Beds, objects and metadata

01 · of 9

What is a bed?

A bed is channel-based audio: mixed to a fixed speaker layout, with one channel for each speaker.

A 7.1.2 bed in an ADM master: the two heights sit overhead at the sides. IAMF’s 7.1.2 moves them towards the front (dashed).

Channel-based audio is the kind you already make. Each channel belongs to one speaker at a fixed place.

In an immersive mix, that fixed part is called the bed. The ITU uses the word in its own examples, for an ambience.

Why keep a bed at all? The ITU says the hybrid way of working keeps the existing channel workflow. Some sounds go into channels; others stay separate, as objects.

The ADM calls channel-based audio DirectSpeakers: each channel feeds a speaker directly.

Read the name to count it. 7.1.2 is seven at ear level, one LFE and two overhead: ten channels. Speaker layouts covers where they go.

SoA bed is the stem mix you already know, carried inside the immersive one. A 7.1.2 bed is ten channels.

02 · of 9

What is the bed in a Dolby Atmos ADM file?

Ten channels at most, on tracks 1 to 10, in a fixed order. The file places a 7.1.2 bed’s two heights overhead, level with the side surrounds.

Dolby’s tools fill the first tracks with the bed. The EBU records their profile as allowing:

“bed configurations up to and including 7.1.2 (tracks 1-10)”
EBU Tech 3393, annex A.2.1

Smaller beds are allowed too: 2.0, 3.0, 5.0, 5.1, 7.0, 7.1, 7.0.2 and 7.1.2. There is no 7.1.4 bed. Height beyond the bed comes from objects.

Track	Channel	Where it is
1	L	Front left
2	R	Front right
3	C	Centre
4	LFE	Low-frequency effects, no set position
5	Lss	Left side surround
6	Rss	Right side surround
7	Lrs	Left rear surround
8	Rrs	Right rear surround
9	Lts	Left top surround, overhead
10	Rts	Right top surround, overhead

Where are the two heights?

Dolby’s tools and files call them Left and Right Top Surround: Lts and Rts. Its home speaker guide calls the pair top middle.

The file places them at the far left and right, halfway between front and back, at the top. That is overhead, level with the side surrounds.

The ITU’s common-definition 7.1.2 puts its top side pair at the same point. IAMF is the odd one out: its 7.1.2 takes the top front pair from 7.1.4.

So one name, 7.1.2, can mean heights at the sides or in front. Check whose 7.1.2 you have before you judge a mix.

A master can carry more than one bed. Extra beds share tracks 11 to 128 with the objects.

Dolby’s beds use Dolby’s own channel and pack definitions, not the ITU’s common ones. Their IDs won’t match the common 7.1.2 pack.

How does Dolby’s profile differ from today’s ADM?

Dolby’s profile follows the ADM’s first edition, from 2015. Today’s edition is BS.2076-3, and the EBU’s production profile builds on it.

	Dolby Atmos ADM files	EBU ADM Production Profile
ADM edition	BS.2076-0, 2015	BS.2076-3, 2025
Programmes	One	Up to 32
Tracks	Up to 128, with at most 118 objects	Up to 128
Positions	X, Y, Z only	Angles or X, Y, Z, one system per programme
Kinds of audio	Beds and objects	Beds, objects, HOA, matrix and binaural
Bed layouts	Dolby’s own, up to 7.1.2	The ITU’s common definitions, up to 22.2

Dolby’s profile also narrows object metadata. Size takes one value for width, height and depth. Diffuse is on or off, where the ADM allows anything between.

Spread is written as size. Panners may call that control divergence, but the ADM’s own divergence field isn’t used, and nor is screen locking.

Some settings, such as the headphone render mode, live in Dolby’s own metadata outside the ADM.

One difference stops some tools. Each stream in a Dolby file points to two definitions; today’s ADM allows only one. The EBU says:

“The file must be processed to remove this double reference to be able to be used with the renderer specified in ITU-R BS.2127.”
EBU Tech 3393, annex A.2.1

Reading an ADM BWF master compares the profiles in full, and shows how to tell which one you have.

