts. Measure the noise floor in the parts without speech as an RMS level (root mean square, the signal's average level), and match it by measurement (steps below). Filmed location and interview sound is cleaned before the cut (see Editing → Filmed footage and interviews).

3. Place each on-screen line in its shot's space (see Sound design and mix → The placed voice). The voiceover stays close and dry.

4. Duck the music under speech. Write it as keyframes on the music track, so the levels stay editable in the timeline you hand over. Starting values: about −10 dB; the dip starts 6 frames before the speech and recovers over the 12 frames after it; gaps under 12 frames count as one stretch of speech (steps below).

5. Ambience low and continuous across cuts. A flat bed may drift (see Sound design and mix → Placing effects).

Then measure how the mix translates (see Sound design and mix → Checking the mix on small speakers), and listen to it on the cut before you approve it.

Do it yourself

Measure the noise floor of a speech take (ffmpeg or Audition)

1. Pick a speech-free stretch of the take.

2. Read its RMS level from ffmpeg's astats filter, or from Audition's Amplitude Statistics. The command is in Sound design and mix → Stems → Do it yourself; give it a file that holds only the speech-free stretch.

Duck the music under speech (Premiere)

Premiere's Essential Sound panel can set the depth but not the two ramps or the gap rule, so finish them by hand:

1. Mark the music clips as Music in the Essential Sound panel, tick Ducking, and duck against Dialogue.

2. Set Duck Amount (Reduce By in some versions) to −10 dB and click Generate Keyframes.

3. Move the keyframes so each dip starts 6 frames before the speech and recovers 12 frames after it, and join dips whose gap is under 12 frames.

Sources

• Higgsfield Studio's The Cully Hill Boys and Hell Grind (ambience under the voices, dipping when someone speaks); 750mph, Ellis McGourlay, the anchor of the mix decided early (LBB); SVT's audio requirements, speech first (spec).