Topic: Very high-end desktop and Piano modelling

If one could create a very high-end desktop - something along the line of:

Core i7-7700k - liquid cooled and over-clocked if needed
16GB or more memory
PCI Express high-end audio card (e.g. RME HDSPe or similar)

Would it be possible to implement 'piano modelling algorithms' that can take advantage of such compute capacity for even much-more realistic sound? It may be reasonable to assume that this would be paired with a high-end 'audiophile' speaker system or a high-end headphone.

There may be market for such boxes if it could get us closer to *the* sound.

Just to clarify: I love Pianoteq 6 sound and play it everyday - so not criticizing the current sound - just wondering what 'more' can be done with extra compute capacity.

Thanks,
Osho

Re: Very high-end desktop and Piano modelling

I recall from somewhere up on the Modartt site (I think a set of video interviews with Philippe) that the original Pianoteq models took extraordinary lengths of time to calculate notes. The inference I took from that was that the model had to be quite simplified in order to be able to function in real time as a playable instrument. Sometimes I wonder if the upgrades from major version to major version aren't partially the advances in computing power allowing Philippe to add  one or two more elements of the original/complex theoretical model into the shipping  product.


In this video, Philippe talks about simplifying the model so that it runs in near real-time rather than 2 days: https://www.youtube.com/watch?v=DcztjEg7vHg

Last edited by tfort (19-09-2017 17:42)

Re: Very high-end desktop and Piano modelling

Yeah. This is a lot more interesting than all this Raspberry Pi nonsense. ;^)

Greg.

Re: Very high-end desktop and Piano modelling

skip wrote:

Yeah. This is a lot more interesting than all this Raspberry Pi nonsense. ;^)

Agreed. There are plenty of compromised piano sounds already!

Osho

Re: Very high-end desktop and Piano modelling

It was proposed years ago. It was also proposed that the audio export function, that do not need to be real time, could render the sound using a lot more complex algorithms able to creat 100% perfect sounds.
But I presume it would require redesign the software, the algorithms from scratch, almost like create a new software.  Also, a perfecter sound generation rivaling with a real time sound would stimulate people to spot more deffects in the real time version, years ago when the sound engine wasn't so refined as today in V6.

The initial concept of pianoteq was to create a modelled pianosoftwate that people could run in a medium capability laptop. If they push the CPU requirements to a top i7, many people would not buy.

oshogg wrote:

Would it be possible to implement 'piano modelling algorithms' that can take advantage of such compute capacity for even much-more realistic sound? It may be reasonable to assume that this would be paired with a high-end 'audiophile' speaker system or a high-end headphone.

Osho

Last edited by Beto-Music (19-09-2017 15:37)

Re: Very high-end desktop and Piano modelling

If there were to be advanced model that required a computer with multiple cpu's and huge amounts of RAM, there would be people interested--recording studios, clubs, concert halls, as well as a few people who just be willing to make the investment or out of curiosity build the computer.

Last edited by Jake Johnson (20-09-2017 14:21)

Re: Very high-end desktop and Piano modelling

Yes, and the less resource-hungry versions would of course still be available.

Greg.

Re: Very high-end desktop and Piano modelling

skip wrote:

Yes, and the less resource-hungry versions would of course still be available.

Exactly. Those who want the high-end-spec version will need to pay extra. This could be combined with much better audio export in non-real-time to make it a good deal for recording studio, prosumer and 'enthusiasts'.

As Apple has shown, real money is going after high end and not the low end.

Osho

Last edited by oshogg (20-09-2017 04:18)

Re: Very high-end desktop and Piano modelling

That video was posted in 2009 and might be even older.  There have been advances in computing that aren't just the main CPU since then. Nvidia for one allows for multiple GPU's (your video card) together in a workstation "supercomputer" that can be used for complex math and other problems - like deep learning/neural network processing.  GPU-Accelerated Computing: http://www.nvidia.com/object/what-is-gpu-computing.html
Even back in 2009, SOS magazine was writing about CUDA-accelerated convolution reverb plug-ins.  And at that, you don't need 12 GPU's just to do heavy lifting, a single or dual GPU common for gaming PCs would almost certainly give a boost to the physical modeling department.

