Building a world from room tone: sound design for documentary

Every November the furnaces come on, and every November I hear them in footage before I notice them in the room. A forced-air system that cycles every twenty minutes will lay a low rumble under half an interview and leave the other half clean, and the edit will lurch between the two whenever you cut. The fix is a minute of recording you made, or failed to make, on the day.

Most documentary soundtracks are assembled from far more pieces than anyone in the audience suspects, and nearly all of them are scraps the recordist was wise enough to collect. This piece follows those scraps from the location to the delivery drive. If you need the basics of microphones and setup, the guide to shooting documentary interviews covers lav, boom and the chair. Here the subject is everything around the voice.

Room tone is the floor

Every room sounds like something, even when nobody speaks. Ventilation, a fridge two rooms away, traffic filtered through glass, the particular ring of plaster walls: together they make the room's tone, and the audience hears it under every word without knowing it. When an editor cuts out a cough or tightens a long pause, the gap has to be filled. Digital silence sounds as if the film has stopped breathing, and even viewers who know nothing about sound feel the drop.

Only that room, recorded through those microphones at the same gain, fills the gap without a seam showing. So at the end of each setup, before anyone stands, record room tone: everyone in place, lights and fans running, boom and lav exactly where they were. Sixty seconds is the minimum I accept. Say what it is out loud at the head of the take so the editor can find it ("room tone, kitchen, after the second interview"), then hold still.

Record it again whenever the room changes. If the furnace kicks in halfway through, you need tone with the furnace and tone without it. The same goes for rain on a skylight or a window someone opened. A dialogue editor can match almost anything, provided the matching material exists.

Ambiences: the sound of a place

Room tone repairs dialogue. Ambiences, which some editors call atmospheres or atmos, build the world around it: a harbour at six in the morning, a school corridor between bells, wind moving through a stand of cottonwoods. They are what makes a cut to an exterior feel like arriving somewhere.

Record them long. A one-minute ambience looped under a three-minute scene gives itself away, because the ear catches the same gull and the same car door on every pass. Aim for two to five minutes per take, nobody talking, with the microphone placed where the audience will imagine standing. Record in stereo if you have the means, and come back at a different hour if the place changes character. Name each file with the place and the time.

On a small shoot on the Sunshine Coast, I spent twenty minutes at the end of a dock at dawn while the rest of the crew loaded the van. That one recording ended up under four scenes, including two shot somewhere else entirely, which is a choice I come back to below.

Wild sound and wild lines

Wild sound is anything recorded without picture. The clunk of a particular truck's tailgate, a kettle on a particular stove, the tags jingling on the collar of the dog you filmed all morning. Grab them while the camera rests. They cost a few minutes on location and rescue scenes later, because a library effect never quite matches the truck your audience can see.

Wild lines are trickier. A wild line is a phrase someone repeats for the microphone after the sync take was spoiled, perhaps by a floatplane overhead or a radio in the next room. In fiction it is routine. In documentary it is a small reconstruction, and I treat it as one. My rule is that a wild line may only repeat what the person actually said, in their own words, with a full understanding of why you are asking. I use them for practical phrases, a name or a set of directions, and never for emotional moments. If someone said something raw while a bus went by, the bus stays, or the line goes.

Foley, and where honesty begins

Foley is sound performed to picture after the shoot: footsteps, the rustle of a coat, a cup set down on a table. It takes its name from Jack Foley, a Hollywood sound effects artist of the early talkie years. Documentaries use it more than audiences assume. Silent archive film needs something under it. Scenes shot on a long lens from across a field arrive with no usable sound at all. Nature films have long used foley for animal sounds that no microphone could have caught at that distance.

The ethical line, in my view, runs between foley that clarifies and foley that invents. Footsteps on gravel we can see someone walking on, at a level that matches the shot, tell the viewer nothing false. A door slam that never happened, a sob laid under a silent face, gunfire added to archive of a quiet street: those change what a viewer believes took place. The same test applies to the ambience I borrowed from that dock. Using it under a different dock is a craft decision. Using it to suggest a place you never went is a lie. The editing guide's section on the ethics of the cut makes the argument for picture, and the reasoning carries straight over to sound.

Dialogue editing, the invisible job

In the picture edit, audio usually follows video, so every cut is also a hard audio cut. The dialogue editor's job is to make those cuts disappear, and most of the work is unglamorous. Each microphone and each speaker goes onto its own track, alternated in a checkerboard so every clip can be treated separately. Clips get handles, extra seconds of audio beyond the cut, so crossfades have material to work with. Gaps get filled with the right room tone. Lav and boom get compared line by line, and the better one wins.

