Physically modeled pianos (Page 1 of 2)

Roland V-Piano?  Physis Piano? The dead Pianoid? I know no other ...

I know only 2 physically modelled software pianos: Pianoteq and Pianoid.
There are also few hybrids (samples+modelling): True Pianos, Pianissimo, Supreme Pianos (sound magic).
And few hardware examples: Roland V-Piano, Physis Piano

The V-Piano came out 3 years after Pianoteq was first released. I suspect it was at least inspired by Pianoteq. Physis is the other system I know of. They're both tied to hardware (you have to buy their controllers). There are also hybrid systems around. https://www.samplemodeling.com is supposed to be one of the best ones; they mostly do wind instruments and they do them very well from everything I've read and heard. It's empirical/mathematical-modelling (not ground-up physical-modelling) that is trained on recordings of real instruments. They don't do pianos but besides wind they do bowed string instruments. Their focus seems to be instruments where the player has control of the sound continually.

Phil Best has a youtube channel and he regularly demos new Pianoteq products. Interestingly he uses a V-Piano as a controller but seems to prefer the Pianoteq engine. All his videos (including his non-Ptq plug ones) use Pianoteq. Maybe it's worth contacting him privately to ask for his opinion on the differences. https://www.youtube.com/channel/UChw5_u...8yzyTHLq1A

Roland offers their modeled pianos (and others of their so-called SuperNatural instruments and sounds) in their newer keyboards (such as the RD-800) and in their rack-mounted tone generator, the Integra-7.
Note, added later-- EvilDragon has responded a little later in the thread that Roland currently uses a hybrid approach for their "SuperNatural" instruments based on sampled sounds with added modeled or filtered convolutions or effects. Thanks, ED.

I have the Integra-7 (which I bought specifically for the modeled pianos and their string resonance), but PianoTeq's modeling is quite superior to Roland's in my opinion. I play a Yamaha CP5 digital piano, which has a couple beautifully sampled grand piano sounds (but without sympathetic string resonance modeling, so they sound rather flat and sterile compared to Roland's or PianoTeq's), several rack-mounted tone generators including the aforementioned Roland Integra-7, but in almost all cases (unless I simply don't feel like turning on the computer and want to quickly play using the digital-piano or rack hardware) I greatly prefer PianoTeq (I use version 5).

I also have an older GEM (GeneralMusic) RealPiano Expander, one of the first hardware tone generators to have modeled string resonance for piano, and it still sounds very good, but both it and Roland have been eclipsed by PianoTeq (IMHO).

PianoTeq's modeling is already "virtually perfect," and unlike sampled pianos (even those that include modeled string resonance) the users' ability to customize the various parameters that influence the piano model and the resulting sound in PianoTeq is extremely versatile and just amazing. 

Steve

I'm not sure if the word modeled is being used right.

For me true modelling is the mathematical emulation of piano sound, and not just get a sample and try to apply filters on it to try distort, prolong, alter somehow. A true modeled piano do not depends of samples to have a note, but recreates a note mathematically.

So this talk of modeled piano that distort a recored note to create 3 ou 10 variations, it's purê bull...

The model can be an abstract mathematical model that has the flexibility to reproduce the acoustic behaviour of physical instruments. Such a model seeks to capture all the essential characteristics of the sound source in as simple a way as possible without worrying about irrelevant details of how the sound was produced. If you're thinking of interpolating between samples and combining them in various ways to mimic resonance etc, then that's not what such models are about. (Although creating sympathetic resonance out of individual note samples will involve some kind of mathematical model, however crude.)

I think "sample modeling" was a poor choice of terminology for that approach. That model does not store any samples. It is trained on recordings of real instruments, and adjusts its internal parameters to best match the instruments' behaviour. It's a bit like teaching a neural network to speak by providing it with examples of people speaking. The network does not store recordings/samples of the speech but learns how to speak through training as it creates its own internal model for speech.

It's very much a valid approach, and there is no reason why it cannot be as authentic as a physical-modelling based scheme. And for wind instruments that can be much more complex than a piano (with rapidly changing chaotic turbulent flows), it is probably the only way to go. I suspect that even Pianoteq, after getting the physical model (PM) right, encodes the PM in a similarly abstract mathematical way in order to be able to be run in real-time on a PC. Dr J Chabassier's 2003 thesis talks about modelling the striking of a single note on a piano and how calculating one second of actual time in the simulation took 300 CPU hours on a supercomputer (I may remember the numbers wrong, but it's that kind of ballpark figure). So there is no way a full virtual piano is being modelled on your PC as you play Pianoteq. It's some mathematical distillation of the essence of the instrument/physical-model that you're interacting with.

