EBU R-128 meets K-System v2, a possible future for the loudness debate (Loudness War)

[Compyfox]

This is a judgement about "Loudness" without any clear criteria in the system about "Dynamic Range". It allows a program with a Dynamic Range of 6dB to be "Loudness normalized" against a program that has a Dynamic Range of 12dB.

Now imagine they're the same program - just produced at 2 different dynamic ranges. I'm not sure how this has fixed the Loudness Wars which I thought were about Dynamic Range and over-squished music. It appears that Loudness Normalization still allows me to squish my DR to 6dB as long as the Loudness Normalization algorithm can turn down the volume (normalize) to K-14 for example.

...

I'd like to suggest that "Dynamic Range" be added to the debate.

You found the current weakspot of this whole debate. But the thing is, you can not define a certain "Dynamic Range" as each production is different.


The idea behind the "Loudness Normalisation (Playback)" is that people (musicians and engineers alike) drive back on loudness by themselves one way or another. Why? Because the customers decide which production to go for and which not. It's a subjective thing.

And this(!) was the clear reason why I said "why not offer another interim solution to get there in the first place, so the transition is not as drastic?!". And here it is, K-System v2. It can act as a backbone, and the original idea was intended to do the same. The current trend with Console type emulations (-20dBFS and -18dBFS reference level) also go that route.

However people still stick to peak normalisation rather than loudness normalisation.


It's a vicious cycle. And I fear that even though the "Alliance" might be successful, people would still do whatever the hell they want. So our job, as engineers and interested parties, is to educate as many people as possible. Tell them to drive back their output, rely purely on "sound" again rather than "loud equals better", stick to certain metering rules and do not trust the "loudness normalisation" systems alone to cover your mistakes.

Thankfully, some engineers already realise that, and teach the very same.


But it's again shifting over to the consumers, rather than the industry (as a whole) saying "STOP! It's enough already".



"We have the medium, so why shouldn't we abuse it?!", "nah, don't let bits and HDD space go to waste!", "it's still too quiet! Limiting let's every production sound fat and rich".

Wrong thinking IMO. And I could crawl walls up and down these days yet again. Every other Music Magazine is doing one "Loud Mastering" article after another. This is nonsense.



I just ran my REDD/Altec&VCC combo demo (see the WAVES REDD Console Strip thread in the Effects section) through ToneBoosters Barricade and EBU Loudness, pushed it into K-16v2/K-14v2:

I had a PLR (dynamic range) of over 15dB. Granted, no vocals that might steal some decibels. But I only setup Barricade to cut off the peaks at -1dB TP. And you know what? The limiter barely needed to respond. 1-2dB gain reduction with the peaks only at K-14. This is nothing that you clearly notice. And that production sounds punchy as hell, though a tad muddy (since I was lazy while mixing in the first place).



I really ask myself: Why is it so difficult to realise, that mixdowns at a loudness that you're used to from CDs from the 90ies, are not bad but actually superior?

Unless we talk about the "iPod generation" - those people that were born in the 90ies and never actually grew up with the Tape/Vinyl to CD transition. And engineers that overdrive their gear on purpose since it sounds "way more fat and warm".

[paterpeter]

I'd like to suggest that "Dynamic Range" be added to the debate.

You found the current weakspot of this whole debate. But the thing is, you can not define a certain "Dynamic Range" as each production is different.

I wouldn't call it a weak spot. I see dynamic range from a different persective. Why is the dynamic range of loud master so small? Not because mastering engineers want it to be small. What they want is their mix to be loud, ie. very close to 0dbFS. If the engineer only wanted the DR to be small, they could easily compress the hell out of the track, but keep its RMS at -20dbFS. The small dynamic range is, of course, a direct "by-product" of the effort to bring up the RMS as nothing can go above 0dbFS (obviously). Every mastering engineer knows that DR adds life and punch to the sound, but they accept the low DR as a trade-off to achieve their actual goal. That's basically why tools like Slate FG-X exist - to increase the perceived DR of a track, even if it's "loud".

Now image that there was something above 0dbFS, but that area was only reserved for transients and short peaks and the RMS isn't allowed in there. Mastering engineers could still master rather close to 0dbFS, maybe a bit closer even, as the peaks will go to the area above 0dbFS. They don't need to squash any transients because they can happily "evade" above 0dbFS. So they can freely, musically(!) decide how much of the transients/peaks they want to keep. All mixes would sound comparably loud (a weighted RMS of close to 0dbFS or so for loud parts) but the quality would be better because of the new peak-headroom.