I would bet that such a unit could be used to make a much more complex algorithm run in real-time.  That being said, I'd be very curious in hearing what might be considered the "perfect algorithm's sound" compared to the PianoTeq sound running on a reasonably priced 2017 PC.  I'd also love to see how many could pick out the nuances or even tell the difference at all.  Maybe sans reverb one could find some differences, but that wouldn't be a practical use for it, so if with reverb you could barely tell, then - why do we need a more high-rez algorithm today?

This article in 2013 described some research into physical modeling on a GPU - http://queue.acm.org/detail.cfm?id=2484010.

Re: Very high-end desktop and Piano modelling

Actually, now that I think about the video - I'm not so sure that a "simplified model" is a worse model.  By injecting the pre-calculated piano knowledge into the physical modeling, he's probably taking away the need to re-calculate 99.6% of the model each time around.  Instead of having to make a dumb physical modeler that could make the sound of anything at all from scratch, this one only models a piano.  So you can't use the Pianoteq engine to make violin or guitar sounds, it just wouldn't work.  I assume that all of the components as well as the integration of the components can be pre-calculated and saved in a format that will massively accelerate the process.  If you consider the size of the install, it's probably 90% pre-calc data that's compressed efficiently.

The only analogy I can think of is this -- it's like using a hammer to nail stuff or a nail gun.  The nail gun is only good for quickly putting nails into wood, but does it amazingly well -- by injecting the knowledge of a good carpenter using a hammer perfectly. Whereas the hammer is not so great at efficiently putting nails into wood - it takes much longer to do it, but can also be used to break up ice or do demolition work, or hit someone in the head. :-)

Re: Very high-end desktop and Piano modelling

dstorfer wrote:

That video was posted in 2009 and might be even older.  There have been advances in computing that aren't just the main CPU since then. Nvidia for one allows for multiple GPU's (your video card) together in a workstation "supercomputer" that can be used for complex math and other problems - like deep learning/neural network processing.  GPU-Accelerated Computing: http://www.nvidia.com/object/what-is-gpu-computing.html
Even back in 2009, SOS magazine was writing about CUDA-accelerated convolution reverb plug-ins.  And at that, you don't need 12 GPU's just to do heavy lifting, a single or dual GPU common for gaming PCs would almost certainly give a boost to the physical modeling department.

I would bet that such a unit could be used to make a much more complex algorithm run in real-time.  That being said, I'd be very curious in hearing what might be considered the "perfect algorithm's sound" compared to the PianoTeq sound running on a reasonably priced 2017 PC.  I'd also love to see how many could pick out the nuances or even tell the difference at all.  Maybe sans reverb one could find some differences, but that wouldn't be a practical use for it, so if with reverb you could barely tell, then - why do we need a more high-rez algorithm today?

This article in 2013 described some research into physical modeling on a GPU - http://queue.acm.org/detail.cfm?id=2484010.

The problem with using GPUs for audio processing is added latency. It's definitely not good enough for serial processes like calculating how an instrument can sound. It sorta works with convolution because FFT bins can be calculated in parallel. Still, you cannot avoid added latency (10 ms for processing a 512 sample buffer is a lot - which means you have 20 ms latency total - 10 ms to process it, 10 ms to play it back, and this is literally unacceptable for realtime playback - also this is without accounting for round-trip latency of the system).

In other news, most native convolvers are ZERO latency these days...

Last edited by EvilDragon (20-09-2017 08:06)
Hard work and guts!

Re: Very high-end desktop and Piano modelling

dstorfer wrote:

[...]That being said, I'd be very curious in hearing what might be considered the "perfect algorithm's sound" compared to the PianoTeq sound running on a reasonably priced 2017 PC.  I'd also love to see how many could pick out the nuances or even tell the difference at all.  Maybe sans reverb one could find some differences, but that wouldn't be a practical use for it, so if with reverb you could barely tell, then - why do we need a more high-rez algorithm today?

+1

Make it more expensive and let it eat more hardware power, so people will think it's more valuable.

