Chapter 33 — Key Takeaways

The one-page card to screenshot before an audio-post session. Definitions, the order of operations, the dialogue chain, loudness targets, and the mistakes that sink a mix — mostly tables, by design.

The core idea

Audio post is where "sounds professional" is won — the payoff of sound is half the picture. You do not start until the picture is locked, you build the mix dialogue-first, and you master to a loudness target, not past it. The picture never changes; your command of its sound does.

Definitions (first-defined this chapter)

Term Definition
Audio post All the tasks that turn raw timeline sound into a finished soundtrack: organize/sync, clean, level, score, design, and master.
The mix The balanced blend of all sound elements — every level, tone, and placement set so nothing fights and the dialogue is always clear.
Dialogue cleanup Removing everything from the recorded voice that is not the voice — clicks, rumble, hum, hiss, room ring.
Noise reduction The tool that learns a "noise print" (ideally from recorded room tone) and lowers steady background noise while keeping the voice.
EQ (equalization) Raising/lowering specific frequency bands to shape a sound's tone — cut mud, add presence, roll off rumble.
Music bed Music laid under the dialogue and picture to set mood and pace, sitting low and ducking under the voice.
Sound effects (SFX) Added non-musical sounds — ambience beds and synced spot effects — that build a believable space.
Loudness How loud a program feels over time (not its peaks), measured in LUFS.
LUFS mastering Setting the whole mix's overall loudness to a destination target while keeping true peaks under a safe ceiling.

The order of operations (memorize this)

# Step Do Rule
1 Organize & sync audio into stems (dialogue/music/SFX/ambience) separate so you can control each
2 Dialogue cleanup de-click → high-pass → de-hum → noise-reduce reduce, don't resurrect; less is more
3 EQ & level cut mud, add presence; compress; ride cut before boost; peaks ~-12 dBFS
4 Music bed choose, license, place, duck dialogue is the boss; music serves
5 SFX & ambience ambience bed + synced spot effects ambience hides edits; sync to frame
6 Final mix & master balance, automate, master to loudness ~-14 LUFS, true peak ≤ -1 dBFS

The mix priority order (loudest → quietest)

Layer Typical level vs. dialogue Job
Dialogue the reference; peaks ~-12 dBFS carries meaning — always clear, always on top
Spot effects (SFX) ~-18 dBFS accents punctuate actions; synced to frame
Music bed -24 to -30 dBFS, ducks under lines mood and momentum; lifts in gaps
Ambience -28 to -34 dBFS, continuous establishes place; hides edits

If any layer ever makes a word harder to understand, it is too loud — no exceptions.

Dialogue cleanup guide (in order, gentlest first)

Problem Tool Setting / note
Clicks, pops, bumps de-click / manual dip / cut+patch do first; patch gaps with room tone
Low rumble (traffic, HVAC, handling) high-pass filter ~80 Hz; back off if voice thins
Tonal electrical buzz de-hum notch 50/60 Hz + harmonics
Steady hiss / drone broadband noise reduction learn print from room tone; ~8-10 dB, then stop
Room echo / reverb de-reverb can reduce only, never remove — fix on set
Harsh "s"/"sh" de-esser tame 5-8 kHz only when it spikes

Over-processing = "underwater," swirly voice. The cure becomes worse than the disease. Do less; mask residue with a slightly louder ambience bed.

Dialogue EQ (subtractive first)

Region Move Why
~80 Hz high-pass remove sub-bass rumble
200-400 Hz cut 2-3 dB clear the "mud"/boom (lavs especially)
500 Hz-1 kHz small cut open "boxy," small-room sound
3-5 kHz boost 2-4 dB presence — intelligibility, "cut through"
~10 kHz high-shelf +1-2 dB "air," polish

Cut problems before boosting good parts. Toggle the EQ against raw to confirm you improved it. Then compress gently (~2:1-3:1, 3-6 dB) and ride the big swings by hand.

Loudness targets (master to these)

Destination Integrated target Peak ceiling Note
Streaming / social (YouTube etc.) ~-14 LUFS -1 dBFS true peak the book's Project 3 target; platforms normalize to ~this
Podcast / spoken word ~-16 LUFS -1 dBFS mono deliverables common
Broadcast — Europe (EBU R128) -23 LUFS -1 dBTP strict, enforced; ±1 LU
Broadcast — North America (ATSC A/85) -24 LKFS -2 dBTP LKFS = LUFS under another name

Treat exact numbers as current-and-revisable — confirm the destination's spec (Chapter 36). The principle is fixed: measure integrated loudness, master to target, never cross the peak ceiling.

dBFS vs LUFS (don't confuse them)

dBFS LUFS
Measures instantaneous peak perceived loudness over time
Ceiling / target 0 dBFS = clipping destination target (e.g. -14)
Used for leveling dialogue, avoiding clipping mastering the whole program

Music: the licensing rule

Use only music you can prove you are licensed to use. Credit is not a license; "not for profit" is not a defense; automated systems block/mute/demonetize infringement.

Legal source Cost Watch for
Royalty-free / production library one-time or subscription keep the license/receipt
Creative Commons free follow the exact terms (attribution? non-commercial?)
Commissioned original pay a composer clean ownership
Platform-cleared library free on that platform may not transfer off-platform

Full legal treatment — sync rights, releases, copyright — is Chapter 38.

Top mistakes → fixes

Mistake Fix
Starting audio post before picture lock lock the cut first; every fix is frame-keyed
Music too loud / fighting dialogue duck the bed; dialogue is the boss; pick a sparser track
Unlicensed "temporary" song license first — never build the mix on it
Over-cranked noise reduction ("underwater") back off to ~8-10 dB; mask with ambience
Boosting EQ before cutting subtractive first
Dead silence at cuts ("blinking") lay a continuous ambience/room-tone bed
Spot effects off-sync place frame-exact; nudge to the action
Over-limiting to be "loud" platforms normalize — master to target
Judging only on studio headphones QC on phone speaker + mono + a car/TV

Craft principles surfaced

  • Sound is half the picture — this chapter is its payoff; the finish decides "professional."
  • Fix it in pre, not in post — pointed at your own workflow: lock, then finish sound; and cleanup reduces, it cannot resurrect.
  • You shoot for the edit — room tone recorded on set becomes the noise print, the edit patch, and the ambience seed.
  • Motivate every choice — every effect, cue, and bed serves the story or comes out.

Pre-flight checklist (run before you export)

Check Pass if…
Picture locked no pending edits; every sync still holds
Dialogue clarity every word clear on a phone speaker at half volume
Dialogue level peaks ~-12 dBFS, steady; no volume-knob moments
Noise floor low, no hum; voice not "underwater"
Music licensed (receipt kept); ducked under every line
Ambience continuous; no cuts "blink" to silence
Effects synced frame-exact; below dialogue
Loudness integrated ≈ target (-14 streaming)
True peak ≤ -1 dBFS
Mono check dialogue survives; no effect vanishes

If any box is unchecked, you are not done — that box is exactly what a viewer will hear and you will not be able to name.

Project increment

Production Checkpoint (Project 3): mix Project 3 — stems; clean and level the dialogue; add a licensed music bed and an ambience bed with synced effects; master to about -14 LUFS, true peak ≤ -1 dBFS; QC on a phone and in mono. Keep the stems (Chapter 36 exports to delivery spec; Chapter 39 folds it into your reel).

Anchor

The Café Scene got its soundscape: cleaned/matched counter dialogue, a looped-room-tone café ambience, the door bell and espresso SFX, a ducked warm music bed — mastered to -14 LUFS. The scene never changed — your command of it did.