That isn't possible, of course, as there isn't such an area above 0dbFS. But that can be easily fixed by moving this reference point down the digital scale. And that's what those EBU standards and K v2 do. E.g. with K-16 v2, the "0dbFS" point of the imaginary scenario above would be around -16dbFS, and the peak headroom "above 0dbFS" would be -13dbFS to -1dbFS.

This is of course exactly what this whole topic, the standards and the discussion in general are all about and nothing new. My point is: there is no need to define a strict standard for the crest factor/DR IMO. The standards only need to make room for it by suggesting a maximum RMS value/range. In the same way nothing forces you to use the frequency range from let's say 10KHz to 22KHz if you don't want to - but the 44.1KHz audio format gives you the room for it.

If there is room for DR, mixing and mastering engineers can use it in any musical way they see fit.

tl;dr:
Low dynamic range is not the goal but the negative by-product of current loud masters. If the standards suggest maximum RMS values that are low enough, there is enough room for transients/peaks. How much of this room is used by the mixing/mastering engineers is then a question of musical preference (instead of technical limitation), so the DR shouldn't be standardized in any way.

I don't know if that makes much sense... but I hope so.

[kylen]

Whew...lot's to read here Compyfox. It's taking some time to digest. I don't think many people here would argue that you're detail oriented so good thing you're keeping track of all this.

I'm not saying that "Dynamic Range" be constrained in any way. My thought right now is it is an artistic device to a point - it's not my judgement for someone elses production. Although I do get pissed off when some genius clips my consumer player DACs, that used to happen but not so much anymore.

But I am looking for a commonly accepted way to measure Dynamic Range. ToneBoosters might do that with it's EBU Loudness I'm still getting down with it, it is not very expensive. SPAN gives me DR also, it's free. Brainworkx meter is back up to $99 (just missed the Christmas sale!) so that's out. Right now I can say - "...this song has a dynamic range of 3dB" and I think most people will think it is probably squished. True enough I can 'normalize' that in my DAW to K-14 but eech...

That's all I can say for now about some type of catagorization of crest and dynamic range. I'm thinking this doesn't really fit in with the 'Music Loudness Alliance' primary mission of playback 'normalization' in broadcasting and other public utilities & media.

The 'Music Alliance' tools and recommendations along with EBU and other broadcasting governance gives me tools that have features like 'True Peak', etc. that are actually -1dbFS or something of that nature in 'EBU mode'. The spec says EBU mode can be turned off - so the measurement standard would have governed mode vs raw mode so to speak.

Anyway - I'm sure this will all be good for broadcasting & audio but for the loudness wars I'm not sure yet.

So more reading and thinking, lots of specs on the sites you've posted...good debate you guys.

Like you've already said Compyfox - some step forward is better than not doing anything.

P.S. Yes - sorry 'bout the dBm stuff - hangover from my board days...
P.P.S. - Free Orban meter for Win/Mac: The software accepts two-channel stereo inputs and displays instantaneous peaks, VU, PPM, CBS Technology Center loudness, ITU BS.1770-2, EBU R 128 loudness, and Reconstructed 8x Over-Sampled Peaks: http://www.orban.com/meter/

[Compyfox]

... They don't need to squash any transients because they can happily "evade" above 0dbFS. So they can freely, musically(!) decide how much of the transients/peaks they want to keep. All mixes would sound comparably loud (a weighted RMS of close to 0dbFS or so for loud parts) but the quality would be better because of the new peak-headroom.

That isn't possible, of course, as there isn't such an area above 0dbFS. But that can be easily fixed by moving this reference point down the digital scale. And that's what those EBU standards and K v2 do. E.g. with K-16 v2, the "0dbFS" point of the imaginary scenario above would be around -16dbFS, and the peak headroom "above 0dbFS" would be -13dbFS to -1dbFS.

I think you want to suggest to maybe introduce 32bit float media distribution.

The problem is however that DAC's are still locked to 0dBFS. Everything that goes above would mean damaged/destroyed signal. So even if you have a song that has an RMS value of 0dB (RMS), and uses a peak headroom of +20dB (since we're in a floating point environment), it would clip until infinity.

Shifting the loudness down, or using reference levels in the first place, simply does the trick indeed.

If there is room for DR, mixing and mastering engineers can use it in any musical way they see fit.

Definietly agreed.

I don't know if that makes much sense... but I hope so.

Makes perfect sense. At least to me.

The 'Music Alliance' tools and recommendations along with EBU and other broadcasting governance gives me tools that have features like 'True Peak', etc. that are actually -1dbFS or something of that nature in 'EBU mode'. The spec says EBU mode can be turned off - so the measurement standard would have governed mode vs raw mode so to speak.

Anyway - I'm sure this will all be good for broadcasting & audio but for the loudness wars I'm not sure yet.

