Chapter 33 — Exercises
Audio post is a doing skill far more than a reading one. You can memorize every frequency band and every loudness target and still produce a muddy, uneven, over-loud mix — because the ear is trained only at the timeline, comparing before and after, over and over, on real material. These exercises put you there. Do them in DaVinci Resolve's Fairlight page or in Audacity (both free; see Appendix E), or in any editor with an audio meter, an EQ, and a noise-reduction tool. Wear closed-back headphones for the work, and check the important ones on a phone speaker — the device most of your audience actually uses.
Exercises are graded by effort and depth:
- ⭐ Warm-up — a few minutes; locks in a concept.
- ⭐⭐ Core — the real practice; do as many of these as you can.
- ⭐⭐⭐ Stretch — deeper or more ambitious; where the biggest gains are.
Model solutions and critiques for the starred and odd-numbered exercises are collected in Answers to Selected Exercises at the back of the book. Try each one before you read its answer — a mix you got wrong teaches more than an answer you only read. Above all: bypass every plugin often and compare to the raw. Almost every audio-post mistake is too much of something, and the only reliable guard is the A/B.
A. Hearing the mix (Eye & Ear Training)
No software — just your attention, trained on audio the way earlier chapters trained your eye. Keep your findings in the Frame Log's listening column.
1. ⭐ The volume-knob test. Watch five different videos to the end — a film scene, a news package, a corporate promo, and two creators — and note every time your hand moves toward the volume: up because dialogue vanished, down because something blasted. Write what caused each grab and which §33.3/§33.6 tool would have prevented it. Constraint: you must find at least one failure in the amateur examples. Self-review: how many of the grabs were leveling problems versus mastering problems? (Answer provided.)
2. ⭐ Find the bed. Pick any polished video with people talking over music. Consciously listen past the dialogue to the music bed. Is it ducking under the words and lifting in the gaps, or sitting static? Roughly how far beneath the voice does it sit? Write one sentence. (Answer provided.)
3. ⭐⭐ Name the noise. For one full day, whenever you encounter amateur video (a random upload, a social clip, a friend's vlog), name the specific audio flaw out loud: room echo, electrical hum, broadband hiss, uneven levels, distortion/clipping, or music-too-loud. Do it ten times and tally which flaw is most common. You are building the diagnostic vocabulary of §§33.2–33.4, and you will find that two or three flaws account for almost everything.
4. ⭐⭐ The layers of a place. Watch a scene set somewhere busy — a café, an office, a street — with your eyes closed. List every distinct sound layer you can separate: dialogue, room-tone/ambience, distant murmur, individual spot effects, small Foley, music. How many layers is the "simple" scene actually made of? (Answer provided.)
5. ⭐⭐⭐ Sound-off, sound-on, sound-only. Take one 60-second scene and experience it three ways: picture with the sound muted, sound with the picture blacked out, then both together. Write a full paragraph on what the sound alone told you that the picture did not — location, mood, offscreen events, emotional steering, whose point of view you were in. This is the entire argument of the chapter compressed into one exercise; keep the paragraph.
B. Reading soundscapes (Read the Sequence)
6. ⭐ Picture it. Re-read FIGURE 33.2 (the stems mix). Without looking back, write the five track types in their priority order, loudest to quietest, and give a one-phrase job for each. Then write the approximate level of each relative to the dialogue. (Answer provided.)
7. ⭐⭐ Diagnose the field. In FIGURE 33.5 (the Café soundscape build), the espresso machine is added at layer 3. Name the two jobs it does — one obvious, one hidden — and explain the hidden one in terms of edits and masking. Then connect it to a scene in Case Study 1. (Answer provided.)
8. ⭐⭐ The missing layer. Look at FIGURE 33.5. The scene already has ambience, spot effects, Foley, and music. Invent one additional sound layer that would deepen the Café Scene, say where it sits in the level order (relative to dialogue), what it adds to the story, and what action or off-screen event motivates it. (Answer provided.)
9. ⭐⭐⭐ Write your own build. Choose a 30-second scene from something you admire and write a layer-by-layer build table like FIGURE 33.5: what you would add, in what order, what each layer does, and roughly what level it sits at relative to dialogue. This is the exact planning a sound designer does before touching a fader — do it before you ever open the timeline on your own projects.
C. Cleaning dialogue (Clean This)
The heart of the chapter. Use real recordings — your own from Projects 1–3 are ideal, and their flaws are the ones you most need to fix.
10. ⭐ Kill the rumble. Take any location dialogue clip. Add a high-pass filter and slowly sweep its cutoff up from 20 Hz until you hear the voice start to thin and lose its chest, then back it off. Note the exact frequency where the low weight was gone but the voice was untouched (usually near 80 Hz; nearer 70 Hz for a deep voice). Self-review: did removing the rumble also reduce any hum, or is that a separate problem? (Answer provided.)
