Okay to be clear, you are doing three things-
1) identifying the offending peaks
2) reducing the level of those peaks by some amount
3) raising the level of the entire file into the "now unused top part of the scale" where you've made more room via lowering the highest peaks.
The first two steps are
limiting. The third is
normalization, regardless of if you are doing all this manually or using a limiter and normalization algorithms to assist you.
Okay, now the details-
When doing so over the entire file, nothing is altered other than the "saturation". More often than not, I'm raising the level of an entire file (set) by + "Xdb" (or lowering if I'm balancing out multiple tracks to blend a certain way...more live mics or more soundboard...).
^This describes manual normalization.
Saturation is a magnetic tape term. Good magnetic tape saturation meant finding an appropriate balance between low-level tape hiss noise at low saturation levels and non-linear compression effects at high saturation levels. Digital audio is linear and does not suffer (or benefit depending on your perspective) from those distortions. It's just
digital level, measured in terms of dBfs (decibels, full scale) where 0 dBfs is the highest level possible without clipping distortion. Normalization raises the level of everything equally to closer to 0dBFS. It's good practice to leave a small buffer of a dB or so at the top, with the highest peaks topping out at a few dB below full-scale, say -2dBfs or whatever, since that can avoid something called
intersample-overs, where the reconstructed analog waveform my occasionally clip even though the highest sample value to either side of that peak is actually below 0dBfs.
I don't uniformly drop those spikes to match the next highest levels, I reduce them by 1db - 3db depending on their severity, essentially maintaining their higher level, just not as a "run-away" spike. Does that make more sense? I know I'm thinking of what I do (I'm at work now), but may still be leaving out some fine details. When I do this, I do it individually as to maintain their relation to the rest of the music. I don't just say, "OK, I'll reduce these few spikes to X", but rather I may reduce one by 2db, and another by 1db, allowing me to bring the entire file up another couple db's. So, if I used the normalization feature instead, wouldn't that all be done at a single level, say 1db for example?
^
Okay so this is the
limiting steps (finding and lowering the peaks), not the
normalization part (raising the level of everything) which is applied after the peaks are lowered. The
threshold control on the limiter is the detector which determines what will be affected. Only the peaks which exceed the threshold setting
might be affected (
might depends on other settings of the compressor/limiter. If limiting,
might becomes
probably or
definitely (which is partly what differentiates limiting from compression). The threshold is only a sensor. It does not affect the sound itself. It's just doing the identification of peaks part. Part 1) in the list above.
The limiter's
compression ratio and time-constraint settings
(attack time, release time, and on a digital limiter-
look ahead) affect how the peaks which are detected as exceeding the threshold are reduced and by what amount. Limiting implies a high compression ratio and a very short attack time to more aggressively control excursions above the threshold. A short attack time catches the shortest onset peaks, (
look ahead essentially enables the ability to achieve an zero or immediate attack time, catching everything).
The
ratio determines how much reduction in level is applied above the threshold. Unless the ration is set to
infinite, it won't crush all peaks to exactly the same peak level, unless its a "brickwall limiter" intended to absolutely never let anything exceed a certain peak value (which in essence is infinite compression with zero attack time above the threshold). Instead it turns down the level by the ratio amount. So a larger peak and a lower one, both of which exceed the threshold value, will still peak at different levels, but both will be reduced by a percentage of their original value.
Here's the thing- It can be hard to wrap your head around all these settings to determine exactly what the limiter (or compressor - they are the same thing essentially, only differing by degree in settings) is doing. It can do what you describe, but when you go in and manually find each peak and manually reduce it, you know exactly which peaks are being targeted and by how much they are each being reduced. The trade off is between the effort to do that manually verses developing the skill of how to set the limiter to do the same thing. Those who really know their way around compression and limiting and do it well (not the loudness war casualties) will often use several passes instead of just one pass of compression and or limiting, catching only the very highest peaks with a more aggressive setting while perhaps using a less aggressive setting with a slightly lower threshold on a seperate pass to sort of "feather" the reduction and make it more transparent.
None of this is to argue that you should do it this way. Do what works for you and you are comfortable with. It is only to explain that limiting and normalization can be done manually or with limiting and normalization tools, but is essentially the same process. Like musical performance and like taping itself, the quality of the end results depends more on the operator and how one uses the tools available to them rather than the tools themselves.