The Room, Not the Microphone

Beerends and de Caluwe, "The Influence of Video Quality on Perceived Audio Quality and Vice Versa," Journal of the Audio Engineering Society 47(5), 1999, states its result in one line: "the video quality dominates the overall perceived audiovisual quality in nonconversational experiments."

The two effects that paper measured are not the same size, and the larger of them runs in the direction almost nobody quotes. In the 1999 experiments, video quality contributed about 1.2 points on a nine-point scale to how good the audio was judged to be. The effect running the other way, audio quality moving the judgement of the picture, was about 0.2 points in the same 1999 work. Degrade the picture and listeners marked the sound down. Degrade the sound and viewers barely moved the picture.

Three facts travel with that citation every time this book uses it, and they are printed here rather than left for a hostile reader to find. It was published in 1999, which makes it 27 years old as of

1. It measured codec artefacts, the compression damage of the late 1990s, on short test items in

a laboratory. And it is not about YouTube, which did not exist when the experiments were run. It is cited because it is the nearest real anchor anybody has produced on the question, not because it settles it.

It also cannot be turned into a share. A mean opinion score on a nine-point scale, moved 1.2 points by one variable and 0.2 by another in the 1999 study, is a measurement of two effect sizes in one experiment. It is not a percentage of anything, and there is no arithmetic that converts it into one.

The claim this book has to answer says the opposite, says it as a percentage, and attributes it to a literature. In the version that circulates on microphone product pages, in gear-review video descriptions and in the opening two minutes of a great many channel-starting tutorials, audio is seventy to eighty per cent of the viewing experience, and studies show it.

No study says it. This volume's research pass, closed 5 September 2026, located no paper, no dataset, no instrument and no author behind that figure. What it located instead was the claim sitting on pages that sell microphones.

That is a microphone manufacturer's blog post, "Video vs. audio — what's more important?", retrieved 5 September 2026. The page asserts a research literature and then cites none of it: zero studies, zero sample sizes and zero dates, on the page as read on 5 September 2026. The research it gestures at is not named, so it cannot be checked, and the nearest paper a reader can actually open is the 1999 one printed at the top of this chapter, which measured the asymmetry running the other way.

The percentage travels because a percentage is portable and a craft ordering is not. A sentence telling somebody to move the microphone closer before buying a camera is about geometry, it ends the conversation, and it sells nothing. A figure that assigns audio a share of the viewing experience fits under a product photograph, survives being repeated by people who have read none of the underlying work, and converts a sequencing rule into a division of something that was never divided. The 1999 paper produced two effect sizes on a nine-point scale. It did not produce a split, and a split is what the claim needs before it can be quoted.

One further paper is normally produced alongside this debunk, and it is not cited in this volume. Perceptual Video Quality Assessment: A Survey, submitted 5 February 2024, is a methodological survey of video quality assessment; it does not report that the video modality is weighted above audio in joint quality-of-experience studies, and audio-visual quality assessment appears in it as one listed application area among several. It was fetched and read on 5 September 2026 and dropped. A book that quietly deletes a citation that failed is doing the thing this one is complaining about, so the deletion is printed.

What survives all of that is the advice, and it survives on different grounds from the ones it is usually sold on. The grounds are asymmetric repair. A blown highlight and a cool white balance are recoverable to some degree at the desk, and a room recorded into a microphone from three feet away is not: the reverberation is baked into the same waveform as the voice, arriving a few milliseconds behind it, and no plug-in unmixes them cleanly. Fix audio first is a scheduling rule about which mistakes are cheap to undo. It is not an evidence claim, and this chapter is not going to pretend it has become one. Whether the published work on speech intelligibility and listening effort supports the rule specifically for talking-head video is something this volume did not establish, because no such study was retrieved in the research pass closed 5 September 2026. As of September 2026, this was unresolved.

What would close that gap is not exotic and it is not expensive. One production, cut twice, with the audio degraded in one version and the picture degraded in the other by amounts a panel has rated as equivalent, shown to viewers who do not know which arm they are in, scored on watch time as well as on a rating scale, with the sample size, the degradation levels and the field dates published alongside the result. That is an ordinary experiment. It would cost a fraction of what the gear discourse it would settle turns over in a week, and as of September 2026 nobody had published it.

The other reason the advice holds has nothing to do with perception research. Audio is where the cheap failures are, and they are failures of geometry rather than of equipment.

A microphone in a small untreated room is picking up two things at once: the voice arriving directly, and the same voice arriving a fraction of a second later off the walls, the ceiling, the desk and the window. The ratio between those two is what a listener hears as "in a room" or "in a bathroom", and it is set almost entirely by how far the microphone is from the mouth. Distance governs the first term. It does nothing to the second. That is the whole mechanism, and it is why the most consequential variable in an amateur recording chain is not on the invoice at all.

