Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards Why all pianos are sampled not modeled?

I mean, almost all.

Let's look at top Kawai, Yaham and Roland Pianos.
Kawai - no sampling in DP, Yamaha - CP series
and some grands (even AG has samples sounds as I know).
Roland - V-piano (horribly ugly) and V-piano grand only.

Excluding these few horribly expensive pianos (except the CP series which offers modeled piano in price rage about 3000$ in Europe, the CA95 costs about 4-4,5k$ and HP-507 5k $ to have some price comparison) there is nothing to buy with modeling.

I always wonder, why there is no piano on the market that:
- would be in full DP cabinet
- have top action like PHAIII or RG from KAWAI
- have MODELED piano sound, not samplet
- and some few other software features.

The situation is even worst, because I can buy some so higly developed instruments like Clavia for much less than CA-95 or other piano of this class, but it looks awful and the keybord would be some Fatar, not good piano-inspired action.

However, each of the manufacturers have in their companies some kind of VST instrument, there are many companies that do it as independent like Pianoteq.

And, I cannot understand why even the TOP low range (HP507 etc) pianos are and most probably will be in a long time available with sampled pianos, not modeled. For people like me, and also many of us - who plays on acoustic, the sampled piano isn't tha same quality as modeled.

Resonance plates, and other magic stuff - why just not put the good sampled piano soft instead of some strange marketing tools?
Each loudspeaker manufacturer know, that resonance is one of the worst thing, while piano manufacturers are trying something elase to us.
Where is the sense?
The think that piano lovers are deaf and will don't see the difference?

KawaiJames,
Maybe you are able to speak about this in company
Can you imagine the CA-95 with modeling instead of sampling?
It would be total all-pianos killer, with some marketing of coruse, becase there still some countries in which all people are saying "buy yamaha" and "casio is good in clock not in pianos". And price between HP507 and NU1.

And - each person who has money and mainly - space and condition to have acoustic, will always choose acoustic,
so there is nothing to be afraid for acoustic piano sales.

Waiting for your opinions

In my opinion, modeling is just not there yet. Even the atrociously priced V-Piano sounds somewhat unnatural and synthetic. IMO, the best thing in modeling world right now is Pianoteq's Blüthner, and even it is miles away from the high-end sampled libraries.

I agree, modeling still comes up a bit short. But then, so does the sampled sound in digital pianos.

So, kapelli, my answer starts with a question: For Yamaha or Kawai, what would be the point of replacing mediocre sampled sounds with mediocre modeled sounds? There would be a tremendous cost to convert. But how would a manufacturer recoup the costs, given that the buying public hasn't exactly been clamoring for the $6000 Roland V piano (modeled)? (To be fair, the modeled sound in the V is pretty good.)

I think modeling must improve before it can become widely accepted. After that manufacturers might consider changing.

But this digital piano market changes VERY slowly. I wouldn't expect wonders to come cascading down upon us at the next NAMM show.

false statement.

Yes. OP is wrong in his/her original statement. V-Piano is fully modelled. Roland's other products using their Supernatural sound engine have some association with modelled technology although it is unclear what exactly. The electric piano presets on the RD series stage pianos are fully modelled.

The RD-1000 from 1986 was fully modelled and that technology continued as Roland's staple sound engine across all their DPs for some years.

The Yamaha CP series (1, 5 and 50) claim to use some sort of modelled elements. I think again the electric pianos are modelled on the CP1, 5 and 50.

The Viscount Physis Piano is another fully modelled piano with acoustic, electric and other keyboard instrument sounds, all apparently fully modelled.

Modelling is the future but there are sonic flaws intrinsic to the technology at the moment, but they will be overcome I have no doubt.

Most pianos are sampled for the same reason that when you call a company and get an automated service it's a recording of a human, rather than a synthesized human voice: It's cheaper, easier, and sounds more natural.

Synthesized piano sound has the same problem as synthesized human voice: it's a complex problem and the solutions we have tend to hang around in the uncanny valley. Recordings are obviously limited but they are cheap and easy, and they sound pretty much right.

