Sound is in the script, not added to it
Three of the film’s turning points have no change of image. They are sound events, written into the screenplay as scenes in their own right, with their own page count. A financier reading the script sees them and asks what happens in them. The answer is that the audience’s understanding of the situation changes, and nothing visible occurs.
Writing sound into the script rather than the post schedule changes the economics of a small film substantially. It means the sound designer is in development conversations, not handed a locked picture. It also means we can cut visual material with confidence, because we know what is carrying the scene.
The ringing, and why it is the spine
Margarita’s tinnitus is established in the first movement — after a nearby impact, before the miscarriage — and it never entirely leaves the film. It is the single most useful device in the sound design, for three reasons.
It is subjective without being surreal. Every audience member either has experienced ringing in the ears or immediately understands it. It puts us inside her hearing without any of the signalling that a conventional subjective-sound treatment requires.
It can carry information. The pitch and level of the ringing tracks her state. It does not do this dramatically — a change of a couple of decibels and a few dozen hertz across a scene. Audiences do not consciously notice and reliably respond.
It is where the voice lives. This is the essential one. The boy’s voice is not laid over the ringing; it emerges from it. In the mix, the tinnitus tone is the carrier, and the voice becomes audible as the tone resolves into it.
Why that matters
A voice that arrives from nowhere is a ghost. A voice that resolves out of a sound already established as living inside her head is something the audience cannot classify — which is exactly the state we want them in.
Building a siren that means something
Air-raid sirens are a problem for a sound designer because they are already so loaded. Used plainly, they announce war film and flatten everything around them into genre. We approached them as a musical element with a fixed set of rules.
Distance is dramaturgy. The siren is never close. It is always at least two buildings away, filtered through glass, concrete and distance, so that it functions as weather rather than as an alarm. What changes across the film is not its volume but the reverberant space it arrives through — and the room tone tells you where the walls have gone.
The interval matters more than the sound. We treat the gaps between sirens as the composed element. Long silences after a pattern has been established are the most reliably frightening thing in the film, and they cost nothing.
Never on a cut. A siren never begins on a picture cut. It always starts mid-scene, which denies the audience the small comfort of a sound that behaves like an edit.
The problem of the child’s voice
Everything in this film risks sentimentality, and nothing risks it more than a dead child speaking kindly to his mother. The line where he tells her she is the best mother in the world can be unbearable in the wrong reading, and not in the way we want.
Four decisions manage it.
The voice is recorded in a real room, not a booth. Booth-clean dialogue floats. The boy is recorded in a domestic space with its own reflections, so he is acoustically somewhere — the same somewhere she is.
No processing beyond that room. No pitch shift, no reverb bloom, no whisper double. Every one of those is a signal that says supernatural, and the film cannot afford to answer that question.
He is often unremarkable. Much of what he says is mundane and some of it is funny. A child who is only ever profound is a device; a child who complains about something is a person.
He is never mixed louder than the room. He sits inside the ambience, at the level of a person speaking quietly nearby. The audience leans in. That lean is the effect we are after.
The frequency of dread
The low end of this film is almost entirely non-diegetic and almost entirely inaudible on laptop speakers, which we accept. Distant impacts are built with content down to around 30Hz, and in a properly calibrated room that content is felt in the chest rather than identified by the ear.
What matters is when we withhold it. There is a stretch of roughly eleven minutes in the third movement with no low-frequency content at all. Coming after ninety minutes of consistent sub-bass presence, its absence produces a distinct physiological unease that no audience member reports as a sound observation. They describe the scene as tense. It is not the scene.
Silence, and the last eighteen minutes
There is no score in the final eighteen minutes of the feature. There is very little score anywhere — a few sustained textures in the second movement, mostly to smooth transitions that the picture edit could not carry alone.
The final sequence has room tone, the ringing at its lowest level in the film, and two performances. Nothing else. Removing music from a climax is a decision that gets argued about in every development meeting, and the argument against it is always that the audience needs to be told how to feel.
Our position is that by minute ninety the audience knows exactly how to feel, and that the most disrespectful thing available to us at that point is a cue.
The location sound brief
Sound-led films are made or lost on production sound, and most of the work happens before the shoot. Our brief to the location recordist has four items.
Room tone from every space, for at least ninety seconds. Not thirty. Ninety, at every significant change of state — windows open, windows taped, door closed, room emptied. This is the single most valuable material we collect and it costs the schedule almost nothing.
Boom first, radio as backup. A boomed voice in a real room carries the room with it. Radio mics produce a clean, close, acoustically detached voice that has to be put back into a space artificially, and it never quite goes.
Wild lines for everything spoken off-camera. Cheap on the day, and it removes an entire category of ADR that would otherwise be recorded months later in a different acoustic.
Record the building. Pipes, lifts, the stairwell, the boiler, traffic through glass. Our library of the actual locations is the source of most of the film’s ambience, and it is why the interiors sound like themselves rather than like a sound-effects library.
Designing an absence
The most difficult sound problem in the film is not the sirens or the voice. It is the miscarriage sequence, which has to convey a catastrophic event happening off screen without either explaining it or aestheticising it.
What is in that scene: a corridor’s room tone, a neighbour knocking four times, a pause, the neighbour leaving, and Margarita’s breathing changing in a way that takes about forty seconds to become audible as distress. There is no music, no impact, no dramatic sound event of any kind.
It was assembled over three weeks and rejected in eleven versions. Everything we tried adding — a distant siren, a dripping tap, a low tone — turned an event into a scene about an event. The final version works because it declines to perform.
The mix, and the small-room problem
A film built on low-frequency content and quiet dialogue faces an obvious difficulty: most people will not first encounter it in a properly calibrated cinema.
We mix for a cinema and then check in three other environments — a domestic television, a laptop, and headphones — and make a specific decision about what each one is allowed to lose. The laptop loses the sub-bass entirely and we accept that. What it must not lose is intelligibility, and that constrains how quietly the boy can be mixed relative to the room.
This produced the only substantive disagreement in the sound department. The designer’s instinct was to keep him at the edge of audibility, which is the correct artistic answer and produces a laptop version where a crucial line disappears. The compromise sits a few decibels above ideal, and we would make the same call again.
Two things we tried that did not work
A recurring musical motif for the boy. A short, four-note figure that appeared whenever he spoke. It was beautiful and it was fatal — it told the audience how to feel about him, which converted an open question into a sentimental one. Removed entirely after one screening.
Diegetic radio broadcasts. News bulletins heard through a wall, carrying information about the progress of the invasion. Efficient exposition and completely wrong for this film: they gave the audience a vantage point above Margarita’s, and the whole construction depends on us knowing nothing she does not.
Both were cut late, after real work had gone into them. The general lesson, learned repeatedly on this project: the elements hardest to abandon are usually the ones doing the audience’s thinking for them.
What this costs, practically
A sound-led film is not a cheap film, but it is cheap in an unusual place. We spend almost nothing on production sound complexity — no radio mics on six actors, no crowd, no vehicles. We spend a great deal on post: a long design period, a proper mix stage, and a designer involved from development.
For a production our size, that trade is the best value available. Sound work happens on a schedule we control, in a room, with no weather, no daylight window and no location fee. It is the one department where a small film can genuinely compete.