Record an Ambient Location Bed for a Commercial Sound Sketch

A location bed is not a sound effect that got long. A door latch, a cart wheel, a coin dropping — those are objects. You can go find them one at a time and you will know when you have them. An ambience bed is a relationship: the thing the room does all the time, the things that interrupt it, how far away those interruptions sit, and how much time passes in between. Capture the relationship and the sketch sounds like somewhere. Capture a pile of good individual noises and it sounds like a library with the lights on.

So the working answer is short. Listen before you record and decide which relationship this commercial needs. Take a permitted position you can hold still. Verify what the recorder is actually getting, not what your ears are getting. Stay long enough for a repeating pattern to show itself. Then keep the untouched file and prepare any cleanup or looping on a separate copy, labeled as what it is.

That is the whole method. The rest of this is where the decisions actually get made.

One thing before we start: the laundromat below is constructed. It is a worked example of the method, not a session I ran. No location was recorded, no audio was heard, and no permission exists for it. Treat its numbers as stipulations, the way you would treat a diagram.

1 — Listen for what makes this a place

Stand in the space with the recorder off and try to answer four questions.

What is always there? That is the bed — ventilation, refrigeration, road hum, rain on a metal roof. It is usually boring, and it is usually the thing that makes the recording work.

What comes back? Not what happens once. What happens again, and roughly how often. A chime, a compressor cycling on the roof, a train.

What is far away? Distance is information. A sound two hundred meters off tells the listener the room has an outside.

What is the shape of the gaps? The quietest stretch of a place is often the part that identifies it, because it shows you the edges of everything else.

Then ask what the sketch needs. These are not the same question. A striking passing sound — a motorcycle, an argument, a good accident of timing — may be the most interesting thing in the room and completely wrong for the scene. If the director wants a working morning, a motorcycle says one person's morning and the room says everybody's. The brief tells you which relationships are load-bearing.

A public field-recording tutorial from Simon Fraser University's Sonic Studio handbook treats listening and recording as connected decisions — what you attend to shapes what you choose to capture, and the recording arrives with the context you documented around it. That is the useful part. It is not a claim that a recorder produces a neutral copy of reality; nothing does. The page's linked audio wasn't played for this piece, and its material on microphone arrays isn't relevant to a single fixed position, so neither is used here.

The constructed example. A fictional pitch wants the felt space of a small laundromat on a weekday morning, for a twenty-second spot. The fictional relationship is this: two exhaust fans running continuously; a washer drum humming under them; a spring door chime that arrives irregularly, four times in the fourteen minutes of the take; a rooftop unit on the building behind, cycling on for about ninety seconds roughly every six minutes; carts and coin drops in between; and one genuinely quiet stretch where nothing but the fans is audible. The chime is the event. The fans are the place. Neither is much without the other.

2 — Choose a safe, permitted recording position

Position is a compositional choice. Standing at the door, the chime and the street dominate and the room becomes a doorway. Standing against the back wall, the fans and the drum dominate and the room becomes an interior with weather outside it. Both are true. They are not the same recording.

Pick the position that matches the perspective the sketch is listening from, and then hold it. If you move the microphone between the quiet stretch and the chime, you have made two recordings and glued them. The comparison you were building is gone.

Check permission before you set anything down. This is not paperwork in the way of art; it is the difference between a file you can use and a file you can't. Being audible in public does not make identifiable speech an automatic asset for a pitch. If someone's conversation is intelligible in your take, that is a thing you have to decide about deliberately, with whatever rules apply where you are — and those rules are yours to check. A tutorial about technique doesn't settle them, and neither does this piece.

Write down the conditions while you still remember them: time, weather, wind, which machines were running, how many people came through, how far the microphone sat from the door and from the machines, and anything that interrupted you. Twenty seconds of notes saves a take two weeks later.

And keep the exercise small. This does not require trespass, an argument with a manager, or a new microphone. A phone and a pair of earbuds will demonstrate every decision in this article. If the shopping is doing the work, the listening isn't.

3 — Check the signal that is actually being captured

Put the headphones on. Feed the room into them and listen to what the recorder thinks it heard, which is not the same as what you heard standing there. Rooms are forgiving. Cards are not.

