The Limits of Physical Modeling in Pianoteq?

Hi Igor,
Maybe I'm not looking in the right places but I've never seen anyone attempt to use the argument that Pianoteq's licensed models are an instrument in their own right, and should be treated as such.
That line might make sense a little bit for a model of generic piano sound but Pianoteq doesn't just model "German grand" or whatever, Modartt models in fine detail specific products.

I wouldn't pretend it sounds 100-percent real all the time that would be silly. Certainly no sampled pianos or modelled piano do that. They can however be extremely good.

If you have been here from version 1 you will have heard all the upgrades. You will know it keeps getting more realistic. It couldn't keep getting more realistic if it was 100% real in the first place. Hell there is plenty wrong with many or perhaps most or even arguably all stereo recordings of pianos - and surround sound doesn't correct this FWIW.

I suspect if Pianoteq wasn't constrained by commercial realities of having to limit CPU requirements to a wide variety of average real-world computers in our homes it could be even better. Even then if the base machine became a top of the range brand new i9, or it was being rendered on a top of the range GPU using the latest methods it may still have a way to go.

Little doubt people will still be having these conversations when Pianoteq 9 and 10 arrive.
Don't underestimate the power of confirmation bias as well.

Maybe one day they can add an offline rendering mode?

I will add that when I listen to the excellent demonstration tracks on the Modartt website the piano always strikes me as the most realistic instrument in the full mockups.
I find orchestral strings need much more work to get there than piano. I suppose this is inevitable given the controllers we have available.
Getting to a 100-percent convincing piano sound (at least compared to a stereo recording) is a relatively realistic goal, especially when they are so close now.

There remains something of a hint of it's own character to the sound that occasionally makes me recognise that I am listening to Pianoteq specifically, just as sometimes when I am watching a beautifully captured movie on TV that I recognise the type of digital cameras they probably used to record the footage.

I find that hint of its own character far less detracting than the somehow dead stitched together sound of sampled instruments.

Pianoteq evolution if's similar to video games and movies CGI evolution. Games was limited by the processor speed and used simple polygons graphics to create characters and sets. When processors got bettet they created graphics with more poligons or even more advanced ways to render 3D images.

  From the film Jurasic Park to today there was a very huge advance in processor's speed, but the result in realism wasn't exactly proportional. For example a double processor speed did not created double realism. Indeed it takes a lot more speed to the tiny few details that make CGI very realist. And even today it's not 100% perfect, since Peter Cushing CGI in Star Wars prequel wasn't perfect.

  Many people  don't understand pianoteq core engine it's almost a miracle. It's in a technology able to analyze the mathematic involved in all components of real piano physics, find redundant elements in the calculations, and resume the equations to a degree it will be able to render in real time. Without it one single bass note sound would take days to compute.

    If the average home PC was 20 times faster it could help get more realim (the few tiny details that some few people notice as missing), but only if the pianoteq engine was reworket to compute more variables. And it's nof just add more variables to compute, but also find out what variables and combination of variables are more important to get what humans perceive as a natural piano sound. It's a very delicate puzzle indeed.

    That's why some people, includig myself, think there could be nice to have two pianoteq engines, one for real time playing, and other adjust to compute much more variables, renderig the sound even than not in real time of taking 30x more time, and could be used in recording studios, like performing in real time and rendering the advanced version later.  Or perhaps even a intermediary engine, like activating a turbo set, for people witb i9 processor.

Yes, that's what I'm thinking about: maybe, instead of creating new models of pianos, guitars, etc., it makes sense to work on this?

The performance would not change.
Your performance is captured in midi data.

The midi data is then used by the Pianoteq engine which calculates the sound output based on your input data instructions to the mathematical model.

We are merely talking about the difference between our low latency playable instrument and an offline version that can do many times more mathematical calculations because it doesn't need to do so immediately (on the fly). 

Hopefully that makes sense?

well, jazz pianists are misinformed and snobs too, believe me.  at least i am

here's how i use the software, to give some idea of why i want this.  i play with ptq on one computer with my keyboard.  that's of course real time and effectively a live performance.  then i do post-production on another computer to produce an audio file.  this means editing, choosing a preset to render it with, tweaking effects, etc.  this preset is usually totally different than what i recorded it with.  usually i'm recording with a "player" preset so it feels like i'm actually at a piano, but then the track is a "recording" preset.  then there is the final render.  if this last step had a palladium-level option that would take all night and heat my house in the process, i'd be fine with that.