Turning off the EBU Specs and going "raw" would mean no oversampled peak meter (to get true peak metering) and turning off the weighting filter of the RMS meters. But this would mean that these meters would not be EBU R-128 meters anymore.


The EBU specifications don't mention Dynamic Range in the first place. They don't have to even. The huge Dynamic Range is a, in this case, positive by-product. Along with barely any need for limiting.

Audio streams at -23 LUFS (movie mixes) can have resulting peaks of over 9dB to even using up the full available headroom (see EBU R-128 testfiles). Which would mean, if the signal is at average at -23dBFS (with the EBU R-128 ballistics), your Dynamic Range in the process could be 20 due to rouge peaks and the dBTP might barely be at -3dBTP. Still save by specifications.

The EBU R-128/ITU-R BS.1770-2 meters focus on loudness (and loudness fluctuations measured over a certain time) in the first place. And since the reference levels are set fairly low, you can expect huge dynamic ranges.

This is what the designers aimed at again in the first place. Simplifying and porting it over for measuring music is simple. Though we at least need a reference level shift in this case.

Porting these productions/mixdowns/edits over to playback schemes (the "Alliance") would also be simple - since the loudness is adjusted, not the dynamic range. Which in turn would mean "peak limiting" again.

So more reading and thinking, lots of specs on the sites you've posted...good debate you guys.

Like you've already said Compyfox - some step forward is better than not doing anything.

Agreed.

P.P.S. - Free Orban meter for Win/Mac: The software accepts two-channel stereo inputs and displays instantaneous peaks, VU, PPM, CBS Technology Center loudness, ITU BS.1770-2, EBU R 128 loudness, and Reconstructed 8x Over-Sampled Peaks: http://www.orban.com/meter/

Hm... I don't see any reference level calibration capabilities for the EBU meter. Unless it is locked with the CBS meter (as the page hints at). I might take a closer look. Though it seems to be stand alone (like Pinguin Audio and PAS Products).

Still a nice find. Thanks for that.

[kylen]

I just ran my REDD/Altec&VCC combo demo (see the WAVES REDD Console Strip thread in the Effects section) through ToneBoosters Barricade and EBU Loudness, pushed it into K-16v2/K-14v2:

I had a PLR (dynamic range) of over 15dB... But I only setup Barricade to cut off the peaks at -1dB TP. And you know what? The limiter barely needed to respond. 1-2dB gain reduction with the peaks only at K-14. This is nothing that you clearly notice. And that production sounds punchy as hell, though a tad muddy (since I was lazy while mixing in the first place).

I think this is the best solution for me. I already am using Barricade as a dbFS overload type protection against transients and just push into it a half dB every few bars or so - just so I can tell I'm touching the ceiling at some point(s) during the program. Transparent, as you say nothing you clearly notice.

I think after all of this I would characterize myself primarily as a 'Crester' with a secondary interest of 'Broadcast & Playback Loudness'.

But I can easily switch into EBU compliance as the Loudness topic progresses. My ability to manage my own personal DR (Dynamic Range) taste (squishy or full) will not need to change.

Yes - the Orban is standalone (not VST) although it responds when I run Reaper as it can listen to my audio interface (spidif). I just found it last night after reviewing your metering threads from last summer and looking at various tools. I might run a test tool like that for my DR needs after I render (it takes about 20% cpu on my I3) or one of the other methods you've posted. While remastering I have Barricade and found an old copy of TT Dynamic meter (might get bx_meter if it ever goes on sale again not sure if I like it yet). SPAN doesn't give me the DR I'm looking for (it says Max crest factor but the numbers don't make sense) so that's out. Melda Loudness and ToneBoosters seem to match the LRA (Loudness Range) but those numbers I'll have to learn to appreciate.

Anyway - back to the matter at hand. Loudness Equality across all media!!!
[Phase 2]

EDIT: BTW is there a master glossary somewhere for items that the Loudness Alliance has set out? I didn't find one looking at the EBU website either - thought you might have a link handy. Thanks.

[Compyfox]

I think after all of this I would characterize myself primarily as a 'Crester' with a secondary interest of 'Broadcast & Playback Loudness'.

Ignore the (playback) loudness normalisation for the time being. Just work at K-14v2 and K-16v2, and you automatically get a huge dynamic range value. Not to mention you're currently kind of future proof.


The thing is however, will the clients accept such low loudness (as of this moment, not yet)? That is currently one of the main points while debating "it's too quiet and not competitive".

It'll take a while and a lot of patience to get that sorted.

Yes - the Orban is standalone (not VST) although it responds when I run Reaper as it can listen to my audio interface (spidif).

Hm... reminds me of RME DigiCheck. This software can listen to the playback port of my RME setup. Definitely need to check that tool out.

