K-Meters are now "Dynamic Range Meters"?

[mystran]

Actually I was going to post something more meaningful here as well:

Any single number is not going to do any good, because really there's two different things into this "loudness" thing: loudness of the loudest parts, and musical dynamic range.

Basically, if you took running short-window (say 100ms or 500ms or whatever) RMS value (with some frequency dependent weighting if you want) for a given song, then the highest peak for that is decent indicator as to how loud the loudest part is, and basically tells you how hard it was pushed against the 0dBfs ceiling. Beyond a certain point, you can't raise this value without increasing distortion, because this will depend on the signal's crest factor at that particular point of the song, and you can't really lower crest factor without distorting the thing.

However.. for the purpose of musical dynamics, the mean variance (or something similar) of the short-window RMS signal (again possibly weighted as per above) is much more interesting value, because that's going to tell how much the loudness varies from one part of the song to another. It doesn't really matter if the peaks are at -20dB and crest factor is 10 if the whole thing is compressed flat enough that the running RMS reads the same throughout the whole song: there's no music dynamics left anyway. So

Now, you obviously can't get high RMS average for the whole song without both compressing AND pushing it hard against the ceiling, but pretending that these are the same thing will do absolutely no good.

When it comes to crest factor of the loudest passages, as long as it isn't audibly distorted or too lifeless (like some of loudest albums on market) is shouldn't matter much... but here it's also trivial to fix the thing by turning the volume knob. Some production techniques (ducking to basses and the like) will obviously push the crest factor down (and hence loudness up if you normalize peaks). The worst examples here (IMHO) tend to be rock bands trying to match the levels that dance music can reach: it's just not going to happen, because the dance typically ends up with lower crest factor anyway. Some standard for this sort of loudness could be a good thing, but you don't really need multiple values for that.. you just figure out how much crest factor is acceptable, and then force nobody to ever print a CD where a short-time RMS of the loudest part reads higher than the standard. That could actually be a good idea, but .. well I wish you luck.

But even if we managed to get everyone to agree on some RMS ceiling (rather than the current 0dBfs or so peak ceiling) and got rid of the crest factor race, there's still the question of musical dynamics.. that's what really matters when somebody is listening in non-optimal conditions, and where louder CAN be better to a certain degree, and some reduction CAN be a good thing (up to a point). This is also something that volume knob won't do any good (one way or another): if it's got no (musical) dynamic range, you can adjust the volume knob to make it uniformly loud or uniformly soft. If it's got too much (musical) dynamic range, you can select whether you want the softest parts inaudible, or the angry neighbors calling the police when the loudest parts hit the speakers. Too much is too much obviously, but at least you can control musical dynamics without distorting signal (too much anyway).

Now, it's perfectly understandable that radio stations and the like want to reduce the musical dynamic range somewhat, because most of us listen to radio a lot in very non-optimal situations. It is also understandable that somewhat reduced musical dynamic range is a good thing when you listen to your iPod in public transport or whatever.

But the real problem with the loudness wars is that not everyone involved understands it's a tradeoff: more flat music will be easier to listen in non-optimal situations, and well done compression will help bring lower level details audible even in somewhat noisy situations... but take it too far and there are no more details, because it will all melt into an uniform goo.

The other real problem is that not everyone involved seems to understand that crest factor of the (musical) peaks is a separate issue.. and this one is trivial to compensate with the volume knob if you've got a clean signal to begin with. For radio stations again, it would make sense to run somewhat hot to get better SNR, but here it's nothing to do with perceptible loudness being better as such.. rather just weaker signals sinking to noise sooner, thus reducing the area where the reception quality is acceptable (and the stupid cramming of too many stations into too little spectrum resulting in stupidly narrow modulation bandwidths at least around here is the main reason I don't personally listen to radio at all)... but that's not an issue at all when it comes to CDs or MP3s or netcasts or.. well actually anything other than analog radio (unless you count the other obsolete analog formats)...

So the sum of it all? Decide yourself, I don't care to.. but IMHO people should stop confusing crest factors with musical dynamic range: control the latter as you see fit (that's the artistic choice part really.. and done well one can have stuff that appears to have much more dynamics than it actually has, which is a kind of win-win when you want to listen to it in a noisy environment with crappy headphones when someone's kids are crying on the next seat; done badly things will sound flat), but stop confusing it with the crest factor, which helps no-one, and just makes everything more distorted.

