Aug 18, 2011
STOP PRESS – Why the Loudness War HASN’T reduced ‘dynamic range’ !
I’ve just read a truly excellent analysis of the effects of the Loudness War on the music we listen to in Sound On Sound magazine, and – hold on to your hats – according to Emmanuel Deruty:
“There is no doubt about it: contrary to general belief, there has been no obvious decrease in loudness range due to the loudness war, and brickwall limiters have not reduced the loudness range in music production.”
In fact, the article argues very convincingly that “dynamic range” is so poorly defined that’s it’s pretty much useless as a way to discuss the effect of the so-called “war”. It goes on to look at the issue in great detail, trying to find out – if it’s not reduced dynamic range that makes Loudness War casualties sound bad, what is it ?
So, as the organiser of Dynamic Range Day, does this mean I’ve been barking up the wrong tree ? That we don’t have to worry about the Loudness War after all, or that it’s NOT making modern music sound worse ?
No.
It does highlight the point that using simplified language like “dynamic music sounds better” isn’t strictly meaningful in a technical sense, but that phrase – and the whole Dynamic Range Day site – is somewhat simplified to help make the issues as clear as possible. And the article itself goes on to conclude that the over-use of limiting and compression does tend to cause
“…reduced crest factor, envelope modifications… and in the worst cases, distortion. Common sense suggests that although there is nothing wrong with these characteristics as such, they shouldn’t be on virtually all records“
Which in a nutshell, is the main message of Dynamic Range Day. We shouldn’t feel obliged to apply extreme processing to all music and styles in the mistaken belief that it will make our music sound better on the radio, or more “competitive”.
And the fact remains – the easiest way to avoid all those problems is to simply use something like the TT Dynamic Range Meter and avoid going beyond an average of DR8 in your music. Sure, strictly speaking the TT Meter is measuring “crest factor” rather than “loudness range” as defined by the EBU 3342 specification – but really, at this point we’re just in the land of semantics !
The article still contains some fascinating analysis, though – the discussion of exactly why so many people think Metallica’s “Death Magnetic” sounds terrible is particularly interesting, and concludes that it’s actually a relatively unusual special case – a combination of very low crest factor and very low RMS variability – all the time.
I’m a little concerned that all this will be mis-interpreted – mainly because the headline conclusion that “the loudness wars haven’t reduced dynamic range” may mask the underlying point that much music genuinely does sound worse when crushed and distorted – what works for rap or death metal probably won’t sound good on mainstream pop like Justin Bieber, or singer-songwriter Adele !
And even though “Death Magnetic” was a “perfect storm” of factors, just because it was a special case doesn’t mean this isn’t still a real, damaging trend in modern music production values – as the article says,
“Obviously, limiting does something ‘wrong’ with the signal, otherwise people wouldn’t be complaining so much”
However these reservations are only about the way the article may be (mis-) interpreted. They can’t take away from the fact that it’s a superb piece of analysis and comment, and if you really want to understand the issue of the Loudness Wars, it should be required reading. You can find it here in full:
‘Dynamic Range’ & The Loudness War
Meanwhile the fight goes on, perhaps with some revised technical language choices in future !
Related posts:
.gif "Dynamic Range Foundation - Active Member")
Hi Ian,
thanks for commenting the article. The risks of it being misinterpreted are in fact quite high, and that’s why I brought it to your attention on Facebook (Lea is me using my mother’s account LOL). Btw, I mostly listen to metal music and hate the sound of Loudness War metal. Drums lack punch and vocals lack clarity when everything’s squashed and distorted – after all, reality is still dynamic during death metal concerts, too! I wish all drums and vocals in metal albums had the same punch and clarity as in Europe’s self-titled debut album from 1983 (I know that’s not metal: just a very well done dynamic production).
Hi Andrea/ Lea !
Sadly even live engineers are starting to use Waves plugins to try and get closer to the sound of the CDs…
Ian
Well, the article as you describe it surely sounds like someone that likes to listen to himself and loves to contradict for contradiction’s sake (or for semantics’, for that matter) – and I know what I’m talking about since I used to be very similar (hopefully not so much anymore).