Pianoteq is perfect. It is an elegant piece of software, making all piano samples dispensable while giving you incredible control. It sounds perfect and better than many real instruments. I'm always looking forward to new piano (or other percussive instrument) models or refinements of the algorithms or new ideas, but, really, there is plainly no need for any fundamental increase of computing effort.
My point of view, remember, you don't have to agree

Pianoteq Pro 8.0.0, Organteq 1.6.5, MacBook Pro 16" i9, Mac OS X 13.0.1, Universal Audio Volt 4, Logic Pro X 10.7.5, FM8, Absynth 5, The Saxophones/Clarinets, Reaktor 6 and others

Re: Very high-end desktop and Piano modelling

Jope wrote:
dstorfer wrote:

[...]That being said, I'd be very curious in hearing what might be considered the "perfect algorithm's sound" compared to the PianoTeq sound running on a reasonably priced 2017 PC.  I'd also love to see how many could pick out the nuances or even tell the difference at all.  Maybe sans reverb one could find some differences, but that wouldn't be a practical use for it, so if with reverb you could barely tell, then - why do we need a more high-rez algorithm today?

+1

Make it more expensive and let it eat more hardware power, so people will think it's more valuable.

Pianoteq is perfect. It is an elegant piece of software, making all piano samples dispensable while giving you incredible control. It sounds perfect and better than many real instruments. I'm always looking forward to new piano (or other percussive instrument) models or refinements of the algorithms or new ideas, but, really, there is plainly no need for any fundamental increase of computing effort.
My point of view, remember, you don't have to agree

With all respect, I definitely disagree. While Pianoteq is great and I use it everyday, it certainly is not 'perfect' IMHO. If it was 'perfect', there would not be any more release of Pianoteq and Moddart is probably already working very hard at developing Pianoteq 7.

Pianoteq is a complicated software model of an intricate physical process - there are 'optimizations' and 'approximations' to make sure that it works well on a typical low to mid-end system. And, if there was an option for those who have high-end computing equipment (and who are willing to pay for that option) - that would be another step closer to the real piano sound.

Thanks,
Osho

Re: Very high-end desktop and Piano modelling

Would more complex algorithms really mean more realistic sound?

The inherent limitations of Pianoteq could lie in the fact that models will always be approximations of pianos. And some features can't be efficiently modeled. So there will probably a level of complexity that if we go above...

I view it as that, is it worth to spend 2X more resources for 5% more accuracy... with an extra feature which could be run on only the 10% fastest computers?

I would believe that it could be more efficient to sample some parameters... what dstorfer term as pre-calculation could be substituted by the word sample (because samples are just that in one way or another). Even CPUs have instructions which are used instead of just calculating from scratch.

I might be wrong, but if companies use as standard an ARM device as reference, they'd be sure that the software will universally run in every computer. This is a better option than comming up with hyper complex softwares which are sucking all the resources and have features that most computers won't efficiently run.

I view those softwares as dinosaurs..., why would Modartt imitate resource sucking dinosaurs  and run at risk of becoming extinct. :b

Last edited by Lucy (21-09-2017 00:51)

Re: Very high-end desktop and Piano modelling

dstorfer wrote:

Maybe sans reverb one could find some differences, but that wouldn't be a practical use for it, so if with reverb you could barely tell, then - why do we need a more high-rez algorithm today?

What exactly do you consider to be "sans reverb"? For example, do you consider the following (real) recording to be without reverb, or with reverb?
https://youtu.be/jGGwwDuO9vs 

Is "without reverb" an anechoic chamber?

Greg.

Last edited by skip (21-09-2017 00:55)

Re: Very high-end desktop and Piano modelling

Pianoteq works according a technology of mathematically reduce/ equations based in the natural behavior of piano ellements. Without th ability to resume the equations a computer would take weeks instead os micro seconds to produce a grand piano note calculated according the laws of physiscs & acoustic.

But the reduce of equations it's not perfect, it take many things out, like digital video compression take details out from HD image. To get a natural sound as possible the delicate point is to know what remove and what left, in thounsands of elements. This take time, and that's why pianoteq got better and better for quite a long time with almost no CPU requirements increase. Only after V3, in V4, the CPU needs got more hungry.

Pianoteq don't need to be 100% complete perfect, but just enough close to perfection to convince nearly everyone. Our ears can't distinguish perfection from something really very close to it.


