Yeah, my reply was just kind of catch-all response.  Sometimes I get lazy. Like when I normalize tracks to get an initial track balance. :)

I like to think most signals have an attack component which produces a short higher peak value followed on by an rms component.  The distance this peak is above the rms component depends on the nature of the sound.  In my work flow I like to keep all the rms levels the same and let the peak levels sit above that constant rms level at various random levels. eg some sounds might reach -4 dB in my system but the rms component will still be down at say -14 dB FS. And it does not matter because it will still get through your system untouched right to the end stereo mix and sound fine.  Some transient peaks might only be 4 db above rms and reach -10 dB FS at a -14 db FS ref rms level.  Some sounds have no transient edge or peak at the start and they just ease into the rms component.  Other sounds are fast and too fast for the VU.  Keep your eye on peak metering then.
 
Setting all the peaks to a constant value for me still sets the rms levels on each of those sources at various points below and around the ref level.  They will all over the place.  The rms is the meat of the sound and is closely allied to how loud something appears to be.  It is very basic theory.  Why not make it the constant and ensure all your tracks have a matched rms reading.  As long as there is plenty of headroom available for very transient sounds (and the -20 dB FS ref level ensures this) adjusting tracks to match up rms wise just sounds like a good idea.  And it is something we did before in the analog days.  Peak metering was not so big then and we just let the analog mixer/tape systems absorb the peaks the best way it knew.
 
When all your rms levels are correct on tracks then the fader is around unity.  You still have plenty of gain above and lots of control under.  Setting multiple faders to roughly the same position yields almost a perfect balanced mix as well.  Sending multiple tracks to a buss for example only requires all the faders to be around -3 to -4 or so for the correct ref level to be read on the buses.  The more tracks that are being directed to a buss (eg 6 vocal harmonies to a vox harm buss) just means pulling the group down slightly more.  When rms levels are all the same on tracks being assigned to a group, the faders then all end up at the same position eg around -6 dB or so.  Easy to see and move.
 
You can even mix into VU meters and add in various components watching the needle swing build up towards a full mix just hitting 0 dB VU on the output.
 
For metering I have real VU's which are nice for sure and even the best VST's only come very close in terms of needle ballistics.  But the Klanghelm meter is good.  Cheap and very effective.  Adjustable in response a little as well making it easier to line up to a real VU movement.  Studio One can put its meters into rms reading (eg full scale 0 dB VU = the db ref level below eg -12, -14 or -20 dB FS) which is handy and I like using it quite a lot.  But I still like the VST as well because you can insert them where you need them and just see how loud any signal is at that point. 
 
I can get the Klanghelm meter to move in a very similar way to the VU's.  I am a fan of watching how the needle swings over tracks, buses and the main stereo mix.  When everything is perfect the needle moves a certain way.  Poor dynamics control on tracks/buses shows up in the needle movement eg wild!  Compressors even when used lightly alter the movement of the needle very quickly.  Sometimes the wrong way.  Watching the ballistics can keep you setting your compressors better so the tracks still sound dynamic and punchy but the needle is just swinging up to 0 dB VU and not overshooting much at all and falling back real nice too.
 
PSP make a nice VU too called the Triple Meter.
 
As you can see I am fan of rms monitoring and let the peaks look after themselves approach.  It seems to work very well for me and I have been happy using this method for a long time.  It is something we did before and it is easily brought forward into the digital multitrack medium.  The sound is better to me now because there is no noise and the transients sound better to my ears too.
 

I doubt you'll ever end up with a satisfying mix in which every track is even close to being either peak- or RMS-normalized.
 
We just don't perceive all instruments equally, so there can be drastic level differences that perceptually come across as equal in volume. You can't, for example, RMS-normalize an electric rhythm guitar against the bass; the former would swamp the latter even though the waveforms looked similar in amplitude. That's just good ol' Fletcher-Munsen at work.
 
Peak normalization is even less relevant, since no two tracks are going to have the same peak-to-rms ratio. Turning up a string section to hit some arbitrary peak value will make it much louder-sounding than turning up a percussion instrument by the same amount.
 
However, setting the level of your highest-peak track first is a good starting point because it's going to establish the ceiling for everything else. For most genres and styles, that's going to be the kick drum - or whatever instrument serves that role, be it a real acoustic bass drum, an 808 sample or taikos in an action trailer. If that track's highest peak is, say, -12 dB, then because you know nothing else will peak higher you can therefore be confident of having adequate headroom for mastering.

