Oct 1, 2009
How Spotify will end the Loudness Wars

This post ended up as a bit of a mess – if you read on you’ll see what I mean ! To find out why, read the next one.
I’m leaving it here because there are some sites linking in, and some people will find it useful, but if you really want to know what I think about Spotify and the Loudness Wars, please read my guest post on RecordProduction.com instead:
How Spotify will end the Loudness Wars
And now the original post for those who really want all the gory, messy details !
*** IMPORTANT NOTE ***
The short version is – I was wrong in this post.
It used to be called “How Spotify is contributing to the Loudness War”, but that was wrong and misleading.
Spotify’s “Volume Normalisation” works exactly as advertised, and doesn’t add any unnecessary limiting to already high-level tracks. This feature is a great thing, in my opinion, and hopefully in the long term will help make the “Loudness Wars“ irrelevant.
I want to apologise to Spotify and their programmers for posting incorrect conclusions about the way the software works, and criticising them as a result. It was hasty and unjustified, and I regret it.
In my defence, this is an issue I’m passionate about, and nobody’s perfect (as this post proves !). But still, my bad.
I’ve decided to leave this post up, so that anyone who has already read it can see it was wrong. Please read it anyway, if you’re interested – there is still plenty of useful information here. I’ve edited it with corrections where necessary, and will put up a new post explaining exactly how and why I drew the wrong conclusions soon, which some people may find interesting and/or useful.
***
I love Spotify. I need that to be plain and clear, before we start.
BUT I have a big problem with the way they are presenting their music, and the potential impact it has on the “Loudness Wars“.
Currently, Spotify’s influence is unquestionably bad – in my opinion they are actively making things worse. That’s not to say they planned it or intended it that way, but it’s true, nonetheless.
[In fact, their influence will be a good thing, in the long run, I hope]
To understand why, we need to look at a little-known feature buried in Spotify’s preferences, called “Volume Normalisation”. I didn’t even know it was there, until Production Advice user @freefallmusicco pointed it out to me.
In a nutshell, this feature is intended to make sure all the songs in Spotify play at a similar level. This is a good thing ! The last thing we want is to be constantly adjusting the volume control while we try to work/sleep/whatever. So far, so good.
The problem mystery
For some reason, I have always had a problem with Spotify’s sound.
When I listened to the new U2 album before it was released, I had my fingers crossed for a great-sounding, not-smashed-to-hell album that I could write a glowing review of on my mastering blog, but I was disappointed. Later though, I heard a friend’s copy, on CD and although I still didn’t love the production, it sounded much better to me.
In the same way, I listened obsessively to Imogen Heap’s new Album Ellipse while I waited for my CD copy to arrive, but was unexpectedly astonished when it did – astonished at how much better it sounded when I loaded it up to my iPod. More life, more depth, more groove – strange…
(For my geekier readers, I was especially surprised because Spotify uses the Ogg Vorbis codec for their files, which are pretty good, especially when compared with mp3s at a similar bit-rate. I didn’t expect 224kbps AAC to sound that much better)
And then I happened to see another tweet, suggesting that the “Volume Normalisation” setting actually made things sound worse, by ‘squashing’ or limiting them. This sounded strange to me, so I took a listen. To my astonishment, this was right !
When I first listened, I agreed with this comment. With hindsight, I was making several schoolboy errors, and should have known better – see my next post.
Spotify’s “Volume Normalisation” feature is adding extra limiting to everything it plays - even the songs that are already loud.
Which is mad and here’s why:
How Volume Normalisation SHOULD work
[And does !]
- To make all songs play at a consistent level, you have to turn some of them up, and some of them down
- But some older songs are very quiet in comparison to modern CDs. If you turn the modern stuff all the way down to match, everything will be very quiet
- So Spotify have compromised and turned lower-level songs up, and higher-level material down a little
- When they boost the level of the quieter tracks, they use a limiter to stop them distorting
So far so good. The lower level a song is, the more you have to turn it up, and the more limiting you have to use. The higher level a record is, the more you need to turn it down.