Therefore I don’t think I will even read the original article – I very much prefer just supporting the mission that you and other fine engineers are on!
Wow, awesome article!
However I have to call out HUGE bullshit:
“Such crest‑factor values are comparable to what can be found on tracks from Kanye West’s My Beautiful Dark Twisted Fantasy, or 50 Cent’s Get Rich Or Die Tryin’. Those are stylistically loud urban music albums with really strong percussive elements that articulate the writing, and are better suited to low crest‑factor values than Metallica’s constantly buzzing guitars. They are also comparable to tracks from MGMT’s Oracular Spectacular or Congratulations, two albums with a sound so distinctive that a constant use of the second loudness paradigm and/or dynamic compression artifacts is not a problem at all. But Metallica’s ‘classic’ sound simply doesn’t easily allow for sonic extravaganza.”
Not a problem at all? Have they even listened to them? MBTWF and MGMT are punishing and EXTREMELY loud. They push the envelope a LOT harder than anything mainstream by 1-2db. They are trying to downplay the issue and that brings up another thing. Why should it be “ok” for them and NOT Metallica if they choose? Double standard. I often see metal or rock or such talked about in terms of dynamic range, NEVER pop or popular music, like it’s trash and it’s ok if it sounds bad, it’s just made like fast food. That is not ok at all and a big part of the problem, why indie and smaller groups have this too… everybody copies what works commericially! If antyhing, popular music has to held to a higher standard at least technically for this reason alone.
Personally I LOVE distortion though… but not UNIFORM distortion. And I only mean creative distortion. There is a huge difference between jacking one instrument to hell and back versus crunching the whole thing at once! That is what I mainly hate, when the entire sound suffers at once and moves away from the original master. Otherwise, I do not mind crunching, dirty, distorted records. It is most noticable usually BECAUSE they are so dyanmic to start with and you are hearing reductions everywhere. Take already distorted on purpose selectively and you woudln’t be able to hear much difference becuase the waveforms aren’t getting as aggressively shaped.
I have also heard such overtly distorted records with DR of 11-12. There is just so much to start with (huge peaks and such) that reducing to commercial levels of today is WAY overkill.
Hey there
I’m the author of the article. Just a few comments if I may.
Ian Shepherd => thanks a lot for your positive review! it’s great to read you. Indeed no, you haven’t been barking up the wrong tree, at least IMO. But ‘dynamic range’ isn’t really the enemy here. Rather, one troublesome aspect would be the ‘crest factor’ – or ‘peak salience’, rather. One other aspect may be the shapes that are assumed by loudness over time in songs. It has to be verified, but it might be possible that they’re getting ‘squarer’.
John Richard Dominy => when you say “The wav looks like a flat bar”, are you aware that what you’re saying is that the peak levels are stable? and, quoting Bob Katz, “peak levels have (almost) nothing to do with loudness”. As a way of consequence, stable peak levels have (almost) nothing to do with low loudness range. There is in fact more loudness range in Lady Gaga’s “The Fame Monster” (median LR over the album = 4.5 LU) than there is in The Cure’s “Pornography” (median LR = 3.8 LU), Nirvana’s “Nervermind” (2.0 LU), or Motorhead’s “No Sleep till Hammersmith” (3.6 LU). The median LR over the corpus I used is 5.1LU. “The Fame Monster” is indeed over-average in terms of LR!
Kahlbert => if you had read Ian’s review, you’d know that he and I are going in the same direction, and agree with each other. Knowing that, how on earth could he be right and I wrong? Plus, if you don’t read the article, how do you know whether it’s BS or not?
Greg => you raise an interesting point. “Why should it be “ok” for them and NOT Metallica if they choose?” Two main points:
- Argument 1: Because in Metallica, there is a loud guitar that features an already static dynamic envelope (low peak salience / crest factor, low RMS variability). Put a limiter on top of that and you run into trouble. I know it’s only a guess, but I think the original material (before mastering) in MGMT was much less stable: if I’m correct, it featured a higher peak salience / crest factor, and a higher RMS variability. The same applies for MBDTF. An interesting point: the death metal band Dagoba handles very low RMS variability and low peak saliences by lowering the guitar and pumping up the drums in the mix (compared to “traditional” metal mix balance). This way, spectral variability is much higher, and makes it up for the low RMS variability: it sounds huge and powerful, not uselessly noisy as it’s the case for Death Magnetic. Maybe it’s what Metallica should have done.