... and found an old copy of TT Dynamic meter (might get bx_meter if it ever goes on sale again not sure if I like it yet).

IIRC, you can (still?) crossgrade from DR-Meter - though you will miss out on the offline DR measurement tool. Did you contact Brainworx regarding upgrade paths?

Melda Loudness and ToneBoosters seem to match the LRA (Loudness Range) but those numbers I'll have to learn to appreciate.

Haven't used the MLoudness plugin for a while. Liked the concept, but elder posts on KVR showed that the plugin lacked certain precision. Namely the True Peak meter. I lost track to be honest if that was fixed, since my main focus were indeed on the ToneBoosters plugins. And currently, the TB EBU Loudness/Compact plugins are indeed a great deal.


Though I remember that the LRA readouts on MLoudness was dnymic (read: didn't stay on "lowest peak", reset itself after a while), while the PLR readout for EBU Loudness goes by "held lowest dynamic range" until you reset the whole VST.

Both do measure the crest factor/dynamic range like TT_DR-Meter/bx_meter. Just without the meter but rather give you final numeric readouts. Much like "highest peak" (on hold). Simple, yet effective IMO. Nothing much to learn here.

Anyway - back to the matter at hand. Loudness Equality across all media!!!
[Phase 2]

That, and maybe debating why certain peope want to stick with high loudness (which is a subjective/marketing thing, rather than a technical advantages), or if there are concerns that certain loudness values are too low.

I'm really curious about that. Currently, the debate is between you, paterpeter, Burillo and me. That is a bit... disappointing. Though monitoring the hits of the thread shows that there is a certain interest.

EDIT: BTW is there a master glossary somewhere for items that the Loudness Alliance has set out? I didn't find one looking at the EBU website either - thought you might have a link handy. Thanks.

I got ton of additional information through contacting the developers themselves. But if you want to read the official statements and whitepapers, just follow the links on page 1.

If there will be updates, I'm sure you can find them at the source.

[paterpeter]

... They don't need to squash any transients because they can happily "evade" above 0dbFS. So they can freely, musically(!) decide how much of the transients/peaks they want to keep. All mixes would sound comparably loud (a weighted RMS of close to 0dbFS or so for loud parts) but the quality would be better because of the new peak-headroom.

That isn't possible, of course, as there isn't such an area above 0dbFS. But that can be easily fixed by moving this reference point down the digital scale. And that's what those EBU standards and K v2 do. E.g. with K-16 v2, the "0dbFS" point of the imaginary scenario above would be around -16dbFS, and the peak headroom "above 0dbFS" would be -13dbFS to -1dbFS.

I think you want to suggest to maybe introduce 32bit float media distribution.

It was actually a very lengthy and clumsy explanation why I consider approaches like the K-System the realization of a fictitious "better world" for audio engineers (at least I suppose, as I'm no pro)

The thing is however, will the clients accept such low loudness (as of this moment, not yet)? That is currently one of the main points while debating "it's too quiet and not competitive".

Yeah, that's obviously the big question. I guess this will only happen if RMS normalization (at "low" RMS values) is widely used (and inevitable like you said), ie. by radio stations, clubs and media players. If that's the case their tracks will sound worse compared to high-DR material, potentially causing loss of sales. That's probably the only way to encourage a change. But that requires cooperation of many parties.

The weird thing is, that most listeners don't even realize what the problem is. For example, my girlfriend always gives me a weird look if I start complaining about the squashed sound of radio stations. You know, when the drums in a sparse verse are loud and clear and totally collapse to a wimpy nothing as soon as the chorus starts and guitars kick in etc. Driving in a car with me can be exhausting

[DuX]

Thank you Compyfox for starting this thread! However, I'm too tired of discussing it with [mostly] pigheaded people... I'm one of those "weirdos" who always mastered everything [that I could] at kinda K-14. I found it a pretty safe solution for everything, and the transients get preserved 99% of the time. Also, any VU-meter can serve as a K-14 loudness meter, so that makes my life as a half-hardware, half-software guy easier. If everybody just at least stuck to the 0 as a maximum peak with a VU-meter [average audio window ~300ms] everyone should be alright. But of course - people are people...

I just hope some standard will catch up with music. Any standard, just give us one, and people accept one. I still vote for good old VU-meter as a reference and *very* light limiting.

ITB, I now mostly use Sleepy Time Record's VU-meters for -18dB and -14dB reference as they're soooo smooth. -18dB as a reference for tracks, and -14dB as a reference on the master buss. I cannot thank enough d.bop for making those plugins. OTB, I use the peak meter on my TC BMC-2 which is not a true peak meter but DIN standard one. Great DA also. I couldn't live without this little bugger.

Cheers!

