When you are making a multisampled instrument, when it comes to the pianissimo samples, do you normalize them to make them just as loud as the forte samples? I assume the amplitude envelope will take down most of the volume when it sees the velocity is below some threshold like 32 or something. See what I\'m getting at?
Basically when you record each level soft, medium, and loud, should each sample be normallized to the same volume ? To get a good signal to noise ratio? Is this how the pros sample stuff ?
I believe the whol point of sampling at different velocity is to catch different timbres of the instrument and not volumes.
Your goal is to have the smoothest transition between layers you can. Hopefully the timbral transition between layers isn\'t too abrupt since we should be using enough layers to capture the timbral change of the instrument, so you want the samples to have similar energy. So as a general rule, yes you should normalize.
If the samples have grossly different timbres or transients, the perceived loudness could be quite different after normalization. Anyway you can match the comparitive volumes in the editor, IMO it\'s best to give the users as much of the 16 bit sample resolution as possible, especially since normalization at the mastering stage at 24 or 32 bits shouldn\'t damage the samples.
Imagine the counter example: You have a ff sample at 0db, f at -12 db, mf at -18, and pp at -30 db. In this case, you would want to play them back without velocity scaling the amplitude. The sample volumes would then match at the transition points, but there would be no volume dynamics within a layer. Also, your pp would only be using 11 bits of resolution. If your users wanted to build a patch from the pp layer, they\'d have trouble getting proper volume out of it, and it would definitely sound digital and grainy when boosted.
In my first 2 libraries I have normalized every sample and made individual volume for each of the 6000 samples! I wanted to have full control. But now on the Expansion Packs and my next libraries I will do it another way. This is a bit of a secret right now, so I would rather not tell about it. It is something that speeds up my work...
Well... If you\'re going to normalize in the end, I\'d suggest jsut adjsuting the volume trim when you record softer samples. Instead of raisiing the noise floor after the fact. You\'ll get better resolution...especially if you\'re recording at 16 bit still. Which is why I\'m not too crazy about 24 bit sustain samples. The noise floor problem can be worked around
still its more convenient not to have to adjust trims