Given the current state of software samplers, it is virtually impossible to trigger samples with accurate dynamics (linear decibel curve) required to mirror real world acoustics. At this point in time, there is only one sampler on the market that can accurately render these dynamics. Even this sampler has its limitations because it addresses linearity in a global sense, rather than on a note-to-note basis.
In the November 2005 issue of Sound On Sound (www.soundonsound.com) on the accompanying SOS DVD002 disc there is an article (with 25 audio examples) by Ernest Cholakis called "Sample Velocity".
Any other sampler or virtual Instrument than the ones presented in this article can also have its velocity versus real output mapped out and graphed. Please contact me at email@example.com on how to generate such an audio file. The audio file will then be analyzed and the results posted at the Numerical Sound web site.
This sounds interesting, but:
Did you compare the impact of sampler velocity curves to the differences resulting from the variety of MIDI-keyboards? Every real piano feels a little bit different, you always have to adabt to a new one.
The graphs that you present are quite linear in the range from 25 to 127. There might by a significant influence of the software sampler used, but I suspect that this influence counts when you test it against different keyboards.
All the Art Vista sample libraries and Virtual Grand Piano has a calibrated dynamic response which is exactly identical to the response of the physical instrument. It is difficult to achieve it with GigaStudio, but not impossible - it just takes time. In Kontakt it is very straight-forward as each velocity value can be mapped directly to a specific dB output.
Sorry for writing without reading the article, but I almost jumped out of my chair. We put great pride in this aspect of sampling accuracy and it has always been a trademark of the Art Vista libs.
My reference was actual acoustic instruments - piano, guitar, and drums (audio examples included) - they all exhibit linear decibel curves - which is what samplers/keyboard must match - as can be seen in Sampler C. The sampler need at least a 40-48db range but also have to be programable on each note.
I agree with you that one can approximate the correct dynamic curve if all the individual sampler values are individualy entered - but this is needless complexity for developers especially with large instruments such as a piano. This feature should be done right by the software sampler in the first place - this is the only real world acoustic reality. If you want I can analyze and post graphs of your sampled pianos ?
Some of the great synth's such as the DX7 and Minimoog had exponential envelopes and linear decibel velocity response ...
Great article, Ernest! I thought it was really revealing - I can see what effect response has on what you play. I spend my time on so many samplers that I always thought there was response differences, now you just quantified it.
BTW, to all, my good friend scientist Ernest here writes a wonderful article called Lost In Translation for Sound On Sound for their November issue... I thought it was such a good idea I wrote one for Electronic Musician also called Lost In Translation, in the November issue. That are unrelated, except for the title and that it involves samplers.
Amazingly, it's a total coincidence. Two friends write indentically named articles for major magazines unbeknowest to each other, titled the same thing. You shouldn't have heard the phone conversation when Ernest was telling me about the article - I thought he was talking about MINE! "Whos on first, why's on second..."