I'll be brief. I was thinking a lot about what should be a good way to (objectively) judge if a sampled piano has good velocity layers. Clearly, you need to hear it to find out if you like it, but I think there is a way to design the layers or test them in a following way.
I divided the space as follows:
The numbers are in dB's. So, if you assume fff is the loudest and you look at a single note, you need to obtain samples of each note that would have peak powers to be at lower levels as described above.
So, if you take F1 note played as fff and measure -17dBFS digital peak power level (e.g. 35ms RMS). To get to ppp you would have to sample a note that was played in souch a way to produce -50dBFS digital peak power.
It is possible that you could compress ff and fff a bit and would not require them to be 6dB apart and I could live with that. Might be reasonable. Say, you put them 3dB apart so that fff is a bit distorted and ringing version of ff, just with about 3dB higher peak level.
So far I tried this method on Akoustik Piano and found that it correlates to my subjective listening tests.
It would be nice if developers/designers would publish their data with their libraries so we would know better what to expect prior to spending money. For example, they could publish their dynamic range based on 35ms RMS peak powers between softest and loudest notes. They should have at least 30dB range there, I guess.
For comparison, Akoustik Piano from NI has about 15-17dB (pedal up) and 8-9dB (pedal down). Far, far away from 30dB. (Steinway D, note F1)
The appropriate dynamic range depends on the controller and the musical environment. It's difficult to control velocities on a non-weighted controller so too wide a dynamic range would be annoying - there would be no consistency in the playing. Also, if we are talking a typical heavily compressed background track, a large dynamic range would cause one to go from too loud to disappearing in the mix.
Seems to me you can make the dynamic range anything you want if you can adjust the velocity curve. That's the easy part since it can be done with a basic multiplication in software. The harder part is to match the timbre to the velocity and getting smooth timbre transitions. That's got nothing to do with db's and everything to do with "sounding loud" and "sounding soft".
In a real Piano, (i've a Schimmel and sometimes a real Steinway B Grand Piano) , the PPP to FFF depend more on sound colour than DB's value.
It's unreal, to have a big amplitude in Db in sampled Piano programming.
In Akoustik you can set your own db preset :
The Dynamic Range parameter allows the total dB range of the piano, from the quietest to the loudest note possible, to be determined. This function can also be used to modify the dynamics of the whole instrument track without compression or adjusting the sequencer’s note velocities."
The way a piano maps realistically to a sampler has much more to do with the way the microphone hears the piano than what actual dynamic range a particular piano has.
The intervening air in the room is a huge aspect. In an anechoic chamber at two feet distance, a piano will have a much wider dynamic range than in a concert hall (or even the same anechoic chamber) at twenty feet. Some of the energy will be converted to heat in the air before it ever reaches your ears.
So, yes, even acoustically, it is a matter of timbre more than any given amount of dynamic range. The recording and playback process makes it only more so.
A really important thing to consider also is that microphones hear very differently than ears. They have no processing capability, where ears are connected to the most sophisticated processing environment on the planet--the human brain. So, microphones are NEVER designed to pick up exactly what we hear. They are designed to pick up a musically interesting signal that we respond positively to when we hear it played back via vibrating pieces of cardboard.
Some microphones are designed to eke out every detail of the surrounding air (your average super-sensitive large diaphragm condenser). Others are designed to take a smaller "sample" of the air, and therefore they have completely different characteristics (a tiny diaphragm mic like a test-mic or an Earthworks-style condenser). Some microphones are designed to purposefully be insenstive to either a frequency range or an amplitude threshold. These types give yet another specialized and musical result.
Microphones are chosen like paintbrushes--for their musical effect. Accuracy, beyond a certain point, is never the issue in a microphone. Almost all microphones are leaps and bounds more accurate than the best speakers. Tone is why microphones are chosen, not accuracy.