What are you referring to in the above statement? I cannot find a post on this thread that talks about anything like that.

In 2009 Yamaha put out three models of hardware digital pianos for stage use that have physically-modeled piano instruments. This was to compete with the Roland V-Piano. These included the Yamaha CP1, the CP5, and the CP50 which are still listed on the Yamaha website. But they have not been big sellers. Yamaha seems to have more financial success with entirely sample-based "workstation" digital pianos.

http://usa.yamaha.com/products/musical-...mode=model

http://usa.yamaha.com/products/musical-...mode=model

Yamaha calls their virtual-modeled pianos "Spectral Component Modeling Technology". In addition to the purely digitally-modeled virtual piano instruments,  these keyboards also feature sample-based instruments for other sounds (organs, strings, DX7 and whatnot).

It's my understanding that the high-end model CP1 has an entirely physically-modeled piano, while the lower-priced models may employ a hybrid approach.

OK - those are home pianos. I was talking about SN pianos in Integra and JP-80. Not all aspects were modelled there.

From the Roland US website--
http://www.rolandus.com/blog/2013/06/04...ernatural/

What is SuperNatural?

Roland instruments with SuperNATURAL technology adds powerful sound modeling to the equation (to sampling). Each SuperNATURAL instrument— piano, violin, trumpet, etc.—has its own specialized sound engine because, just as the sounds of a piano, flute and timpani are very different, so are their related modeling needs. Each SuperNATURAL engine is based on pristine studio samples of the original instrument; however, that’s where the similarities to PCM-based (sample-based) gear end.

(image of INTEGRA-7 With SuperNATURAL Technology)

First, SuperNATURAL doesn’t use sample looping. We’ve also developed technology that creates models of the characteristics and responsive aspects of each instrument. The SuperNATURAL engine handles how those parameters react to performance dynamics and nuances from a controller such as a keyboard, just as they would in the original instrument. As a simple example, volume and brightness in SuperNATURAL instruments respond to nuances in playing force without any velocity switching and inherent stepping sounds between sample layers. SuperNATURAL goes much deeper. Looking at the orchestral strings scenario mentioned earlier, SuperNATURAL modeling translates the player’s natural use of playing styles such as legato and staccato into appropriate variations in bowing technique and associated sound. Further, use of a pedal or switch will seamlessly invoke tremolo or pizzicato styles.

--- end of passage from RolandUS.com ---

Note from me (Stephen Doonan)

I could not find (in an Internet search) a definitive answer about whether Roland's new LX and HP pianos use only modeling, or sample-plus-modeling to produce the sounds. The information I found merely stated that the LX and HP pianos (as of 2015-1026) use "the newest" or "most advanced" version of Roland's "SuperNatural" sound engine technology.

The Integra-7 rack-mounted tone generator I have, which uses the above mentioned "SuperNatural" engines to create the piano sounds (as well as other instruments) does not sound nearly as good or authentic (to my ears) as the PianoTeq modeled pianos (with no samples). Roland's V-Piano uses completely modeled sounds (no samples), but because I don't own one I can't make any comments or comparison between the V-Piano and PianoTeq, although it's difficult for me to believe that Roland's models could compete favorably with PianoTeq's. Perhaps someone else with experience on both could comment about or compare the two.

I much prefer PianoTeq's modeled pianos to Roland's "SuperNatural" ones.

It certainly seems that way. The marketing "information" seems more intended to obfuscate than to clarify.

After listening to quite a few demo videos of Roland's V-Piano (which uses only modeling, no sampling), it seems to me that PianoTeq offers as much or more in terms of quality of sound and user-configuration of the parameters affecting the sound than does the V-Piano, without having to buy the extremely expensive hardware. I think it might be appropriate to say that the most value-for-money approach would be to buy a MIDI keyboard for the action (there are so many relatively inexpensive MIDI keyboards available with very good, responsive key-action) and PIanoTeq for the sound.

If it's 100% modelled then why do they have V-piano and SN going as separate projects? Or they've given up on the V-piano (which hasn't had any major updates for many years afaik)? The guy in the video sounds like your typical skilled Aussie salesman who tells you anything you want to hear.

Have you played the piano?

Modeled....
Hmmm, are analogue circuits that produce wave outputs NOT "modelling" sound production ?

