Audio for streaming: loudness, Atmos and stems

Sound is not the cinematographer's department, but a streaming delivery succeeds or fails as a whole. A beautiful HDR master is rejected if its audio is out of sync, too loud or missing a stem, and many sound problems start on set with timecode, slating and the camera department's habits. This lesson gives cinematographers an awareness-level view of streaming audio delivery: how loudness is measured and specified, what mix formats and stems are, how music and effects tracks support localization and how picture and sound stay in sync from the set to the final package. The figures are common published examples; the re-recording mixer and post supervisor follow the current specification for each production.

Why loudness is specified

Before loudness standards, television commercials and programmes were often mixed as loud as possible, and viewers had to keep adjusting the volume. Loudness standards fix this by measuring perceived loudness rather than peak level. The measurement method in ITU-R BS.1770 weights frequencies the way people hear them and averages over time with gating that ignores silence. Results are expressed in LKFS (loudness, K-weighted, relative to full scale), which is numerically the same as LUFS. A loudness unit (LU) is a relative difference: going from −27 LKFS to −24 LKFS is a change of 3 LU.

Common loudness targets

Different services and broadcasters use different targets, and they measure in different ways.

Dialogue-gated measurement looks only at passages with speech, which keeps dialogue consistent while letting music and effects have wide dynamic range. Programme loudness averages the whole mix, so a quiet drama and a loud action film are levelled in a different way. Services publish their own targets; for example, Netflix's sound specification has described a −27 LKFS dialogue-gated target with a true-peak limit of −2 dBTP. Confirm current values before relying on them.

True peak

Loudness measures the average; true peak measures the highest instantaneous level, including peaks between digital samples that can appear after conversion and encoding. True peak is expressed in dBTP. Streaming encoders can raise peaks slightly, so specifications set limits below full scale, such as −2 dBTP or −1 dBTP. A mix that exceeds the limit can distort when encoded or played on some devices.

Mix formats

Streaming services usually ask for several mixes, depending on the title.

• Immersive mix: Dolby Atmos is common for premium originals. Atmos combines a channel-based bed with audio objects positioned in three-dimensional space, and is rendered by playback devices to the speakers available.
• 5.1 surround: six channels (left, centre, right, left surround, right surround and low-frequency effects).
• Stereo: two channels for headphones, laptops and televisions without surround, sometimes made as a dedicated mix rather than an automatic fold-down.
• Near-field or home mixes: some productions make versions tuned for home listening rather than cinemas.

The deliverables list specifies channel layouts, file formats (often 48 kHz, 24-bit uncompressed audio) and which mixes are required.

Stems and music and effects tracks

Stems are sub-mixes of the final mix, separated by category so that the sound can be adjusted or replaced later.

• Dialogue stem: all dialogue, including additional dialogue recording (ADR) and group voices.
• Music stem: score and source music.
• Effects stem: sound effects, Foley and backgrounds.
• Music and effects (M&E): a complete mix without original-language dialogue, used by dubbing studios to create foreign-language versions. A good M&E also includes effects that were recorded with dialogue, such as footsteps or rustle, which must be recreated if they were only on the dialogue tracks.

When the stems are summed at unity gain, they should match the full mix. QC checks this.

Sync and timecode

Picture and sound must stay in sync through every stage. Lip sync errors of about one or two frames are noticeable to many viewers, and services reject deliveries with drifting or offset sync. Sync starts on set:

• Jam-sync timecode between cameras and the sound recorder at the start of each day and after power changes.
• Slate every take clearly, with a visible clap or a smart slate, so editors can check sync.
• Record the project frame rate and timecode rate consistently, and avoid mixing drop-frame and non-drop-frame timecode.
• Log any takes with sync problems in the camera and sound reports.

In post-production, the conform and audio teams keep sync through the edit, VFX and mixing, and QC checks sync at several points in each episode.

Audio for accessibility and localization

Delivery also includes audio for accessibility and other languages.

• Audio description: a separate mix or track in which a narrator describes important visual information between dialogue for blind and partially sighted viewers. Lesson 8 covers accessibility requirements.
• Dubbed versions: created by dubbing studios from the M&E and translated scripts.
• Dialogue lists and scripts: used by translators and dubbing studios, with timecodes.

The cinematographer's role here is indirect, but visual storytelling decisions affect audio description. Scenes that rely on silent visual details give describers more to describe; very busy sequences may leave little room between dialogue for description.

Where the camera department fits

The camera department supports good sound delivery through simple habits:

• Keep timecode jammed and checked, and report problems quickly.
• Slate clearly, including tail slates and second sticks when needed.
• Note camera speed changes and high-frame-rate shots in reports, so sound and editorial know which shots have no usable sync sound.
• Keep camera noise low, including fan modes and lens motors, and cooperate with the sound department on blocking and placement.

These habits make the conform and mix faster and reduce the risk of sync failures at QC.

Common audio rejections at QC

Audio QC reports for streaming deliveries often repeat the same handful of problems. Knowing them helps the whole team, including the camera department, avoid them.

• Loudness outside tolerance, often because a mix was made to a broadcast target and delivered unchanged to a service with a different target.
• True peaks above the limit after a late gain change or a music replacement.
• Sync drift in long takes, usually from timecode problems on set, mismatched frame rates in the conform or a frame-rate conversion.
• Missing or mislabelled channels, such as a swapped centre and low-frequency effects channel in a 5.1 file.
• Stems that do not sum to the full mix, often because a late fix was made to the mix but not the stems.
• Incomplete M&E, where effects recorded with dialogue disappear when dialogue is removed.
• Clicks, dropouts and digital distortion introduced in editing or file conversion.

Each of these can delay delivery. A short conversation between the DP, sound mixer and post supervisor during prep about timecode, frame rates and slating can prevent the most common sync problems before they start.