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

That's because the note/key you hear is C2. Also, I don't like how certain aspects of the string resonance behaves on the synth/DP when I play certain classical pieces, so what I recorded is my current settings. If you are interested and when I have some time later, I can record again with much higher/louder resonance settings.

I'm confused. What have we learned here?

Probably that even the current best DPs and software pianos can not reproduce the resonance particularities of a real grand piano very accurately.

About the missing fundamental, it is because the soundboard can't produce much sound at a very low frequency. Even if the string does vibrate at the fundamental frequency (and other partials).

The others frequencies doesn't surprise me. If you see the steady sound, after the attack, you will only consider few frequencies (fundamental, and its multiple). But if you consider the quick attack you will have to consider other frequencies which may excitate the corresponding strings. This should explain the enhanced spectrum.

(The spectrum of a product - steady piano sound + envelope - is the convolution of the spectrum of available signals (partials of the string + something like a noise. ).

As you can see in my OP, the fundamental of C1 is clearly present in all four pictures, though at an amplitude 20 to 25dB below the 2nd partial. You are correct regarding soundboard resonance. The same effect occurs in speakers. Frequencies below the resonant frequency of the driver (soundboard or speaker cone) are not reproduced well.

What we have learned so far is that the modelling of an acoustic piano, and I include the necessary manipulation of sampled sounds in this statement, has come a long way, but is not yet a high fidelity simulacrum of an acoustic piano. This is not to disparage DPs. Quite the contrary. I see no particular need for a DP to be limited to an acoustic piano sound. It is an instrument in its own right.

That being said, there is much concern and gnashing of teeth here in the DP forum over how realistic the sound of one's DP is compared to the acoustic piano.

Well, that explains it. I didn't listen to the sounds, just analyzed them.

I do have an MP7 (which now resides at my girlfriend's apartment), and an LX17, but I don't have any recording equipment. The best I can do, is put the piano's line-out into the laptop's line-in.

The LX17 has a settings called full scale resonance (FSR), and damper resonance (DR).

- FSR controls how loud the resonances of pressed keys are. for example, silently press and hold a C-chord an octave below middle C. Then hit the same chord starting on middle C, and you'll hear the lower chord resonate on the pressed keys. FSR controls how long and loud those resonances are.

- DR controls the 'sonic wash.' If you press a key with the pedal up, the sound is much thinner than when you press it with the pedal down. DR controls how far the 'wash' travels up and down the keyboard (how many keys it affects), and how loud it is.

I'll try and get a decent recording, but I doubt it.

I am happy to analyze any files. It is interesting to see the different approaches to sampling and modelling.

Recording wise, as with the HP605, there's a USB port you can use and then "record" to that, I think as a WAV, but that's all in the settings. You just use the "start/stop" button for the recording.

If I was going to sample piano, I would try to dampen any surrounding strings to avoid any inharmonic resonance so I would aim for the cleanest possible single notes, because as pointed above, this is not a sum game. Resonance has to be simulated in order to resemble real situation. Hence only few top end DP get anywhere close. Most of the low to high DP are just very clean, you can hold the sustain and it won't get into a mush territory. Even on my supernatural roland with resonance all the way up, it behaves very predictable and clean. I don't find it a bad thing though, unless you need to switch from acoustic to DP often. Then the acoustic unforgiveness may become a problem.
I think my Casio uses rather simple sympathetic resonance - it doesn't really play the actual samples for the sympathetic strings, but use some sort of simple wave generator.

Ideally, there should never be any requirement for sampling. Only modelling can recreate
1) the actual interactions of the strings with each other:
2) the interactions that occur when multiple strings vibrate the bridge both vertically and horizontally:
3) the many resonant modes of the soundboard: and
4) the plate/rim vibrational modes.

This of course does not address hammer velocity, hammer velocity partial amplitude response, keybed noise with respect to key velocity, damper noise with respect to damper pedal velocity and position changes, and ....and....

To accurately model the sound of a single string requires enormous computing power, so sampling is an inexpensive and effective alternative.

Any attempt at sampling can only capture an infinitesimally small number of the potential combinatorial interactions that occur in the ordinary course of playing the piano.

What? Only modeling can recreate the interactions? What do you think an acoustic piano does? It creates those interactions.

Proper sampling would capture all of that ... if only there was enough will to capture a rich set of every note with both pedal up and pedal down.

The will seems lacking, so we're stuck real sampled lone strings and fake resonance simulations.

Modeling is no better. It's fake lone strings with fake resonance simulations.

We need to agree on what the word 'model' implies.

The Acoustic Piano is a mechanical device, a 'model' if you will, that creates from a variety of devices - keys, whippets, shanks, hammers, strings, bridges, soundboard, plate, rim, and dampers - a set of interactions from vibrating strings.

The Digital Piano Simulator is an electro-mechanical device, a 'model' if you will, that creates from a variety of devices - keys, shanks, sensors, sound generators, damper sound modifiers - a set of interactions from mathematical algorithms.

The level of fidelity of the DPS compared to the AP is limited only by the processing power. This level of fidelity is now achieved in the Flight Simulators that I trained on in the airlines. They were so real that my first flight in the 'real' aircraft was with a full load of paying passengers. This level of fidelity is achieved through independent multiple parallel processing.

Whereas a model starts from a tabula rasa, a sample set starts from a number of severe limitations and distortions - a full slate if you will.