The computer world was not always predominantly "digital".

Arguably Roland's SA was analogue "modelling" in the 80s - - their "marketing" copy
is SOME attempt to tie what they do now to what they did then, i.e. they are trying to
show continuity - evolving development if you like.

I could be interested in a rack module of the Physis piano engine, assuming there was an app
to access the settings. 
Right now the difference in street price between their K4 and K4-EX is
about $700, which might be OK.

No, I don't think it's going to be anything like Pianoteq. The Owner's manual is silent about anything groundbreaking or new, it just says "Piano Sound: SuperNATURAL Piano Modeling" in the specs. You'd think they'd at least briefly highlight the advantages of the product if they truly had something special or new. Yeah, it does not just play back a sample, but processes several samples to create a sound and makes also some adjustments based on your settings. Any half decent sampled piano will do that. I admit that I am not speaking from experience with that particular model, but I believe I have seen enough Roland marketing BS to be sceptical of anything more than incremental improvement on the basic SN scheme you get with less expensive models. But really, no one should listen to some random opinionated person like myself on the Internet. You should play and listen to the piano yourself and make up your own mind.

From my favourite old electronics engineering textbook (public domain copy from archive.org):

@Piet De Ridder
I was going to have a go at you for being so being so mean as to make one poster delete their post sharing their personal experience with the LX-17... Then your user name rang a bell: "P. De Ridder". You do some of the demo tracks for Pianoteq, so I presume you know what you're talking about more than most of the rest of us. Still, I think you could phrase things more diplomatically.

My personal bias against Roland is since having played their V-accordions.  The marketing hype gave me high expectations. The experience of playing the instruments, including the top expensive models, left me with disgust and in disbelief. Their advertising and marketing was extremely misleading and I believe I speak from experience not ignorance (but that's also what an ignorant person would say, which is why one should not trust what they read online or buy instruments without trying them first).

I would have a hard time to judge.
If I am right I hear the sound of a grand in the LX piano which is something different from the model B. Something to consider is also the speaker system in the Roland which enhances the model and makes a huge difference when you hear the piano sitting in front of it and when you record it (I own a Roland myself so I have some experience with it). In the end it would just be an opinion based on my likes and dislikes.  .__.

Forgive me if I say some nonsense below.

But I think there is moddeling level 1, modelind level 2 and modeling level 3...

-Before modeling level 1 we may have modeling level zero, that would be the simplest technology used in some digital pianos, as looping and key release samples to simulate key interruptoion, some algorithms to pedal sustein or repedaling.  No big deal, but antway it atempts, in a simple way, to change, distort, prolong samples to fit it in a playable digital píano.

-And we have modeling level one, like the GEM Pre Mega, trying to recreating more velocity layers for notes in a way to try convince some people, and sympathetic ressonance effects like string ressonance, and pedal ressonance.

-The New Supernatural Roland series may have a modeling tone generator, good timbre. But it if it have no option to change hammer hardness, string length, soundboard impedance etc... it's a strong signal that this is not a full modelling of separated elements (hammer, string, soundboard, harp, cabinet) that makes the piano as we know.
  That's why I put it on modeling level 2, as I presume it's a physical modeling just trying to imitates very well the soundwave behavior and the changes involved while in interaction, and trying to simulates string ressonance, pedal resonance.

-Now pianoteq and V-piano have a modeling that emulates not just a core sound, but a resultant of many modeled separated elements that have as result the piano sound. That's why pianoteq and V-piano allows to change hammer, strings, soundbaord etc...

Can someone show me a proof Roland have one single new digital piano (with exception of V-piano and V-piano grand) that allows to change hammer and strings ???

Yamaha modeled series appears to have such controls.

Hello,

I don't want to take a position or to compare Roland and Pianoteq, I haven't tried the LX serie, and I don't know whether it is full modeled or not...

But for you information, except Single note volume and Single note character, these are exactly all the same parameters available in my Roland HP-507.

Based on what I’ve read and what I’ve heard people (who are supposed to know) say, the LX is every bit as much modeled as the V-Piano or Pianoteq.
Why they didn’t upgrade the V-Piano instead is a valid question — painfully valid, if you’re a V-Piano owner — but I’m guessing that maybe the new software won’t run on the V’s hardware. (Something not unfamiliar to people who’ve been using Apple computers for the past few decades.)