Lucy wrote:

Would more complex algorithms really mean more realistic sound?

The inherent limitations of Pianoteq could lie in the fact that models will always be approximations of pianos. And some features can't be efficiently modeled. So there will probably a level of complexity that if we go above...

I view it as that, is it worth to spend 2X more resources for 5% more accuracy... with an extra feature which could be run on only the 10% fastest computers?

I would believe that it could be more efficient to sample some parameters... what dstorfer term as pre-calculation could be substituted by the word sample (because samples are just that in one way or another). Even CPUs have instructions which are used instead of just calculating from scratch.

I might be wrong, but if companies use as standard an ARM device as reference, they'd be sure that the software will universally run in every computer. This is a better option than comming up with hyper complex softwares which are sucking all the resources and have features that most computers won't efficiently run.

I view those softwares as dinosaurs..., why would Modartt imitate resource sucking dinosaurs  and run at risk of becoming extinct. :b

Re: Very high-end desktop and Piano modelling

Hi Beto, when you say reducing equations you imply that there is an equation to be reduced. But some things aren't reducible to mere equations. Sound is being taken, studied and patterns are being found which at the given resolution can be modeled... and others I assume are being randomized, since it is assumed they're random in reality. If everything could have been represented by mere equations, there would be like dozens of companies offering a product with the exact same sound as Pianoteq... because everyone would have derived the same exact equations. Your comparison with compressed images while in a sense might be accurate I think, a better analogy might be vectorial images. Why aren't they replacing pixel by pixel reproductions? Inherent limitations of models? While they take less space, there is a threshold when they aren't efficient anymore because they start requiring to much horsepower for differences most won't even perceive.

Also at the point the most efficient way would be to use both. See the 3D industry and the use of non-vectorial textures... the advantage of that is that vectorial formula based objects serves as containers... to the pixel by pixel skin images, patterns directly photographed. Vectorial and models give a level of freedom at the expense of more CPU cycles.

ARM CPUs could therefor be a good reference... if they can be compiled on them and the basics could be run on those devices... then the softwares aren't relying too much on some complex algorithms and some instruction sets specific to some CPUs in particular.

Beto-Music wrote:

Pianoteq works according a technology of mathematically reduce/ equations based in the natural behavior of piano ellements. Without th ability to resume the equations a computer would take weeks instead os micro seconds to produce a grand piano note calculated according the laws of physiscs & acoustic.

But the reduce of equations it's not perfect, it take many things out, like digital video compression take details out from HD image. To get a natural sound as possible the delicate point is to know what remove and what left, in thounsands of elements. This take time, and that's why pianoteq got better and better for quite a long time with almost no CPU requirements increase. Only after V3, in V4, the CPU needs got more hungry.

Pianoteq don't need to be 100% complete perfect, but just enough close to perfection to convince nearly everyone. Our ears can't distinguish perfection from something really very close to it.


Lucy wrote:

Would more complex algorithms really mean more realistic sound?

The inherent limitations of Pianoteq could lie in the fact that models will always be approximations of pianos. And some features can't be efficiently modeled. So there will probably a level of complexity that if we go above...

I view it as that, is it worth to spend 2X more resources for 5% more accuracy... with an extra feature which could be run on only the 10% fastest computers?

I would believe that it could be more efficient to sample some parameters... what dstorfer term as pre-calculation could be substituted by the word sample (because samples are just that in one way or another). Even CPUs have instructions which are used instead of just calculating from scratch.

I might be wrong, but if companies use as standard an ARM device as reference, they'd be sure that the software will universally run in every computer. This is a better option than comming up with hyper complex softwares which are sucking all the resources and have features that most computers won't efficiently run.