11. ⭐⭐ Learn a noise print. Take a clip with audible background hiss or drone and a moment of room tone (no voice). Learn the noise profile from the room-tone section and apply broadband noise reduction three times — gentle (~6 dB), moderate (~10 dB), and heavy (~20 dB). Listen to all three back-to-back. Where exactly does the voice start sounding "underwater" or "swirly"? Keep the version that is clean but natural, and write down the dB value. (Model answer discusses the trade-off.)
12. ⭐⭐ The full cleanup pass. Take one minute of your worst usable dialogue and run the whole §33.2 order in sequence: spot-fix transients, high-pass, de-hum if needed, then a modest noise reduction learned from your room tone. Do not EQ or compress yet. Compare to raw. Write down what improved and what stubbornly remained (some things — like room echo or clipping — will resist, and that is the lesson). (Answer provided.)
13. ⭐⭐ Room-tone patch. Find a dialogue edit where the background "blinks" — drops to dead digital silence at the cut. Lay a bar of room tone underneath, across the cut, and crossfade its edges so the background is continuous. Listen before and after. This single trick separates amateur from clean; note how a cut you could hear becomes a cut you cannot. (Guidance provided.)
14. ⭐⭐⭐ The rescue challenge. Deliberately record dialogue badly — across a hard, empty, echoey room with a fan running — then try to rescue it in post using everything in §33.2. Document exactly how far you can get and where you hit the wall. Constraint: you may not re-record. The lesson is not the rescue; it is feeling in your own hands why Chapters 14–15 insisted you fix it on set. (Answer discusses what post can and cannot do.)
D. Tone and level (Level & EQ This)
15. ⭐ Subtractive first. Take a clean voice clip. Find the "mud" by boosting a bell filter +6 dB and sweeping it slowly across 150–500 Hz until the voice sounds boomy and boxy; note the frequency, then cut a couple of dB there instead. Compare to the raw. Explain in one sentence why cutting the mud is better than boosting the highs to compensate. (Answer provided.)
16. ⭐⭐ Presence and air. On a dull, "behind-glass" voice, add a gentle boost at 3–5 kHz (presence) and a small high-shelf at 10 kHz (air). Toggle on/off against the raw. Then deliberately overdo both until the voice turns harsh and spitty — note the exact point where "clear" became "painful" — and de-ess it back to comfortable. (Model answer provided.)
17. ⭐⭐ Even it out. Take dialogue with an obvious loud/quiet swing (someone leaning in and then sitting back). Fix it two ways on two copies: (a) by hand, drawing clip-gain automation up on the quiet parts and down on the loud; (b) with a gentle compressor (ratio ~2.5:1, a few dB of reduction). Level both to peak around -12 dBFS. Which sounds more natural, and when would you reach for each? (Answer provided.)
18. ⭐⭐⭐ Match two shots. Take two dialogue clips of the same person recorded differently — a lav and a boom, or two rooms, or two distances. Using only EQ and level, make them sound like they belong in the same conversation, so a cut between them is seamless. This is dialogue "matching," the audio cousin of the shot-matching from Chapter 31. Self-review: cut rapidly between them — can you still hear the join? (Model answer provided.)
E. Music and the bed (Cut This — music)
19. ⭐ License audit. List every piece of music in any three videos you have made or plan to make. For each, write the specific reason you are legally allowed to use it — a library license (with receipt), a Creative Commons license (naming the exact terms you meet), original/commissioned, or a platform-cleared library — or flag it red as a problem to fix before publishing. (Answer provided.)
20. ⭐⭐ Pick the bed. For one 60-second scene, audition three different licensed tracks under it. For each, write whether it fits the scene's energy and arc (not just its genre) and whether it leaves the presence range (3–5 kHz) clear for the voice. Choose one and justify it in a sentence. (Model answer provided.)
21. ⭐⭐ Duck it. Place a music bed under dialogue at a level where it competes with the voice. Now duck it two ways: (a) draw manual volume automation down under each line and back up in the gaps; (b) if your tool has it, set a sidechain from the dialogue onto a compressor on the music. Note the actual level difference between "under a line" and "in a gap" (often 6–10 dB). Which method sounds more natural on your material? (Guidance provided.)
22. ⭐⭐⭐ One theme, three moods. Take a single piece of music and, using edited sections of it (or a stems version), score three different beats of a piece — an upbeat opening, a low emotional point, and a resolved ending. Make one track feel like three by how you deploy and level it. This is the Up "one evolving theme" lesson from Chapter 1 applied to your own timeline. (Answer discusses approach.)
23. ⭐⭐ Recreate it (a famous cue move). Pick one well-known moment where music and picture change together — a beat drop landing on a cut, a score swelling as a character realizes something. Recreate the structure of that move (not the copyrighted track) with your own footage and a licensed cue: land your musical change on your edit point. Self-review: does the cut now feel intentional rather than arbitrary? (Guidance provided.)