[X2]

I found this helps

http://www.orban.com/meter/

[Compyfox]

Yeah, that's obviously the big question. I guess this will only happen if RMS normalization (at "low" RMS values) is widely used (and inevitable like you said), ie. by radio stations, clubs and media players. If that's the case their tracks will sound worse compared to high-DR material, potentially causing loss of sales. That's probably the only way to encourage a change. But that requires cooperation of many parties.

Currently, the Loudeness Normalisation schemes are only active (at least in theory) in a broadcast environment. It's a bit different with webstreams (rules are still bent) - but it's effective since August 2012.

It will take a while until it will be ported over to other medias and playback systems as well. Though different rules apply for live shows for sure.

I just hope some standard will catch up with music. Any standard, just give us one, and people accept one. I still vote for good old VU-meter as a reference and *very* light limiting.

Actually, a plain VU meter is non-weighted and has the problem with emphasis on lowend. The EBU R-128/ITU-R BS.1770-2 specifications evade these problems and work (IMO) great for mastering content.

If you'd only use a plain VU to utilize the "K-System" specs, you're still off a couple of dB's. It's a guess-timation. K-System v1 uses 600ms rise/fall, K-System v2 uses the EBU R-128 specs (weighting filter, reference level, color codes).


I still recommend using these three meters at least:

Recording/Post Production:
VU with ANSI C16.5-1942/British Standard BS 6840/IEC 60268-17 standard, reference level at -18 dBFS
Digital Meter with color codes (green up until -18dBFS, yellow from -18dBFS to -9dBFS, red from -9dBFS onward)

The mix bus "can" have a red zone from -6dBFS and up.


Mastering (Music):
EBU R-128 meter, with K-System v2 specs.


A DIN (5ms integration time) or EBU (10ms integration time) PPM could be used as well, but then you are not allowed to go past 9dB.

Example: Your mixing console has a reference level of -18dBFS, then the 0-point on your consoles meter bargraph usually mean: -18dBFS = 0. -9dBFS would then mean +9 on the bargraph. In case of a Behringer MX2004/2442 (which use bargraphs similr to PPM's with 5ms), the signal peak "per channel" should not exceed 9, or the yellow zone.

Readouts on PPMs can differ, as you can set them up to your needs of couse.


But this is why I highly advise to use both a VU and PPM in combination. Or within a DAW a VU and Digital Meter. With a globally accepted reference level of course.

ITB, I now mostly use Sleepy Time Record's VU-meters for -18dB and -14dB reference as they're soooo smooth. -18dB as a reference for tracks, and -14dB as a reference on the master buss. I cannot thank enough d.bop for making those plugins.

Again, the Sleepy Time Records VU is fixed to 300ms. I currently use Klanghelm VUMT, which can be setup to my needs (no weighting filter available!). PSP Vintage Meter wasn't updated in a while (doesn't work on my rig anymore). The next nearest VU/PPM VST I'd recommend is zplane's PPMulator. It also comes with an EBU R-128 meter.

orban.com/meter

I just tested this little tool. And I'm pleasantly surprised.

Installed, it only eats 7MB on my HDD, works right out of the box with my mainboard built in RealTek Audio Driver and has easy to understand visual readouts.

A bit of a turnoff is the lack of calibrating the VU. I think it's locked to -18dBFS, but that's fine with me. The CBS/ITU meters are connected in terms of setup (Display, CBS Loudness and ITU BS.1770 belong together). If one wants to setup this meter to K-16v2 (for example), simply set the meter scale to "relative", the meter range to "EBU +9 Scale" and the Reference Level to -16LKFS (see difference LUFS and LKFS). The ITU BS.1770 type needs to be BS.1770-2. Then you're good to go.

Though I do recommend to switch the PPM to 5ms rather than 10ms. And the "reconstructed peak" is (to my understanding) a True Peak meter.

Can't make much off of the logs, and the peak hold can't be set. But this thing is for free and made by the people of the the infamous ORBAN Broadcast Compressors (Optimod). Definitely a nice gift one way or another.

[DuX]

Thank you for the info Compyfox! Yeah, I figured that normal VU meter doesn't have a weighting filter, and that often I have to trust my ears to match the loudness. I didn't know that EBU R-128 uses a weighting filter! That's great news!

Now where to find a good EBU R-128 meter... I wonder if RME Digicheck conforms to EBU R-128. I think I read that they implemented it in V5 of Digicheck. There is a freeware one at Audication, and there's Jeroen's TB R-128 loudness meter. I like the PPMulator, too, it works so smooth. Thank you for finding the free Orban meter. One of these should work. Probably Digicheck, Audication one, or Orban because I'm a little 'dry" currently...

edit: Digicheck 5 has EBU R-128 meter! yay!

