Skip to content
Elephant Audio
More delivery specifications
Standard reference

AES TD1008 — Streaming Loudness

What loudness does AES TD1008 recommend for streaming?

AES TD1008 recommends a Distribution Loudness of −18 LUFS for assorted and speech content and for interstitials, −16 LUFS for track-normalised music, and −14 LUFS for album-normalised music such as on-demand services. Upper tolerance is +1 LU where speech is measurable, +2 LU where it is not, and +0.2 LU for music and interstitials. Maximum true peak is −1 dBTP at the codec input of lossy-encoded streams.

Source

Transcribed from the published standard, read 2026-09-28. Check it yourself

−18 LUFS
Speech / assorted
+1 LU where speech is measurable
−16 LUFS
Music, track-normalised
+0.2 LU
−14 LUFS
Music, album-normalised
loudest track, on-demand services
−1 dBTP
true peak
measured at the CODEC INPUT

One stream, several targets

Every other recommendation in this reference sets one number for one kind of delivery. TD1008 does something different: it accepts that a single service carries speech, music and advertising, and sets a target for each rather than averaging them into a compromise nobody wants.

That is why a podcast with music beds and a music album can sit on the same platform without one of them sounding wrong. The document is written for distributors, not for mastering engineers, but the targets are the numbers a mastering engineer is being measured against.

Table 1 — Distribution Loudness by content type

Read directly from the document. Each line is one row: content type, target, upper tolerance, measurement method. The tolerance is the part most summaries drop, and it is not the same for every row:

  • Assorted, speech is measurable — −18 LUFS, +1 LU, Dialog Integrated Loudness.
  • Assorted, speech is not measurable — −18 LUFS, +2 LU, format-specific (Table 2).
  • Music, track-normalised — −16 LUFS, +0.2 LU, integrated loudness.
  • Music, album by loudest track — −14 LUFS, +0.2 LU, integrated loudness.
  • Interstitial — −18 LUFS, +0.2 LU, integrated loudness.
  • Virtual assistant — −18 LUFS, no tolerance stated, integrated loudness.

Track-normalised and album-normalised are not the same target

This is the distinction the document exists to make and the one that is almost always lost. Where a service normalises each track independently, the target is −16 LUFS. Where it normalises an album as a whole — using the loudest track to set the gain for the set, which is how on-demand music services preserve the relative levels an artist intended — the target is −14 LUFS.

Two decibels is not a rounding difference. Mastering an album to the track-normalised figure gives away level the service was going to let you keep; mastering a single to the album figure spends dynamic range the service is going to take back.

The true-peak ceiling is measured somewhere specific

The document asks that maximum true peak not exceed −1 dBTP AT THE CODEC INPUT of lossy-encoded streams. The location matters. Lossy encoding can raise peaks above what the source file measured, so a file that reads exactly −1 dBTP before encoding is not guaranteed to hold it afterwards.

That is the same reasoning behind Spotify advising −2 dBTP for louder masters, and behind EBU R 128 limiting true peak rather than sample peak.

What the footnotes cover

The scope notes decide which row applies to your delivery:

  • Assorted covers radio-style streams, musical concert performances, and podcasts containing speech, music and/or effects elements.
  • Dialog Integrated Loudness applies when speech, music and effects are produced with reference to a speech anchor.
  • Virtual assistant loudness is integrated loudness of the device's output preceding its volume control.

The source

Read from the AES technical document itself:

Common questions

Is AES TD1008 a requirement or a recommendation?

A recommendation, and it is addressed to distributors rather than to the people delivering files. No platform rejects a master for missing these numbers. Its value is that it is the only document setting targets per content type on a mixed stream, so it describes what a well-behaved service is trying to do to your audio.

Which figure applies to a single released to streaming?

−16 LUFS if the service normalises each track independently, −14 LUFS if it normalises by album using the loudest track. The document separates the two because they are different jobs; a master optimised for one is two decibels wrong for the other.

Why is the true-peak ceiling specified at the codec input?

Because lossy encoding can push peaks higher than the source file measured. Holding −1 dBTP before encoding does not guarantee holding it after, which is why the recommendation names the measurement point rather than just the number.

Next

This page is the lookup. Broadcast and Streaming Loudness Specs for Commercials is the explanation behind it.

Delivering to streaming?

Every deliverable leaves here measured on its own, against the target that applies to it, with the report attached rather than one figure quoted for the whole set.

Talk about a master WhatsApp the studio

Studio credit

Elephant Audio

Audio production · Sound design · Music production

LahoreEngineering sound since 2018

Last updated:

Talk about a master