SoCount the bed first, and read 7.1.2 heights as top side, not top front. Expect Dolby’s own channel definitions, not the ITU’s common ones.

03 · of 9

What is an object?

One sound kept separate, with metadata that says where it is and how to play it. The metadata can change over time.

The ITU puts it like this:

“Object-based audio is an audio representation in which elements of the content are separate and accompanied by metadata which describe their relationships and allow a renderer to generate signals most appropriate to the playback system.”
ITU-R BS.2051-3, annex 1

The audio itself is ordinary. Without its metadata, an object plays as plain mono, and you can EQ it like any track.

An object needn’t last the whole programme. It can appear, move, change and disappear.

Objects suit a voice, a spot effect or a fly-by: sounds whose place must hold on every system. They also allow listener control, such as a commentary you can turn down.

There is a third type, scene-based. It stores a whole sound field, not speaker feeds. Ambisonics is the example, and all three types can sit in one mix.

SoUse an object when a sound’s place, its movement or separate control matters. If it is everywhere at once, it belongs in the bed.

04 · of 9

How can a mix be ten channels and a hundred positions?

A master can hold a 10-channel bed and over a hundred objects. A delivery to listeners carries far fewer, so masters are cut down.

A master and an IAMF delivery, to scale. The bed stays; objects are what gets cut.

The ADM sets no limit. Profiles do. Dolby’s, as the EBU records it, allows beds up to 7.1.2, up to 118 objects and 128 tracks in all.

Ten bed tracks and 118 objects make 128. The EBU’s own production profile also allows 128 tracks, and 128 objects at once.

Objects can share a track if they play at different times. So a long programme can hold even more, just not all at once.

Deliveries are smaller. IAMF’s largest profile allows 28 channels in a mix. The ITU’s emission profile caps tracks at 16 or 28 in Levels 1 and 2. Level 0 sets no limit.

Cutting a master down to fit is what the EBU calls squeezing. Each object can carry an importance from 0 to 10, and the least important can go first.

SoMix big if your master format allows it, but know your delivery limit. Mark the objects you can live without as less important.

05 · of 9

Where is an object, and how big?

Its metadata holds a position, in angles or in X, Y and Z. Optional hints cover size and how to play it.

One front-left place, written both ways. The signs run in opposite directions.

Polar positions are angles. Azimuth 0° is straight ahead, and positive angles go to the left. Elevation is positive upward.

Cartesian positions sit on a cube from −1 to +1. X runs left to right, Y back to front, Z down to up.

A file should use one system per programme. Mixing them can cause problems, the EBU warns.

Hint	What it tells the renderer
Width, height, depth	How big the sound is. Width and height in degrees in polar form, depth 0 to 1; all 0 to 1 in Cartesian.
Diffuse	From 0, a direct sound, to 1, a fully diffuse one.
Divergence	Splits the sound into a pair either side, like spreading a centre across L, C and R.
Channel lock	Snaps the sound to the nearest speaker, for when clean playback matters more than the exact place.
Zone exclusion	Keeps the sound out of some speakers, such as the sides or rears.
Screen reference	Ties the sound to the picture, so it can follow the screen size at home.
Gain	A level change. The ITU prefers the audio at the right level instead.
Importance	0 to 10. When objects must go, the least important can go first.
Head-locked	Keeps the sound fixed to your head when head tracking is on.
Headphone virtualise	Says whether headphones get a binaural render or plain stereo, and how much room sound.

SoRead a position in the system it was written in. Mind the signs: positive angles go left, positive X goes right.

06 · of 9

How does an object move?

Its metadata is cut into timed blocks. Each block says where the sound should be by its end.

TrySwitch how the last block moves.NoticeThe target never changes: 60° at 7 s. Only the route there does.

Each block has a start time and a duration. Its position is where the sound should be by the end.

So the EBU recommends starting a moving object with a zero-length block. It fixes where the sound begins. The ITU’s own rule holds the first block’s position from its start.

