What still stands between Pianoteq and a professional recording?

I think what stands between Pianoteq and absolute realism is the fact that it has to be rendered in real time, and notably on a multitude of different budget CPUs and platforms.
Just as what stands between any sample instruments and absolute realism is the baked in pre-recorded sound of the limited samples.
As the CPU base line gets better so the Pianoteq engine will benefit from continuous simultaneous tweaking for further refinement (within a given version of the physical model) and the greater complexity of the latest model.

Ultimately Pianoteq is an incredible simulation, which in part must be economic in a smart way, IOW throwing away or not attempting to replicate the least important, or almost inaudible, or masked sonic details in order to work in real time on a normal consumer CPU.

I don't suppose they would ever tell us if they have a no-holds-barred version of the physical model that's only works on high-end CPUs.
Also I  think an offline rendering mode would be frustrating for ordinary users (why can't it sound this good when I play it normally, blah blah..) and only satisfying for geeks.

So perhaps we can take it as a testament to how skillful their work is that you now find yourself asking these questions, expecting that elusive nth degree of realism to be cracked.

Your recordings of your personally tweaked piano presets sound fine, and like numerous others are good enough to fool many people I suspect.

Actually I don't think so, at least not immediately, not for quite some years. It's already better than the vast majority of sampled pianos. It's more a matter of faith, or prejudice, or expectation bias.

Certainly the better the programs get, and the closer the sounds get tonally to the real thing the more people will shift to physical modelling from cumbersome sample libraries.
The biggest problem with virtual instruments to me is the move by some to completely programming in the piano roll rather than playing the virtual instruments.

Regarding any instruments that have been synthesized I would rather hear the odd hint or betrayal of synthetic tone with an expressive human performance than a completely programmed piece, for the most part. At least that's what I believe at the moment.

Who knows at some point the listeners may even prefer a slightly artificial recreated tone over the original instruments since all the imperfections have not been captured - "Even better than the real thing" as it goes.
No doubt that happens already as people are dissatisfied with their home recording efforts of real grand pianos.
Certainly hardware reverbs have that appeal for many people over real hall impulse responses.

Try adjusting in the [Design] menu:
[Impedance] down.
[Q Factor] up.
to suit your taste, and maybe:
[Energy] to about: 0.01; 0.02, etc.

My 2 cents: I'm lucky enough to own a real grand piano (a 6' Gunther, from 1951) in the same room as my monitors for Pianoteq. Switching from one to the other is easy and fast. I tried several times to emulate my Gunther in PT, with some interesting results but still far from perfection. But I have just one thing to say: use the "condition" slider... A real piano is NEVER in perfect condition !

Sorry, one more thing: use big monitors, a real piano delivers an impressive bunch of dB's !!!

I can still hear what I am talking about in that recording. It's a digital/metallic ring, and it honestly kind of hurts my ears. You can hear it pretty clearly in the three notes of the melody at 1:37 in that video, and the sound carries. It's a metallic sound, like the piano string was hit with a metal hammer rather than a felt one. I can listen to the real recording at a much higher volume without discomfort to my ears.

You should do a side by side comparison of the Pianoteq 6 recording and the real recording of Lugansky, and try to pay attention to exactly *what characteristics are different.* I definitely notice that the real Steinway is far more resonant itself, and it's not the resonance of the room alone because the nature of the sound indicates that the phase shift of the resonance is very close in time to the original tone, with much lower, more harmonic overtones making up that resonance, which is what makes the piano sound big.

I really think most of the issue stems from how they simulate the hammer interacting with the string. There is a bit of a harshness that comes through in Pianoteq. And, at the end of the day, we're trying to simulate something that is high quality when it's tranquil and resonant, meaning that we're dealing with mostly low level, harmonic overtones making up most of the sound. So, I think that it should be computationally efficient.

The next issue is that that the way that the sympathetic resonance evolves in Pianoteq needs more focus, so that the correct tones resonate a the right volume to make the piano sound large, clear and clean.

This isn't a simple "here's a quick fix" type of post; I've spent hundreds of hours 60-70% of my days since march last year adjusting this, tuning every parameter, adjusting every parameter, sometimes editing the individual overtones of notes, and before that I spent hundreds of hours tweaking the default Steinway in Logic Pro to try to make that sound better, as well as the harpsichord. And before that, I was always very nitpicky about the characteristics of pianos, including nice Steinways, and payed attention to which recordings had the pianos I liked the most, and that's something that has persisted for the better part of 15 years now.

What I'm doing in this post is trying to offer that experience and perspective as feedback to the engineers so that they have an idea of what to focus on. I'm not trying to say that I'm some master at it, especially since I've never done piano maintenance or building.

Edit: also majored in physics.

This is not a challenge but merely a question. Do you hear a metallic harshness in the following recording?

