normalization seems to be personal preference type thing- What it does is analyze the waveform
say the "loudest" peak is -2.3dB and the you have a -3.9db, a -5.8 dB, etc- you normalize to 0 dB
the -2.3 dB is raised to 0, the -3.9 goes to -1.6 dB, the -5.8 goes to -3.5 db etc- - your loudest peak goes to your setting and everything else raises the same level (but not to 0)... this happens with no coloring (at least in theory) the tune just generally gets louder...
compression only looks at those portions of the waveforms that exceed the threshold, and compress that portion by whatever ratio you have decided - so say you've set your threshold at -10 dB (I'm pulling this number out of the are as an example)... at a 2:1 ratio that -2.3 dB peak will be 8.3 dB over the threshold- the compressor will reduce that by 4.15 dBs (a 2:1 reduction... a 4:1 would reduce it by 6. 25, etc)- the attack/hold release all have an impact on how the sound is essentially colored- it may make it fatter, it may make it muddy/remove clarity, but the waveforn is changed at all areas that exceed the threshold , whereas normalization is pretty much a level increase across the board. Normalization is nothing more than extreme compression generally starting at 10:1 (and more, with a max output)
If you look at the waveforms, normalization just gets "bigger" across the duration of the tune/waveform, Compression has the higher peaks "smoothed out", and limiting tends to just flatten the tune.
over compressed/brick-wall limited tunes tend to look like a block with not a lot of definition
It is this definition that gives the song it's dynamics
so why not just normalize, since it doesn't "change" the waveform beyond making it "bigger"? Some of the peaks may just plain and simply be to strong to begin with- an over-zealous drummer, plosives on the vocals, the bass player smacks a string, the keyboard is not well balanced across the range of posible notes- Normalizing just makes these that much louder and possibly stand-out more than you'd care... and maybe there are quiet parts that should have stayed where they were... and the biggest issue (IMHO) is many playback systems begin to distort at ~ -2 dB... and the left channel is louder than the right (or maybe your master faders aren't calibrated properly?!?)
distortion very bad...
Now, if you normalize and then compress, having raised the overall waveform(s) during normalization, you run the risk of compressing a lot more than you might normally desire, and kill the dynamics, as well as processing more of the material.
Limiting tends to make it mainly one level, which for me is fatiguing to listen to after a very short while
In ProTools there is a "Gain" function which can be applied across the board (much like normalizing) or applied judiciously to those spots that were just a bit too quiet (especially on vocals where the vocalist tails off making the final syllables hard to determine, and really strong peaks can be "trimmed"-
Compression I usually apply to the individual tracks (or by using an aux bus), because I may not want to compress everything, which applying to the mastering bus does (nor compress something already compressed. And if you should send your material out to a mastering studio there's a very real possibility that there will now be compression upon compression, upon....
On my master faders, if I'm not sending out for mastering, I may use a multi-band compressor with pretty tight bands , because what the master fader sees is the "summed" total sound of all the other tracks, reverb, perhaps some "exciting" and some stereo widening
I leave dithering to my bounce to a final 16 bit .wav or MP3
as with everything, what sounds good to you and me may differ greatly, and for a variety of reasons, so never is my way the "one right way" (this ain't religion!)