Microphones fail differently, and the failure modes sort by type rather than by price.

The cheap USB condenser is the standard first purchase and the standard first disaster. A condenser capsule is sensitive by design, and the desktop stands these ship with put it twelve to eighteen inches away, which is far enough that the room arrives at nearly the same level as the voice. What it records is the whole space: the air handling, the refrigerator two rooms away, the keyboard, and every first-order reflection off bare drywall. Multi-pattern models make it worse, because a switch left in omnidirectional or stereo is a microphone told to listen to everything. This is the origin of the bathroom sound, and no amount of processing afterwards removes it.

The dynamic microphone is the correct answer in an untreated room, and it is correct for an unglamorous reason. It is less sensitive and its cardioid pattern rejects what is behind it, so at two to six inches from the mouth the voice is enormous and the room is not. Its failure mode is the distance requirement itself. A dynamic used at eighteen inches is a bad condenser. It also needs gain, and the gain has to be quiet, which is the mechanism behind an entire accessory market and behind at least one manufacturer eventually building the preamp into the microphone.

The lavalier solves distance by attaching to the speaker, and in solving distance it solves the room. It is the only microphone in this list that is close by construction rather than by discipline. Its failure modes are all physical and all recoverable only by reshooting: cloth rustle against the capsule, placement too low on the chest, which produces a dull chesty tone, a transmitter nobody switched on, and one channel dying mid-take with no second recording anywhere. That last one is why onboard recording on the transmitter matters more than any tonal characteristic in this category. A wireless link is a radio path across a room with a metal desk, a router and a phone in it, and the recording written to the transmitter is the only copy that never travelled over that path. A take is not lost when the link drops. It is lost when the link drops and there was nothing writing locally.

The shotgun indoors is the misuse this hobby repeats most often, and it is worth stating mechanically rather than as a preference. An interference-tube microphone is directional because off-axis sound arrives at the slots along the tube out of phase and partially cancels. Reflections in a small room do not arrive off-axis. They arrive from in front, having bounced, so the tube passes them, and what comes back is a hollow comb-filtered tone that is harder to fix than plain room sound. A shotgun on a camera hot shoe three feet from a presenter in a bedroom is the specific configuration that produces it.

Six fixes, ranked by effect per dollar, with the two prices checked in US pricing on 5 September 2026 and the acoustics stated as mechanism rather than as measurement, because no controlled test of any of them on a talking-head recording has been published by anyone:

The first line is free and beats everything under it. Sound from a point source falls off with distance, and each halving of the distance raises the direct level by about 6 dB while the reverberant field in the room stays where it is. Twenty-four inches to six inches is two halvings, so roughly 12 dB of improvement in the ratio that decides whether a recording sounds close or sounds like a room. Nothing on a price list does that, and it costs a microphone stand being moved.

The second line is free and nobody does it. A refrigerator compressor and a furnace fan are continuous, broadband and low, which is exactly the profile that noise reduction handles worst without taking the bottom out of a voice. Switching them off for a twenty-minute take costs nothing and cannot be done afterwards.

Moving blankets, at roughly $20 to $30 each in US pricing checked on 5 September 2026, work and they work in a limited band. They are thick enough to absorb high-frequency reflections and not thick enough to do anything below roughly 500 Hz, which is where a small room's worst behaviour lives. Two of them, on the wall the presenter faces and the wall behind the presenter, address the two earliest and loudest reflections. That is the honest description of what they buy.

Rockwool or rigid fibreglass in a fabric frame, at roughly $20 to $40 per panel in US pricing checked on 5 September 2026, is the same idea with the thickness the physics requires, and it absorbs broadband rather than only at the top. Leave some wall bare. A room absorbed everywhere sounds unnatural on a voice, and the target is a controlled room rather than a dead one.

Thin egg-crate acoustic foam is the correction this chapter exists to make, because it is the product most creators buy first and it is the one that can leave a room measurably worse for the purpose. Absorption depends on thickness relative to wavelength. An inch or two of open-cell foam absorbs efficiently above roughly 1 kHz and progressively less below it, so it takes the top off the room while leaving the low-mid energy that produces boxiness entirely intact. The result is a room that has lost its brightness and kept its boom, which is a worse starting point for a voice than the untreated wall was. This is checkable in ninety seconds against any published absorption coefficient chart for the material, and it does not require trusting this book.

