AES Loudness Defined!

[Hugh Robjohns]

Always behind the times... the AES have finally come out with their recommendations for Loudness normalisation on streaming services.

https://www.aes.org/technical/documentD ... ?docID=729

There aren't any real surprises. Individual music tracks should be normalised using BS1770-4 Integrated Loudness with a target of -16LUFS. Where whole albums are being normalised the loudest track can reach -14LUFS.

Anything primarily speech based (podcasts, talk radio etc) should aim for a target of -18LUFS (measured only on the dialogue sections).

The expectation is to lower the targets by 6dB in the years to come to bring it fully in line with R128 etc for audio-for-video sound tracks.

[Drew Stephenson]

I'll have a proper look later when I'm not on the phone, but is there a quick explanation of why albums have a different target to tracks?
My gut response was that if you were going to do that you'd want a larger dynamic range for an album than a track, rather than the other way round.

[RichardT]

Hi Drew, I’ve think they’ve done this precisely because the dynamic range of an album can be wider than that of individual tracks. If album normalisation were set to -16, and the dynamics of each track respected, then quieter tracks might be too quiet. The previous recommendation has been that tracks should not go below -20.

Personally I aim for about -16 across an album as a whole but make sure not to go below -20, and as it happens I usually do have a track that needs to be -14 or so if the relative dynamics are to work.

[Wonks]

It's applicable when doing 'album normalization'. So the loudest track is normalised to -14LUFS, and then the same level of gain reduction is applied to all the tracks on the album, to keep the relative track levels the same. It presumably is designed to stop very quiet tracks on the album from being too quiet.

[Hugh Robjohns]

...is there a quick explanation of why albums have a different target to tracks?
My gut response was that if you were going to do that you'd want a larger dynamic range for an album than a track, rather than the other way round.

Loudness normalisation targets and dynamic range are different things.

The document states:

Music Normalization Techniques
Two methods of music normalization are currently practiced (see Table in post above):

Track normalization: Normalize each track to the “track normalized” loudness specified in the Table.

Album normalization: Normalize the loudest track of the album to the loudness of the “album-loudest track” specified in the Table. Then apply the same normalization gain or attenuation to the remaining tracks. This preserves the relative loudness between the tracks and maintains the artist’s intent.

Basically, it allows the loudest track in the album to be slightly louder than 'normal' to maintain it's intended mix loudness relative to 'standard' material, while also helping to make sure any gentle tracks aren't too quiet.

[Drew Stephenson]

Thank you all, that makes sense.

[Lophophora]

Always behind the times...

To be fair, the AES was the first to come up with recommendations being written specifically for music streaming services with the TD1004 in 2015. And the TD1008 is more thorough than the R128 S2. I don't have a stake in either of these organizations... just saying.

[Hugh Robjohns]

Fair point...

I've noticed over the years that the AES tends to wait for a 'standard' to evolve across the industry, and then they publish a 'standards document' to confirm it. There are some key exceptions, of course, but that seems to be the general trend.

[claz]

The expectation is to lower the targets by 6dB in the years to come to bring it fully in line with R128 etc for audio-for-video sound tracks.

Does this have any practical implications now in loudness target considerations for releasing tracks to streaming services?

[Lophophora]

Does this have any practical implications now in loudness target considerations for releasing tracks to streaming services?

I think it really shouldn't. The recommendations are for normalization algorithms, not music producers / artists / engineers. If anything, it will give mastering engineers more leeway to retain dynamics instead of having to compete for loudness.