But you never need to use a limiter for high-level songs
That’s because to get the high level in the first place, the CDs have already been heavily limited. So there’s no need for any more processing. To make a track quieter, you just turn it down. No risk of distortion, no need for a limiter.
But Spotify is using a limiter – all the time. The Volume Normalisation setting is adding extra limiting, and crushing yet more life out of everything.
Surely that can’t be right ?
No, it’s not ! Spotify’s “Volume Normalisation” does use a limiter, but only on lower-level, dynamic material which needs to be boosted, and would otherwise clip and distort. This is exactly as it should be.
I was so astonished, I thought I must be wrong, so I used the ultimate arbiter for audio comparisons – the null test. The concept is simple.
If I give you an apple but then take it away from you, you have… no apple. But if you have an apple and change it in some way – say, take a big bite out of it – when I take it away from you there is something left. You have a big bite of apple in your mouth. The same thing works with audio.
Take an audio file, take away another file that’s exactly the same and you get nothing, ie. silence. But if you change one of the files before you take it away again you’ll hear something left over – the audio equivalent of the bite that was taken out of the apple.
(Audio nerds – to do this yourself, invert the phase of one file and sum them together)
So, I did this test on a Spotify track, comparing with and without the “Volume Normalisation”, but matching the levels first. Since it was a pretty high-level song to begin with, Spotify should have simply turned it down, and when I null-tested I should have heard nothing.
In fact, this is what I heard:
Please install Flash plugin
There are two things going on here – overall we can hear some trebly stuff, so one of the versions sounds brighter than the other - ie. there’s an EQ difference. But more important are those little bits of kick drum and snare that are pulsing through – the bits that have been bitten out of the original apple.
That’s the limiter working, crushing all those subtleties out of the mix - and remember, this is a song that was pretty high-level to begin with. So it doesn’t need to be limited at all – just turned down. [Which is in fact, exactly what happens]
So WHY does Spotify do this ?
[It doesn't ! There was a mistake in my listening and reasoning, as I'll explain in the new post]
I don’t know. I’ve asked them, and they haven’t told me, only Spotify say that “Having normalization on as default seems to provide the best experience for the average user” – well, yes, that’s because most people (like me) don’t want to keep adjusting the volume.
But wouldn’t the average user be even happier if the songs weren’t unnecessarily crushed ?
There are plenty of Spotify users who thought they heard this problem and didn’t want it. In fact, they were reacting against the limiting of lower-level tracks being boosted, or were making the same mistake that I did. And they persuaded Spotify to allow Volume Normalisation to be disabled, as you can see in this thread:
http://getsatisfaction.com/spotify/topics/sound_is_very_flat_and_empty
But disabling the feature altogether is throwing the baby out with the bathwater. We could have the best of both worlds – consistent volume and no crushed “flat sound” - but Spotify aren’t choosing to do it this way. [And, we do]
What about the loudness wars, then ?
The biggest shame about all this is that As more and more people use services like Spotify, and hear consistent playback levels, the Loudness Wars could become irrelevant. [And, hopefully, they will.]
There’s no point in “doing a Death Magnetic” if it will just be turned down automatically when people listen to it. Just as the absolute level of a CD is irrelevant on the radio, so it would be is when listening via Spotify with Volume Normalisation enabled.
Instead though, the current version of Spotify’s “Volume Normalisation” feature is making things worse by squashing even more life out of perfectly decent-sounding music.
So people’s enjoyment isn’t as great as it might be, and they gradually become even more desensitised to the overly crushed “Loudness War” sound.
Persuade Spotify to fix it !
I want Spotify to stop this happening, and I hope you do too. “Volume Normalisation” is a great idea – potentially it could defuse the Loudness Wars forever. But not the way it works at the moment.