I view those softwares as dinosaurs..., why would Modartt imitate resource sucking dinosaurs  and run at risk of becoming extinct. :b

Last edited by Lucy (21-09-2017 06:37)

Re: Very high-end desktop and Piano modelling

Lucy,

i could be reading you incorrectly, but are you suggesting that there is something "more" to the production & representation of sound in Pianoteq than "mere equations"?  or rather that there is something more to the sound artifact of physical models than can be represented by the same??  either way, don't such notions rather undercut the very project of Science and its isomorphism with the logic of "equations" as a functional tool for understanding the world?  and (tongue-in-cheek here) i wonder how far one would get in building a physical piano without the use of at least some maths...?

i'm curious however as to what those "patterns" & "randomizations" that go into Pianoteq's modeling could possibly be if not "mere equations" (though i'm not sure that "modeling" is really the proper word for what Pianoteq is doing, but that's perhaps a different topic) ?  as Frege and Russell might have it, isn't it all "equations", independent of the source that's generating the sound...?  (ie the representation of all phenomena can only & fundamentally be understood through the truth function of logical propositions.)

further, the reason why there aren't "dozens of companies offering a product with the exact same sound as Pianoteq" is simply because "they" haven't in fact derived those same "equations".  the folks at Modartt strike me as a uniquely bright crew with some amazing proprietary content/maths in the form of those very "equations"/code that constitute our beloved virtual instrument(s)... no mean feat for "mere equations"...

edit: PS i name-dropped Frege & Russell, but i'd be sadly remiss to not also give a shout-out here for Carnap and, if you're into soporific reading, his Der logische Aufbau der Welt

Last edited by _DJ_ (21-09-2017 11:42)
Matthieu 7:6

Re: Very high-end desktop and Piano modelling

Hammer noises on Pianoteq are samples right? Is there other samples hidden Inside?

Suppose that I record each notes from an acoustic guitar by not pinching the notes but hitting them. Then with a software, lets take audacity I change pitches as to cover the entire range of a piano. Then duplicate each notes into 2 or 3 same notes with slight differences in frequencies... to mimic the fact that the piano has 2-3 strings per notes... I can go on to add some other transformations to craft a Piano. I can then with a series of Fourier transformation and other constructions emulate all the given required effects.

Now, if I have a computer strong enough... all those transformations could theoretically be made real time... but the basis would be the guitar. The core notes recording could be taking, about 5 MB (guitars have a limited range of notes compared to a Piano)

Then I add a series of Piano makes which all have that guitar in common. Why taking the guitar? Because a simple instrument could be crafted easily... a company can not just take an existing Piano and build over it without disclosing it... and each Pianos have a make and due credit has to be given. And it would take a lot of resources to just built a whole piano from scratch to just record its notes. But just taking any 6 string instrument on an easily crafted wooden box...

I am not claiming Modartt did it this way, I am simply saying that's the way my gifted son will design it. Audacity is open source... and he would have access to all those transformations. Could I market it? because I can not make money out of codes my son has borrowed from Audacity. But... the wooden box, was built by his father... this undisclosed box is all it takes to market the product..., because I have patent over what my child father built.

That's the analogy I made with those bitmap textures or skins... and 3D vectorial models serving as containers.

See if someone one day steal some codes from Pianoteq... if they're mere equations... they can always get away because Fourier transformations or other constructs can all be derived from simple observations... where they might be caught is if they also steal a possible recording of that wooden box... because simply put those aren't formulas... they are directly recorded from an acoustical instrument... which was handcrafted by them.

The problem with sampled Piano's... is that they have all their makes... they're also heavy with all the recorded notes... and for this reason they already take a lot of resources... transforming those instruments on top of that will make it very difficult to compute for most computers (even if it wasn't, using few notes of unprocessed string, such as those of a guitar gives more freedom)... the same could not be said of... well of few notes... which are transformed by formulas...

Is it this way it is done by Modartt? I don't know!!!... again, that's how my gifted son would have design it.

As for the suggestion that ''the representation of all phenomena can only & fundamentally be understood through the truth function of logical propositions'',... didn't Kant in his critic of pure reason claimed this isn't wholly true?

Anyway, I'll be taking a few week pause from the forum...

Last edited by Lucy (22-09-2017 01:17)

Re: Very high-end desktop and Piano modelling

Physis claim to do a LOT of real time computing in their ......hmmm, I think it is a 6 core special purpose processor.
I don't know what Roland claim for the V-Piano.

How much of the installed image of Pianoteq is executable code vs how much is data would give an idea of how much computing is actually done in real time.
There are tools available that can do this, I'm not suggesting that anyone reverse engineer pianoteq to find out.