Pianoteq can be better than real pianos.
1) It can have mathematical perfection.
2) it is obviously superior to the pianos most of us can afford.

While I do believe the tonality of the given piano and the reverberation characteristics in the environment of the piano are things that contribute to how the musician plays nonetheless a fine performance can be captured and later manipulated whilst still keeping the quality of the performance.

Maybe you could think of it more like seeing your favourite movie originally recorded on 35mm film rendered at say 720p, 1080p  then 4K. The latter transfer takes longer to create but captures more of the detail. It's the same movie.

You would be watching the same movie on all three versions. The difference being the 4K transfer (all things being equal) should be the best. It should be by far the most realistic and most detailed, that is if you viewed it on a large enough screen.

Same deal with the entirely theoretical offline rendering of this mathematical model called Pianoteq.

Sorry, I guess it's not.

The comparison is made ceteris paribus. I mean, I'm not comparing an acoustic grand piano in a concert hall to a Pianoteq recording, but an acoustic grand piano recording to a Pianoteq recording. All my reasoning is based on this.

My easy solution is to de-tune that fine grand piano.  It suddenly sounds woody. like you're playing it closer up if the vol is sufficient..
I've never heard any acoustic piano close up that sounds anything like a concert recording.  Because it isn't, is it?  A concert recording is for the audience only and not the player!
Once one gets over this, you'll start to enjoy this product much much more!

Thank you for raising this very interesting topic! I want to share two observations:

First, theoretically, I suppose in one sense the "end game" of Physical Modeling could be objectively established -- there is a sound sample (like a Steinway D, or Grotrian...etc.) that is being physically modelled. The goal to have the sound waveform generated by the physical model to closely approximate the sound sample waveform. The "end game" will be achieved if the physical model waveform is identical and superimposable to that of the original sound sample. Now in reality, I doubt this could be achieved even with two separate sound samples of the same key of the same acoustic piano (because no two notes can be identical?) and the question is how close is close enough, or which kind of deviation our ears are more sensitive to and are likely to feel "synthetic".

Second, I also felt that with Pianoteq 8, the limits of physically modelling has been reached, but in an opposite sense. I felt that with Pianoteq 8, the sound as well as the playability is so close (in some ways superior) to an acoustic piano that I do not find anything lacking in playing Pianoteq as my primary piano. I also had experiences in the past after playing mediocre acoustic pianos thinking "what a frustrating experience, I wish I could be back home playing Pianoteq". I have very picky ears and I do recognize the "synthetic" nature of the piano (or more like "clinical" nature), but most of the time it is due to the Pianoteq model being too "perfectly tuned" or lack of reverb added. No acoustic pianos are like that (unless an acoustic piano is placed in an anechoic chamber). A couple of clicks in the "random" button or sliding the "instrument condition" slider a bit to the right, plus a high quality reverb provides for me a natural piano sound that I am not confident that I can consistently pick out from acoustic piano recordings in a blinded test.

My two cents...=)

Roger

It's not clear if you and your colleagues tried PTQ version 8 already.
Many people think it's a considerable improvement over v7.

(I find this topic very interesting and of great value as I have no experience of mechanical pianos for several years now)

First day on the forum - actually joined an hour ago - and find this to be a fascinating conversation. There are so many different thoughts and threads of thought, so I would like to throw in another perspective. Have not taken the time to fill in my profile yet, but the short version is that I have been playing piano for 60 years, am a Juilliard-trained composer, and MIT-trained acoustician. And I do own an old Mason Hamlin Grand piano that does sound and feel better than any digital anything, but from my perspective, that is not the point. The perspective I would like to throw in is the reality of logistical constraints.

I can not bring my Mason to gigs or to recording sessions or play it when my wife is sleeping. And even though I do own a pile of microphones and recording equipment the time to get it all set up and tweaked just right directly subtracts from my joy of being a musician and a composer.

The Pianoteq pianos, in terms of realistic return on investment, and I do not mean just money, I mean effort in and music out, is simply fantastic. I say this, having been perhaps the first designer of a digital piano with an integrated sound system in the early 1980s when we could only afford to put under a megabyte of ROM in the instrument.

