Spatial Audio: Three Systems Wearing One Name
Home
Yiwu, Zhejiang, China
Mon – Fri 9:00 AM – 6:00 PM · Sat 9:00 AM – 4:00 PM
[email protected]

What Is Spatial Audio: Three Systems Wearing One Name

Alan Xiao, Founder of Sourcing Pioneer
Alan Xiao Founder of Sourcing Pioneer
· Updated November 11, 2026 · 10 min read

Spatial audio puts sound around you. It is a result, not a method.

Three different systems can produce that result, and each one needs different content, different hardware and a different licence.

That is why the phrase on a specification sheet tells you almost nothing on its own. A phone, a soundbar and a pair of earbuds can all print “spatial audio” and do three unrelated things.

This article separates the three systems, then shows what the claim usually means and how to test one in ten minutes.

Studio still life of over-ear headphones and a small speaker on a beige surface, with a soft shadow
One word on the box, three different machines inside.

The Short Answer

Spatial audio is any processing that makes a listener hear a sound at a place. Not inside a pair of speakers, and not inside their own head.

The cues are place and distance. A sound can come from the left, from behind, from above, or from the next room. Ordinary stereo only places it along a line between two speakers.

What changes is not the number of channels. It is whether the system knows where each sound is supposed to be, or is only guessing from what is in the file.

That distinction is the whole article. A system with position data can place a sound behind you. A system without it can only widen what is already there.

Three Systems Wearing One Name

The first system carries each sound as an object with coordinates. The second carries a scene and decodes it later. The third carries two channels and filters them for your ears.

Object audio. Each sound arrives with a position, and a renderer places it in whatever speaker layout the room has. This is the model behind Dolby Atmos.

The file holds 128 tracks with metadata for position and loudness over time. A bed of channels carries ambience and dialogue, and the rest are free-moving objects.

Because the mix is rendered at playback, a cinema, a soundbar and a phone can all play the same file and place the objects differently. That is the strength of the model and the source of the confusion.

Scene audio. Ambisonics does not send speaker feeds at all. It sends B-format, a speaker-independent picture of the sound field, which is decoded to whatever speakers the listener owns.

The omni channel in that format is scaled down by about 3 dB so it carries similar energy to the others. A with-height scene still plays on plain stereo, folded down without losing the content.

The trade is resolution. A first-order scene is described as slightly blurry, with a small sweet spot, and higher orders add channels to sharpen it.

Binaural filter. The third system takes ordinary two-channel audio and filters it so each ear receives what it would have received in a real room. No position data is involved beyond what the mixer imagined.

This is the one most consumer products mean. It is cheap and it works on any stereo file. It is also the only one of the three that can be added to content never mixed for it.

SystemWhat travels in the fileWhat the player needsWorks on an old stereo mix?
Object audioTracks plus position dataA renderer and a licenceNo
Scene audioB-format, speaker independentA decoder for your layoutNo
Binaural filterTwo ordinary channelsA filter and a head modelYes, with limits

Nearly every argument about spatial audio is one person describing the first row and another describing the third.

Your ears are two, but the cue set is not. The head blocks high frequencies on the far side, and the arrival time differs between ears.

Those two cues define a cone rather than a point. Anywhere on that cone, the time and level differences are the same, so the brain cannot tell front from back on those cues alone.

The pinna solves it. Its folds reflect the sound a fraction of a second late, and the pattern of notches depends on where the sound came from.

The full filter, from free air to eardrum, is called the head-related transfer function. It is not a gentle curve.

It generally lifts the band between 2 and 5 kHz, with a primary resonance of about +17 dB near 2.7 kHz. That peak is one of the main elevation cues you own.

Diagram of the cone of confusion, with identical arrival time and level for many directions, and the pinna notch that breaks the tie
The lift near 2.7 kHz is an average. The notch is what breaks the tie.

Here is the part that matters for a buyer. That filter is different for every head, every ear and every pair of shoulders.

A product therefore ships with an average head. It is measured on a dummy, usually in an anechoic room, in 15 or 30 degree steps, with the gaps filled by interpolation.

Interpolation is where front and back get confused. The published description of the filter says so directly: even at small steps, filling in between measurements can produce a front-back error.