- Argument 2: Metallica’s sound was defined by albums such as “Master of Puppets’ and “Justice for All”, it seems to be a curious decision to take a Metallica album that include similar arrangements to “MoP” and “JfA” and compress it as if it were a style created in 2005 (they probably wanted to sound “modern”?). On the other hand, MGMT is some kind of twisted revival of psychedelic music, their sound is not “classic”, they can afford to be much more daring IMO. In the case of MBDTF: rap has always been very much compressed (even King Tubby, which has a clear filiation to rap, was already awfully compressed in 1981….), why not trying to get the trend further? I see no bad taste there. The extreme compression on the guitar loop and crunchy vocals in “Gorgeous” is delightful IMO.
Generally speaking, my point with this article is the following: the loudness war causes trouble. But the trouble doesn’t lie in loudness range decrease, because it remains stable. Then what is the problem? Know your enemy: if you want music to be made more listenable, know what to fix first. Struggling against a pretendedly low loudness range whether the problem is entirely elsewhere won’t lead audiophiles anywhere…
Erratum: my sentence ““The Fame Monster” is indeed over-average in terms of LR!” is wrong. For those who are curious, in EBU3342 distribution, I’ve found quartile 1 to be at 3.4LU, median at 5.1LU, and quartile 3 to be at 8.2LU. “The Fame Monster” is near-average, not over-average…
Emmanuel,
Sorry if I offended you! I may have gotten a wrong impression from reading the review only. I’m just so sick of listening to overcompressed music, and even more sick of having to even argue about the issue!
Still, if we’re “on the same side” on the actual topic (semantics aside), I’m certainly fine with that. The rest you may blame on me being not a native speaker. Peace?
All the best – and to good sounding music!
Kahlbert
Germany
@Emmanuel
I still think they are way too loud. Guess we’re entering a matter of taste though. Actually the whole argument makes a matter for taste anyways. I guess I am more sensitive to things like MGMT vs Death Magnetic. You talked about the two differnet kinds of loudness methods and I guess I’ll mention another one: I’d rather take crispy and jagged over flat/cleaned up sounding. That’s just me Reminds me of when you deblock video – at what point do your video deblocking/dering/etc settings clean it up so much that you are losing “sharpness”… technically losing blockiness, but you know what I mean. Everything looks too smooth to me. I don’t have a degree or anything but I believe one can easily hear the difference between high compressor action vs high limiting. High compression sounds more smooth yet worked, whereas more limiting I think sounds more sharp yet distorted. I think the trend is definitely near more compression too, a lot of times you see very high RMS pieces with overall flat shapes yet no TRUE flat edges. Limiting would do the opposite. But what do I know lol.
to Kahlbert => sure! Alles gut.
to Greg => I agree with you entirely, each and every point – I might add that it’s something curious that a band like MGMT, whose arrangement are quite colorful and interesting, appears to be so preoccupied with “graying” and “flattening” everything during mastering (especially obvious in their 2nd album), even though it’s part of their style and it often leads to interesting effects
Hi Emmanuel,
thanks for taking your time to come over here and comment/answer our questions. I’ve been trying to read your article but it seems to be way out of my technical grasp and I therefore feel compelled to put a query the answer for which is most likely already there (I simply need a “for dummies” version). If, as Ian said, the TT Dynamic Range Meter doesn’t really measure the dynamic range of a song, why does there seem to be a correlation between loudness, squashedness/flatness, distortion and low values in decibels measured by the TT DR Meter?
Also, why is it ultimately inaccurate to describe the dynamic range of a song as the difference in sound pressure between the quieter and the loudest sound it comprises?