The reason the cheap end of the table stops at roughly 500 Hz is a wavelength problem and it is the same problem in every small room. A porous absorber works on the air velocity of a passing wave, and velocity is highest a quarter of a wavelength from a hard surface. At 2 kHz a quarter wavelength is under two inches, so a blanket on a wall sits where the energy is. At 200 Hz it is around seventeen inches, so the blanket sits in the wrong place and the wave passes through it largely intact. That is why the fixes that are cheap are also the fixes that only work at the top, why the boxiness in a bedroom recording is the last thing to go, and why the honest small-room answer is to get the microphone close enough that the room contributes less rather than to try to remove the room.

Then the distinction that no product page draws, because drawing it would cost a sale. Absorption is not transmission loss. Every item in the table above changes what happens to sound inside the room. None of them changes what arrives from outside it. Blocking sound requires mass, decoupling and sealed air gaps, and it is a construction job rather than a purchase. Nothing at hobby price stops the neighbour's leaf blower, the aircraft, the road, or the person upstairs. Creators buy foam expecting that, and the disappointment is structural rather than a matter of having bought the wrong foam.

The ranking above is mechanism, and the ranking is the part with no evidence under it. The acoustics are standard and the ordering follows from them, but no published test measures what any of these six changes does to a viewer's judgement of a talking-head video, and none of the craft sources this volume read publishes a measurement of its own. What the chapter can defend is the physics and the cost. What it cannot offer is an effect size.

Light is the other half of a room, and the doctrine is older than the platform and cheaper than the gear discourse admits.

The standard arrangement is three positions, and only the first is mandatory. The key sits about 45 degrees off the lens axis and slightly above eye level, which puts the shadow of the nose on the cheek rather than across the mouth and gives a face some modelling. The fill sits opposite, at 50% to 70% of the key's brightness, per The Post Flow's key-light guide as updated 9 August 2026, which is a trade craft source that publishes no measurement of its own. Fill below that range leaves a face contrasty in a way that reads as harsh on a small screen; fill at parity with the key removes the modelling the key was placed to create. The third position, a light behind the subject aimed at the head and shoulders, exists to separate the subject from the wall and it is the one that can be dropped when the wall is already darker than the face.

The reason budget lighting disappoints is arithmetic, and it is the same arithmetic in every case. Diffusion costs two or more stops, per The Post Flow's key-light guide as updated 9 August 2026. Two stops is a quarter of the light. A 26-watt edge-lit panel pushed through a softbox is therefore delivering something closer to a nightlight at any working distance, and the response most people have is to move it closer, which changes the framing, or to raise the camera's sensitivity, which adds noise the codec then has to spend bits on. There are two honest ways out. Use the panel bare and near, where it is soft because it is large relative to the face and close to it. Or buy enough raw output that it survives the modifier, which is what the higher-wattage fixtures in a gear list are actually for.

The three fixture types answer that arithmetic differently and they are bought for different reasons. An edge-lit panel is already a large emitting surface, so it is soft at close range without a modifier and it plugs in and works at desk scale. A COB fixture puts a single bright emitter behind a mount and is far brighter, but bare it is a hard point source and it needs the modifier the extra output exists to pay for. A softbox kit on stands gives the largest soft source per dollar and asks for floor space most rooms with a desk in them do not have. None of the three is better. They are three answers to whether the binding constraint is money or floor space.

The cheapest visible upgrade after the microphone is not a fixture at all. A lamp inside the frame, a strip behind a shelf, a monitor glowing off-camera: practicals do not light a face and do not replace a key. They put something in the background other than a wall, which is the part of the frame the key was never aimed at, and it is the reason a shot reads as arranged rather than as a room with a light pointed at it.

There is a free version of the same physics, and it sits in most rooms already. A window is an enormous soft source. Facing one at about 45 degrees from three to five feet away is a key light with better colour rendering than anything at this price, and a sheet of white foamcore on the shadow side is the fill. Its failure modes are that it moves across the day, it changes colour temperature with the weather, and it makes batch filming across an afternoon inconsistent — which is a scheduling constraint rather than an image-quality one, and it is worth knowing before it produces four videos that do not match.

The two halves of this chapter meet at one component. High-wattage COB fixtures are actively cooled, and a fan two feet from a presenter is an item in the noise floor list along with the refrigerator and the air handling. A light chosen without listening to it is a microphone problem bought with a lighting budget, and it is the one purchase in this room that can undo the free fix at the top of the table.

Move the microphone to six inches. Switch off the refrigerator and the air handling. Record sixty seconds of ordinary speech, saved under today's date, and play it against the same sixty seconds from last week's file. If the difference is audible on the laptop speakers that recording will mostly be heard on, the rest of this chapter's price list is optional.