[Compyfox]

The RME DigiCheck got the EBU R-128 meter with one of the last updates indeed. You can also change the reference level down to -20dBFS, which is indeed neat.

The Audication EBU meter can't be setup in terms of reference level. Neither can be the Steinberg one (which only works in Cubase 6.5/7 and Nuendo).


So if you're "dry", stick with RME Digicheck for the time being. If you got some pocketmoney to burn, take a closer look at ToneBooster's EBU Loudness which only costs 15EUR. Though the feature set is definitely unbeatable.



You should however read the white papers of both the EBU R-128 meter (which clearly mentions the weighting filter, and how the loudness is measured), on top of my white paper regarding the K-System v2.

You can't use a tool effectively if you don't know how it works.

[DuX]

I actually checked EBU's white papers some time ago, but it wouldn't hurt to throw another look. I like to know how stuff works, too. R-128 is a little harder to comprehend than a common VU-meter.

[paterpeter]

Just for the sake of it I ran the Beatles' Let it be through EBULoudness using the K-16v2 preset. Here are the results:


The loudness range is HUGE compared to typical contemporary songs. Now, obviously this version does not adhere to K-16v2 as it is too loud at its loudest parts.

I'm a bit confused about how to "correctly" put the track back to K-16V2.

1. Reducing gain by -4.2dB will give an integrated loudness of 0LU. But that would still mean that some loudness peaks exists in the "red" part. The graph only shows short term loudness, the momentary loudness gets as high as almost +9dBLU.

2. If instead the gain is reduced by 6 to 8 dB, the loudness peaks stay within the amber zone as suggested, but the integrated loudness is considerably less than 0dBLU (e.g. -2 to -4 dBLU).

I suppose the way to go is the second approach, right? The integrated loudness is then simply a property of the track, ie. it's not the goal to stay very close at 0dBLU. Is that correct? For tracks with large (macro-)dynamic range this will therefore result in a lower integrated loudness than tracks with fairly constant loudness.

[Compyfox]

Just for the sake of it I ran the Beatles' Let it be through EBULoudness using the K-16v2 preset. Here are the results:

From the looks of it, it's one of the many remasters?

I'm a bit confused about how to "correctly" put the track back to K-16V2.

1. Reducing gain by -4.2dB will give an integrated loudness of 0LU. But that would still mean that some loudness peaks exists in the "red" part. The graph only shows short term loudness, the momentary loudness gets as high as almost +9dBLU.

This is another problem that needs further work:
For individual music, Integrated Loudness does not work. The Integrated loudness shows the loudness over a certain course of time. It's made for broadcast streams only, telling you how loud a program is overall. While not taking into consideration the gating of the measurement at -10LU and -70LU.

It does work however, if you have a whole CD analysed. Like... 10-15min at least, ideally 20-30min. Then you can work on adjusting tracks so that they won't sound too jumpy through the course of the whole set.

But again, I wouldn't advice going for it for music. At least not individual tracks.

2. If instead the gain is reduced by 6 to 8 dB, the loudness peaks stay within the amber zone as suggested, but the integrated loudness is considerably less than 0dBLU (e.g. -2 to -4 dBLU).

That should be the case, yes. So simply ignore it for the time being and rather focus on the ML values.

I suppose the way to go is the second approach, right? The integrated loudness is then simply a property of the track, ie. it's not the goal to stay very close at 0dBLU. Is that correct? For tracks with large (macro-)dynamic range this will therefore result in a lower integrated loudness than tracks with fairly constant loudness.

Again, this is a problem for music, but not for broadcast streams where music is not really an issue but an "added effect".


This is why I wrote in my white paper, that we need some sort of indication as Meta Tag as read-out for playback devices in terms of what's going on. Though this system is not perfect yet either.

I suggest to offer two values:
1) the K-system or at least the meter calibration used (in our example: -16LUFS)
2) the maximum loudness measured with that particular reference level.

The playback system then only needs to adapt accordingly.


This is also no perfect solution, but analysing one song alone, and then going by the Integrated Loudness doesn't work. Short Term Loudness is just an indication for "jumping parts" while broadcast and whether or not you're still within specs (3s), certain songs (like from Pink Floyd) also have an ungodly low SL on average. Momentary alone doesn't work either, and there is no "average" measurement like it is handled with "Integrated Loudness".

A mathematical middleway would be great, but as with Integrated, how does the system know if the production is just once going towards the red zone but stays overall in the green zone, or if it's constantly in it? There are just no rules set for this, and this will be a debate for years to come.