..and I don't know if I had a point.. but I must have lost it by now. In the mean time I'm going to keep researching into how to make a one-knob-only ultra-mega-maximizer because obviously there's market demand.

[mystran]

But seeing from the specs of it, it can only measure peaks (volume) in dBFS, and not loudness (RMS).

But the point is, since when does RMS measure loudness? It has always been a poorman's (simple & cheap CPU-wise) way to roughly measure it, but it's not very good.

Agree on it not being very good, especially if you don't do any frequency weighting.. but it's still shitloads better than pure peak measurement.

[Cordelia]

I didn't make myself clear. What I mean is that the transition works perfectly in context, and still the ballad is loud enough to sound fine when singled out for play in a CD changer, or in an iPod. But if it were reduced in volume by 6dB simply because the next song on my CD has less dynamic range, it would be too quiet to be heard clearly when played before or after other material from a different CD.

Not quite correct ( if we assume the DR-model would work ).
All other overcompressed material would be reduced in volume, too,
so the difference would be the same.
The hardware volume would be raised by 6 db and everything would be as before,
with the significant difference, that material with more dynamics
( and thus more 'listening pleasure' ) will not sound lower in volume.

Not true in this example since the ballad was not over-compressed (DR16) but being judged by the song that follows it on the CD. The rating/level change is not being done on a song by song basis.

Then there will be the learning process :
Your DR8 songs will be DR11 for the next release and You will have less problems
linking it to DR16 material, when mastering.

I didn't master this release myself, it was mastered by Jonathan Wyner, an excellent ME (David Bowie's back catalog, for instance) who has fought the loudness wars.

Regarding DR14 as standard :
I guess that's a bit over the top.
With todays technologie, something like DR11/10 would be a better solution.
But AFAIK that's not decided, yet.

The most important part of all this is, to stop Dr4-2 releases,
as they're hurting the industry by reducing sales
and turning poeple away from music !

I agree.

AND
regarding artistic freedom :
Sorry, but this is the biggest joke, ever.
1) Since 2 decades, radio has taken away all your artistic decisions,
by smashing songs to the bone, making silent intros sounding much louder than the chorus, etc. etc.
2) If you artistically decide, to completely distort and smash Your songs,
so that no one can listen to them, longer than a few minutes ...
Don't worry, the effect will stay exactly the same, even if reduced by 8 db !
3) for many, many mastering engineers, the DR system would bring back their artistic freedom !
Which has been taken away by record companies, demanding certain RMS values.

Bye, Jan

I agree, mostly.

[tonAP]

It's a measurement vs psychoacoustic thing

That's the hard stuff

I'd like to believe that, but look, I can get LeqA & BS.1770 to 'agree' on the loudness of some signal, and then for another signal they're 15dB apart. Mmmh, I know that loudness perception is subjective, but 15dB apart?
Make an algo that auto-gains or does something else critical (compression) according to that, and 15dB is not a detail.
To my ears (and it's of course not just subjective but also depends on the playback device (speakers)), LeqA looks more 'accurate'.

he,he they (Dolby) say that ITU say that BS.1770 algorithm is better for music measurements, while Leq(A) is better for speech-based material (?)
Sure: 15dB apart seems too much..

[mr]

Considering how the DR algorism works, can you cheat the system?

[donkey tugger]

2000?

[Compyfox]

Basically, if you took running short-window (say 100ms or 500ms or whatever) RMS value (with some frequency dependent weighting if you want) for a given song, then the highest peak for that is decent indicator as to how loud the loudest part is, and basically tells you how hard it was pushed against the 0dBfs ceiling.

This is what the K-System actually does, with a measurement "short-window" of 600ms, which resembles more like our ears.

Some production techniques (ducking to basses and the like) will obviously push the crest factor down (and hence loudness up if you normalize peaks). The worst examples here (IMHO) tend to be rock bands trying to match the levels that dance music can reach: it's just not going to happen, because the dance typically ends up with lower crest factor anyway.