Then comes the cleaning: clicks, lip noises, a chair creak under a key sentence. Go gently with breaths. People breathe, and a voice with every breath removed sounds like a machine reading a script. When the picture editor hands over the cut, ask for an AAF or OMF export with generous handles. It saves days of rebuilding.

Noise reduction and its artefacts

Noise reduction tools are remarkable now. Broadband denoise, spectral repair, de-hum, de-reverb and the newer dialogue isolation tools can rescue recordings that once would have been thrown out. They also leave fingerprints. Push them too far and voices go watery and phasey, as if heard through a fishbowl. Sibilant consonants start to lisp. Background hiss turns into a faint chirping warble, sometimes called musical noise, which distracts more than the hiss it replaced. Heavy de-reverb can strip the room away so completely that the speaker seems pasted onto the picture.

A few habits keep this in check. Reduce rather than remove, since a little honest air sounds better than a vacuum. Several light passes beat one heavy one. Compare before and after at matched levels, since a change in level alone can fool your ears into hearing an improvement. Listen in context, with music and ambience underneath, where mild noise often vanishes on its own. And fix what you can at the source. Mains power in Canada runs at 60 hertz, so a buzz at 60 and its multiples usually points to a cable or grounding problem you could have found on the day.

The mix: dialogue first

A documentary mix starts with the voices. Bring every line of dialogue and narration to a consistent, comfortable level before a single music cue goes in. Then the ambiences and effects, then the music, which sits under the speech and gives way whenever someone talks. Music that fights a voice in the middle frequencies makes people reach for subtitles, and some simply stop listening.

Check the mix on bad speakers. Most of your audience will watch on laptops and phones, so play it on both, and on a cheap pair of earbuds, before you call it done. If the budget allows one sound expense, make it a day or two in a properly treated room with someone who mixes for a living.

Loudness for broadcast

Broadcasters set loudness specifications in their technical delivery requirements, and they check them. A spec usually names an integrated loudness target, measured across the whole film, and a true-peak ceiling the signal must never cross. Many broadcasters specify around minus 24 LKFS integrated, for example. You will also see LUFS; the two terms describe the same scale under different names. Some specs measure everything, some measure only the passages with speech, and festivals and online platforms set their own targets or none at all. Ask for the spec sheet before the mix, not after.

Mix toward the target with a loudness meter on the master from the first day. Turning down a finished mix that came out too loud will meet the number, but a film mixed hot and then pulled down rarely sounds as good as one built for its target. If you need versions for a cinema, a broadcaster and the web, make each one deliberately and label it clearly.

Stems: deliver the parts as well as the whole

A finished mix is one file. Stems are the same mix split into its parts: dialogue, narration, music and effects, each on its own track, which sum back to the full mix when played together. Make them even if nobody has asked. A broadcaster may need to trim two minutes. A music licence may run out and a cue may need replacing. A distributor abroad may want a dubbed version, which needs an M&E (music and effects, no dialogue) so local voices can be laid on top.

Keep the music cue sheet with the stems: every cue, its duration, how it was used, and who wrote and published it. Broadcasters ask for it, and rights organizations rely on it to pay composers. Then archive the whole mix session, because the version you never expected to need is the one somebody will request in three years.

A sound checklist

• Scout by ear: find the fridge, the furnace cycle, the clocks and the ventilation, and switch off what you can
• Room tone at every setup, at least sixty seconds, gear running, mics and gain unchanged, and again whenever the room changes
• Ambiences of two to five minutes, nobody talking, files named with place and time
• Wild sound of every object that matters in a scene
• Wild lines only as true repeats, explained to the speaker, and never for emotional moments
• Foley that clarifies what we see, never sound that invents an event
• Dialogue split by mic and speaker, with handles, gaps filled with matching tone
• Noise reduction in light passes, compared at matched levels, judged in context
• Dialogue set first, music under the voice, the mix checked on a phone and cheap earbuds
• The loudness spec in hand before the mix, and metered throughout
• Stems, an M&E, the music cue sheet and the archived session delivered with the master

Next time you walk into a room in November, stand still before you unpack anything. Wait for the furnace to cycle off, and listen to what is left. That faint hum, different in every house, is what the whole soundtrack will rest on.