First, not every company has the ability to do a quality modeled piano. Companies have their own unique skills and technologies. Second, not everyone agrees with your premise that modeling, at least in its current level of development, is better than sampling. I am among those who think today's fully modeled pianos (Pianoteq, Roland V) are inferior to many sampled pianos. They have certain advantages, but overall, I prefer numerous sampled pianos.


I'm not sure what you're saying here, but it sounds like you're equating the replication of natural piano resonances with undesirable speaker resonances, which is not the case at all. On an acoustic piano, the resonances are part of what makes it sounds as good as it does, and so a lot of effort is often put into trying to recreate those phenomena in an electronic recreation. In fact, one of the first "modeled" pianos, from GEM, used samples for the main piano tones, and used modeling to recreate additional aspects of the sounds, including the resonances.

Modeling of sounds is a complex problem and computationally very expensive to do right.

That's because you'd have to model all of the strings plus all of the interactions between all of the strings, plus all of the interactions of the strings with the resonator (the wood and corpus of the piano or grand piano).

You'd also have to do it for the complete harmonic series of each string/each tone at least until you get to some arbitrary volume/loudness cut-off (if a harmonic is x db below don't model it)

This requires lots of processing power to do "right" (if you can actually do it right with current hardware and synthesis algorithms) which might be prohibitively expensive (just think about what the V-Piano or similar systems cost even though the sound is not that great IMHO).

Current digital pianos already pack a lot of punch with regards to DSP (digital signal processing) power just to integrate basic modelling features (like sympathetic resonance of strings or whatever is super natural in Rolands Supernatural sound engines).

After a certain price point you'll have to ask yourself if the benefit is worth the cost and effort, i.e does the modelled piano sound SIGNIFICANTLY better than a sampled one. If not why bother when a sampled solution is both cheaper and less complex (and it is). If it does sound significantly better then are enough people willing to spend the additional money on that.

You also have to consider space (has to fit inside of the digital instrument), power (should probably work on 120V/240V and without a fan) and complexity of the solution (SW footprint, testability, shouldn't crash).

Right now most of the instruments with modelling are sold at a price point where I could either buy a real piano (granted not a very good one though), a device with significantly more features (e.g. Kronos X, Nord Stage etc.) or a sampled piano that sounds comparable (or even better YMMV) than a modelled one at a significantly lower price (I could buy 2 1/2 Kawai MP10 for one V-Piano)

I also could buy nearly all of the DAW software out there and a good 88 key weighted action keyboard.

With large unlooped samples the sound is "good enough" for the majority of people and modelling doesn't seem to be "there yet" as far as sound and price is concerned. Also most people looking for the "real deal" tend do buy a real piano eventually.

So to make a long story short. You compete against much cheaper sample based systems on the lower end, you compete against real pianos on the higher end and modelled systems don't seem to be "there yet" with regards to both price and "sound experience".

btw. when instrument manufacturers talk about resonance they don't mean undesired resonances due to resonace frequencies of certain parts or due to "resonance cascades".

They talk about intended resonances of for example the strings with the resonator or instrument corpus (which offers for example amplification of the sound) or of the strings with strings that vibrate at one of the harmonics of the basic tone (sympathetic resonance).

Both are processes that would need to be modelled/replicated in order for the instrument to "sound right", since it's what the real instrument also does.

So to reference another thread, if I'd have the choice between a V-Piano and a controller with a great piano action plus a SW solution with modelled or sampled pianos I'd probably go for the controller plus SW.

Or I'd go for "everything but the kitchen sink" boards like the Nord Stage or the Kronos X.

All those solutions would still be less expensive.

I'm not too into modeling cause I can't warm up to the sounds. But you'd think someone would build a modeling monster just to show that it can be done. Kind of like the first Moog synths. It appears no one has ever demonstrated it's possible?

