Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards Sampled VS Modelled Pianos ???

it's a bit of a misconception that a sampled piano with 10 velocity layers ends up having 10 timbres. It ends up with 127 timbres in reality. Each velocity will sound unique because the samples are blended steplessly. By adding more samples, all you do is give a more realistic reprensation of how these in between steps are supposed to sound like.

I'm thinking of a circle with 10 points that still looks pretty angular. If we add more points, it gradually starts to look rounder and rounder. But does it really need a high number of points in order to look round? When does adding more points become a waste of resources instead of an aesthetic benefit? Consider that the points wouldn't produce angles but form a curve in between similar to the blending in sampled pianos, you wouldn't need many points at all.

I'm pretty convinced that 20 to 30 samples are that point where adding more just doesn't make sense. These changes in timbre would be so minor compared to the already changing timbre achieved with the blending that it would just be a waste of resources.

I agree that there will be a point at which the steps are no longer noticeable, but just be aware that specialists in any field are trained to a very high level of awareness, and attuned to tiny differences. What is good enough for you or me, may not be for someone more specialized. It depends on your target market (and the cost-benefit ratio of the technology employed) as to the cutoff point.

What is good enough for you or me, may not be for someone more specialized.

MonsterCable?

Sure, but I don't see software pianos being a product tailored to these extrordinary people, whoever they might be.

Nice try! But not a good analogy. Some things have no proven benefit, or make no discernible difference. However, in the area we're discussing, I think we can all agree that 2 velocity layers is inadequate (even with layer-blending). On the other hand 200 (for example) may be beyond anyone's ability to detect. So that must mean that somewhere in between those numbers will be the point at which the average person ceases to be able to hear a difference. However, not everyone will conform to the average. Those who have trained their ears may be able to detect changes way beyond where the ordinary person notices things.

As a related analogy, performers go to extraordinary lengths to play instruments that meet their demanding standards, even though audiences would be completely unaware of the subtle difference between such devices.

I agree - at the moment. However people have speculated as to whether at some point in the future, a classical concert might be performed on just such a piano. Obviously, it won't be unless the technology matches the pianist's expectations. Even for private practice purposes, professionals will require there to be no detectable difference that could possibly affect technique. So to suggest that we've gone as far as we need in terms of development may be premature.

I doubt digitals will ever fully replace acoustics. This would mean that all instruments in an orchestra would need to be digital too, why else would the pianist be the only digital guy? I really don't see that happening.

The only reason we even have digital pianos is because of money and space issues. Nobody is seriously asking for a digital violin, and I mean one where the sound is generated entirely digital. Since neither money nor space is usually an issue for serious concerts (and that's the only place where you would find such incredibly demanding performers), I don't think we will ever see such a development.

Unless in some distant future they will have a digital instrument that will actually top the sound of an acoustic with some alien technology that we cannot imagine yet. At that point performers might be willing to let go of acousics in favor of even more expressive possibilities.

I agree with this, although I might not be quite as emphatic about the future - we never manage to predict it accurately!

One thing to bear in mind is that recording artists - even some classical and jazz pianists - may be more willing to embrace this kind of technology if it can rival the real thing.

A sampled piano with even 1 level can play MIDI velocities 51, 52, 53 (in any shape you want) monotonically increasing loudness for each step. It doesn't have to lurch any more than a modeled piano would.

Besides that your example doesn't apply to playing a piano. On a piano the sound does not glide continuously from velocity level 51 to 52 to 53 as you would in car. You discretely have to strike a key (or keys) in a sequence from 51 to 52 to 53. Each strike has its own attack and decay. Therefore it lurches with the attack of each separate note by definition. And besides lets see you repeatedly strike 3 keys in succession with velocity levels 51, 52, 53 on purpose while playing normally. If you do it, it will be completely by accident. You can't control the pressure of your fingers that exactly. So your playing is going to "lurch" anyway for these reasons.

You seem to have misunderstood me two ways, maybe I wasn't clear, or the natural imprecision of my car analogy made things worse instead of better. I'll try to clarify.

To answer MMM's assertion, "Can any performer exert such a fine level of control?", I was making the point that you don't need to be able to have fine enough control to produce a velocity on demand to benefit from its existence. To use your example here, it is indeed valuable that there is a MIDI velocity of 52 between 51 and 53. You may not be able to reproduce 52 on demand, but if you were playing something in that velocity range, and every time you happened to hit on 52, it refused to play it, but instead substituted 51 or 53, you could conceivably notice that a slight an unevenness of response in that range, compared to the smoothness of response of a real acoustic piano.

Or to make the point even more clear, let's say that you didn't have fine enough control to ever produce anything between 50 and 55 on demand, and you used that argument to eliminate MIDI velocities 51, 52, 53, and 54. Now you'd hear a jump between 50 and 55, and that would almost certainly be audible to many people as they played up and down through that general velocity region. The point, again, simply, is that the value of a velocity point's existence is not predicated on the player's ability to produce that velocity at will.