Look, I don’t know anything about all this under-the-hood stuff, and I’m not in the least interested either. Sampled or modeled, or some hybrid in between, … I couldn’t give a toss. Nor do I have any affection for Roland.
The only thing I know is that, when listening to the LX, I hear something which, in my opinion is very much deserving of our appreciation and respect. Praising Pianoteq on the one hand, and at the same time dismissing the LX as a failure, is, in my view, a dishonesty very much unbecoming to anyone who claims to have a genuine interest in virtual pianos.

_

I don't know how many companies committed to adding physical modelling to their intruments, but it seems to be they way everything is heading, whether fully modelled or hybrid. It seems to fall into two categories, either sold as software separate from an instrument, or built into the instrument.

If Modartt is anything to go by, physical modelling seems so much easier to update and advance.

I remember this talk  on physical modelling:
https://www.youtube.com/watch?v=dUcNzPhZdwk

It all looks good from here.

That was the most honest sales pitch yet: basically it's spmewhere between sampling and physical modelling. Note how he called it "behaviour modelling" and not "physical modelling". That means modelling the aspects you cannot capture by sampling individual notes (resonance mainly).

Sure, but it's most likely using an empirical model where each individual note's samples are fitted to a generic formula that describes the sound as a mathematical function of hammer velocity and time (e.g., sum of overtones with time-varying amplitudes and frequencies). Such a function can be parametrised using a vastly smaller number of parameters (many orders of magnitude smaller) than the amount of data samples would take up. (This is because, except for the very short attack and release stages, the overtone amplitudes and frequencies are slowly and smoothly varying, and also because there are strong correlations between different overtones that can be exploited with a bit more clever maths/stats.) It's quite feasible that a gigabyte of samples could be reduced to a few or few 10s of kilobytes worth of floating-point numbers.

So indeed the samples need not be stored on the instrument because they always can be recovered by applying the mathematical formula that was fitted to match them at selected velocities. Once one has such mathematical description for how individual notes respond, the rest of the model will emulate more complex interaction effects, most notably sympathetic resonances, in the same way that it's been done on earlier SN piano models. So the LX-17 would not use a near complete physical model like Pianoteq or the V-Piano, and would still have a sampling-based approach at its core.

That this is the general approach behind this "fully modelled" instrument is strongly suggested by the fact that one cannot alter many aspects of the instrument's characteristics, like hammer hardness, length of strings, number of strings per note (like on the V-piano), just to name a few. In fact the LX-17 has no user tweakable model parameters beyond what the LX-15 had, and those parameters also have the exact same settings options on the two models.

Furthermore, the LX-17 manual, online description and marketing make no mention or give any detail of a physical model of the guts of the instrument, unlike the detailed info we got from Roland about the intricacies of the V-Piano's physical model when it was released. All the behavioural modelling aspects demoed in LX-17 sales videos have already been around in earlier models (although they are often presented as if it was the first digital piano ever to have them).

I'm not saying the LX-17 can't be an excellent piano, but it still smells very much like a sampling-based approach at its core with behaviour modelling on top regardless of the 'no samples on board' claim (which I don't doubt).

It would be great to hear from Philippe on how Modartt develop their models, but maybe he wouldn't want to give away too many trade secrets. If it was only done from recordings I assume they'd be a lot more sophisticated recordings than just individual note samples and capture the sound under a wide variety of realistic playing conditions. Then it's a much more complicated problem of fitting the computer model to the behaviour but can be done with clever optimisation and a lot of number crunching. But I'd think they'd take as much into account as possible, and collect as much real data about the instrument as possible, in as many ways as practicable, including actually measuring physical properties of various components more directly.

In reply to your second question, it's for reasons I've outlined in my previous post. Users have the exact same model parameters they already had access to in previous not "fully modelled" SN instruments. And the lack of any detail from Roland about the modelling. If they had something groundbreakingly new, or even something approaching the V-piano, they'd surely make a big song-and-dance about it and detail some new features of the model (and give the user more control than in previous models).

But the only thing I see Roland's reps show off in demos about the "new" modelling technology is sympathetic resonance, which has been around for yonks in DPs. Like this guy starting his spiel at 1min 19sec into the video:

https://www.youtube.com/watch?v=3B6QIH676OA&t=1m19s

(The shop owner then goes "Wow that's impressive!" - either he's playing along, or he's clueless.) And it's not just this rep, but they all do it: presenting nothing but long-implemented features as if they were new to these latest LX models.

Not really. V-Piano DSPs don't cost that much. Roland made it expensive to pay off years of R&D. Without that, I'd say a normal street price for V-Piano should be 2-3k.