Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards Modartt releases ANT. PETROF 275 concert grand for Pianoteq

In another Pianoteq thread somebody posted a Modartt patent. I've shared my understanding of the patent there and I will repeat that here again. I might be wrong, but anyway. It seems full modeling where the piano is represented as a finite set of differential equations describing each of its physical characteristics and solving that in real time isn't possible with current computing power. Which is why the piano sound in Pianoteq is rather "synthesized" by summing up sine waves representing the partials (overtones). Since the piano sound is not static one but has an attack followed by a sustain, they use tables with how the partial amplitudes change with time (the sustain portion of the sound). They use prerecorded attacks that are superimposed over the synthesized portion. And the resonance stuff is modeled by exciting the partials. In order to prepare the tables with the partial amplitudes and how they change with the sustain, they use recorded sound of a real piano that has been scanned/analyzed in their labs. Once again, that's my understanding. With that in mind, it makes sense to call this approach rather "data & modeling" because they start with some samples of the real piano which they analyze in order to prepare the tables and also include prerecorded attack sound.

That's definitely different from how I originally thought Pianoteq worked. So there are samples after all (though probably just a fraction of a second long I imagine).

Although the patent clearly stated that the attack is sampled it’s not necessarily true they followed that patent entirely or that’s the only patent they have. So they might have discovered a way to also model/synthesize the attack sound as well.

Yep, I was under the impression that they had a separate set of algorithms to model the attack against the reference piano.

If there are samples shipped with the product, I imagine they'd be easily discoverable in a package teardown.

I read some years ago that for action noises, there were samples.

According to their FAQ page:

Do I need a dedicated disk drive?
No. Since Pianoteq does not store any samples on the computer, there is no need to invest in a dedicated disk drive. Pianoteq occupies just 40 MB whereas other virtual pianos need to frequently access many sample files (usually several gigabytes) from a fast and stable disk drive.

If they use any samples at all, are they not effectively using false advertising on their FAQ page?

It should be kept in mind that sampling in order to obtain volume levels for the attack is not the same thing as actually replicating a sampled sound. If they do this, they are simply using measured data to provide as parameters in the model, and thus their approach is still full modelling (but it could be argued that it is not strictly "physical modelling", but rather a combination of physical and statistical modelling).

So essentially, there is no algorithm calculating the attack volume levels?

Yeah, I am a Pianoteq fan but I do have to agree that the Petrof Bach recording sounds rather ... weird.

In what way exactly? I pulled the gun and bought it last night (it's the easiest way to creep in a piano in the house without being interrogated by my CFO) :-)....I actually found it rather nice sounding...but bear in mind I am digitally bred through and through in terms of playing on a piano and the only time I played a non digital piano is during those times when I am passing by one in the odd tube station in London (and when doing an ABRSM exam & that's it) - so I am totally agnostic to real piano sounds... :-(

*chuckles*

Haha, I can believe that

Very difficult to put into words, I am not sure whether it is a problem with PTQs realism per se or whether the player perspective is just not ideal for a piece like that.

It sounds too direct, way to aggressive and the notes blend into each other quite eerilie and it sounds like the piano is stuck in a tin can.
80s MIDI sound really is an apt description because it does remind me of these ancient soundtracks a bit.

Pianoteq Petrof Bach recording - direct link
Compare these to the above, maybe you will be able to hear what I can't express very well. (seriously, how does one convey sonic impressions in words? >.< )





PS: I'm not Piano connaisseur by any means. I only get to dabble on a grand when I have my lessons.

I don't know the answer to that question; my elucidation was of a general nature based on what CyberGene (who seems to have insight into how Pianoteq actually works) wrote.

Thanks Granyala - just heard both, the YT links sound natural. Aside the fact that the PT link is a from VST I also feel it is really in your face in terms of gain - I had to lower the volume to listen to it throughout..

