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

On what do you base this assertion? Before you state these unequivocal, simplistic remarks, please be sure of your facts.

I hate to see these topics digress into generalities about the nature of sampling and the nature of modeling.

I'm guessing that few here have any clear idea of how modeling works, so it's not useful to build conclusions upon unsubstantiated premises. Likewise for sampling.

It would be meaningful if the discussion were to focus on the products in each of these categories.

I don't find much meaning, though, in statements about the methods and technologies, given that these are semi-secret/proprietary, and given that few here have any means to dig beyond a first-level understanding. (You just can't get that from a vendor's press releases, advertising copy, or web page self-aggrandizement.)

Isn't it obvious by now? He has absolutely no clue what he's talking about. After posting silly facepalm images and defending an irrelevant point about optical illusions and how such phenomenon are not possible in the audio domain, he is now resorting to generalized and insulting statements about the type of people who use sampled pianos. Not much more needs to be said I think.

Did you happen to take a glance at the Subject of this thread? ;-)

ok, I'm outta here, good luck with your samples.

Here's a big fat Raspberry! I'm just a jazz schump and not as serious as you obviously are. By the way, I think you just put down the sound of a modeled piano with your assumption that samples have a "nice sound."

I would get a cold - detached feeling if I only could play a piano that might have control but sounds like Mr. Roboto. But to each his own.

I haven't counted but it appears that you are in a minority here who thinks Modeling at it's current form doesn't hold a candle to a real piano that's been recorded and made playable by people who can't own a concert grand piano.

It also appears that it's your clan that keeps having to defend your precious expensive V every chance you get.

Once again, a thread on this subject has descended into vitriol spewing ultra-partisanship. I don't get it.

I think it's great that manufacturers are experimenting and developing the science of modeling. Have they perfected it? No. But given the snail's pace of change in electronic instrument technology this past twenty years, we're lucky it's taking place at all.

In the meantime, we have access to a slew of sampled offerings - either through the power of VSTs or the convenience of hardware. Do they provide a flawless substitute for the real thing? Not quite. But are they viable? Undoubtedly.

The internal combustion engine has been around well over a century. It's polluting and inefficient. The new technologies involving batteries and fuel cells are likely going to be much saner alternatives for an overpopulated planet. Can they do everything the oil-based engine can do? No, not yet. But should we welcome their development? Absolutely.

Does it help to have 127 velocity samples, rather than 10 or 20?

Disregarding velocity as it pertains to timbre, for just a moment ... If the total dynamic (loudness) range is, say, 50 dB, then 127 velocity levels produces about 0.4 dB change per step. Given that it's difficult to hear the difference between two sounds that are 1 dB apart even under the best conditions, does having 0.4 dB steps help? I don't think so.

Bringing velocity-dependent timbre back into the picture: I think likewise about the velocity samples number. I think 3, 4, or 5 isn't enough. But isn't ten enough? Surely twenty must be enough. But 127? Can anyone distinguish that many timbre gradations?

I think this is vendor specsmanship. It takes real work to sample more levels when building a sampled library. But it takes little effort to give a modeled piano more levels, up to the limit of MIDI granularity. The result is a "superior" spec that might attract a buyer's attention, but offers no real improvement.

I don't think we need 127 levels of loudness. Fifty would be plenty.
And I don't think we need 127 velocity layers, either.

I wonder about the work done to "improve" either of those. Isn't that a cheap way to avoid the hard work of making real improvements?

Isn't that what Mr Fourier is for with his transforms?

Whatever the pros & cons of sampling v modelling, what matters in the end is the music-making, not the technology behind them. Don't know about any of you people, but I buy a DP based on how it responds to my (inept or otherwise ) attempts at getting it to produce lovely (when appropriate), soothing (when appropriate), expressive (always appropriate) music; or conversely, ugly, menacing, aggressive sounds (when appropriate). If it doesn't respond as I expect it to, I don't buy it. End of.

I am not claiming that it is necessarily ideal to have 127 velocity samples. I was just saying that I think that would be the only way to capture the same detail of velocity transitions that could be accomplished with modeling. Whether that should be the goal, or whether there are other problems with doing that, or whether it is worth the effort and expense, are all other questions.


To create a truly smooth gradation between 50 audio points, I think you should be unable to hear the difference between any two adjacent points (but it's fine to be able to hear the difference between any point and another that is two points away). So yes, based on your numbers, that would be what would be required.