Well but isn't it the same for Rock Music, where the lowend is so filled up with stuff, that it's not even funny? Especially if we look back at times before Sidechaining was popular (and I still tend to not use this technique either).

Some standard for this sort of loudness could be a good thing, but you don't really need multiple values for that.. you just figure out how much crest factor is acceptable, and then force nobody to ever print a CD where a short-time RMS of the loudest part reads higher than the standard. That could actually be a good idea, but .. well I wish you luck.

Well if I read the manual right, it'S doing that already:

DR Display Bar: Color‐coded display of the dynamic range which is used.
Values of less than DR8 are displayed with the color red. Values of DR14 or more are
green. Values between the two are displayed in various shades of yellow.
The DR bar shows the average difference between peak and RMS and corresponds to
the so‐called crest factor.

The thing is however, it responds (to my experience) stronger to bass intensive material, and lowend distortion. Meaning: Rock tracks even if in K-12 to K-14 (which I'd call fairly clean) still generate DR-8 readouts with the plugin, and in best case DR-10 to DR-9 readouts with the standalone plugin.

But even if we managed to get everyone to agree on some RMS ceiling (rather than the current 0dBfs or so peak ceiling) and got rid of the crest factor race, there's still the question of musical dynamics.. that's what really matters when somebody is listening in non-optimal conditions, and where louder CAN be better to a certain degree, and some reduction CAN be a good thing (up to a point).

While agree on that thing (and the following paragraphs), I think the K-System worked a bit for that purpose already. This is why there is K-12: best of both worlds. You get loud productions (to be heard everywhere) while not being distorted so that you hurt your ears unnecessarily.

So the sum of it all? Decide yourself, I don't care to.. but IMHO people should stop confusing crest factors with musical dynamic range: control the latter as you see fit (that's the artistic choice part really.. and done well one can have stuff that appears to have much more dynamics than it actually has, which is a kind of win-win when you want to listen to it in a noisy environment with crappy headphones when someone's kids are crying on the next seat; done badly things will sound flat), but stop confusing it with the crest factor, which helps no-one, and just makes everything more distorted.

Well... this can work like a two edged blade:
Sharp side 1: Create clean cut mixes (with sidechain, EQ cuts, etc, while not using any form of distortion, including compression) and loose out on the life of a track (do we really need sidechaining? I worked back in the day without, too)

Sharp side 2: Mix as you feel like, get the feeling portet, but you have less dynamic... er I mean crest factor and stuff sounds just "undynamic" and muddy.


Finding the middleway is a serious issue here. And that's not easy to be done for sure. A gun pointed at your chest is not really helping in this case either.


EDIT:
On a sidenote, I wish I could listen to some unmastered material (before pressed to it's final limits and on to CD). I'm fairy sure some of that goes as high as DR-10. Happened to me recently at least with a track by a colluegue (zircon) from his last album. The impact sure was different compared to my demo-mixdown and his with used T.Racks as mastering suite.

[tonAP]

.. but it's still shitloads better than pure peak measurement.

But peak meters aren't there for loudness measurement, they are there for.. er.. peak measurement! While RMS *should* give you a loudness idea.
Question: is there a loudness metering system better than RMS?

[Cordelia]

I just wanted to take a minute to thank you, Compyfox, for your clear headed analysis and the time you've taken out to study/test this.

[jonnyG]

I guess everyone's aware of Fletcher Munson curves and all the subsequent work on measuring the perceived relative loudness of different frequencies at different volumes?

The rub is that any standard would have to make a random guess at what volume the listener was subject to, the frequency response of the playback system (for instance a heavily subby track wouldn't move any air on a set of crappy computer speakers, yet may bend the needles on an RMS meter), oh, and don't forget that younger listeners may have an extended high frequency hearing response. Talk about can o' worms.

Anyway, couldn't this take the final say about how a record actually sounds away from the artist?*

*Edit: obviously record companies have been doing just that for decades...

[tony tony chopper]

Now, it's perfectly understandable that radio stations and the like want to reduce the musical dynamic range somewhat, because most of us listen to radio a lot in very non-optimal situations.

at the same time, radio's already play special, shortened radio-edit versions of songs - so it's not really where it applies, nothing would change for them

[_dada_]

What I mean is that the transition works perfectly in context, and still the ballad is loud enough to sound fine when singled out for play in a CD changer, or in an iPod. But if it were reduced in volume by 6dB simply because the next song on my CD has less dynamic range, it would be too quiet to be heard clearly when played before or after other material from a different CD.