Between blocks, the renderer glides. The ADM doesn’t say exactly how, so renderers can differ.

A jump flag moves the sound at once instead. Give it a length, and the move takes that long.

The example is the EBU’s own: 20° at the start, then 30°, 40° and 60°.

SoMovement is a list of timed positions, like automation breakpoints. A static object needs just one block.

07 · of 9

What does the renderer do?

It reads the audio, the metadata and the speakers in front of it. Then it works out a feed for every speaker.

The object’s intended place never changes. Only the speakers that reproduce it do.

Objects always need a renderer. It turns each position into a gain for each speaker, for the room in front of it.

Beds are simpler. A bed channel goes straight to its matching speaker. If there isn’t one, it is folded down or panned from where it should be.

On a rig with no LFE, the ITU’s reference renderer drops the LFE channel.

ITU-R BS.2127 is the reference renderer, for loudspeakers. Headphones are separate: the ITU lists them as their own output, System Z.

Renderers differ. The EBU notes that different algorithms may sound subjectively different.

SoObjects adapt to the room; beds stay on their layout. Monitor through a renderer like the one your listeners will have.

08 · of 9

What survives delivery as IAMF?

Beds and Ambisonics, in every profile. Objects only in IAMF 2.0’s newer profiles, and only their position.

IAMF wraps audio in audio elements: channel-based, scene-based or object-based. Mix presentations say how to render and combine them.

Objects arrived in IAMF 2.0. Only the Base-Advanced, Advanced-1 and Advanced-2 profiles carry them, and only the Advanced ones mix them with a bed.

An IAMF object carries a position, on the same axes as the ADM. Size, diffuseness and the other ADM hints have no IAMF field.

Movement is fixed in the file too. IAMF moves an object in steps, linear ramps or curves; the ADM leaves the glide to the renderer.

On headphones, each element says how to play. In mode 0, IAMF sends a bed to stereo, and renders objects and scenes binaurally. IAMF and Eclipsa Audio™ covers the rest.

SoIf your delivery is IAMF, check the profile allows objects, and don’t rely on size or divergence surviving.

09 · of 9

How do I hear and check it?

Play the master through a renderer, on speakers and on headphones, and look at where every object actually goes.

OrbitOpens an ADM BWF master and plays its beds, objects and Ambisonics. Everything renders to one 7.1.4 bus, and binaural and stereo come from that.
Orbit ProEverything in Orbit, plus export from ADM to IAMF. Its routing graph decides how beds and objects are carried, such as folding objects into Ambisonics.
Orbit LitePlays ADM BWF and IAMF files, for a quick listen before a delivery leaves.

Your DAW’s renderer and the ITU’s reference renderer are good second opinions. Renderers differ, so check on more than one if you can.

For the file itself, chunk by chunk, see Reading an ADM BWF master.

SoOpen your master, watch every object move, and listen on speakers and headphones before it leaves.

The words

Glossary

Bed: The channel-based part of an immersive mix, made for a fixed layout such as 7.1.2. A working term; the ITU uses it but doesn’t define it.
Object: One sound kept separate, with metadata that says where and how to play it. The metadata can change over time.
Scene-based: Audio stored as a sound field rather than speaker feeds. Ambisonics is the example.
Metadata: Data that describes the audio: what it is, where it goes and how to play it.
DirectSpeakers: The ADM’s name for channel-based audio, where each channel feeds a speaker directly.
Top surround (Lts, Rts): Dolby’s name for the two heights of a 7.1.2 bed. The file places them overhead, level with the side surrounds. Dolby’s home speaker guide calls the pair top middle.
Block: One time slice of an object’s metadata. In the ADM, an audioBlockFormat, with a start time and a duration.
Renderer: The processor that turns audio and metadata into feeds for the speakers or headphones in front of it.
Squeezing: The EBU’s word for cutting a master’s objects and tracks down to fit a delivery format.
Importance: A 0 to 10 rating on an object. When objects must be dropped, the least important can go first.
Audio element: IAMF’s unit of audio: channel-based, scene-based or object-based, plus its metadata.

Sources