Thanks in advance,
Andrea
@Andrea
They are reformulating the dynamic range meter, it should be out next month. It is incoporating the EBU R128 new loudness measuring but besides, that, I don’t know. It’s by Algorithmix so they know what they’re doing but it might answer some of your questions in action.
One thing I can say is that the DR range scores of almost EVERY piece I have plugged into it (the kind like Lady Gaga where the loudness is linked to chorus vs verse sections) are about the same as the highest RMS value present. I believe DR numbers (for a piece, not real time) comes from the 20% loudnest sections then looking at their RMS numbers. So the correlation of those characteristics is because they are present in high RMS pieces usually (wait is it high or low… higher number because it sounds louder? or low because its’ a lower number?)
@Andrea
The TT DR meter measures some kind of “crest factor”. See http://www.sfxmachine.com/docs/loudnesswar/metrics_for_quantifying_loudness_and_dynamics.pdf for confirmation. The crest factor should IMO reformulated as “peak salience”: to what extent the local peaks are above the rest of the audio content.
One of the things a limiter do is to reduce the peak salience. This way, the audio content can be made louder without any clipping. There is a correlation between peak salience and loudness in the sense that lower peak salience + gain = more loudness.
Distortion (let’s say clipping, rather) is a form of brutal limiting. Whereas a limiter will look ahead so it can decrease peak levels without audible harmonic distortion, clipping will “cut” the peak plain and simple. This is where the famous video at youtube.com/watch?v=3Gmex%204hreQ is misleading: it suggests that limiters remove the peaks, whereas they don’t. Clipping does, not limiters.
There is a correlation between clipping and loudness in the sense that peak “cut” + gain = more loudness.
The TT DR meter measures the peak salience, so it’s only natural that it shows low values when in presence of a limited / clipped signal.
Limiters naturally reduce both peak salience and loudness range — however, it’s perfectly possible to have audio signal with both low crest factor / peak salience and high loudness range. Take a square wave, and automate it with Pro Tools using only square automation curves: extremely low crest factor / peak salience, high loudness range.
There is no correlation between CF/PS and DR, even though both decrease when a limiter is present, that’s why the TT DR meter is terribly misleading: it makes you believe that all limited audio content lack DR, whereas they lack CF/PS instead.
Hi Emmanuel,
thanks for your reply. So, basically (please correct if I’m wrong):
* “dynamic range” just doesn’t have the meaning it is generally given by anti-Loudness War activists such as me, as CF/PS should be considered instead.
* the TT DR Meter still provides reliable figures as it actually measures some kind of CF/PS, albeit calling it DR.
* (I’m going out on a limb for this one) Clipping is not a collateral effect of excessive limiting / hypercompression, but rather a kind of compression that may or may not be (deliberately) chosen?
I’m sorry for all of these trivial questions but I’m trying to understand the whole Loudness War thing better despite my lack of a technical background
@Andrea
* “dynamic range” just doesn’t have the meaning it is generally given by anti-Loudness War activists such as me, as CF/PS should be considered instead.
=> exactly
I guess peak salience is better since crest factor is a notion that’s more suited to simple waveforms
* the TT DR Meter still provides reliable figures as it actually measures some kind of CF/PS, albeit calling it DR.
=> basically true
although, the problem is, no one does know for sure what the TT DR meter is measuring! but whatever it’s measuring, it’s related to peak salience
* Clipping is not a collateral effect of excessive limiting / hypercompression, but rather a kind of compression that may or may not be (deliberately) chosen?
=> not really
clipping is… clipping, as explained in http://en.wikipedia.org/wiki/Clipping_(audio)
you remove the loudest samples, harmonic distortion appears
as a side-effect, it does diminish the signal’s peak salience
it’s usually deliberately chosen
some people put “clipping” as one of the factors of the loudness war – it’s true that when you clip the kick for instance, it makes for some kind of “flatness” that may recall some side effects of limiting
but I personally disagree with that: to me, clipping is deliberate audio processing , it’s sound design, and it’s difficult to blame sound design the loudness war, it’s where it gets dogmatic IMO
The DR range meter takes the upper RMS readings (think it’s 20%, heard that once on PleasurizeMusic) and then uses that as DR score for a piece. I have the DR meter on 24/7 in Foobar and I have used the offline one for years, and it’s how it works. It uses AES17? (DBFS+3) for it’s readings.