To the other issue you allude to: As you say, the normal behavior, even for a single layer sample, is to "increase loudness for each step." If this was, in fact, the only way a piano's sound changed with velocity, there would be no reason for more than one layer. But also, as velocity increases, the timbre of the note changes, as does the the shape in which the sound falls off immediately after the hammer attack. If you record a real piano striking a single note very hard, you cannot make that sound like the same piano playing the same note very softly merely by lowering its volume, or even by lowering the volume and closing down on an EQ filter. This is why (at least in lieu of more sophisticated modeling) multiple sample layers helps. You're right that the volume changes exist in MIDI pianos no matter what, it is these other changes that require more than just simple processing to achieve additional realisim.

How many sample layers would be required to reproduce those transitions as perfectly as they occur on a real piano? I don't know. (You would need 127 of them to do what a model can do, but that doesn't mean you necessarily need 127 of them.)

When dewster ran his DPBSD test on the up-to-8-layer Korg Kronos, he heard 6 distinct points where the timbre shifted as he gradually increased velocity on middle C. On a real piano, you would not be able to hear any. The more "data points" exist along the continuum, the smoother the transitions can be. I don't know if that difference--although clearly audible--really is crucial to making a digital piano feel more real to play, but I am not willing to say it makes no difference, either.


Right, but the fact that all those velocities are represented--even if you can't hit each one on purpose--helps make it natural sounding. Again, to exaggerate the difference, if every one of those hits were rounded to 50 or 54, it would not sound as natural as a bunch of smaller variations around 52.

In fact, since our fingers are capable of an infinite number of possible velocities between even 51 and 52, some rounding is already taking place. The question is how much rounding is permissible before someone can recognize that, gee, for some reason, this just doesn't seem to feel as natural as playing a real piano. If your argument is that nobody would ever be able to tell if 52 were missing (i.e. if the only options were 51 and 53), then by logical extension, we have no need for 127 velocity levels... we could eliminate every second one and have just 63 velocity levels (say, all the odd numbers) and no one would be able to tell the difference in how the boards played. It would be an interesting experiment, but I would not assume to know the answer without some actual experimentation.

Giagnotad, even much smarter people have done this mistake.

"There is not the slightest indication that nuclear energy will ever be obtainable. It would mean that the atom would have to be shattered at will."
--Albert Einstein

Maybe it's just that I'm from a different planet to (seemingly) just about every other person posting here, but isn't the proof of the pudding in the eating? In other words, if you can't hear the seams when you play with varying dynamics (no matter how many different steps), why worry about it?

We all know that CDs record in binary digital code, and therefore by definition, what you hear from a CD recording has something missing compared to hearing the live performance. In fact, a LP vinyl record has more information stored in its grooves than a CD ever could, just as a photo taken on transparency (slide) film compared to a digital photo, no matter how many pixels in the latter. But I'd far rather listen to a CD than an LP of the same performance (my huge LP collection has been gathering dust for decades, as I gradually replace them with the same performances on CD when the record company reissues them), simply because even disregarding the crackle & pop and low rumble of the LP, the CD sound is that bit clearer and cleaner, and I can't detect any missing information inherent in the digital format. (But the sound from a MP3/iPod is a different matter....).

As for digital pianos, it seems to me that sampling v modelling is redundant if you can't hear the difference in behavior between the two when you play, i.e. the way the sound changes depending on how hard you strike the keys etc (leaving aside the interaction of resonances and sympathetic 'string vibrations' for the moment). For me, I can easily detect the restricted dynamic range of all sampled DPs at ff: there's a ceiling beyond which you can't go. (And I should add that I'm no Russian bear when it comes to hard-hitting pianism - for that, listen to someone like Andrei Gavrilov or Grigory Sokolov). The sound gets 'stuck' at that particular timbre and volume no matter how much harder you hit the key, which is of course not how acoustics behave - even small uprights. But if you don't ever hit the keys that hard in the music you play, you'll never notice there's a ceiling there. And the same applies to the timbral change as you change dynamics - if you don't hear the steps, does it matter whether there're 10 or 256 steps? Modelling technology provides stepless change in timbre and volume, amoung other things, but if the pianist can't hear these attributes, are they really relevant to him?

The new DSLRs are revealing flaws in lenses that were not apparent when using film. It's often suggested in photography circles that 8 MP pretty well spelled the end of film for resolution supremacy - the latest Nikon FF body is 36 MP.

No wonder film is done.

The sad thing about CD technology is that it could have been so much better.

I've seen it stated that the moon is made of cheese. But I'm doubtful.

Bennevis: I'm with you on this. If you can't tell the difference, there is no difference. Machines can't tell you how a piano feels and sounds. But your fingers and ears can.

I think you underestimate the powers of human perception, especially when trained and honed.

But I agree that once we no longer perceive a difference then it doesn't matter. However, 127 steps is perhaps not much set against an infinity of possible gradations. Do you remember what computer monitors looked like when they could only resolve 256 colors? Why did anyone bother to increase it to 16 million if it was unnecessary, as it requires far more powerful graphics capability?

Because 256 colors aren't nearly enough to reproduce an image.

Then human perception can distinguish way more than 127 steps.

I just thought I'd throw this into the fray.

Depends on what kind of steps, I'd say. Steps of volume? Steps of ever so slightly different timbre in a piano? Possibly, but I doubt it.