I probably don't need to remind you of the ongoing loudness wars in music production and delivery, with CDs getting louder and louder and radio stations in a race to make their station louder than any other station. Now more and more music is being delivered by music streaming services like Spotify and iTunes Radio. What these services do regarding the loudness wars matters more and more and so the Loudness Petition Group has been set up by Eelco Grimm, Bob Katz, Matt Mayfield and Ian Shepherd with the express aim of bringing peace to the loudness wars.
There is a petition on Change.org that you can sign to persuade the music streaming services to agree to adopt a common standard. In this article, we are going to look at the Loudness FAQ from the Loudness Petition Group which expands on the issues and covers some of the misconceptions and urban myths that have already grown up around this whole subject and as someone coming from the broadcast sector with the challenges that we have and continue to face with loudness normalisation of broadcast TV content there are a number of issues that act both as echoes and challenges for everyone to consider from what the Loudness Petition Group are proposing.
But back to the Loudness FAQ and the first question.
Q: We don't want our audio to be processed by anyone. What's different about your petition?
A: The goal of the loudness petition is exactly the same as yours, only better because it puts all artists on an even keel. In order to keep the artists’ intent, loudness normalisation must simply apply attenuation (if necessary) without any other processing, and not add compression or unnecessary peak limiting. Does that answer your question?
Q: You guys seem out of date. Haven't the streaming services already implemented loudness normalisation?
A: No, unfortunately, every streaming service has problems. Some are not normalised by default. Streaming needs to be normalised by default or it's useless.
One service uses such a high target it causes artists to overcompress their singles to match that service.
Another service adds serious and egregious peak limiting on top of an artists' creation.
Another service is not normalised at all. So this is a current and serious problem.
We intend to lobby the services to make these changes, and there is already a movement in this direction. Wish us well, please!
Q: What about your target, isn’t it too low? Maybe if the recommendations were a bit more reasonable would help. -16 max to -20 min LUFS is just too soft for pop/dance, IMO. Sometimes I end up with a mix (EDM) that is already at -12 or -14 LUFS before mastering without clipping. And with another 2-4dbs of compressing/limiting, it sounds denser and better. Why shouldn't I do it?
A: You can. It's in our petition.
Streaming services, please let artists choose how their music should sound! Some producers choose to master loud as part of their art. But other creators feel forced to master too loud (over-compress), compromising their vision because more dynamic music would sound lower than the “competition.” They mix and master to the level of the loudest streamer, adding unwanted distortion, pumping and clipping. This situation intensifies the Loudness War in music production and discourages creators from producing programs which are dynamic, open-sounding and impactful.
As an artist, you can compress or process as much as you want. But with loudness normalisation, dynamic music can co-exist with compressed music.
Q: I agree with that. Just saying -16 LUFS may be too soft a target, considering the playback devices/setups around.
A: People smarter than you and I have already researched the performance of -16 LUFS music and a range of portable devices, and it's quite acceptable.
Q: I think that streaming services should target at -10 to -12 LUFS. Anything below that sounds wimpy in small speakers, laptops, etc. that most people listen from. I see streaming services as radio... sound quality isn't the 1st priority, rather accessibility across devices (even shitty ones).
A: Apple uses -16 currently, and with standard earbuds, everything is plenty loud with the volume nowhere near full. Just because the reference level is low, doesn't mean you have to listen to it quietly. Just turn up your phone and enjoy music. The lower target permits music with dynamics to coexist with music that’s more compressed. A higher target would not permit that and that’s not fair to the artists’ freedom of expression.
Q: What LUFS target value would you recommend for digital streaming releases such as Apple Music and Spotify?
A: It's all in the AES recommendations document. Here are the key recommendations...
It is recommended that the Target Loudness of the stream not exceed -16 LUFS: to avoid excessive peak limiting and allow a higher dynamic range in a program stream.
It is recommended that the Target Loudness of a stream not be lower than -20 LUFS: to improve the audibility of streams on mobile devices.
It is recommended that short-form programming (60 seconds or less) be adjusted by constraining the Maximum Short-term Loudness to be no more than 5 LU higher than the Target Loudness: This ensures that commercials and similar short-form content are consistent with the stream loudness.
It is recommended that the maximum peak level not exceed −1.0 dB TP: to prevent clipping when using lossy encoders.
You can read the full AES recommendations document here.
Q: What do you think would be the right loudness for our mixes and masters?
A: The right loudness includes the peak to loudness ratio that sounds good to the artist. Streaming services need to reproduce the artist's intent without changing the dynamic character that he or she intended. That's the "right loudness"!
Q: How would the streaming service know which peak to loudness ratio is good for the music. A mastering engineer can set this ratio, but it can be damaged by the streaming service.
A: By using a sufficiently low target, the peak to loudness ratio of the vast majority of music produced today will be preserved. That's the message we are conveying to the music streaming services in our petition!
Q: Normalising already mastered recordings causes more damage than the awful loudness war itself! A tender acoustic ballad shouldn't be as loud as a piece of fast rock music, should it? The streaming services shouldn't be allowed to dictate the rules, simply because they are only temporal media (one can bet they are not going to exist in 10 years!) and far from offering the sound quality we all insist on!
A: Album normalisation will help to alleviate the phenomenon you've cited and improve the performance of streaming normalisation. There's a lot of work to be done and getting all the services to normalise is the important first step, hence the petition.
Bob Katz calls what you are referring to “The Acoustic Advantage”. This issue happens with classical as well as popular music and has occurred since the dawn of recording! Long before normalisation came on the scene. Normalisation (in its current form) cannot cure the acoustic advantage but it can improve the situation enormously.