F. Building the world (Five Ways / Sound design)
24. ⭐⭐ Five ambiences, one shot. Take one neutral interior shot of a person and lay five different ambience beds under it in turn: a silent office, a busy café, a quiet library, an outdoor patio, a hotel lobby. Watch how the same picture changes meaning and location with each. Which "reads" most clearly, and why do the specific beds beat the generic ones? (Answer provided.)
25. ⭐⭐ Record your own SFX. Record five of your own sound effects (a door, a cup, footsteps, a light switch, a phone buzz). Sync at least three of them frame-accurately to actions in a clip. Then deliberately nudge one a third of a second off and feel how wrong even a small sync error is. Self-review: at what offset did the mismatch become obvious? (Guidance provided.)
26. ⭐⭐ The three-layer soundscape. Complete the 🎬 On Set assignment from §33.5 if you have not: under a 20–30 second clip, build (1) a continuous ambience bed, (2) two spot effects synced to on-screen actions, and (3) one Foley element — all clearly below the dialogue. Then mute and unmute the three layers together. Self-review: does the scene go from "a void with talking" to "a real place"? If not, raise the ambience or fix the sync. (Model self-critique provided.)
27. ⭐⭐⭐ Design a place that isn't there. Take a shot of someone standing in a plain, quiet room. Using only ambience and spot effects — no picture changes, no new dialogue — make it sound like they are (a) in a storm-lashed cabin, then (b) on a busy subway platform. Prove that sound builds the world the picture only implies. (Answer discusses layering.)
G. Mix and master (Settings Drill)
28. ⭐ Read the two meters. Put a loudness meter and a dBFS peak meter on a finished mix. Play it through and write down (a) the integrated LUFS and (b) the maximum true peak. State which one tells you about clipping and which tells you how loud the program feels. (Answer provided.)
29. ⭐⭐ Master to spec. Take one finished mix and master it three times: to -14 LUFS (streaming), to -16 LUFS (podcast), and once more with a -1 dBFS true-peak limiter engaged so you can watch the peak behave. Note how the same mix needs a different master level for each destination while the balance never changes. (Model answer provided.)
30. ⭐⭐ The over-limiting trap. Take a dynamic mix and crush it with a limiter to be as loud as possible (say -8 LUFS). Then loudness-normalize both the crushed version and a clean -14 LUFS version to the same playback loudness and compare them at equal volume. Which sounds better, and what does that prove about the "loudness war" on platforms that normalize? (Answer provided.)
31. ⭐⭐⭐ The four-system QC. Master a mix to -14 LUFS and play it on four systems: good headphones, laptop speakers, a phone speaker, and a car or TV. Then sum it to mono and listen again. Write down every problem each system reveals — thin dialogue on the phone, a doubled effect vanishing in mono, muddy low end, and so on — and fix them. This is the professional's final gate; a mix that survives all five is real. (Model checklist provided.)
H. Diagnosing failure (Fix the Mix)
32. ⭐⭐ Fix the described mix. A described mix: "The dialogue is clean but jumps from loud to nearly inaudible between every shot; a pop song plays at almost the same level as the voice; there is no background at all, so cuts click into silence; the whole thing distorts on a phone but is fine on headphones; it was mastered to -8 LUFS." Name every problem and prescribe the specific fix for each, in the order you would tackle them. (Answer provided.)
33. ⭐⭐ Fix the plan. A friend says: "I'll finish the audio last, after I upload the video, by just adding a trending song in the app." List four things wrong with this plan, sorted into legal problems and craft problems, and say what they should do instead. (Answer provided.)
I. Interleaving and synthesis
Mixing prior chapters into this one so the whole book stays connected.
34. ⭐⭐ Shoot for the mix. Using Chapters 14–15, list five things you would do on set on your next shoot specifically to make this chapter's audio-post session easier. For each, name which post step it serves (noise print, presence, headroom, room echo, wild SFX). (Answer provided.)
35. ⭐⭐ The finish, in order. You have a locked cut. Put these finishing steps in the correct order and justify it: master to LUFS, color grade, noise-reduce dialogue, color correct, add music bed, add lower thirds. (Draws on Chapters 31–32 and 34.) (Answer provided.)
36. ⭐⭐⭐ Teach it back. In about 200 words, explain to a beginner why "you master to the target loudness, not past it," using loudness normalization and one concrete platform target. Teaching a concept clearly is the strongest test that you truly own it. (Model answer provided.)
37. ⭐⭐⭐ The full finish on Project 3. Complete the Production Checkpoint: mix Project 3 end to end — stems, dialogue cleanup and leveling, a licensed and ducked music bed, ambience and synced effects, mastered to about -14 LUFS with a -1 dBFS ceiling, QC'd on a phone and in mono. Then write three sentences on what the sound added that the color grade could not. (Guidance provided.)
When you have done these — especially the cleaning, leveling, and mastering ones — you have crossed the last hidden line between amateur and professional video, the one the audience hears but cannot name. Move on to Chapter 34, where we add the visual-information layer: titles, lower thirds, and motion graphics, built with the same restraint that keeps a music bed under the voice.