It is important to separate modeling from sampling. Sampling is always going to sound and play better than a modeled instrument, for the fact that the samples are picked from well known top flite instruments. So, the impression is that you are emulating your dream concert grand.
After playing awhile, you start pushing the "this and that piano" buttons, hoping to get "todays'" sound, but oh well, the same sound comes out of those paper speakers.
Modeled pianos, have parameters that can continually evolve with your playing style. I have noticed new users initially dislike the metallic sound of some instruments, so they complain. With the knowlege that the user has control over many characteristics of each key, the sound becomes subjective and the ear hears music.

Modeling is to sampling what playing is to listening to a recording.

That statement is both bold and wrong

There is simply no "world formula" that completely and entirely describes the sound of a piano and that can be calculated by a computer.

If you want something that is as close as makes no difference to the real thing you'll unfortunately have to buy a real grand piano.

The fact that both modelling and sampling are handled by a digital computer with finite resolutions is enough that neither approach will ever sound 100% identical to a real piano.

Modelling and sampling are simply different ways of approximating the sound of a real piano both with their own strengths and weaknesses and focused on different tradeoffs.

So to go with the classic physics motto "All models are wrong yet some models are useful" the question is which approach is (more) useful to approximate the sound of a real grand.

Right now the jury hasn't given its verdict yet. Sampling "wins" due to cost and as of today usually "better" sound.

This might change at any time in the future however.

Nothing will change the fact though that modelling no matter how great it might become will always be an aproximation as will sampling.

If bandwidth, computing power and storage space is no object then one will be able to design a piano sound with modelling or sampling that will sound identical to each other.

So it's entirely up to which system is the better approach given the state of the art and the constraints of the technology. (and the willingness of the intended audience to part with their money)

On the contrary, you totally misunderstand my simile.

A modeled DP generates sound when you press the key, just like an acoustic piano. The quality of the sound is not pre-determined from a previous recording, just as an acoustic piano doesn't have its own finite 'store' of sounds to trot out when the pianist plays.

A sampled DP has prerecorded samples, the number of which is finite and therefore limited, no matter how much processing occurs afterwards to produce the eventual sound.

As for which sounds better, well, that's totally subjective.

Isn´t it the way, that for instance Roland put together in their SN piano sounds the best of both worlds? The difficult to model, but nice to sample attack phase of the tone is coming from a sample, and the difficult to sample, but nice to model sustain and resonance effects are added by modelling? Isn´t this the reason, why the V-piano hasn´t seen a successor for a long time, because its full modelling is not as effective as the combination of both worlds?

Oddly, it's the attack phase of the tone that for me separates full modeling from sampling plus modeling, which is why, for me, the V-Piano is so amazing in its sound generation and the way it emulates an acoustic piano. It gives you precisely what you want in terms of attack, from a gentle stroke of the key to a sharp accent to staccato to staccatissimo, and everything in between - all entirely dependent on, and only on, the way you strike the key.

This isn't a meaningful distinction, though. There is no difference between 5 and the sum of 2 and 3. One required you to compute it on the fly while I provided you with the other.

For both modeled and sampled pianos, for a given configuration and input velocity, the resulting sound will be exactly the same every time. In other words we could record the sounds produced by PianoTeq's computations (or the V) and make a sampled piano out of them. If we do the decent Kontakt scripting that the good sampled pianos have, we would have an instrument that I don't think people could tell is not PianoTeq (or the V).

The advantage I see (and that you implicitly point out) to computing the sound on the fly rather than having it already stored up is superior tweakability of the sound. Modeled pianos unquestionably are more tweakable, but that only matters if we agree that a major problem of sampled pianos is that they are not sufficiently tweakable. The latter is a statement I, at least, do not agree with.

Certain interactions and resonances can be easily added to modeled pianos, which is nice. That was one of the primary original arguments for modeled pianos. But we have observed these interactions being added to sampled pianos, so sampled technology does not inherently preclude them.

In either case, the differences in the technology are nothing like the difference between playing and listening. The computer is thinking on the fly, but its calculations are completely deterministic, so they may as well be pre-computed or sampled.

I'd take roland's V-piano over any sampled piano today without any hesitation

Sampled pianos are sooner or later will be replaced completely by modeled ones that's why they're offering so many of them right now because companies want to get rid off what they have in warehouse.