Do any exist?

They don't exist, at least I could not find any.
There are modeled pianos like V-piano or Physis and then there are sound modules like Roland Integra 7, but they are hardware units.

The only thing I've ever seen that might come close to being considered a competitor is Sound Magic's Pianos.

http://www.supremepiano.com/

When I use to visit kvraudio.com I saw a couple reviews that said they sounded great.

Here's what their website says about the pianos...

Our Industrial Best NEO Hybrid Modeling Engine offers you the best from both the sampling and modeling worlds! The Authentic Sound of Sampling combined with the Playability of a Modeling Piano makes for a piano experience that you have to try to believe!
By utilizing the powerful NEO Hybrid Modeling Engine, Supreme Piano 2 offers the best from the two worlds, Realistic Sound from Sampling and Playability from Modeling. NEO Hybrid Modeling Engine leads virtual pianos to an entire new level by providing an instantaneous feeling, truer and richer in sound and it sounds alive!

It's some sort of hybrid of samples and modeling, so I guess it's not really a competitor in that sense since Pianoteq is all modeling. However I think they push the fact that it is still a small filesize compared to sample libraries and a lot less resource intensive.

They have demo versions of their pianos on their website, but I think some keys are disabled as well as sympathetic resonance.

Here is a link to some audio clips from on of their pianos called "BlueStone Piano" (I think it's supposed to sound like a Steinway)

http://www.supremepiano.com/demo/demo2.htm

Pretty much all DP's these days, software or hardware and using the hybrid samples+modelling approach - although they all use different ratios of the two. I wouldn't say SupremePiano is unique in that sense.

Yes, I agree.

I suppose it reminds me of TruePianos that was also small in size and less resource intensive, but used some sort of mix of samples and modeling. I remember when it first came out I had heard many rumors that it was all modelling like Pianoteq, but I believe their website now debunks that claim.

Can you tell which pianos besides Roland are using modeling? I am not talking about modeling the resonances, but sound modeling like in SN pianos.

In hardware, I think it works out this way:

Roland V-Piano is the only fully modeled piano.

Other Roland SuperNaturals use a hybrid sampling/modeling approach, as did GEM, and as do the Yamaha SCM models (CP1, CP5, CP50).

For the generation of the actual acoustic piano sounds, the rest (Casio, Kawai, Korg, Kurzweil, Nord) don't use modeling at all (though Korg does use modeling in their electric piano sounds on the Kronos).

Sure PianoTeq has competitors: all the sampled software pianos and also to some degree the hardware pianos. Only the V (apparently) shares PianoTeq's purist approach and doesn't use sampling as far as we know, but that's behind-the-scenes stuff. Anything that takes MIDI and produces piano-like sounds is a competitor.

We wouldn't say that Canon dSLR cameras don't have any competition just because the other companies' cameras use CCD technology while Canon uses CMOS would we? (My information on this maybe out of date but let's assume this is true for discussion purposes)

Roland is the only one I know of that uses a sample for the attack and modelling for the decay of each note. Pure modelling is only V-piano, Pianoteq and Viscount Physis. The rest are using varying degrees of resonance modelling, possibly key/pedal sounds too (not sure about that).

It can be more than that... for example the Yamaha SCM models things like the apparent softness/hardness of the hammer, and where on the string the hammer is striking. It even allows the user to alter these parameters.

I'd be interested to know whether that is considered modelling in the way we usually understand it, or whether it's filtering. The PT & V-piano style modelling is a wave-generating algorithm which tries to replicate all the various physical factors in a piano. But if the Yamaha is using a sample as a substrate, then changing other things on top of it, is it really developing a new wave or filtering/EQing a recorded waveform? If it is truly physical modelling, then we are talking about additive/subtractive wave combination. Seems like that might be harder than modelling the whole thing in terms of authentic tone. That makes me suspect it might be a filtering algorithm calling itself modelling. Of course, I could be wrong, given I have totally failed to look into it and cant be bothered right now!

Is it super hard in terms of cost of instrumentation and engineering talent to model a piano with computational physics?