Two listeners with the same earbuds will not hear the same result. One gets a stage in front. The other gets a sound that flickers behind the head.

There is a consolation. Substituting another person’s ears blocks localization at first, and the auditory system adapts over some weeks.

That is why a bad first impression is not always the final one. A ten-minute demo is a fair test of the effect, but a poor test of your own fit.

Head Tracking, And What It Buys

A fixed stage and a moving head are two different experiences. Head tracking decides which one you get.

Without it, the stage is glued to your ears. Turn your head and the whole scene turns with you, because the filter never learns that you moved.

With it, a small sensor reports the turn, and the renderer rotates the scene the other way. The sound stays where the mixer put it, and your head moves through it.

That is what makes a virtual screen and a virtual speaker feel external. It is also why the effect is easier to demonstrate than to describe.

Three things can go wrong, and all three show up as the same complaint. The stage drifts, the stage lags, or the stage jitters.

Drift is a sensor that has lost its reference over minutes. Lag is a loop that is too slow, so the scene arrives after the head. Jitter is a filter that changes gain as it rotates.

A good implementation hides all three by re-centring when the head is still. A poor one asks the listener to press a button, which is the tell.

There is also a design choice behind the tracking. Some products lock the stage to the room. Others follow the head and keep only a hint of movement.

Ask which one the product does. “Spatial audio with head tracking” can mean either, and the two feel nothing alike.

How Object Audio Actually Ships

The model is simple. The shipping details are not, and they are what a buyer usually gets wrong.

In the cinema, the format this model comes from carries 128 tracks. A bed of 9.1 channels holds ambience and centre dialogue, which leaves 118 tracks free for objects.

At home the file is squeezed. A technique called spatial coding reduces the same mix to at most 16 concurrent elements, grouped and re-grouped as the content moves.

Games sit between the two. A spatial format used there carries 32 active objects on a 7.1.4 bed, with 20 of them dynamic.

So the number on the box is not the number in the file. A phone playing the same film is rendering at most 16 elements, whatever the studio mixed.

The speaker count is written as three numbers. The last one counts overhead or upward-firing speakers, so a 7.1.4 system is a 7.1 layout with four of them.

The range runs from 3.1.2 at the simple end to 24.1.10 at the largest. A soundbar with two up-firing modules is the small end of the same system.

Up-firing modules work by reflection and have a placement rule. They belong within about 0.9 m of the speaker they sit on, or the reflected image lands in the wrong place.

Under all of this sits the ordinary copper question. The cable and connector between the box and the speaker still has to carry the channels the renderer produces.

Diagram showing a mix of bed channels and position-coded objects entering a renderer that adapts to three different speaker layouts
The same positions, landing in four different speaker counts.

The Content Dependency

This is the sentence most product pages leave out. Spatial audio needs a mix that was made for it.

An object mix carries positions. A scene mix carries a sound field. Both are decisions taken in the studio, and neither can be recovered later from a stereo master.

What a binaural filter can do to a stereo master is guess. It can widen, externalize and add a sense of room. It cannot know that the helicopter was meant to pass behind you.

The industry has tried the shortcut. One well-known television series upmixed its existing 5.1 mix for a disc reissue rather than remixing from the stems.

That is the honest precedent. Upmixing is a processing step applied to finished audio, and it is not the same as a mix made with positions.

So the question for any product is what it does with what you feed it. A renderer needs object metadata. A decoder needs a scene. A filter needs nothing and promises less.

The transport matters here too. The radio link that carries the audio to a wireless headset has to pass the encoded stream, not a re-encoded stereo copy of it.

That is a common failure in cheap hardware. The phone holds an object mix, the link re-encodes it to stereo, and the headphones apply a filter to the stereo. The result is spatial in name.

Ask for the signal path, not the badge. Which format arrives at the headphones, and where in that chain does the object data stop?

Five Questions For A Spec Sheet

A specification sheet can answer all of this in five lines. If a supplier cannot, the badge is decoration.

One. Which of the three systems does it implement? Object, scene or filter. The answer decides everything below it.

Two. Does the device decode the stream, or does it receive audio that was already rendered? A decoder can use its own head model. A receiver cannot.

Three. Does the position data survive the link? Name the codec at each hop, from the source to the driver.