It's the same with visual colors. I'm sure you've seen an illustration of a color spectrum where you can easily see the change from one end to the other, yet you cannot see the difference between any pixel and the one right next to it. The result is a smooth gradation. If you could actually see a change from one pixel to the next with the naked eye, the "stairstepping" would get in the way of a the illusion of a seamless transition from color to color. So I think the smallest programmed change over time/distance should fall below the threshold of perception.

Also, note that the ease with which you can hear a 1 dB difference is variable. It is easier to detect at some volume levels and some frequencies than at others. At some frequencies and volume levels, even a smaller than 1 dB change can be easily detectable, at least by some people. Here is an interesting page, where you can test your ability to hear volume differences of smaller than 1 dB.

http://www.audiocheck.net/blindtests_index.php

Also, sometimes people can hear that something sounds "different" even if it's not exactly registering to them as "louder." This is used to explain why people will generally think that the louder of two otherwise identical sound sources sounds better even if the volume difference isn't enough for them to tell you that they think it is louder. A meter shows that it's louder, but subjectively, a person may simply think it sounds "better." Or someone may simply be able to tell that two things sound "different" even if it's so close that they can't tell you which is "louder." It is still affecting their perception of the sound, even if it isn't being expressed in the words you expect.

Getting back to the idea of using 127 steps for 50 db of range at .4 dB per step, there's another problem there, in that you assume that evenly spacing them that way will yield a realistic playing response, which I'm not sure is the case. You might need a specific amount of velocity change to create a smaller audible change in some areas and a larger audible change elsewhere in the overall velocity range.


I don't know what the number is, but I suspect it is a lot more than you think. Think of a simple low pass filter adjustment on a synth... imagine how coarse its operation would sound if it had only ten or twenty positions! And that only adjusts one parameter of the sound's timbre (essentially "brightness"), and in a piano tone, it is more than simple brightness and volume that varies with velocity... there is the behavior of hammer attack and subsequent decay, and how those harmonics vary over time.

I'm not saying you need the "infinite" variation of a real piano to create a convincing piano sound, I'm just saying that I think we are capable of hearing a lot more differences than you suggest.

A very well thought-through post, anotherscott.

Given that DP manufacturers try to get concert pianists to endorse or at least validate their products, there may well be people out there with the kind of experienced ears that can detect these minute changes. I've seen it stated that 127 steps is still too crude a number for such refined players. This is one of the reasons why digital recording has gone beyond 44.1khz (CD quality), even though the quantization at that level is "inaudible". The closer we get to a completely analog curve, the better, imo.

I am personally inclined to view claims that layer transitions in VST's are audible with skepticism simply because I can't hear them. I have reasonably good ears and I've played a good bit with several VST's and I can't decipher one layer from another in any reasonable one. Certainly they are not something that bothers me either in playing or listening.

On the other hand, when the timbre is off (fake, or whatever), I hear it right away and it bothers me a lot. That's why sampled pianos are better. It's as simple as that for me.

Should they have more timrbal layers and velocity steps? Sure, why not? I just got a new computer that is supposed to be a lot faster than my other computer. Can I tell the difference without a stopwatch and processor-intensive program? No. But that doesn't mean I don't want computers to keep getting faster.

It's not purely a question of hearing. A concert pianist is almost subconsciously trying to give extremely subtle nuances of expression to a piece. If s/he perceives (either through hearing, touch, or a combination of all the senses) that the instrument is not responding with this almost infinite precision, s/he will become frustrated.


Although tone perception is very subjective, I agree with you. We seem to have an almost innate ability to detect the tonal signature of a real acoustic instrument, and for some reason we find a recreation displeasing. That's why more work needs to be done on piano models.

Maybe we could say that current modeled pianos are stuck in the piano version of the uncanny valley.

But there are two questions here. What can the piano do? And what can the performer do?

Can any performer exert such a fine level of control? I suspect only the greatest masters can do so, and I'm not even certain of that. Regardless, that level of talent won't be playing a digital piano, eh?

More to the point ... I don't notice ANY shortage of dynamic control or of timbral control with a library having a mere ten sampling layers.

If Galaxy wanted to offer me a twenty (or 127) layer version, I'd have to ask them why they're not working on other, more important improvements.