The real issue for me is the piano is the most musically powerful of all instruments and I say this as a person who has professional musical  experience spanning the concert stage to farmers' markets. And since pianos are expensive, heavy, and get out of tune and I know I am not telling this group anything new, you all know this.

But as some of you have been mentioning, there are an infinite number of different musical-acoustical contexts that impact the way things sound. Yes, I agree with the consensus here, that Pianoteq sounds less real but is spectacularly expressive, in fact proably more so than a real piano. As a composer, musician, acoustician and electroacoustical engineer, I have spent decades, as many of you have, messing around with this stuff. But I still compose, record, and perform and do not want to spend all of my time being an engineer and moving man, and all of my money on conventional pianos because almost no one in the world can tell the difference. Pianoteq is good enough that the vast majority of the world can not really tell the difference and furthermore having a distinct sound is the desire of all musicians for if you sound like everyone else you can not make it anymore except in snob dominated circles but this is not where the action is. Most of the world is far more interested in authenticity of artistic expression than in technical details so while I applaud and admire all of you for expressing the diminishing returns with good points on every side being raised, I have decided, now that I am 70, I want to spend the rest of my time being a musician, and a composer, more than an intellectual and for my purposes Pianoteq really hits it out of the park, and yes it is getting better and better.

So, no disrespect intended and I love this forum's conversation, and the intellect of everyone expressing themselves, but hey we do have a miracle on our hands here. We get more than 90% of an excellent piano in a far more convenient and affordable package. And I, for one am totally delighted even though I am as calibrated to excellence as most of you, and by the way, after a decade at Bose I became in charge of sound and music worldwide at Apple Computer, now the world's largest sound and music company. And in fact, the world's most valuable company and the reason is sound-centric computing.

The real world makes tradeoffs between art and science and money and in this regard, I am totally impressed with Pianoteq and the community it has spawned.

So keep up disecting the details, but I am thankful and appreciative for the hours I spend with Pianoteq and have performed dozens of times in
the last two years and not a single person in a single audience ever said anything other than wow that's amazing.

Nice to meet all of you.

I also think Pianoteq 8 is great, amazing even. I just think there's an upper limit of how much any hand-tuned program with so many inter-dependent variables can possibly reproduce such a complex signal. Breaking things down to their base components and combining them is in some sense the opposite of what the machine learning revolution has done, and while, as a programmer and an amateur piano player I have much respect for Pianoteq, I stand by what I wrote: it may get 90% of the way there to sounding like a real piano to most people, but that final 10% is where things get really nasty and I simply think it's probably too complex for a human-written software to achieve. I don't say "definitely", but if we do get an algorithmic solution to the problem of realistic sound synthesis, it would be as a consequence of reverse-engineering the machine learning solution, rather than building it from the ground up like Pianoteq has done. Or perhaps a combination of both top-down and bottom-up, who knows. But, I do fully expect to see a competitor enter this space using pure end-to-end machine learning which uses GANs. GANs are like a chicken vs fox game of evolution. You have a discriminator that evaluates if something passes as the real thing, which gets better at saying yes/no as the generator gets better at trying to fool it. Eventually the generator gets so good the discriminator can no longer tell the difference between a real piano recording and the synthetic output of the model. That's when you ask "can humans tell either?" There's a good chance they won't be able to, and that's the end game. But, of course, AI art hasn't quite reached that level yet in many areas. Still, entire professions like radiology are being replaced by fully-automated machine learning tools because it's more accurate and makes less mistakes than humans do.

I can tell the difference between Pianoteq and a real piano recording, it's very close now, but every increment in Pianoteq software maybe gets a teeny tiny bit better and we are perhaps at the point of inflection where to pass a critical threshold of realism, ML is needed, if only to help tune the parameters for the model, but perhaps to replace it entirely with something that contains the essence of what a piano does when it generates music.

To be perfectly honest, I'm kind of surprised such software doesn't exist yet. Maybe just in a lab in academia. I should ask my friend who's got a Ph'D in ML-based music synthesis what the current state of the art is. I'm almost certain his response won't be Pianoteq, sorry. I say this as a huge fan of PTQ, I think it will be superseded at some point, possibly soon, who knows.

Sound is hard to pin down and every end-user has a different setup. When I hear a live acoustic piano sound, I basically hear the pitch speed, duration, and volume. This is what I hear as a timbre. This is accompanied by the mechanics, the condition, the room, and the feedback from the piano.