Not quite correct ( if we assume the DR-model would work ).
All other overcompressed material would be reduced in volume, too,
so the difference would be the same.
The hardware volume would be raised by 6 db and everything would be as before,
with the significant difference, that material with more dynamics
( and thus more 'listening pleasure' ) will not sound lower in volume.

Not true in this example since the ballad was not over-compressed (DR16) but being judged by the song that follows it on the CD. The rating/level change is not being done on a song by song basis.

Yep, I know about CDs being averaged. Room for improvement, there, too ...
What I wanted to point out, is, that the difference between Your songs DR16 > DR8, will stay the same, when lowered by 6 db.
AND other songs will not sound louder, as they would have been lowered by 6 or more db, as well ...
( assuming the 'Pleasurize Music Foundation' has taken over the music world )

Then there will be the learning process :
Your DR8 songs will be DR11 for the next release and You will have less problems
linking it to DR16 material, when mastering.

I didn't master this release myself, it was mastered by Jonathan Wyner, an excellent ME (David Bowie's back catalog, for instance) who has fought the loudness wars.

Yep, and sure enough he has done a good job, not putting Your songs in the DR4 range !
Nevertheless ... 10 years back, he would most probably have mastered it to DR12.

What I meant by "learning process", is:( again, assuming the DR system is an accepted standard )
the consumers and thus the record companies would learn, that music with higher DR values, sounds better and respectively sells better.
So a year later, when You may release Your next CD, Mr. Wyner would probably choose DR10/11,
for most of the material.

Just to clarify


bye, Jan

[jonnyG]

What I meant by "learning process", is:
the consumers and thus the record companies would learn, that music with higher DR values, sounds better and respectively sells better.

I wish I shared your optimism, but often we're talking about people listening to low bitrate, lossy compressed formats through none-too-posh headphones in often a less-than-ideal environment. The only way to make this idea work is to market it heavily and will the record companies' accountants think they'll recoup the outlay? I doubt it. Chances are it'll just have a lot of consumers hitting the "Loudness" button too.

Let's not forget that everyone here is involved more-or-less with music production. A poll among music consumers may well reveal different demands of their music.

[Cordelia]

What I mean is that the transition works perfectly in context, and still the ballad is loud enough to sound fine when singled out for play in a CD changer, or in an iPod. But if it were reduced in volume by 6dB simply because the next song on my CD has less dynamic range, it would be too quiet to be heard clearly when played before or after other material from a different CD.

Not quite correct ( if we assume the DR-model would work ).
All other overcompressed material would be reduced in volume, too,
so the difference would be the same.
The hardware volume would be raised by 6 db and everything would be as before,
with the significant difference, that material with more dynamics
( and thus more 'listening pleasure' ) will not sound lower in volume.

Not true in this example since the ballad was not over-compressed (DR16) but being judged by the song that follows it on the CD. The rating/level change is not being done on a song by song basis.

Yep, I know about CDs being averaged. Room for improvement, there, too ...
What I wanted to point out, is, that the difference between Your songs DR16 > DR8, will stay the same, when lowered by 6 db.
AND other songs will not sound louder, as they would have been lowered by 6 or more db, as well ...
( assuming the 'Pleasurize Music Foundation' has taken over the music world )

Then there will be the learning process :
Your DR8 songs will be DR11 for the next release and You will have less problems
linking it to DR16 material, when mastering.

I didn't master this release myself, it was mastered by Jonathan Wyner, an excellent ME (David Bowie's back catalog, for instance) who has fought the loudness wars.

Yep, and sure enough he has done a good job, not putting Your songs in the DR4 range !
Nevertheless ... 10 years back, he would most probably have mastered it to DR12.

What I meant by "learning process", is:( again, assuming the DR system is an accepted standard )
the consumers and thus the record companies would learn, that music with higher DR values, sounds better and respectively sells better.
So a year later, when You may release Your next CD, Mr. Wyner would probably choose DR10/11,
for most of the material.

