Advanced loudness normalization, New whitepaper from the Music Loudness Alliance Options

[Notat]

I'd like to share a new whitepaper that discusses loudness normalization techniques. The whitepaper has been posted to Music Loudness Alliance website and describes what we believe are improved techniques for loudness normalization. The techniques are applicable to computer and portable file-based playback. For portable applications, hearing loss prevention (as legally mandated in some countries) is integrated with normalization.

We believe BS.1770-2, the revised album normalization, variable target level and the hearing loss prevention features are potentially useful improvements for ReplayGain. With the publication of this paper, these ideas are being put into the public domain.

The MLA is interested in sharing these ideas to improve loudness normalization implementations to the extent that they may enabled by default, or even always on, in players. Intelligent normalized audio playback improves the listener experience, can help protect people's hearing and eventually may lead to improvements in mastering and production quality i.e. the loudness war.

We are interested in any comments (technical or otherwise) and would like to work with the Hydrogen Audio community to help further the adoption and use of loudness normalization in players.

[2Bdecided]

There are lots of good ideas in here, and I hope it does well.

Album level, BS.1770-2 (EBU R128), on by default - these are all good.


I'm concerned that many audio systems don't have a single digital gain control, but have some analogue gain as well (or multiple digital gains). Where this is true, it's wrong to assume that 0dB FS at a given point in the chain represents the loudest sound a system will ever produce. It would be nice to integrate the loudness normalisation into the only volume control (I suggested something like this in 2001), but you've got to anticipate that this won't always be possible.

Where it is possible, I don't like the idea of giving up the loudness normalisation in an effort to avoid peak limiting or compression (pages 6-8). I'm thinking about how this would work on my portable mp3 player. I'm listening to a classical track that's basically quiet but has a few loud peaks. These loud peaks mean that the top part of my volume control effectively doesn't work. I visibly change the volume control, and the volume doesn't audibly change - because those peaks (which might not even be in the part of the album I'm listening to at the moment) are preventing it from going higher. Then I skip to a pop track (I got bored with the classical track - I couldn't turn it up loud enough to hear the quiet parts!) and BANG the volume control is working and the level jumps up by 12dB. You could put some logic in to prevent the jump, and/or visibly lock out the top of the volume control range while the classical track was playing - but either way I still can't turn up the quiet parts of my classical track (or the quieter parts of any dynamic album) to listen to them properly.

If, instead, you allow peak limiting, then if I really want to turn my mp3 player up that loud (maybe because this part of the album isn't that loud!), I can. You could solve the concerns over the audio quality when the limiter acts in a couple of ways - either restricting the range over which you can push it, or putting a "limit" indicator on the display when the limiter kicks in - people would notice that it sounded better when you reduced the volume in this case. (though this only happens during the loudest parts with the player absolutely maxed out - there might be an expectation that it won't sound as good maxed out, so maybe nothing needs doing)


I like the approach to hearing protection - though given that headphones are replaceable items with hugely varying levels of sensitivity, if the hearing protection is finally something sensible which I might want to keep in place (rather than the current idiotic hard limit which, if it cannot be defeated, means I won't be buying the mp3 player!), it might be worth exploring how headphone sensitivity can be reflected. Maybe dial the sensitivity level into the player, and adjust the calculations accordingly?

If it worked sensibly and didn't get in the way, users in the rest of the world might choose to enable the hearing protection. Some might value the display, even if they disabled the absolute limit.


I don't think getting the portable players to calculate the values is a great idea, but I could be wrong. I don't always have full albums on my cheap little player. Many people don't even own full albums - though I suspect that, for the non picky, just calculating the value over the tracks they do have, while not "right", is probably good enough.


Finally, I doubt audiophiles will accept "always on". "On by default" is fine, but some of them will insist on using the full range of their DAC.


Hope this is some help. What you have is basically a great idea. If you can take away the bits that might cause some people concern, the only thinking stopping you is inertia. While there are valid reasons to object, even trivial ones, people can hide behind those.

Cheers,
David.

P.S. Think about metadata, and how you'll market this solution.

P.P.S. If you introduced some kind of dynamic range management at the same time, you could keep everyone happy. Don't forget that Dolby essentially got all this worked out for movies many many years ago.

EDIT: EBU R128 / BT.1770-2 is slowly taking over the world:
see right panel here: http://tech.ebu.ch/Jahia/site/tech/cache/o...dness-faq-radio
see first external link here: http://en.wikipedia.org/wiki/Commercial_Ad..._Mitigation_Act

This post has been edited by 2Bdecided: Jul 17 2012, 18:10

[Notat]

Thanks for your comments David. Very helpful.

With improved performance of DACs, and advent of digital user interfaces, the variable analog gain stage is seen much less frequently now especially in portable players. Even where this is not the case, it's hard to argue that multiple gain controls for the same signal is of value to the user. The proposal challenges designers to make it look like there is one control even if there are multiple. Either they'll fail and (hopefully) we'll be no worse off or they'll succeed and the listening and human-machine experience will be improved.

I believe there is room for improvement for how the the control responds towards the top of the range. We think, however, the best solution is to avoid the problem altogether and build systems with more headroom. The hearing protection features make this safe to do. One thing we did not want to do is try to tackle the dynamic range management issues. We recognize that dynamic range management in the player is useful for playback in noisy environments etc. but defining how that should work is not something we could take on at this time. You can argue that specifying a limiter is a more tractable problem but based on crappy limiter implementations we see in players, we decided the best approach was to make the system linear.

The focus of the HLP provisions is more on meeting requirements of current (EU) legislation than on actually protecting hearing. True HLP is another piece that we considered but decided we could not, in good conscience, take on. Current legislation calibrates SPL based on factory earbuds. There's no known solution to the loophole of using sensitive aftermarket transducers. While a configurable sensitivity parameter is an interesting improvement. We worked hard, however, to reduce the system to a single knob.

There are no hard requirements specified in Annex 4. The purpose the annex is to make player designers aware that there is no authentication mechanism in many metadata systems. Blindly trusting metadata can enable a different kind of loudness war where content providers fudge loudness metadata to make their media more compelling.