I came across a cool idea from David Glenn recently.  After bringing tracks into a session, he normalizes all tracks to -18dbfs, except the kick, snare, bass and lead vocal, which he normalizes to -15dbfs.  I gave it a try, and wow, on several songs, it resulted in a fairly balanced mix right off the bat, with plenty of headroom, also hitting the "sweet spot" for analog-modeled plugins.
 
Obviously results vary depending on the dynamic range of the source tracks so further gain/volume adjustments are required, but I've found this to be a pretty snazzy way to start off a session. If a track has really high peaks and -18/-15 is too low I'll adjust the gain knob to get the meter to bounce around the -18/-15 mark, then just move all track faders down to -6db or so for more resolution for volume automation down the line.  TONS of headroom for the master bus and at 24-bit no real noise floor to worry about, so I've found this to be a pretty cool idea.
 
Anyone else do something similar?

Yes, this is first thing I do.  I play through the song to see what the peak level is on each track and then adjust the gain to get it to -16.  I put all tracks at the same level initially, but I think I usually bump the kick and snare up a little.

ALWAYS REMEMBER,
any time you apply math to the signal,
you are degrading it.
 
 

I am confused by that comment, since "math" is incredibly generic, and "degrading" a signal would make every VST worthless. Normalization is actually very innocuous in most respects, yet seems to get "blind hate" quite a bit for some reason. I truly do not understand this mindset.

wouldn't necessarily call it "blind hate"... normalization is a wonderful math tool, but it is not exactly musically natural... perhaps that  makes me a hater?

Normalization takes each data point and moves it up or down to meet some arbitrary level. And that's the thing, it operates on each discrete sample. A compressor, which is another way to change apparent level, makes decisions based on trends, so an increase of 6 dB may be reduced to an increase of 3 dB, but an increase of 3 dB is reduced only to 1.5 dB - both changes were reduced by half (assuming a ratio of 2:1). It is the change, not the absolute value, that is modified (the absolute values are modified as a result.)

If you have a quiet track and you use normalization to increase the overall level you will also raise the noise floor. It isn't quite so bad in the other direction, but it is a linear function, and most of what we listen to is not. Audio changes are perceived on a logarithmic scale. I suspect that's one of the reasons that normalization gets a bad rap.

Even gain and attenuation behave differently than normalization. If I apply 6 dB of gain to a 1 V signal I will get a 2 V signal, if I apply 6 dB of gain to a 2 V signal I get a 4 V signal. With normalization every sample is increased by the same linear amount, so a 1 V signal might move up to 2 V, but the 2V signal only moves up to 3V.

Think of normalization as sliding the entire track up or down the Voltage axis instead of the level axis - if that  makes sense.

The "math" comment might have  been a little bit harsh, but the degrading comment is actually spot on, although  I might have used the term "change" instead of degrade. But degrade works if you think in terms of moving away from the original track... except that  sometimes that  is exactly the  object<G>!
 
That's not to say that normalization has no musical (or audio) applications. It just isn't the magic process some folks claim it to be. And in the real world, with S/N ratios approaching 100 db the noise floor isn't the problem it once was, you just have to be aware that you are changing the relationships between relative levels.

Normalization really helps me when I have a whole bunch of tracks with relatively similar levels.
 
For example, if I record 16 tracks for 3 hours and get 35 songs, I would be likely to use normalization.
 
On the other hand, when tracks vary greatly, I'm much less likely to use normalization. For straight acoustic music where compressors are not used to record the original signal, I would probably not benefit much from the normalization process.
 
This process saves a lot of time. I'd compare this to writing papers with speech recognition software. You can crank it out fast, but it's easy to make different kinds of mistakes. You have to read over your work pretty carefully to find the goofy things you might end up with.

I like David Glenn too but I am not sure what sort of metering he is doing.  Peak or rms or both.
 
I believe the signal contains two components, rms and peak.  The peak values vary in relation to the rms value as a result of the transient if there is one, and how it behaves just prior to the signal settling down to its longer rms level over time.  The important level!
 
I find it makes more sense to measure the rms levels of each track and set them all to a standard level eg where -20 dB FS = 0dB VU.   For me this goes a long way further in terms of prepping a mix before you start.
 
I often open the tracks one by one in an editor such as Adobe Audition.  I check rms levels using a VU meter and just add or subtract gain to get most rms levels sitting at the reference.  This will put your channel faders in a very good position in terms of starting and then getting a good mix.
 
I let the peaks reach whatever voltage they have to be above the rms levels.  As long as the transient is not distorted then it will sound great.
 
With fast percussion sounds I tend to use the peak metering more.  Rms is too slow here.  Just ensure things are not crashing into 0dB FS anywhere.  By the time you send a number of percussion sounds to a buss, the rms level on that buss will gradually increase and start to show normal behaviour.
 
Normalisation based on peak levels has no bearing on this approach at all really.  I don't believe it will help you much.  All normalising peaks will do is make every peak reach a consistent level.  But the rms levels of the signal portion below the peaks will still be all over the place level wise on the tracks.  And it is the rms level that is carrying the real how loud is this sound information.
 
When I do the tracking I just ensure the rms ref level is met right there and then and no extra level changes are actually required from that point on.
 
Changing the level of a track to get it right will not alter its quality.  (Well audibly that is!)  I have yet to hear someone pick a track that has had 6 db of gain applied  and then the two leveled out for the AB test. With higher res calculations in the background even on a 24 bit file any calcs being done in 32 bit will be fine. 64 bit even better.
 
You need either real or virtual VU meters to do it this way.  Sonar standard rms meters are not suitable and they also read 3 db lower which does not help.  K sytem is good as it has three ref levels you can work at -12, -14 and -20.  All good for different applications. -18 is another excellent and common reference level.  Apart from very short percussive sounds I find that everything else allows itself to be measured and will respond nicely rms wise giving you meaningful readings.
 
 
 

Jeff Evans
You need either real or virtual VU meters to do it this way.  Sonar standard rms meters are not suitable and they also read 3 db lower which does not help.  K sytem is good as it has three ref levels you can work at -12, -14 and -20.  All good for different applications. -18 is another excellent and common reference level.  Apart from very short percussive sounds I find that everything else allows itself to be measured and will respond nicely rms wise giving you meaningful readings.

What do you recommend for metering?

David just does a simple peak normalization to -18 or -15.  And it's not like he just sets it and forgets it. 
 
I didn't mean to suggest that normalizing all tracks to -18 or -15 will magically create a great mix. I just know that I used to have a lot of difficulty with gain-staging and trying to judge appropriate levels/balance for all the tracks, and this method turned out to be a real time saver.

streckfus
David just does a simple peak normalization to -18 or -15.  And it's not like he just sets it and forgets it. 
 
I didn't mean to suggest that normalizing all tracks to -18 or -15 will magically create a great mix. I just know that I used to have a lot of difficulty with gain-staging and trying to judge appropriate levels/balance for all the tracks, and this method turned out to be a real time saver.

Heh, I was just about to clarify what I do thinking you were referencing my post, when I realized that you are referencing David Glenn's article.