Just to clarify


bye, Jan

You're still missing the point- all the songs on my CD get lowered by 6dB because of one, short, intentionally compressed song. That means that the more dynamic songs on my CD, arbitrarily lowered in volume by 6dB, will definitely sound quiet in comparison to other CDs.

Even if every song on my CD were DR16 except the one mastered intentionally loudly, all other eleven songs would be lowered by 6dB. Another CD, with songs ranging from hypothetical DR13-14 is left unchanged. The song on my CD that has a greater dynamic range than the songs on the hypothetical CD has, for no reason, been lowered in volume, and will now be from 6-10dB quieter than a song with less dynamic range from another CD.

This system is seriously flawed.

The CD I'm referencing was released in 2000, by the way, and I don't feel it's appropriate to imply that Mr. Wyner has in any way contributed to the "loudness wars", has ever done anything but stellar work, or could learn anything new about his profession from a DR meter.

[mystran]

This is what the K-System actually does, with a measurement "short-window" of 600ms, which resembles more like our ears.

If you want to smooth away local transients. Personally I can hear loudness changes somewhat faster than 600ms though so it's not exactly how our ears work, but it can still be useful in order to encourage local dynamics.

Well but isn't it the same for Rock Music, where the lowend is so filled up with stuff, that it's not even funny? Especially if we look back at times before Sidechaining was popular (and I still tend to not use this technique either).

In a whole lot of modern dance music you basically have a kick on beat, and everything else somewhere else, and it's perfectly acceptable (even if we don't go to the electro-house style compression as the defining element style side-chaining) for everything else to somewhat push down on the beat. You can actually do that to some extend without much noticeable ducking because of the over-the-top nature of a typical dance kick. Goes for non-4-to-floor stuff as well.

You can also mess with synthetic sounds a whole lot more aggressively with EQ or clipping or whatever you want, than you can with natural sounds, because it doesn't have to sound natural: it didn't sound natural to begin with.

In the worst (loud) rock music the offenders aren't even lowend content. Rather the typical offenders (as far as my ears are concerned) tend to be guitars that have been high-passed to control the low end mud (which is sane) which as a result have relatively high crest factor (because high-passing tends to make stuff "peaky") and which then are later limited to the point of distortion (hey it's just a few peaks). Since the sounds are often heavy on distortion as an effect already (especially metal guitars tend to have a lot of harmonics), the result tends to be horrible IMD with the rest of the mix. You don't have to deal with that kind of sounds at all in dance if you don't want to: just design your sounds to play nice with regards to loudness from the beginning.

Now I don't produce rock, and I suck at producing dance too.. but that's my personal observation anyway. Maybe I'm wrong.. but the worst stuff I've heard tends to be aggressive rock with guitars limited to distort the whole crap... and if one measures the stuff, one finds that it's about the same as perfectly clean sounding dance tracks.

Some standard for this sort of loudness could be a good thing, but you don't really need multiple values for that.. you just figure out how much crest factor is acceptable, and then force nobody to ever print a CD where a short-time RMS of the loudest part reads higher than the standard. That could actually be a good idea, but .. well I wish you luck.

Well if I read the manual right, it'S doing that already:

Ok fine except for the part that I don't accept the idea that it's called dynamic range, because different signals just happen to have different crest factors and the idea that it has something to do with dynamics is absolutely stupid. "Dynamic range" is already overloaded term used both in musical sense for the difference between loud and soft parts, and in the technical sense for the difference between loudest and softest signal that can be represented. Do we really need a third meaning for more confusion?

While agree on that thing (and the following paragraphs), I think the K-System worked a bit for that purpose already. This is why there is K-12: best of both worlds. You get loud productions (to be heard everywhere) while not being distorted so that you hurt your ears unnecessarily.

Now if people just followed it. Actually.. if we think objectively, you don't really need a system at all.. because if people would just keep from distorting stuff we'd end up with consistent enough loudness levels anyway. In fact we end up with rather consistent levels now as well (unless you compared 70s albums to modern albums) and the main problem is that people seem to want to push further than the sane levels.

Finding the middleway is a serious issue here. And that's not easy to be done for sure. A gun pointed at your chest is not really helping in this case either.

Agree on the middleway, and in all seriousness I don't think any metering system (no matter what it does) is going to make that happen.