To further debate this, I just gave several tracks a spin, ordered by year of release:

K-16v2 test batch
The Beatless - All my Lovin (from a probably 90ies Best off CD, 1963)
Neil Diamond - Cherry, Cherry (1966)
Pink Floyd - Another Brick in the Wall (pt 2, 1979)
ACDC - You shook me all night long (1980)
Bruce Springsteen - Born in the USA (1984)
Sting/The Police - English Man in NY (1988)

K-12v2 test batch
Anthrax - Taking the Music back (2003)
Wolfsheim - Kein Zurück (2003)
Silbermond - Das Beste (2006)
MUSE - Undisclosed Desires (2009)
NOISIA - Could this be (2012)



Here are my results, non compensated and only numeric readouts (no screenshots, sorry):

First batch of songs:
VU - Calibrated to 300ms/-18dBFS
EBU Loudness - K-16v2 preset


The Beatless - All my Lovin (from a probably 90ies Best off CD, but not remastered, 1963)

Code: Select all

VU (avg):    +2 VU (peaking beyond +4VU on the R)
ML (avg):    +0,5 LU
ML (peak):   +1,8 LU
SL (peak):   0,6 LU
Integrated:  -0,4 LUFS
LoudnRange:  +1,6
PLR:         +9,9
True Peak:   -6,5 dB

Compensation: barely, though I would (with a remaster) change the stereo field. Just switching the signal to mono resulted in a different VU readout (avg. +1VU, peak +3,5VU). Though the EBU meter is unaffected by that. With a new mix however, the perceived loudness wouldn't be drastically shifted as with the original recording, just more balanced.

Neil Diamond - Cherry, Cherry (1966)

Code: Select all

VU (avg):    -3 VU (music only), 0 VU to +2 (with vocals)
ML (avg):    -3 to +1 LU (with vocals)
ML (peak):   +2,6 LU
SL (peak):   +0,3 LU
Integrated:  -2,4 LU
LoudnRange:  +5,7
PLR:         +17
True Peak:   -1,4

Compensation: undecided to be honest. The ending is gaining some dB in fact. I'd barely do anything - if at all.

Pink Floyd - Another Brick in the Wall (pt 2, 1979)

Code: Select all

VU (avg):    -2 to 0 VU (peaking up to 4 VU with shouts)
ML (avg):    -4 to -6 LU
ML (peak):   +4,7 LU
SL (peak):   +0,6 LU
Integrated:  -4,5 LU
LoudnRange:  +9
PLR:         +18,8
True Peak:   -1,8 dB

Compensation: at least by -2dB, though then I'd be at -7LU on avg with the music. This track clearly showed me that standard VU's were used for mastering (see -2 to 0 VU on avg), and only "abused" for fortissimo parts (+4 VU). "The Wall" was from 1979, friggin dynamic for that time, and not as washed out as The Beatles.

ACDC - You shook me all night long (1980)

Code: Select all

VU (avg):    +2 to +3 VU (peaking beyond +4VU)
ML (avg):    +2 to +3 LU
ML (peak):   +5,1 LU
SL (peak):   +4,2 LUFS
Integrated:  +2,4 LUFS
LoudnRange:  +4,6
PLR:         +13,5
True Peak:   -0,4 dB

Compensation: judged by the loudest/busiest part of the song, I'd go by -3dB to hover around -2LU to +1LU on ML. Else it's a healthy K-14v2 track, with tendencies to K-12v2. Which gives me an indication that this might be an early  rerelease.

Bruce Springsteen - Born in the USA (1984)

Code: Select all

VU (avg):    0 to 1 VU
ML (avg):    -2 to 0 LU
ML (peak):   +1,4 LU
SL (peak):   +0,1 LU
Integrated:  -1,5 LU
LoudnRange:  +2,4
PLR:         +15
True Peak:   -2,5 dB

Compensation: probably none, maybe +1dB for fortissimo parts at most, but no need.

Sting/The Police - English Man in NY (1988)

Code: Select all

VU (avg):    +2 VU (peaking way above +4 VU)
ML (avg):    +2 LU
ML (peak):   +8,4 LU
SL (peak):   +6,1 LU
Integrated:  +1,4 LU
LoudnRange:  +5,3
PLR:         +15,3
True Peak:   +1 dB (clips!)

Compensation: I just tested a compensation of -5dB. This results in an avg. ML of -3LU to 0LU, and ML peak of +3,6LU. If the track wouldn't have been clipped prior to it's release, I'm sure I would reach up to 18-20dB as PLR value

I am now switching to K-12v2 and the VU to -14dBFS


Anthrax - Taking the Music back (2003)

Code: Select all

VU (avg):    +2 to +4 VU (way beyond +4 VU in fortissimo passages)
ML (avg):    +4 LU
ML (peak):   +6,6 LU
SL (peak):   +5,7 LU
Integrated:  +3,9 LU
LoudnRange:  +2,9
PLR:         +9,3
True Peak:   +1,2 dB (clips!)