Here’s an example. Before CDs were invented, an LP collection of classical music varied subjectively about 6 dB. Since mastering engineers were always trying to maximise the level of LPs to ensure a good signal to noise ratio, a peak-maximised solo harpsichord or string quartet would sound too loud compared to a full symphony. That’s because the peak to loudness ratio of the smaller group is less, and also, the ear tends to like to hear performances at their natural sound pressure level. However, the largest perceptual difference that we found was about 6 dB, which is pretty tolerant. That was the situation back with LPs, so the acoustic advantage has been going on for a long time.
But when the loudness war moved to the digital medium, it became possible to hyper-process material. So the electric groups, who were jealous that the smaller acoustic groups sounded louder, began to apply more and more amounts of processing. The acoustic groups also discovered the loudness race, and in each cycle, it became a vicious cycle until both the acoustic and the electric groups were processed in an extreme way (smashed, super distorted, fatiguing, etc.) in the effort to be as loud as their competitor. At that point the differences between recordings grew from 6 to 10, 14 or more dB and not only became annoying, it became ear damaging, especially to headphone users!
Furthermore, if the streaming services implement album normalisation as the AES document recommends, then any soft cuts from an album will be reproduced at their correct relationship to the loud ones. I would even recommend that album normalisation should be left on all the time since it will approximate producer's loudness preferences for their soft cuts even when the soft cut is from a different album than the current sequence! Instead of maximising even the ballads to the target level.
So in the end, normalisation will return us to the days when our preference differences between small acoustic groups and big or electric groups are no more than 6 dB, and the natural acoustic advantage will be no greater than it ever was in the days of analogue. That will be a wonderful time. Let’s hope for it.
Q: As an enthusiastic supporter of the preserved natural dynamics. If you normalise an already mastered pop song with a cumulative RMS (say) around -14 dBFS down to -20 dBFS aren't you losing one full bit of information, thus cutting most of the oh-so-important fine details? And that's only to match its loudness to a very dynamic acoustic jazz piece, although they shouldn't be equally loud by nature. I am neither arguing nor talking about over-compressed and over-limited heavy metal or 'modern' pop songs - just trying to warn everyone about the damages the normalisation can cause to any well-recorded music, all being in favour for the streaming which is obviously soon to be totally obsolete anyway...
A: You have a misconception of the losses involved when doing digital gain or loss on a given source. People who think they're "losing bits" do not understand the fundamentals of digital audio. You're not losing any audible bits! Instead, you need to think analogue and signal to noise ratio. You need to forget about peak levels because the ear responds to average levels, not peaks.... so you’re not losing any bits! If the volume change calculation is done with high resolution, e.g. in 32-bit float, and delivered to a DAC as a dithered 24-bit signal, then the resulting output signal will exceed the output medium's noise floor if it's normalised to, for example -20 LUFS, which is the loudness level of a great deal of classical music.
Keep in mind that a pop recording with a loudness of -12 LUFS does not have a better signal to noise ratio than a piece of classical music. The listener judges the noise floor by the noise of the DAC, which is already inaudible at the gains needed to reproduce classical music. So if you attenuate the rock song by 8 dB will it be audibly deteriorated? Nonsense! This will be a distortionless attenuation and the original noise floor (now slightly lowered) will still be so many dB above the noise floor of the medium as to be insignificant. Remember that the classical recording with a loudness of -20 LUFS is more than quiet and you never argued that it has to be turned up! The ear judges signal to noise ratio based on the average level, not the peak level. So if you turn the average level of the hyper-compressed pop piece down by 12 dB to match the average level of the classical piece the pop piece will audibly be in no worse state than the classical piece was in the first place.
Q: Surely only if all the calculations are done AFTER the files are converted to a 24-bit format, right? But the streaming services don't do that... Because even if you perform 64-bit deep calculations on a file and then save (or stream) it in a 16-bit form then you have lost information for good. Streaming services can't provide the necessary signal chain simply because each one of them applies their own processing which they are not ready to share in full. It's also easy to check - just listen to their channels wink emoticon.
A: Fortunately, that’s not true. It's not necessary to convert a source file to 24-bit or any format in order to normalise. Normalisation is done at 32-bit float in most computers, in real time. For example, if you start with a 16-bit file, for example, then attenuate with 32-bit float calculations, then dither to 24 bit, then feed the DAC, all will be well. You don’t lose any information. If you start with an AAC file or an encoded bitstream, ideally you should decode it to 32-bit float since the decoders work in floating point. From there we do the loudness normalisation in the PCM domain. And all is well.
Q: Without saving the file before dithering/converting to 24-bit, right? Do the streaming services perform these subroutines in real time?
A: Yes, they do. A number of streaming services have and are quite capable of performing these high res calculations in real time. Apple's OSX is fully floating point and passes the 32-bit float signal between apps quite transparently. iTunes volume control is a 32-bit float and passes its data at full resolution to the master volume control. No resolution is lost at all! So to add a normalisation coefficient at any point in that chain is entirely trivial. If done at the right place there would not even be an extra multiply operation! There are a lot of smart people working at Apple and our musical and data integrity can be kept entirely intact.
Spotify Amend Their Loudness Normalisation Target From -12LU to -14LU
Spotify has changed it’s integrated streaming loudness from -12 LUFS to -14 LUFS bringing their target playback level into YouTube who also use -14LUFS and closer to Apple at -16LUFS and closer to the AES recommendation of -16 to -20LUFS. This is good news as it will allow a greater peak-to-loudness ratio giving tracks a greater dynamic range.
In the second part of this series, we will take a look at the research that Eelco Grimm has just completed and presented at the Berlin 142nd AES convention that he did for HKU Muziek en Technologie, University Of The Arts in Utrecht in the Netherlands, in cooperation with the music streaming service TIDAL.