Run a short check — thirty seconds is plenty — and then play it back. Handling noise, wind working the capsule, a hum you didn't notice because you'd stopped noticing it, a level that is generally too hot or too cold: these are what a short check reliably reveals, and none of them show up in your memory of standing there. A loud event is a different matter. The chime arrives irregularly, so a short check will usually contain none, and catching one is luck. The check's job there is headroom: set the level so the loudest thing you expect from the place lands well below the top, and it will do so whenever it finally arrives — whether or not it arrived while you were listening.

In the constructed example, the first check is forty seconds, handheld, at the doorway. It comes back with the hand on the body of the recorder, a wind gust lifting the low end, and the chime hitting the top of the meter — the first two are what a check is for; the third is luck, since the chime keeps its own schedule. Nothing needs editing to fix that. It needs a stand, a position away from the door, and enough headroom that the chime lands well below the top.

Fix what a change of position will fix. Moving forty centimeters is cheaper than every hour you will spend afterward trying to make a peak disappear. Some problems genuinely belong to post; a badly chosen position is not one of them.

The buttons and menus differ from device to device, and there is no universal gain recipe worth printing here. Your manual covers the specifics. What transfers is the sequence: wear headphones, record a check, listen to the check, correct, then commit.

4 — Hold the perspective long enough to hear the pattern

Duration comes from the pattern, not from a standard. There is no correct ambience length. There is a length at which the thing that makes this place this place has happened often enough to be heard as recurrence rather than coincidence.

For the constructed laundromat, that is fourteen minutes: two rooftop onsets about six minutes apart, four chimes, and one full quiet stretch. Two onsets establish that the roof unit comes back. They do not establish the period with any confidence, and the honest note in the log says so — "two onsets, interval roughly six minutes, unconfirmed." If the sketch really needed the cycle, the right move is a second take, clearly labeled as a second take, not a sentence in the notes pretending the first one caught it.

Resist chasing. The moment you start moving the microphone toward each interesting event, you have stopped recording a place and started hunting effects. Note the events instead, with timestamps. The log is where you keep the motorcycle.

Do not cut anything out of the only copy. If someone walks through the middle of your best stretch, that is an interruption to log, and possibly a reason to record again, but the master stays whole. You can always decide later; you can't decide earlier.

In the example, the take runs 0:00 to 14:00. Chimes at 1:20, 3:55, 8:40 and 11:05. Rooftop onsets at 4:10 and 10:20. A quiet stretch from about 2:30 to 3:10 with nothing but the fans. A customer's exchange with the attendant, audible and identifiable, around 9:15.

5 — Prepare a derivative without rewriting the source's status

Two files leave this process, plus a log.

The master is everything, uncut, with the log beside it: position, time, conditions, interruptions, and the stretches you have decided not to use and why. It stays unprocessed. You will want it the day the director wants a different ninety seconds.

The derivative is a copy. On the copy you make listening decisions.

The central judgment is telling an artifact apart from an event. An artifact is something the recording apparatus did. An event is something the place did. The two seconds at the very start of the example where the stand settles against the table are apparatus — they are about your setup, not about the room, and if you are excerpting from the head of the take, trimming them costs you nothing.

The road rumble under the fan bed is not an artifact. It is the place, and this is where restraint matters. A light low-cut will thin the rumble, and it will also take weight out of the fans, because the fans live down there too. Apply too much and the room gets smaller than it is. The handbook page carries a related caution: removing unwanted artifacts can disturb the continuity of what's underneath. That is a listening instruction rather than a promise of transparent repair. Use it to set up an A/B — the excerpt, then the treated copy — and listen specifically for whether the bed's texture has changed or a join has become audible where the processing stops.

In the example, the ninety seconds from 3:50 to 5:20 contain the fan bed throughout, the chime at 3:55 and the rooftop onset at 4:10. The untreated excerpt keeps a slightly congested low end and sounds like a room with a street near it. The lightly prepared copy is clearer and, if pushed, thinner. Whether the trade is worth it depends on whether the sketch's voiceover has room to sit — which is another reason to check the derivative inside the sketch, not only on its own. A bed that sounds excellent alone can vanish under a music cue, or crowd it.

Then there is the loop.