This is an E-piano Competition MIDI file, originally performed and recorded (to MIDI) by Konstantin Krasnitsky, of Rachmaninov's famous C-sharp minor Prelude, which includes a wide dynamic range including fortississimo, rendered and exported to audio by Pianoteq 7.5.2 using the NY Steinway D Classical preset. The only adjustment I made in Pianoteq was to move the right control point of the default linear velocity-response from 127 to 117.

https://imgur.com/11gwSFd.png

When I'm playing I often move the right control point down to 115 or 110 for some Pianoteq presets, because my MIDI keyboard is a little "hot," registering forte velocities at only mezzo finger pressure.

https://forum.modartt.com/uploads.php?f...nitsky.mp3

Do you hear anything objectionable in the sound of the virtual piano in that recording? Would you make any suggestions regarding altering that sound?
--

I like the simplicity of this suggestion! It reminds me of an idea that somebody proposed a while ago: how about a simple 'tone control' slider in addition to the 'condition' slider? I think it would be appreciated by a lot of users.

When we said non real time computing, we don't refer about the same algorithms, but about an Idea to create new algorithm with more precision to the point real time would not be enough, as the algoritms would be too heavy.

Pianoteq Creation was based in a reasearch about mathematic equation reduction/summarize, of physical model. It try to get all redundant elements to find a shortcut in the computation. Without it a 100% pure calculation of everthing would take a week.

I can hear some of what called harshness in the link to V6 of Op 23 NO5, but it's reduced in pianoteq 7.5, and some real brands also have a similar of it. The video shows velocity curves reaching FFF many times, when in reality FFF it's not so easy to reach.

I wonder if would be a good idea to have a 4th adjust for hammer hardness for pianoteq, or just to work limit a bit it for FF and FFF.

I'm saying that you can likely create more efficient, more realistic algorithms, and that refinement of the existing algorithms should focus on a deeper, more robust sympathetic resonance and a more careful voicing of inharmonicity as it relates to having a metallic vs. woody sound, as well as inharmonicity/harmonicity as it relates to hammer hardness. A more robust treatment of hammer shaping/hardness probably should also be available. I.e. it seems to me that there's more to hammer characteristics that could be added to and refined. I.e. a particular felt and a particular leather might have the same "hardness," but the leather may sound a little bit differently, and perhaps felt would further reduce imparting higher tones than leather.

There's still room for improvement within the current level of computing power. I'll give you an example of something that utilizes clever algorithms to bypass computational limitation: Unreal Engine 5's Nanite system, which basically allows relatively modest GPUs to do real-time rendering of assets with trillions of polygons by efficiently computing what to render. Interestingly, UE5 does something really cool with sound too: it simulates the reverberation of sound as it hits the outer ear, which creates a far more realistic sense of depth. We not only have stereoscopic ears; our ears actually transform the sound due to their shape, which our brains know how to decode into spatial information.

Like, you might think about how since the piano is a continuous shape, you'll have somewhat complicated wavefront as it comes to your ear, especially from the resonating of the wood. But, you could compute a lot of this sort of thing analytically first; that is, on pencil and paper, write out the solutions to differential equations that would give you a transformation to create this effect so that the program itself doesn't have to do so much work.

@Opus32 - Very interesting comments and perspective.

look at this :

https://youtu.be/wb12V-iYzEQ

Original :

The PPP to FFF are better on this original : (perhaps it was a problem with Pianoteq Velocity curve with extracted MIDI Piano ROLL)

https://www.youtube.com/watch?v=327D03P5Xxc

Opus 32 wrote:

”We not only have stereoscopic ears; our ears actually transform the sound due to their shape, which our brains know how to decode into spatial information”.
Bravo!

Interesting this with the shape of our ears due to their shapes.(and ear canals and earflaps and the size of the head and the shape of the head too)

I have been thinking about this in a thread 2018…..I said that the sound is different in different ear canals, and earflaps, and the size and shape of the head have effect on how one feel it sounds…and room acustics and so on…..
Even if two people are standing beside a real piano where the sound come to both ears, maybe not at the same time, and with that rooms acustics, not at same time to both ears - it sounds different for both. We hear different.

With speaker listening I have the speakers positioned so that their angle to  me is 60 degrees. An equilateral triangle is formed. And In order for me to get the right sound image, I try to sit in a fairly well-defined area, symmetrical to the speakers. But I I’m not satisfied.
But with my headphones I’m satisfied with Pianoteq sound (but my ears are old and have tinnitus, noises in both of  my ears).

Maybe binaural in future can reproduce the live experience of sitting at a real grand piano and playing it as seated on the bench so everyone is happy with the sound, regardless of how the ears and the shape of the heads are   

Well, that’s what I’m thinking

Best wishes,

Stig

btw, It’s sometimes easy to say in blind test which one is Ptq and which sampled, not because of what Ptq is missing but because of what sampled don’t have