Again, I think the Pianoteq's claim of 127 is just specsmanship. The samplers cannot match that, so the speccie-techie buyer will glaze over at the thought, without regard to its irrelevance.
I considered that point, but I dismissed it. Granted that some ranges might tolerate larger steps. But if you put larger steps in some places, you have to put even smaller steps elsewhere. I have no doubt that 0.4 dB is overkill, so a still smaller increment has no merit.
You make a conjecture here, but where is the justification? I don't mean to contrary. I just can't reach that same conclusion without some basis.

It has nothing at all to do with control, it has to do with how realistic it seems to play. You don't need to be able to reliably duplicate a specific velocity to be able to benefit from its presence. You may not be able to drive at exactly 52 miles per hour on demand, but if every time you accelerated from 51, the car lurched to 53 because it was incapable of 52, you would notice it.


Heck, I've enjoyed playing a single-layer P95. ;-) But are you saying that your 10-layer piano gives you an experience that is as completely satisfying as an acoustic piano? Even if that's the case, people have different experiences and demands... the fact that you don't notice any short-coming in a given system doesn't mean that no one will. Maybe more layers would be better, maybe it wouldn't. But unless someone gave you a 20-layer version of the same piano to play, it seems to me that the difference (or lack of one) is somewhat unknown.


Again, without hearing the results, that might be a premature question. (As an aside, what improvements do you think would make a bigger difference?)


Pianoteq and a VAX77 gives you 14,000 velocity layers! (I don't know if it matters, though have read numerous reports of people who love playing the combination, for whatever reason.)


You almost got my point there, but it sounds like you just missed it. Yes, if you make the steps larger in some areas, they have to become smaller somewhere else. The issue, though, isn't how inaudible the small steps become as you suggest there... it's how audible the large ones might then have to be. If, for example, you need to "stretch out" the velocity mapping toward the center of the range to make it feel natural to play, then you might have to compensate by having single-velocity-unit differences be far greater (i.e. more audible) in areas that are farther from the center. I don't know if it's important... but like you, I simply thought it was an issue worth considering, I just don't agree with your reason for dismissing it, because I think you looked at the wrong side of the equation.



Well for one thing, I think I provided reasonable evidence that 10 steps of timbral difference can easily be nowhere near enough to represent the limits of the differences we can hear with my low pass filter example... and that didn't even include the variables of a sound that changes over time. From my perspective, my post was full of facts, and it was yours that was full of conjecture!

There are an infinite number of velocities the human hand is capable of between generating an acoustic piano's quietest and loudest notes. On a scope, each would be different, but at some point, differences would not be audible. MIDI gives us an upper limit of 127 tones that can be generated (except in special circumstances, like the Pianoteq/VAX77 combination). Simple volume level manipulation gives us those 127 tones even out of a single layered sample, but that doesn't capture all the ways an actual piano differs as volume is increased. How many layers will capture it with complete authenticity, without modeling (even assuming a blend can be done absolutely perfectly, which is questionable in itself)? I am not suggesting that I know the answer to what that number is; I'm just saying that I don't see any evidence that that magic number is ten or whatever. Maybe, if everything else is done right, even fewer would be enough. Or maybe the sound of ten can be surpassed by the sound of more. I'm not saying I have the answers, but I don't see the justification for your dismissing of some of the possibilities.

Human beings have a surprising range of perception, even if they don't know it, or don't notice it consciously! Advertisers have known the effects of subliminal messaging (such as the classic inserted message frame in a commercial that is not perceived consciously) for a long time. Concert pianists and some lesser players will be able to feel that something is not quite right without necessarily being able to put their finger on exactly what is wrong. This is why it's important, for authenticity, to approach as close as is reasonably practicable to an analog curve - i.e. with as many steps as is feasible, so as to reduce the apparent "sound posterization" effect of jumping from one level to the next, rather than grading smoothly.

Anotherscott is right about this, imo.

There are limits after which we don't feel, see, hear, whatever, any improvement in resolution. It's the job of the marketing guys to convince us otherwise.

I do find it amusing that many here with such strong opinions have such a quiet profile. Guys, at some point you have to stop comparing lenses and start taking pictures.

An example is to enter following in the Google search box
site:https://www.pianoworld.com Sampled vs Modelled

The site:https://www.pianoworld.com part tells google you only want to search the content of piano world.