This is all to say that seemingly objective standards will utterly fail you in the engineering and recording of music and samples. It is about getting a musical result, and to that end, one cannot take one simplistic objective standard and apply it without taking many other objective and subjective standards into consideration simultaneously. Dynamic range is simply one slice of what we "hear" in a piano, whether it is in a tiny room, a huge room, or playing from a sampler.
In real life, the human hear/brain compensate the dynamic, and match it with the very real environment you are playing in (with all your senses).
So, when playing a real piano, you never have an "artificial" sense of the instrument dynamic, but this has nothing to do with objectivivity.
Sampled piano are "recorded" pianos.
Timbre evolution is the most important thing. In term of Db, you can always adjust in your sampler what dynamic fit you (assuming the piano sound haven't been to much processed).
About layers :
Some pianos perform extremely well with 10, 8 layers
For pianos with a very wide dynamic range, 16 is a must.
More than that is a luxury (eg TBO). A very interesting colateral effect (IMHO) is to have very little chance to play twice the same sample. This give the piano more "life".
Been thinking myself about the related matters of mic patterns and positioning. Most sampled pianos out there use large diameter cardiods positioned in stereo-xy or ortf. Cardiods all exhibit a type of compression known as proximity effect (and a reverse effect at a distance) as a consequence of their directionality, as well as an off-axis drop-off in response. Making their use at varying distances problematical. In the typical xy setup, even when ideally positioned, different sections of the keyboard will exhibit different compression and frequency response characteristics, with the middle of the keyboard being especially hard hit being off-axis to both mics. MS sampling helps allot but isn't that common in piano libs.. the one that comes with GS is an example.
If you can do without pure mono compatability, binaural/omni sampling is the closest match to human hearing. With almost perfect dynamic, frequency, and impulse response all across the keyboard at recording distances from 1 inch to 20 feet. But I haven't come across a piano library that documents being sampled that way yet.
However, I was thinking more of a method you could use to ensure you really get the timbre differences! What do you do today to decide if you are getting good timbre differences between layers if you sample a piano?
I used the table with dB range in order to get a variety of timbres when recording raw data! I was thinking that as you play hard or soft (resulting in more or less dB) you would also get timbre differences.
So, once you obtain the raw recording it would be OK to scale the individual samples to push them closer in terms of dB and achieve some "compression" that way. The velocity curve should be able to handle further expression when playing.
IMHO, if you try to hit the target dB's of the peaks within the attack portion as I described initially, you might be able to use it as a checkpoint to make sure you are really getting useful GB's worth of wave files when recording initial raw material. Next, you can apply scaling and all other nice things to make sure you balance the instrument! But, keep in mind! In order to get a variety in timbre you have to have a variety in force with which you trigger the note, hence you end up with a variety in dB of the initial raw material!
I hope I clarified the point I was trying to make... (I would use some kind of dB table only to ensure a variety in timbre when sampling the material!)
So, any comments on this? Do you think it would be useful? Do you think it could get you consistent results with rich timbre differences?
If any developer would like to try it I can help them by providing MATLAB scripts to process samples, obtain peak power measurements, and scale the samples to target power levels for usage in the final instrument...
I'm so desperate to find a real good Steinway samples, that I would do it for free just to get somebody to generate a library that would be "perfect" (timbre rich, perfect consistent tuning, excellent mechanical/acoustic condition of an instrument).
I think if you sat down at a piano with an spl measuring device, you'd find that once you pass a certain point and hit the keys harder, they don't get much louder... but the timber, as you say, changes and the sound gets harsher and rougher. So your db measurements may help with softer strokes, but not so much with louder hits. And your mics and recording technique will determine if you accurately capture the nuances in character.
I'd suggest you think in terms of key-strike force (velocity or weight) rather than just dbs. I think if you can strike a key with 8 or more discreet forces that you'll pretty much cover it. If you're looking for a more systematic technique other than feel, try a grappling hook (rubberize the hook) with a bag of lead shot in place of the handle... you can vary the weight of the lead to get a fairly consistent range of strike forces.