Image may be NSFW.
Clik here to view.
The Audio Engineering Society just released an updated set of guidelines for streaming loudness, code-named TD1008.
But they’re not for you!
They’re not for artists, or producers, or recording, mixing & mastering engineers. They’re not even for music aggregators like CD Baby or Distrokid. They’re exclusively for online radio stations and streaming services like YouTube, Spotify and TIDAL – the services that distribute the audio to our devices and computers.
So, why am I bothering to tell you about them ?!?
Because as always, “knowledge is power” – or more specifically, understanding is power. If you know about these guidelines and understand the effect they’ll have when your music is streamed online, you’ll be empowered to make the best possible decisions about how it should sound.
So, one more time to be clear – you don’t need to follow these guidelines, at all. You can keep recording, mixing and mastering our music at what ever loudness you choose, just as you always have.
But if you care how your music will be heard online – and I think you should, because it accounts for 85% or more of the market at this point – it will be helpful to understand what happens to it once it’s been uploaded, and there’s plenty of interesting stuff in this document about that.
TD…. what ?
TD1008 is the successor to TD1004, first published back in 2015. Its full title is “Recommendations for Loudness of Internet Audio Streaming and On-Demand Distribution”, and you can read it here.
Its goals are fairly simple – to achieve consistency between different services; to avoid “blasting” listeners with unexpectedly loud music or ads and to minimise additional processing like limiting.
It’s 26 pages long and quite detailed and technical, but the good news is that unless you actually run an online streaming service, you probably only need to read the first few pages – and the whole thing is neatly summarised in a single table. You can see it above.
There are a few interesting details in this table which I’ll get to, but let’s start at the top.
Loudness recommedations (for distribution)
For speech-only and “Assorted” content, the recommended online Distribution Loudness is -18 LUFS. “Assorted” means a mixture of speech, music and FX – podcasts and radio-style streams, in other words. So far, so simple – that’s the same recommendation as the original TD1004 document, and the same value applies for “Interstitial” content (i.e. adverts & trailers) and “Virtual Assistant” audio like the voice of Alexa, as well.
But when we move on to music, things get a little more complicated. There are two different recommended Distribution Loudness levels for music, which I’ll talk about in a moment. In practise though, they both give the same overall result – a Distribution Loudness of -16 LUFS. I’ll explain why that’s a different number shortly.
So what IS the Distribution Loudness?
Distribution Loudness is simply the overall loudness value of the stream that your device plays. The idea is that if every streaming service follows the recommendations, there won’t be huge differences when switching between them, and there won’t suddenly be songs or other material “blasting” much louder than everything else.
Distribution Loudness used to be called the “Target Loudness” in TD1004, but this has been updated in the new version to avoid confusion. Many people understandably thought that the “Target Loudness” was something they needed to aim at – the level the music should be mastered to, but that’s not what the guidelines are for.
Streaming services actually adjust the playback loudness of the material for us, to fit their chosen Distribution Loudness, using a process called loudness normalization, so there’s no need for us to change how loud we master (unless we want to).
Why is the recommendation different for speech and music?
…And, why are there two different values for music?
Both these questions can be answered by thinking about the difference between matching loudness as opposed to balancing it.
The issue is that if we match the measured loudness of two pieces of audio, they won’t necessarily sound balanced. For example, if you match the loudness of a piece of speech with a piece of rock or pop music, the music won’t sound loud enough. Our brains expect speech to be a little quieter than the sound of a full band playing. 2-3 dB often sounds about right.
In the same way, if we master an acoustic ballad to the same loudness as a piece of death metal, the ballad will actually seem too loud. This is why mastering engineers don’t match the loudness or EQ of songs when they work, they carefully balance them against each other to give the most satisfying musical result and convey the emotion of the songs as effectively as possible. Some songs are louder and more intense, others are more gentle.
In fact, this also explains one of the biggest objections many people have to normalization in general – it messes up the artistic intent. If you’ve carefully balanced the flow and range of loudness in the songs on an album, the last thing you want is for a computer algorithm to come along and change them all.
The good news is there’s a simple solution to this problem – use Album Normalization, instead of Track Normalization. Don’t try and make all the songs the same loudness (Track) – instead, measure the loudest song on each album and scale all the songs down by the same amount (Album).
Better normalization wins
And it turns out that not only is the Album Normalization method every bit as effective at achieving consistency and avoiding “blasting”, but 80% users actually prefer it. It sounds more natural and musical, and retains the original artistic intent of variations in loudness between songs. In fact TIDAL has been using this method for years now, with great success. Spotify and iTunes have Album modes but don’t yet use this method when playing playlists or in shuffle, but hopefully the TD1008 recommendations will encourage them to change this.
And interestingly, if the loudest songs are played at -14 LUFS using Album normalization, the additional musical variety this results in an overall integrated loudness for the stream of close to -16 LUFS, with the added benefit that louder songs can kick more, and quieter songs aren’t too quiet, even on mobile devices.
So this is why there are two different recommended Distribution Loudness values for music in TD1008. For services that employ Album (Loudest Track) normalization, the value achieves an overall level of -16 LUFS, in line with services like YouTube that use Track Normalization. And these values for music sound natural and balanced in combination with the slightly lower level for speech and other content.
Following these recommendations achieves consistency between services, regardless of the type of content they offer. It stops us being “blasted” by very loud songs, and also preserves the full artistic intent of the original material. What’s not to like ?!
The opportunity
And this is why I say understanding these recommendations is useful, even though they don’t mean we should master at -16 LUFS, or any other specific value. When you’re confident that the normalization will be effective and musical, you can mix and master in the way that suits the material best musically, without having to worry that it won’t be quite as loud as other similar material, or seem to suffer in comparison.
Research analysing over 4.2 million albums (!) shows that over 80% of pop & rock albums have a loudest song of -14 LUFS or louder, so Album Normalization to that level won’t result in extra limiting or other processing for the majority of songs. If you want to master louder that’s fine – the loudest songs will still sound just as loud as anything else. And if you prefer to mix or master with more dynamics, you’re free to do so. (For what it’s worth, the very loudest stuff I master is around -11 LUFS, and has been for years. It sounds great on all the platforms and my clients are delighted with the results.)
Will streaming services listen?
As I mentioned, TD1004 was released in 2015, recommending an overall loudness of -18 LUFS. I helped set up an online petition trying to persuade streaming services to pay attention around the same time, which now has over 10,000 signatures. But as I write this in 2021, most streaming services are still using a Distribution Loudness of -14 LUFS. If they didn’t pay any attention last time, why should it be any different now?
Well firstly I think it’s worth saying that things have changed since then. Both Spotify and YouTube have switched to using LUFS normalization, which is more effective and makes it easier for us to understand and predict what will happen when our music is streamed. Spotify also reduced their default Distribution Loudness from -11 to -14 LUFS, and recently stopped using a limiter to boost quieter songs by default.
These are very substantial positive changes already, but even more significant is that all the major services have been actively involved in helping draft these new recommendations. So not only are they paying attention, but they’re engaged and invested – and I hope it means that they will be adopting the new guidelines soon.
If they do, it’s a real win-win-win. We get more consistency, less processing and more musical results – plus the freedom to prioritise what works best for the material, knowing our artistic decisions will be honoured as closely as possible.
I’m optimistic – please read the full TD1008 recommendation and share this post if you are, too!
Streaming Loudness – AES Recommendations 2021, and why you should care is a post from Ian Shepherd's: Production Advice Subscribe to the newsletter for great content from the archives, special offers and a free interview - for more information, click here