It is my understanding that Pianoteq is not state of the art in pure modeling, and it would be interesting to know why they have not been able to close the gap.

It's a hybrid in the same sense that the Non-V Roland SuperNatural pianos are likewise modeling algorithms applied to sample sets.

Everything is super hard until someone figures out how to do it. ;-)

I have not heard that Pianoteq is not state of the art... or to put it differently, I am not aware of anyone who has done it better. Whether that's because we're still waiting for someone especially gifted to take it to the next level, or because the hardware required to implement it would be too expensive for the average consumer to buy, I don't know.

If I purchase the VPC1 in anticipation of a substantial improvement in pure modeling software, am I making an error in assuming that this can be achieved on a general purpose hardware system?

Yes - my understanding of it is that Pianoteq is precisely and uniquely state of the art in piano modelling. (Roland and Physis also, perhaps, but one gets the impression PT is far more active and interactive in their continual search for raising their standards). Whether you think they make the best s/w pianos (and some do, probably most don't, as we write) is another matter altogether.

To repeat one of my favorite quotes, "Prediction is very difficult, especially about the future." I would say buy something because you're happy with what it does today. Anything better it may do in the future is a bonus. Nobody can assure you that there will be a substantial improvement in piano modeling within any particular time frame, or what it will cost, or whether you'll even necessarily be happy with it.

Some sounds can be seamlessly overlaid onto a sample base because they are separate to the individual note. Pedal noises, resonance are examples. But hammer noise, hammer hardness etc can't. The sample already contains all of the original tonal information of the piano - the hammer used is the hammer used. Anything you put on top of it can only try to alter that. They didn't separate the hammer from the sampled note. So to change hammer type can only be some kind of filtering to simulate, for example, softness in the hammer. There would be no value in trying to physically model such a thing when you are so confined by the underlying sample. Modelling is hard enough in itself without being hamstrung by a sample. At this stage, I'm not buying this as modelling in the sense of Roland Supernatural. Roland has actually generated a wave of its own. Yamaha is just changing the way we hear a sample. I would have to call that filtering.

Owning a CP1 I think I have to agree with ando. I think the basic tonality of what you hear - ALL of what you hear on the SCM Yamahas is sample. There are some adjustable behaviours (hammer hardness for instance) - and that is where perhaps something else comes in. I wouldn't be able to explain any more than that but at the time of their release (and still now as an owner) I always felt their use of the term 'modelling' was a response to a perceived expectation of the market rather than the technical reality behind the products - ie, it's what they thought the public wanted and it was a way to piggy-back off the hype and initial storm of publicity around the V-Piano.

I think Roland's SN is different - a sample segues into a modelled 'thing'. It's not very flexible. On some of the Roland Pianos you can hear the decay portion overlaying the attack stage. The decay doesn't even sound like the same piano as the attack portion (I'm thinking Roland's pitiful 'rock piano' preset, which has a full, un-looped SN type decay). But I WOULD call this a hybrid of sampling and modelling, especially as we are led to believe that the seamless velocity changes of the attack portion are also due to some modelling.

Filtering removes sonic content. What if their algorithm also adds sonic content? (It is possible not just to selectively boost different frequencies, as an equalizer does, but also to generate/add overtones that did not exist in the original sample, as an exciter does... and both of these functions are beyond what would be called filtering.) What if the algorithm changes various aspects of the sound over time, altering things like the slope and rate at which different frequencies decay after the attack (envelope functions)? What if all of these things are calculated in real time, individually for each note, based on the user-entered hammer hardness and location parameters and the velocity of the key being struck, calculated based on algorithms derived from analysis of how the sound of real strings being struck vary in sound based on these same variables? I would call this modeling. It's certainly more than filtering. Not that I know exactly what Yamaha is doing. My point is just that you can start with a fixed sample, and still model a new sound from it that varies substantially from the sample, in ways beyond what could just be called filtering.

"In anticipation of a substantial improvement" SOON is probably a mistake. This stuff never happens fast.