Let’s persuade them to do the right thing and take advantage of this opportunity ! Please spread the word about this post, or write your own – Digg it, Stumble it or Tweet it using the buttons below – email Spotify or get onto their support forums and let them know what we want:
So, after all that, I think Spotify deserve our congratulations and support for implementing this feature, and I’ll be keeping my fingers crossed that other players and streaming solutions implement the same system , so that ultimately we reach the goal that so many of us want -
No More Loudness War
No related posts.


.gif)





Thanks for the article.
Have been waiting for ages, for Spotify to become available in my country(Germany).
Would love to check it out.
I’m not able to reproduce your results. Are both terms of the subtraction audio files extracted in the same way, or is one of them taken from some other source?
More specifically, I do not hear the sound you refer to as ‘trebly stuff’. I do hear the bits of drum that are pulsing through. These are caused by the limiter when volume normalization is turned OFF. However, when volume normalization is turned ON, the limiter does absolutely nothing in my test for that particular coldplay track since the volume is already low enough.
You are subtracting two audio files, so without additional knowledge you can’t tell which one of the terms contains the error. You made the incorrect assumption that the pulsing-drum-distortion is caused by the volume normalization being turned ON, when it’s really caused because of the volume normalization being turned OFF.
The peak limiter is enabled regardless of the volume normalization switch, or you would get hard clipping (i.e. clipping samples above 32767 down to 32767) which would sound even worse. The only thing volume normalization does is to apply a pre-gain right before the peak limiter stage.
Hi Ludde,
Thanks for posting !
Sorry, but this is working in the opposite way to what you describe, in listening terms.
With “Volume Normalisation” ON (ie. with the checkbox ticked) the playback level is LOWER, and the kick and snare drum have audibly less impact, as a result of limiting. This is the “flat, empty sound” described by everyone in the thread on your forum.
With “Volume Normalisation” OFF (ie. with the checkbox un-ticked) the playback level is HIGHER, and the kick and snare have more impact.
Try listening to “Yellow”, by Coldplay, for a clearer example.
But you are saying that the gain is added to the “normalised” version only, and that the limiter is always active ? If so, the limiter settings must be incredibly aggressive, to reduce the overall replay level by 3dB – it would be no wonder in this case that the normalised version sounds bad. It’s also, as my post says, completely unnecessary since to achieve normalisation, all that’s required it to reduce the level of louder tracks somewhat.
Are you saying that there is no adjustment of level based on the original dynamic range, in the “normaliser” ? You’re simply lifting the level of everything and letting the limiter squash it all ?
Ian
PS. Both VN-ON and VN-OFF samples were captured from Spotify, using Wavelab.
PPS. The “trebly” stuff is just high-frequency content that is left over from the null test (remember you need to level-match before nulling). A complete null isn’t possible, even adjusting in increments of 0.1 dB, but most of the signal does cancel, except for the limiter “pulses”.
Ian,
I’d like to find the cause as much as you do, but from my analysis, having normalization on results in less distortion than if normalization is off.
I have a bunch of sample files from my test. Drop me an email and I can send them to you as well as a more lengthy explanation/clarification.
I’d just like to add that you should be careful because when Spotify update their app, at least on mac, they do it in the background, and the user is not notified. When the app is updated, the preferences are reset to default.
Just now I noticed that Spotify had been updated and had reverted to normalisation on, and disabled high quality files for playback!
As a passionate audiophile, you can be forgiven for making the odd mistake. Your honesty is a shining beacon of hope and decency – remind yourself never to go into politics.
Good to provoke this discussion Ian – and to be so honest and transparent about it.
[...] Read the full article (and subsequent corrections) at Production Advice [...]
You should leave the original article without the strikeouts and red, reading it now for context for your other post (the update) is completely impossible.
@Tor That’s a fair comment and something I’ve worried about. I think I will write a new one with only the correct information. I felt it was important to keep this one up to begin with, though. I’ll archive it and link to it from the new one for people who are interested.
[...] How Spotify will end the Loudness Wars [...]
[...] 2. How Spotify Might Defuse the Loudness War from Production Advice [...]