A loop makes a claim the place never made. Someone on the example proposes a forty-five-second bed built by looping 3:50 to 4:35, which contains the chime at 3:55 and the rooftop onset at 4:10. Loop it and the room chimes every forty-five seconds and the roof unit kicks on every forty-five seconds. The actual intervals were roughly three minutes and roughly six. The loop doesn't sound wrong in isolation — that's the problem. It sounds like a place with a strange, insistent regularity that no place has.

If the sketch needs a loop, build it and label it: constructed loop from 3:50–4:35, events repeat at 45s intervals. Anyone working downstream can then decide whether the regularity helps or whether they want the continuous excerpt. Unlabeled, it just looks like what the laundromat did.

What you should have at the end

Three things. An unprocessed recording with a log that says where you stood, when, in what conditions, and what interrupted you. A labeled derivative that isolates the relationship the sketch needs. And a clear note about anything you decided not to use — including any identifiable speech, which stays out of the pitch material and in the log until someone with standing over the question says otherwise.

The test is simple. Play the derivative for someone who wasn't there and ask them what kind of room it is and what time of day it is. If they can tell you the fans never stop and something outside comes back now and then, the bed is doing its job. If they say "nice atmosphere," you cleaned away the reason you recorded it.

Frequently asked questions

What makes an ambience bed different from a long sound effect?

A door latch, a cart wheel or a coin dropping is an object — you can go find those one at a time and know when you have them. An ambience bed is a relationship: the thing the room does all the time, the things that interrupt it, how far away those interruptions sit, and how much time passes in between. Capture the relationship and the sketch sounds like somewhere. Capture a pile of good individual noises and it sounds like a library with the lights on.

How do I decide what to record?

Listen with the recorder off. What is always there — that's the bed, usually boring and usually the thing that makes the recording work. What comes back, and roughly how often. What is far away, because distance is information: a sound two hundred meters off tells the listener the room has an outside. What is the shape of the gaps, since the quietest stretch often identifies a place. Then ask what the sketch needs, which is not the same question. A striking passing sound may be the most interesting thing in the room and completely wrong for the scene — a motorcycle says one person's morning while the room says everybody's. The brief tells you which relationships are load-bearing.

What does a short headphone check actually reveal, and where does it stop?

Thirty seconds is plenty, then play it back. Handling noise, wind working the capsule, a hum you'd stopped noticing, a level too hot or too cold — these are what a short check reliably reveals, and none of them show up in your memory of standing there. A loud event is a different matter. An irregular sound like a door chime will usually not appear in a short check at all, so the check's job there is headroom: set the level so the loudest thing you expect from the place lands well below the top, and it will whenever it finally arrives. Fix what a change of position will fix, because moving forty centimeters is cheaper than the hours spent making a peak disappear later. Some problems genuinely belong to post; a badly chosen position is not one of them.

How long should I record, and what's the risk in looping an excerpt?

There's no correct ambience length. There's a length at which what makes this place this place has happened often enough to be heard as recurrence rather than coincidence. The constructed example runs fourteen minutes, giving two rooftop onsets about six minutes apart, four chimes and one full quiet stretch — but two onsets establish that the unit comes back, not the period, so the honest log entry reads "two onsets, interval roughly six minutes, unconfirmed," and if the cycle really mattered the move is a second take clearly labeled as one. Looping is the other time trap: looping 3:50–4:35 makes the room chime and the roof unit kick on every forty-five seconds when the actual intervals were roughly three and roughly six minutes. It doesn't sound wrong in isolation, which is the problem — it sounds like a place with a regularity no place has. Build the loop if the sketch needs it, but label it, as in "constructed loop from 3:50–4:35, events repeat at 45s intervals."

Is the laundromat session real, and how should identifiable speech be handled?

The laundromat is constructed — a worked example of the method, not a session anyone ran. No location was recorded, no audio was heard, and no permission exists for it; its numbers are stipulations, the way a diagram's would be. On speech: check permission before you set anything down, because being audible in public does not make identifiable speech an automatic asset for a pitch. If someone's conversation is intelligible in your take, that's a thing to decide deliberately under whatever rules apply where you are — and those rules are yours to check; a technique tutorial doesn't settle them, and neither does the article. In the example the audible exchange mid-take stays out of the pitch material and in the log, and the article also notes that the linked audio on the referenced field-recording handbook wasn't played for the piece and its microphone-array material isn't relevant to a single fixed position.