Well sure, you can add to the original sample - and I did cater for that in my post above. My point is that if you are getting that good at the modelling game that you can substantially alter the nature of a sampled sound, you are good enough to model the whole damn lot. The sampled sound and resulting decay has very real evidence of the type of hammer used. If you can model that fundamental characteristic out of it, it is not much of a stretch to model from scratch. Personally, I don't think Yamaha's "modelling" is anything flash. It still sounds the same piano, it's tinkering around the edges. Until they commit to the full modelling scheme, I still think it is mostly marketing speak. Ever seen a DP brochure that didn't sound incredibly advanced and cutting edge? I haven't.

Does it really matter whether a piano is modeled? Or sampled?

I really don't care, because all that matters is the performance. I pick what sounds good and feels good. The tech doesn't matter.

It probably doesn't matter, but if we weren't allowed to talk about the technology once in a while we might as well pack up this forum. I think following what the technology is doing keeps the hope alive that one day DPs will be the great instruments they have always promised to be, rather than merely adequate.

Musicians Friend had Addictive Keys for their stupid deal of the day. Regularly $149 usd, "stupid" price is $29. Gotta say I grabbed it right away. It is sampling technology, but do not know if they also use modeling or not. Great deal - a Steinway grand, Yamaha U3 upright and a Rhodes Mark 1.

6 months ago, I'm not at all sure that the question was answered.
Maybe it was and I didn't find the answer I expected.

I'll try a different question;
HOW competitive is Supremepiano with Pianoteq ?
They appear to use similar methods (we could argue that), but how do they compete at the RESULTS level ?

Supremepiano seems to be about $140 per piano, or 4 for $399, but lets leave pricing aside for the moment.

SupremePiano's PianoOne is a free download, I will get it and give it a trial - next week if I have time.

Hi RB,
Like you, I was excited to try out PianoOne when it was released a while back. I mentioned on the other thread 'tuning electronic pianos' that I believe the future is hybrid sampling/modeling. Although PianoOne claims to do just this very thing, it falls short on both fronts IMO.

First, the sampling is not nearly detailed enough in terms of velocity/tone changes (and at only 160 Mb, what can we expect?). All the notes, whether played at soft or loud volume, have a similar character to my ears. The math engine might be outputting the correct interpolation volume between 0-127 midi values, but the tone or piano color does not change nearly enough as one plays up that velocity scale.

Secondly, the acoustic resonance modeling is not up to the standards of Pianoteq. The sounds and frequencies heard 'in between' the notes, like soundboard noise, sympathetic vibration between strings, damper resonance, etc.) does not sound as realistic as Pianoteq does IMO.

So, in other words, this unfortunately is not how to do my hypothetical perfect hybrid. For that to be pulled off, one would have to physically hand the equations of Pianoteq (which are no doubt closely guarded) over to a reputable sampling company like Synthogy, Galaxy, or True Pianos. Or, hand their beloved samples (which are also very proprietary) over to the Pianoteq team and have them somehow integrate the two. I have no idea how THAT would work, but this IMO is what needs to be done to further virtual pianos to the next level.

I tried heavily-handed to merge PianoOne and Pianoteq inside Reaper so that my keyboard triggered both. PianoOne provided a decent pianistic-attack sound and Pianoteq provided the beautiful resonance. I was partially successful, but the two libraries are too disparate to morph into my 'ultimate' piano. It needs to be done professionally at the software level. And the sample library needs to be in the Gigabyte range, not the Megabyte.

Thanks erichlof,
I was thinking that if "done right" the sample library size might well be quite small, so I had hope.

I have a minor issue with the hassle of lugging a computer over to the keyboard, plugging cables, starting programs every time I just want to play a little excerpt of something or try out an idea.
I know, I could dedicate another computer to it and one day I might, but the rack gear is already there and always on.
It may not be current state of any art, but probably no worse than most hotel foyer grands set in the worst acoustic space possible.
Every time I touch one of those I feel pretty good about my old rack gear

I have picked up several papers that indicate the math to be quite well known and not all THAT complex, but the computational load is claimed to be very high.

That suggests that MOST of the calculation needs to be done when the virtual piano is "built" and the rest done when it is "played". Is that too obvious ?
I usually think of wave tables at about this point, so I'll go off and practice before I get into that.