Personally, I liked the warm setting of the Petrof because it doesn't sound as high as the others I have in PT (namely the D and K2) but that could be because of the way I've only played on digitals and my ears are tune to that...

I was commenting on a Modartt patent I've read. Whether they use it entirely in the production versions of Pianoteq is not clear. I doubt it anyone would know for sure besides Modartt themselves.

What I am really convinced though is that *real* modeling with differential equations isn't feasible with current computing power and whatever Pianoteq have done is rather synthesis. They've modeled resonances into that synthesis. Regarding the attacks, those might be real samples or might be just a percussive sound interleaved with the synthesis.

In my own opinion Pianoteq hasn't changed its inherent character throughout all those years. They are refining here and there and make it even more realistic sounding which is really commendable however the inherent synth-like quality to the sound has always been there. Some can live with it, others are not bothered and there are people who aren't able to even hear it. Honestly, I have lost faith in Pianoteq or any other piano that claims to be physically modeled just because the current computing power won't allow them to do proper modeling. I am not even sure we will ever see proper modeling soon, so it's not Pianoteq to blame or Roland, or Physis. They've actually made pretty convincing pianos that run on modest hardware. In any case, my own, personal and probably very biased taste leans much more towards sampled pianos.

I agree with your assessment there, but I'm not unwilling to consider what they do "physical modelling", although maybe a more useful term to use would be "simulation", since they (I guess) simulate the motion of various parts of the piano (notably the strings and soundboard).

It seems plausible to me that they have been cutting corners in order to keep the CPU requirements low, and that the sound could be improved considerably, if they didn't cut those corners. Maybe they plan to gradually improve the sound over the years by allowing more complex calculations, as CPUs gain more processing power.

The beauty of what these guys at Moddart and elsewhere are doing, is maximising minimum resources (ours, ie those of us who don't have high CPUs and aren't likely to change the habits of a lifetime anytime soon) which will entail, I would think, a little bit o' cheating along the way. One has to cut these chaps a bit o' slack, you know. And, given time, and expertise, the overall picture will develop well within minimum user resources.
The advantages of such thinking are simply many more customers; the Pianoteq product is innovating and not too difficult to get your head around, which will bring in older customers, too.
They'd be crazy to change such a successful formula, but, as part of future development, and to allay such criticism as there always is, a Heavy Version may be brought out. . . . . with a price to match.
I hope they don't. It'd muddy the waters.

After reading all this I decided to give Pianoteq a try once again.

I installed the demo yesterday.

I didn't much like the Petrof mentioned above, but that's not surprising.
I lean more toward the Steinway sound, with a secondary admiration for the Kawai sound.

So I tried their Steinway.
It, and a bunch of others (Petrof, Steingraeber, Grotrian) are built into the demo pack, so it's easy to try them all.

It's still a no go for me.

I did like the "cleanliness" of the tones.
But in all other respects I still find that it just doesn't sound right.
The simulated/imitation aspects come across very strongly, and it just doesn't sound like or feel like a piano.

Play the Blüthner.

These things take time; stay with it, play with it. I never liked acoustics or Pianoteq much until I bought a Roland; then I started to like the acoustics and all their peripheral issues and then Pteq became the obvious choice, since I already had it . . .

A couple of years ago, as a side project, I built my own modelled piano. I read the latest literature, implemented the equations and built a simple AudioUnits plugin of Mac. It was not a pretty sound, but it was a start and gave me more of an understanding of what's required. The main thing I learnt was that the string attack and sustain waves can be well modelled, but what is difficult to capture is the interaction with the soundboard. The most realistic-sounding models were those that took an impulse response of a piano soundboard (literally by hitting the bridge with a hammer!) and convolved that with what was being generated by the purely physically modelled string equations. I would consider this a data/modelling coupling. Maybe Modart needs to do similarly. Just speculation...

I'm thinking I should put what I have on GitHub and turn this into a community project. It would be great to have an open-source physical piano model.