@Greg
Sorry, that doesn’t tell us how it works, in particular how do they translate RMS to “DR”? RMS is a measure of level, “DR” is a measure of range. And I thought AES-17 was a calibration method?
@Emmanuel:
The problem with RMS is that it doesn’t take the peak level into account. Trim down the “master fader”, and you get a lower RMS – however, you won’t get a higher dynamic range (yes, I do think this term is fairly appropriate).
Therefore, the DR meter “relates” (if this is the correct word) the RMS to the peak level of the material (mathematically pretty much a simple substraction) in order to get over this downside of the pure RMS (this is one advantage over the “pure” K-System, by the way).
Actually, I found all this explained very well on the DR site, and I think you can get a decent understanding of what this is all about if you read carefully through the available information. You may be able to question the wording, but the foundation certainly doesn’t try to create any kind of “mystery” as to what they are trying to achieve, and how.
@ Kahlbert
Can you give me the exact URL for those explanations on the DR site?
But judging from your explanations, the DR meter indeed measures some kind of crest factor (ie the difference between peak and rms, as you’ll find on any science dictionary or wikipedia).
Not of “dynamic range”, in other words the dynamic “spread” (see Vickers E., “Automatic Long-term Loudness and Dynamics Matching”, in AES 111th Convention (New York, NY, USA, 2001) or “loudness range” (see SkovenborgE. and LundT.,“LoudnessDescriptors to Characterize Programs and Music Tracks’, in AES 125th Convention (San Francisco, CA, USA, 2008), or SkovenborgE. and LundT.,“LoudnessDescriptors to Characterize Wide Loudness-Range Material”, in AES 127th Convention (New York, NY, USA, 2009), or EBU/UER, “Loudness Range: A descriptor to supplement loudness normalisation in accordance with EBU R 128.”, December 2010.), which measures how much the level (or loudness in case of perceptual weighting) is mobile.
Not the same thing at all! It’s not even a question of semantics.
@Emmanuel:
Well, for starters http://pleasurizemusic.com/en/our-aim explains all of this, about halfway down the page.
It also explains why I won’t go into those scientific wording issues you seem to have. There’s a time and place for scientific discussions and philosophy – and mind you, I’m scientifically educated myself -, but when I am literally not able to listen to any of my favourite artists’ recent albums anymore, I call for a pragmatic approach like the foundation is suggesting.
Wouldn’t you?
Emmanuel and others, if you sign up as an “active member” on the Pleasurize Music site and log in, there is a spec for the meter available for download. I haven’t reviewed it though, so I don’t know if it answers the questions being raised here.
Ian
@Emmanual
Lets say the loudest sections of a song have a sustained RMS of 5.4, such as in the choruses when things are really compressed. The meter looks at the whole thing, and weighs parts with higher RMS heavier since they are louder, and if it takes up 20% of the top loudest parts, then the DR score will be 5 (round up at .5′s). That’s pretty much all there is to it! I have tested this on hundreds of songs and it has operated this way consistently. I don’t think one has to go into it any deeper.
AES17, well that has to be mentioned because there are multiple ways of expressing RMS. AES17 was the “original” way RMS was measured, however when digital came around and added full scale to it, it added another 3DB. So DBFS refers to the added +3, but it’s not expressed as +3, just DBFS. So when you use the “original” it’s called AES17 or DBFS+3 such as in Voxengo Elephant. That way if I say -6 RMS AES17 you can picture it exactly (-6 RMS AES17 would be equivlanet to -9 RMS DBFS, otherwise you might think -6 RMS DBFS which would be even LOUDER than -6 AES17).
Last thing I saw on the manual said, “A histogram
(loudness distribution diagram) is created with a resolution of
0.01 dB. The RMS – an established loudness measurement
standard – is determined by gathering approximately 10,000
pieces of loudness information within a time span of 3 seconds
(dB/RMS). From this result, only the loudest 20% is used for
determining the average loudness of the loud passages.
At the same time, the loudest peak is determined.