1) Each sample set is unique to a unique piano serial number, meaning if only one set is produced, every person's DP will sound identical.
2) That piano, to be worth sampling, needs to be set up perfectly. This means regulation, tuning, voicing, elimination of buzzes, proper alignment of dampers. Each one of these tasks is fraught with potential errors and non-linearities.
3) The microphones must be perfect - no harmonic or IM distortion and perfect frequency response - obviously not possible.
3) The placement of the microphones must be such that no emanated sound is lost due to the mics being in an interference cancellation node or in an additive node and no room modification of the sound can be tolerated.
4) The preamps must be perfect - see mics above.
5) The number of samples must be large - arbitrarily large - infinite in theory - somewhat less for practical use.
6) Some modification of the samples will inevitably be required since none of the above is perfectly achievable.
7) The processing power required to assess and manipulate the trillions of required samples in real time is laughably large and not presently achievable.

"The level of fidelity of the DPS compared to the AP is limited only by the processing power. This level of fidelity is now achieved in the Flight Simulators that I trained on in the airlines. They were so real that my first flight in the 'real' aircraft was with a full load of paying passengers. . ."

Gulp . . .

Yes, a model starts with a full slate. Generally the VST products are produced using high-end pianos. Ones that people pay tens of thousands for ... because they (and I) adore those full slates.

In your numbered items I think you overstate the need. And you ask more from samples than you ask from models!

Your item 1: I don't see the point. I get the same sound from my upright every time I play it ... save for when it goes out of tune. But with a VST instrument I can (and do) change instruments with the press of a button. It's only the same until I decide to change.

Your item 2 is adequately handled with today's sampled offerings.

Your item 3 is not a problem. There is no perfection anywhere. And speakers are far worse than mics, so I don't see why we need worry about mics.
Your item 4 ... same, and even less concerning than mics.

Item 5 is the only one of concern. But there need be no infinite number of samples. There need be only enough to sound better than what we have today. We don't need a sample of every note played under every circumstance. Just a variety wide enough to capture the strongest, most important resonances and other aspects of the piano's character. That's more than what we have today. But tomorrow will (or might) bring more.

Rather than gulp you should be thankful that this is how it works. The pilot who gets into the real airplane for the first time after this extensive simulator training is about the safest and most competent driver you will experience.

No commercial pilot flies the real thing today without extensive training in the simulator. The level 'D' simulator, the standard today, is considered to be simply an example of the model you will be flying. It is no more, nor no less than the real thing. It is simply too expensive to fly the actual aircraft, and, more importantly, the pilot gets to train every possible known emergency procedure, which, in the real airplane is not possible without extensive damage to the hull and engines. We get to make mistakes, learn, and then not make mistakes when confronted with the same emergency.

In terms of how the real airplane feels, responds, looks like from the cockpit, behaves, smells, sounds, vibrates, whatever, in comparison to the simulator, there is essentially no difference, no more so than driving two cars of the same model with different tyres or options.

I ask more from samples than models because samples are not real. They are distorted copies of the original sound. Models can be essentially infinitely variable, and each variation is an undistorted, original creation by the device, the same as an acoustic piano.

Why spend all your processing power manipulatin distorted samples when you can use the same processing power to create undistorted music?

I disagree with your assessment of item 2. I regulate, tune and voice my own piano. In most sampled DPs I have heard, there are badly matched (in terms of inharmonicity) wound bass strings, meowing unisons, poorly stretched temperaments, and improperly voiced hammers on some notes. The fact that most people don't hear this doesn't make it any less a problem.

Quote from a great paper published in 2013:

F. A Virtual Piano
The resulting numerical scheme is stable, under the previously mentioned conditions on the numerical parameters, globally implicit, nonlocal for the soundboard, and uses multi- farious methods. The efficiency of the computer code is optimized through the use of adapted additional unknowns and Schur complements, so that the update of the unknowns of each subsystem is made separately at each time step. A massively parallel computing was necessary, and special attention was paid to the cost of each step, in order to minimize the global computation time. In average, computing one second of sound for the complete piano model (with frequency content up to 10 kHz) takes 24 hours on a 300 cpus cluster (around 86 ms per time iteration for the complete piano).

Here is the link to the paper.

You have it backwards ...The models are entirely fake. 100% fake.

The pianos are 100% real. And the sampled recordings of a real piano sounds real. Imperfect, yes. But very real.

Can it be more obvious?

If not, just listen. A good sampler sounds quite good. Not yet comparable to an acoustic. But quite good.

I only know of one pure modeled piano: Pianoteq. Take a listen. Its v5 is better than was v2. But that has only shifted it from terrible to unacceptable.

That's the difference between real and fake.

I don't think you are actually reading and comprehending my posts. You want to believe that the only way to simulate an acoustic piano is by sampling, whereas I am arguing that modelling is improving, and may someday surpass sampling. I am arguing that modelling doesn't come close to the realism of sampling...AT THIS TIME.

How can we have a constructive discussion (argument) if you are unwilling to see both sides? For example, you blast me with unproveable statements - "...sampled recordings of a real piano sounds real." They are still 100% fake, as much as is a modelled sound. Neither is real. Why can't you accept that. I am providing facts, you are providing opinions. I showed you a modelled piano compared to an acoustic piano in my original post. Was it not obvious that it was imperfect? You have not provided me with a properly recorded sample sample to prove your side of the argument. Why not?