Compensation: Since we talk about "modern mixes", we can't reproduce the PLR values, unless we use specific "unclipper" tools (all of them are still in development IMO). A gain compensation of -4dB to -5dB resulted in a ML of -1,5LU to +0,5LU, and peak up until +2,8LU. Much more enjoyable

Wolfsheim - Kein Zurück (2003)

Code: Select all

VU (avg):    +3 (non busy parts) to way beyond +4VU (busy parts)
ML (avg):    +4 to +6 LU
ML (peak):   +7,4 LU
SL (peak):   +6,5 LU
Integrated:  +4,3 LU
LoudnRange:  +6
PLR:         +8,7
True Peak:   +0,8 dB (clips!)

Compensation: Again, "modern mixes", we can't reproduce the PLR values. I consider this a great electronic/pop/IDM production that suffered from overcompression. I also used -5dB as gain compensation. Suddenly the ML (avg) is around 0LU, and I use the full amber zone for busy parts. This can be further finetuned, but it's a start. Sad for this production, really - there is also noticable distortion thanks to the MP3.

Silbermond - Das Beste (2006)

Code: Select all

VU (avg):    +2 to +4 VU (higher if the drums kick in and at end of song)
ML (avg):    +1 to +4 LU
ML (peak):   +8,3 LU
SL (peak):   +7,3 LU
Integrated:  +3,4 LU
LoudnRange:  +11,1
PLR:         +11,3
True Peak:   +2,6 dB (clips!)

Compensation: "modern mixes", we can't reproduce the PLR values. I also consider this a great pop production that suffered from overcompression. Though here it was intentional (upfront compressed vocals, etc), but it's less noticable after MP3 encoding compared to Wolfsheim and surprisingly dynamic. A gain compensation of -6dB let's this song breathe a bit more. Wish it was released like this in the first place.

MUSE - Undisclosed Desires (2009)

Code: Select all

VU (avg):    -3 to +4 VU (verses), way past +4 (refrain)
ML (avg):    -1 to +2 LU
ML (peak):   +5,0 LU
SL (peak):   +4,4 LU
Integrated:  +1,3 LU
LoudnRange:  +16,6
PLR:         +10,9
True Peak:   +0,3 dB (clips!)

Compensation: "modern mixes", we can't reproduce the PLR values. I chose this song since it's not your usual in-your-face MUSE production, but more dynamic. Still it's a tad too loud (unfortunately, all MUSE CDs are too loud), and sounding a bit dull as well. Could also come from the YouTube rip I did. Gain compensation by -2dB did the trick (ML avg from -3LU to 0LU, ff passages at +3LU), maybe some highshelf as well. But it worked in this case.

NOISIA - Could this be (2012)

Code: Select all

VU (avg):    way past +4
ML (avg):    +4 to +6 LU
ML (peak):   +8,8 LU
SL (peak):   +7,9 LU
Integrated:  +5,1 LU
LoudnRange:  +7,1
PLR:         +8,1
True Peak:   +1,2 dB (clips!)

Compensation: "modern mixes", we can't reproduce the PLR values. I snagged this one directly off of the official music video of Noisia's 'tube page. It's the loudest track I could find (other than Metallica's Death Magnetic) that I still enjoyed listening to. This track uses a lot of bass (drumloops and distorted synth), so I got a constant light show. I first tried a gain compensation of -5dB, then compensated additional -1dB to -1,5dB - much better. I only wish I could repair the transients.

As you can see, different productions, varying readouts. Especially on the Integrated Loudness side of things. Also surprising - already noticable loudness variations from end 70ies to early 80ies (vinyl and to some extend cassette times, first CDs came up in 1982), which went downhill end of the 90ies (era of CD and MP3). Unless certain songs I measured are reprints after the CD came up (which could be possible for AC/DC and Sting). For compensation, I mostly trusted my instincts and went for values that were around "ML average" values.

Unfortunately, you can not measure "avg" values for ML as of this moment. So I hope to see an update for TB Loudness where I could see a histogram for the ML as well instead of SL alone. This makes it easier to judge where the average level of the production was.

But nothing that can't be done through training and trusting your ears.



I'm sad with certain productions however. The Noisia and the Silbermond one for example. With an experimental Declipper I still had on my HDD, I could even get back a PLR of +14,5 for the Silbermond production (K-14v2 results in ML +2,5LU peak, -2,0dBTP, PLR +14,5).

Pressing the song to it's limits good and fine. But at a more suitable loudness (i.e. K-12 or better K-14), it sounds just as good. If not considerable better.