The DR Value is the difference between the peak and the top 20
average RMS measurements (top 20 RMS minus Peak = DR).” Sounds like what I said. Don’t know know about minusing and the peaks but I know how it usually comes up with numbers so I dont think I need to know more.
I meant -5.4 RMS. I always forget to put the minus symbols with RMS. I hope you get what I mean from the above though.
Watching the real time meter and using the offline one, it becomes very easy how it looks at concentrations of high RMS and then uses that as a DR score. In fact almost constantly in the real time meter, the RMS and DR reported are EXTREMELY close, in the range of only 1-1.5db difference. When there are “pulses” such as big bass lines in otherwise unchanging or lower volume music (such as a large majority of J. Lo’s new album, especially “One Love”) then the meter flucuates more and I think you can see the subtracting going on, but it is STILL very RMS and DR connected. What I like about the real time meter too is it shows you like 1ms difference of what you hear… on one hand, it doesn’t take into account real loudness measuring like humans hear (which is a 3s average), but it does show you the highest, highest RMS’s that are contained on anything it’s run on. And it’s from these highest numbers where the DR scores ultimately come from.
Oh and by the way, I looooooooovve the sound of Jennifer Lopez’s “Love?”. Have you heard it? Still very loud mastering but the arrangements the peaks are very flat, things stick out more so they recover WAY easier during declipping and transient reshaping.
I wish I could get into mastering. I just picked up what I know from looking at how things work/sound/look and experience over time.
And I meant DBFS is -3. Gimme and edit function so I can correct these things!
Wait a minute… I think I’m onto something. Perhaps because many songs peak at 0.01db or such, that’s why DR and RMS are so closely related… there is no significant peak to subtract in the first place!
I reckon that’s how it’s supposed to be? AFAIK, digital media achieve full resolution only at 0.0 dbFS, and lose 1bit of actual resolution for each subsequent -6.0 dbFS decrease. Therefore, a signal which peaks top at -6.0 dbFS would be 15bit in resolution. Correct?
It ensues that peaking at 0.0 dbFS while having a high peak salience / crest factor / dynamic range (you name it) would be our best case scenario.
@Kahlbert, post August 22, 2011 at 11:11 am
According to http://pleasurizemusic.com/en/our-aim (thanks 4 the link btw)
“The DYNAMIC RANGE METER displays the inner dynamics of a recording in whole numbers or more precisely the inner grade of compression (micro dynamic). The macro dynamic (difference from pianissimo to the fortissimo in a song) is not considered, because it wouldn´t deliver usable information about the degree of density.”
So they state precisely that they measure “inner dynamics” as opposed to “macro dynamics”. They speak precisely of RANGE, or DIFFERENCE BETWEEN LOUD / NOT LOUD.
This is also what Bob Katz refers to when he mentions “micro dynamics” and “macro dynamics” in “Mastering Audio – The Art and the Science” p. 109.
But then they say:
“For technicians, this is the average cumulative difference between peak and loudness (RMS) over a specific period of time (duration of a song or album)”.
But this is not a RANGE, this is a variant of the CREST FACTOR. Really, this is not the same thing, not the same concept. They made a mistake. It’s neither “inner dynamic” or “micro dynamic”, it’s SOMETHING ELSE.
And they’re NOT CORRELATED to each other. Look at the following picture:
http://private.1-1-1-1.net/productionadvice/CF_TDP3.png
This is “inner dynamics” vs “crest factor”. As you can see, they’re NOT CORRELATED. I tried with other measures of both CF and micro dynamics, they’re still not correlated.
Once and for all (I’ll be much more busy starting tomorrow and will not be able to answer other posts).
1. PEAK MINUS RMS = VARIANT OF THE CREST FACTOR
2. CREST FACTOR AND MICRO DYNAMICS ARE DIFFERENT CONCEPTS
3. CREST FACTOR AND MICRO DYNAMICS ARE NOT CORRELATED
Sorry for the capital letters, but the information has to go through once and for all, or people will keep on acting Don Qixote-like and fight the wrong enemy…
For those who are interested, I posted the distributions for RMS, EBU3341 integrated loudness, high sample density, peak salience, and EBU3342 loudness range at the bottom of http://1-1-1-1.net/IDS/?p=747
e.