Four. Is there head tracking, and what happens when it drifts? A product that re-centres by itself is a different product from one that asks for a button press.

Five. What happens on hardware the buyer already owns? Every system has a downmix, and the downmix is where a demo room turns into a living room.

Two more lines belong next to those. The choice of radio for the link sets the bandwidth available for the stream. And the water and dust rating of a wearable speaker decides whether the product survives the gym that the marketing photographs.

All of it belongs in writing. A specification sheet the factory has to build against turns a badge into a testable line item.

Ten Minutes With A Sample

You do not need a lab to separate a real render from a stereo widener. Five checks cover it, and a sample is enough.

Behind. Play a clip with a known sound behind the listener, such as a door or a voice at the back. Ask where it came from.

A real render places it behind. A widener smears it to both sides, and the listener reports “wider” rather than “behind”.

Above. Do the same with rain or a passing aircraft. Height is the hardest cue for a filter to fake, because it depends on the pinna notch.

Turn. Rotate the head about 45 degrees and hold. The stage should stay in the room. If it follows the head, there is no tracking.

Toggle. Switch the feature off and on. A real render changes where sounds are. A widener changes how loud and how bright they are.

Mono. Sum the output to one channel. A badly filtered signal goes hollow or thin, which tells you the processing is not phase-safe.

Hold. If the stage drifts from the correct position after five minutes of sitting still, the tracking loop is not re-centring.

That is the whole test. It takes ten minutes, needs one sample and one familiar clip, and it will not be fooled by a badge.

Do it before the order, not after. A speaker programme that cannot pass these five checks in a listening room will not pass them in a customer’s front room either.

What The Claim Usually Means

Most of the confusion is in the wording, so here is the translation table.

What it saysWhat it usually meansWhat to ask
“Spatial audio” on earbudsA filter applied to two channelsDoes any position data reach the filter?
“Atmos” on a phoneThe same file, reduced to 16 elementsWhich profile, and does it decode or receive?
“360” or “immersive”A scene or object mix per trackWhich tracks were mixed, and how many?
“Immersive soundbar”Up-firing modules plus virtualisationWhere must the modules sit, and how high is the ceiling?
“Works with any content”Upmixing, not placingWhich processing, and on what source?
“Cinema-grade”A different object budget from the one shippedHow many objects in the file I would receive?

None of these lines is a lie. Every one of them is a shorthand that hides the number the buyer needs.

The habit worth building is small. When a deck says spatial, ask which of the three systems, and then ask what content it needs.

Those two answers decide whether the product is a renderer, a decoder or a filter. Everything else on the page is a consequence of that.

CLAIM CHECK

Send me the product deck and the specification sheet for a spatial audio product you are considering. I will tell you which of the three systems it actually implements, whether the position data survives to the driver, and which sentences on the page are doing no work. No cost, no obligation.

Read My Spec Sheet

Questions Buyers Ask

Is spatial audio the same as surround sound?

No. Surround sound is a channel count, and a 7.1 system has channels but no height. Spatial audio is about placement, which a channel layout can carry but does not have to.

Do I need special headphones?

Not for the filter, which runs on ordinary stereo. Head tracking needs sensors in the headset, and some object formats need a licensed decoder in the product.

Does it work on any song?

Only through filtering or upmixing, which widens and externalizes. It cannot place a sound that the mix never placed.

Is it the same as those eight-dimensional audio clips?

No. Those are a panning effect written into a stereo file by the creator. There is no renderer and no position data, so the movement is fixed before you play it.

Will it work on my existing stereo speakers?

Partly. A scene folds down and an object mix is re-rendered, but with two speakers in front there is no height and very little behind. The effect is smaller than the deck suggests.

Why does it sound like it is behind me when it should be in front?

Front-back confusion, from a head model that is not yours. The time and level cues point at a cone, not a point, and the filter’s ear shape decides the direction. A different product, or a calibration step, often fixes it.

Questions & Comments

We read every one and reply within 24 hours

If a product deck says spatial audio, send it over. I will name which of the three systems it is, and which sentences on the page are doing no work.

Subscribe
Notify of
guest
0 Comments
Most Voted
Newest Oldest
0
Would love your thoughts, please comment.x
()
x