When I refer to tracks all being at a ref rms level I am not talking about the mix in any form.  Or creating any sort of mix by setting groups of faders at the same value etc.. It is all about having all the signal you need to be able to place any sound within a mix from very soft to loud if you want it.  When rms levels on tracks are right then you will have the right amount of range of that sound.  Decent amount at unity, above unity, and lots of softer levels below unity. (for the channel fader that is)
 
Once those tracks are correctly set up with all their rms levels at the ref level, then you are set to do an excellent mix.  The rest is up to you.  How well you balance that mix as the balance engineer.  You will get that balance because you always have the right amount of any level waiting behind a fader (pulled down to silence) ready to come into your mix and establish itself at the perfect level every time.  Set by you of course.  (Other things will improve your balance, cleaning up the arrangement, reducing masking using EQ and other effects etc..This is all good mix techique, nothing to do with track rms levels now!) Except I still involve the VU meter.  What I am aiming for is a real nice mix that is just hitting 0 dB VU nicely and not overshooting or anything like that.
 
I feel getting involved with peak levels first is still a little unnecessary.  And it can be avoided.  Set your rms levels the same instead.  Peaks are only a concern as they get closer and closer to 0 dB FS.  And peaks are very important with those all important fast/percussive sounds that slip past the VU very quickly.
 
If a track or mix had its highest peak at say -12 where is rms level then in relation?  If its 14 dB down for example in the case of that K System ref level, then the rms will sitting down at -26 dB FS.  Which is actually a little low.  Better off having that level at -20 dB FS (K system ref) rms and its peak may reach -6 dB FS instead.
 
The more you think about peaks and keeping them at some arbitrary level the more variable the rms levels will be everywhere.  Because as Dave very well put it, the rms/peak ratios are all different for each sound and track etc..So why not keep rms levels all the same throughout your system and let the peaks vary instead.
 
It is so easy to track and render at a consistent level.  (eg virtual instruments and the levels vary wildly from VST's. They really need to be set at a correct rms level before any rendering takes place) Once everything at track level is at a consistent rms level, there is just less work from that point on now because all your levels are right and ready to go. 
 
People seem to hang onto peak metering and try to meter everything peak-wise when it is actually futile.  They gave us peak metering when the DAW really came to town, they left out rms metering by mistake.  We had it before. It was dumb to leave it out.  Some DAW's have the right idea now.  They are letting you switch the metering into rms mode much more easily.  (eg FSD for 0 dB VU ref levels) You can also easily put it back in and work with it.  It is tried and true and wins every time doing a great mix.
 
The rms way for me is simply the best way.  So give it a go you may just never look back.  It becomes so easy and natural after a while.  I never see a clip light come on anywhere which is telling me all my peaks no matter how high they jump above rms levels on tracks, buses and the final mix are still well clear of 0 dB FS.  The headroom is always built in.  A -20 db or even a -14 db pre mastered mix is a dream to master.  Just so much headroom built in and so much to work with for the EQ, compressor and final limiter too.

wst3
Normalization takes each data point and moves it up or down to meet some arbitrary level.

Doesn't that argument mean that every time we move a fader, we are degrading the material?

cparmerlee

wst3
Normalization takes each data point and moves it up or down to meet some arbitrary level.

Doesn't that argument mean that every time we move a fader, we are degrading the material?

That was my thought too.

+1, I initially replied from my phone and already kicked the bee's nest in passing, but every argument against normalization is identical to increasing a fader's level from unity (or a gain boost anywhere in the chain for that matter). No one seems to get irate for these... same concerns apply since every effect just acts upon what it is fed by its predecessor.
 
Audition has several normalization schemes available (peak volume, total RMS, loudness, perceived loudness), as well as the ability to do noise reduction based off a noise print (do prior to normalizing). Both of these are not present in SONAR (but can be worked around with other options).
 
Of course, numerous other factors also need to be taken into consideration, but the Pavlov's dog response is a bit disconcerting... every tool as its time and place.

I mean we're told often enough that all DAW's sum tracks the same (and that it doesn't matter if levels are hot when summing) because after all it's a simple case of arithmetic to combine waves, so why wouldn't the same thing apply to normalization? 

i think ultimately, the IDEAL situation is that you get your gain staging right from the beginning..
by tracking conservatively...
and then, when you take all the faders across the board, up to ZERO,
you should be peaking exactly where you want to, on the master bus.
 
 

cparmerlee

wst3
Normalization takes each data point and moves it up or down to meet some arbitrary level.

Doesn't that argument mean that every time we move a fader, we are degrading the material?

Technically, yes. When you move a fader you're multiplying each sample by a floating-point value, resulting in some loss of numerical accuracy. However, it's down at such a low level you'll never notice.
 
But that's not the same as peak normalization, which adds integer values to each sample rather than applying a multiplier. If you subsequently de-normalize it, the final result is identical to what you started with. That is not the case if you add N decibels in one stage and subtract N decibels at another stage.