I think Pianoteq has created such a high level of physical modeling of a piano, but I still believe mathematically creating a piano is limitless and subjected to those creating it. Considering Pianoteq, Roland, Piano V, MetaPiano, how many ways are there to mathematically recreate a sound?

Hi Igor, this is a very interesting topic.

In this year I was inveted to play with a orchestra the Grieg's piano concerto, but the theater did not have a piano and neither a sponsored to rent a good one, so I dicided to take my casio px-160 and used the pianoteq (Bluthner model one) to the concert.

Spoiler alert, I really liked, besides I have many problems with the mic, but we can have a good ideia how well the real orchestra mixed with pianoteq. We still working on videos and audio of the concert and if the final mix will be kind of ok I can post over here for the opinions of this great forum.

Since I am not a professional pianist and took me 20 years from my last recital because my change of life (today I am run my own hotel in the Brazilian mountains), I think it will be a very interesting way how well (or not) the real orchestra worked with pianoteq.

My plan is to make a little documentary (around 10 minutes) about this premier because I think it is the first concert made in Latin America, or maybe in America with a real orchestra and a piano software.

Anyway, I really would like the opinion from all of you and I am open for any ideias that I can improve the recording. I think maybe around December 20th the documentary it will be ready.

My best wishes.

Thank you Dazric, but I will not compete this year.

This contest is very nice and always has a lot good people with nice ideas and great talent and for me this contest was great, especially to make music when I was complete out of musical thing. But I am focus more about the premier here in America to be the first piano concert with pianoteq with the real orchestra and besides the concert was November 10th I used the version 7.5.4 at the moment. I would not be legitimate to participate anyway.
Thank you once again and I hope you like the the video. I will be very open to any comments to improve any other presentation that wil come in the future.

Best regards.

The problem we have is Pianoteq is too perfect and acoustic pianos are not perfect. There are many overtones, timbres, register differences in particular in the older instruments, and all kinds of other things occurring with the acoustics that can affect the sound. Play a real piano on a warm humid day and the piano sounds sweet and mellow, but play that same piano on a dry winter day and the piano may sound bright and harsh. Real pianos are made up of felt, copper, steel, plastics, and all kinds of glues that all interact with each other. Older instruments have rattles, buzzes, clicks, and all kinds of other sounds not made up of the ones mentioned above. Pianos also interact with their environment and can set other things rattling and buzzing away. I've had lightbulb filaments vibrate when playing certain notes on my acoustic piano. That was one sound that was difficult to find! Another time I encountered an old clock with a chime mechanism. When I played the E near Middle C, the clock's chime rang out loudly. I spent hours hunting that one down.

A slightly out of tune piano, one that's not quite freshly tuned, also has a warmer sound to it than a perfectly tuned piano. I noticed that recently when my piano tuner tuned my grand. I felt he made the instrument sound cold and distant rather than welcoming. The weather wasn't cold and dry either, so I can't say it was that. The piano sounded unfriendly rather than welcoming to play.

With Pianoteq, we see an amazing product that attempts to model these various aspects digitally but the problem with the digital world is there's only so much data that can be captured. Our sampling is a tiny subset of a larger infinite set of sounds out there in the real world of acoustic instruments. We can adjust the parameters to a certain extent to mimic the real instruments, but we can only do so much because our computer hardware, while still being the fastest ever today for personal use, still can't capture every bit of the infinite sound pallet found in the real world. To paraphrase what was said in the beginning of the thread, who knows what kind of computers we'll have on the desktop to use. (Note: Desktop here refers to personal computer regardless of being a Mac, Linux, tower or laptop).

With that said, I treat Pianoteq as it is. I know it's a digital representation of the best pianos in the world. I mean, how many people out there can wander from a New York Steinway D one day or within the same hour to a Petroff concert grand all without spending more than a couple of hundred dollars or euros? This is a virtue of the digital world. How many of us would dare detune one of these pianos, or play with the hammers and adjust the dampers?

For me Pianoteq is a compromise I need to make because I can't always play my real grand piano. I have an elderly father living with me who sleeps a lot more now during the day, and I sure as heck don't want to disturb him. With Pianoteq, I can plug in my headphones into my Roland and practice and play away on the best grands in the world without bothering anyone else.

Physical Modeling 101: Physical Modeling Explored
https://youtu.be/JklN1yqb6No