Normalization

Hi! 
When I add a new sampleset, is it better to normalize the volumes of diferente velocities? I mean, lets say I have 4 velocity layers. When i recorded they have -20db -18db -16db and -14db. Should i normalize everything to -14db for example? I think that if samplerbox just chooses the note that correspond to the closest velocity you dont need to normalize. However if it chooses the note and aplay a gain that correspond to the specific velocity you need to normalize.

An example to clarify:
I have four velocity layers with the same volume. The last velocity is 127. lets say that i play velocity 120. it will choose the sample that correspond velocity 127. My doubt is: is he gonna aplay a gain automation there that correspond 120? or, it will play always tha same layer with the same volume until you are in the range of the second velocity layer. If it aplays a volume automation i think its better to normalize. 
thanks!

Hi John,
TLDR:
If you define %velmode=accurate you need to normalize, if you define %velmode=sample you need unnormalized samples.

Longer answer:

There are two aspects in velocity.

Most obvious is the volume: the harder you hit key, the louder the sound.
If the velocity samples in your set only address the volume part, you should just use the loudest one and set (%)%velmode to "accurate" and (%)%velolevs to 1. Both default values, so you actually don't have to set them.
This saves loading of redundant samples and samplerbox will calculate a velocity "volume" in range 1-127.
If you would use those samples for the volume, you need %velmode=sample and %velolevs = number_of_levels in you sample set.
In this case samplerbox will pick also this number of volume levels, so volume goes in large steps.
This last approach is less accurate and adds extra / redundant samples to be loaded.

Second aspect of velocity is the changing of the sound / color / character of the instrument depending on the force it is played.
The demo sample of grand piano is an example of this.
In this case you have to define the velocity levels for the sound, but drawback would be steps in volume as such samples are often as is = unnormalized, so you have to use %velmode=sample.
So here I suggest to normalize and define %velmode=accurate in combination with the velocity levels.
With this you get smooth volume and changing of sound character .
Testing with grand piano learned me that 3-4 levels are good enough for the human ear for sound character, but for volume you need at least 7. A big difference in loading time.
Your mileage may vary ofcourse.

I don't quite get your example: if the four samples already have same volume, then they are already normalized.
But if they have -20db -18db -16db and -14db, then they don't have the same volume - not normalized..
Perhaps I misunderstood.

I hope this gives you the idea,
Hans