Now the NEW dynamic range meters will be unveiled shortly. *anxiously awaits*
@Andrea/Lea
I agree with you.
For me 2010 music is much worse than 2003 music.
For example compare “Revolutions per Minute” and “Endgame” by Rise Against (this is melodic hardcore punk). Both are DR6. But the 2003 “RPM” have a bit punch, the 2011 “E” not. Why?
The 2011 album is less limited. But WAY too
over-compressed. When it’s limited, in D/A conversion the peaks cannot wait all the time. They get a little higher. When over-compressed but not limited all the time, the waves go EXACTLY 0dBFS+amplification. In limiting: 0dBFS+something+amplification.
I’m right, am I?
@Marcin
It’s quite the opposite.
Limiters are devised so the peaks are strictly constrained.
Compressors are much more flexible: a compressor used with particular settings, for instance, will let the attacks go through, and mainly deal with the decay-sustain-release parts of the envelope, not with the attacks.
Andrea:”I reckon that’s how it’s supposed to be? AFAIK, digital media achieve full resolution only at 0.0 dbFS, and lose 1bit of actual resolution for each subsequent -6.0 dbFS decrease. Therefore, a signal which peaks top at -6.0 dbFS would be 15bit in resolution. Correct?
It ensues that peaking at 0.0 dbFS while having a high peak salience / crest factor / dynamic range (you name it) would be our best case scenario.”
I’m afraid it’s all a bit (pun intended) more complicated than that. Today, most digital audio is captured at 24 bits, so no one need worry about tickling the meters to “use up” all the bits. As demonstrated in Bob Katz’ book Mastering Audio, the increased resolution is maintained when dithering down to 16 bits for CD masters, a rare case of getting something for nothing, as Bob puts it.In order to maintain headroom for further processing, it’s probably optimal to have your peaks at around -12dBFS or lower when tracking instruments.Depending on the quality and design of your summing bus, you may want to track at even lower levels if you plan on tracking many instruments, because the levels are cumulative. (why they call it “summing”) Common practice in some circles is to normalize mastered files to as little as one tenth of a dB, which is bad news, because a CD with so little headroom will likely trip a limiting circuit when the music is broadcast, with less than optimal results. I’m told by some folks who should know that as little as -.3dBFS headroom can be enough to avoid such a disastrous fate as not being the loudest song on the radio dial, but you can never tell what a particular broadcast engineer might do.HTH
The Sound On Sound article puts a new perspective to the debate, but I think the author’s statistical approach simply ignores a lot of bad mastering; he’s talking about mainly songs that achieved some sort of commercial success, some had good dynamics, some didn’t, but they all had enough. Further, the article presupposes that the EBU standard is the best available measure of dynamic variability, a fact very much not in evidence, to my mind, as it appears not to reveal the same trends as other methods.(and seems designed to allow commercials to blare at intervals while listening to broadcasts, as well) His points about modern dynamics manipulation vis a vis verse/chorus differentiation are well-taken, regardless. The biggest takeaway from the article is that no form of measurement will replace trained ears and good taste in the mastering suite, and it may be merely sustained distortion that the ear objects to, rather than any lack of dynamic range, whether perceived or otherwise. Too bad the author didn’t measure that as well.
@ Mojo Bone
Thanks for explaining
In answer to this
“1. PEAK MINUS RMS = VARIANT OF THE CREST FACTOR
2. CREST FACTOR AND MICRO DYNAMICS ARE DIFFERENT CONCEPTS
3. CREST FACTOR AND MICRO DYNAMICS ARE NOT CORRELATED
And they’re NOT CORRELATED to each other. Look at the following picture:
http://private.1-1-1-1.net/productionadvice/CF_TDP3.png”
First of all, thanks for the article)
Now some considerations…
On this diagram you depict 0,2 ms RMS variability vs. Crest factor, and you equate RMS var. with microdynamics. This is where I disagree. OF COURSE, RMS variability and Crest factor are not the same, BUT the question is what parameter we should associate with microdynamics. To my opinion, 0,2 ms RMS variability is already more “macro”, than micro, because 200 ms integration time allows transients to pass undetected. On the other hand, crest factor deals exactly with momentary peak values, so it reflects “peak” dynamics, which is in the first place effected by limiters. So, if you want to judge about limiter’s action on audio, 0,2 ms integration RMS variation will tell you nothing. Next difference, RMS variability in your implementation is somewhat “percent-gated” full range between min and max values obtained with the same integration time, while CF is the difference between momentary peak value (zero intagration) and average RMS (some hundreds of ms integration). Difference is obvious, I guess.
So, maybe it’s just a question of terms, if you call RMS variability “microdynamics”, we need to call crest-factor, don’t know… “nanodynamics”? The point is, we need to understand what each of those reflects and what tools can be used to change this parameter. I mean, momentary (nano) dynamics is better described with crest-factor and can be treated with limiters or compressors with very short attack/release time; longer-term dynamics can be described with RMS variability, and even longer-term dynamics – with Loudness range (3 s integration and K-weighted) according to EBU-128; to affect this parameters (which of course, have some “resilience” to treatment with limiters) you’ll have to use RMS-compressors!
As we see from diagram 3 in article, loudness war resulted in loss of momentary dynamics, that is reflected by small crest-factor values in 90′s and 2000′s. In the same time, macrodynamics and Loudness Range is not affected by it, because limiters/maximizers, used on final stage of production, work with peaks and do not apply RMS algorhythms. Crest factor (nanodynamics) is strongly affected with limiters, that doesn’t allow transients to pass. At the same time, they leave relatively untouched macrodynamics, i.e. they can remove peaks, but won’t flatten verse/chorus level differences, for example. To reduce macrodynamics, you can use RMS compressor or leveler, which reacts on RMS signal value and leaves peaks alone. I guess this is the reason why Loudness range (as measured by EBU-128 recommendations) demonstrates no decreasing throughout the loudness wars.
To sum up, the EBU 3342 simply doesn’t seem to measure what we see as the effects of the loudness war.
(I even wonder if EBU 3342 is really useful for anything?)
I always was rather reluctant to measure dynamics, as it strongly depends on the definition and window size. I agree with Emmanuel that the poor audio quality coming along with the loudness war is caused mainly by heavy compression, limiting and clipping, which results in an unnatural distribution of the samples, particularly the peak samples (i.e. pushed to the limit).
That’s what you can see with my Ber-SD ClippingAnalyzer (click on my name to get there).
Anyways, the easiest thing is to look at the waveform…
In answer to Mojobone (yep, laaate answer)
“The biggest takeaway from the article is that no form of measurement will replace trained ears and good taste in the mastering suite, and it may be merely sustained distortion that the ear objects to, rather than any lack of dynamic range, whether perceived or otherwise. Too bad the author didn’t measure that as well.”
True!!! The correspondance between signal descriptors and “percepts” in this matter is a really delicate matter. I’m currently in the process of trying to tie loose ends there, but it is not an easy issue, quite far from it
Fed up with the so distorted and hyper-compressed horrors! – I’m increasingly boycotting releases, using freebee ReLife vst isn’t a ideal solution either, gets worse and worse, till no miracle tool helps!…Flat, lifeless, and full of rotten clipping (even in acoustic/folk music), Ouch! – Surely the deafness rates are rising too, End the sick $Loudness War!…BRING BACK MuSiCaL DYNAMIC RANGE!
[...] (Actually, it’s more complicated than that. “Loudness War” sound suffers from limited crest factor, low RMS variability and in the worst cases distortion. We’ve chosen “limited dynamic range” as an intuitive way to describe all this for Dynamic Range Day, but for a more rigorous technical analysis, click here.) [...]
I couldn’t finish this article. I can’t take advice from anybody who puts a space before their exclamation marks.
Rotfl !
Yeah, because that has everything to do with audio quality…
Well then, I’ll just call it “Dynamic Quality” instead, or say that music is more or less “Dynamic”. Problem solved.