Given that any new release is expected to run acceptably well on whatever the customer base already has (say 3 or 4 year old generic PCs) and usually seems to do so with a CPU load of 20% or less - it is reasonable to assume that the amount of actual "computing" done in real time is fairly low.

There is probably NOT a market for a modeled piano that REQUIRES a lot of compute power.
If every major release needed upgrades to computers less than a year old I think PTQ would die very quickly.

Re: Very high-end desktop and Piano modelling

Lucy wrote:

Hammer noises on Pianoteq are samples right? Is there other samples hidden Inside?[sic] ... Anyway, I'll be taking a few week pause from the forum

well, for when you return:

so, again, are you suggesting that samples stand somehow "beyond" your "mere equations"?  look, it's no doubt trivial to say that all sound is a matter of complex wave functions that allow (of necessity both a priori and a posteriori) for corresponding "equational" representation, but surely its this very fact that (back on-topic) a) allows for said sound to be manipulated (with more or less fidelity, depending on the complexity/depth of the representation), and b) thus begs for increased computational resources to manage and manipulate that same greater complexity/depth...?  the physical handcrafting and recording from that source are (practically) irrelevant here, because the "magic" of making the end sound artifact something approaching or exceeding [beyond "reproduction" to "production"!? beyond good&evil!!?? ] its intended model (as your gifted son will doubtless find out) isn't in his father's wooden box, nor even in the quality of the base recorded sample (and its attendant complex of wave functions), but rather in the challenging and aesthetic complexity of that "transformational"/equation aspect you seem inclined to deride.

it all kind of reminds me of the hue&cry over the introduction of digital media in the first place, but look at where we are now (vinyl-loving nostalgiasts aside)... and no doubt the introduction of keyboard instruments that play both loud and soft through the use of hammers raised quite a few hackles at the time (think of Bach's response to Silbermann's early efforts)...


didn't Kant in his critic of pure reason claimed this isn't wholly true?

Kant of course suggests many things in the Kritik (a state-of-affairs not entirely dissimilar to biblical exegesis/hermeneutics), but the whole problem of the schematismus and representation (vorstellung) does indeed provide that reason is bounded (Kant primarily wants to make an argument for the validity of faith, but also grapples with accounting for purely physical impressions), but then Kant, for all his truly celestial profundity, didn't himself arrive upon the "deep-structure" perspective of the far-subsequent logical positivists (though he's certainly the one who opened the door for them and pretty much everyone else.  i think of Kant as kind of like the Bach of philosophy (copernican revolution and all with his [Bach's] arrangement in the horizontal [the obvious all-time master contrapuntalist] and the vertical [establishing functional harmony on a rational basis!])... but, to continue the analogy, Bach makes possible a Beethoven, a Chopin, a Liszt, a (for better or worse; ymmv) Wagner, even "dialectically" perhaps a Schoenberg...)

Last edited by _DJ_ (22-09-2017 04:33)
Matthieu 7:6

Re: Very high-end desktop and Piano modelling

aandrmusic wrote:

Given that any new release is expected to run acceptably well on whatever the customer base already has (say 3 or 4 year old generic PCs) and usually seems to do so with a CPU load of 20% or less - it is reasonable to assume that the amount of actual "computing" done in real time is fairly low.

There is probably NOT a market for a modeled piano that REQUIRES a lot of compute power.
If every major release needed upgrades to computers less than a year old I think PTQ would die very quickly.

totally agree with this, but that doesn't mean that i personally wouldn't pay dearly for both software and hardware on the bleeding-edge if it brought my rig that much closer to the holy grail of "sense of presence"...

on which front (somewhat straying off-topic here) i think as well that speaker setup, not just computation, is a critical factor in creating that pleasing complexity of sound which we associate with physical instruments.  when i leave asia and return to europe (space!), i'm thinking that a setup of five, high-end, powered monitors +sub (i currently use dynaudios and am very pleased with the results), each corresponding to one of the virtual space's mics, vertically oriented and arranged in a 1 meter box plan just above the height of the soundboard (both the virtual mics and their corresponding speakers), will go a long way toward generating the sort of artifact we associate with physical soundboard resonance/radiation... two speakers can only be just that: two-point & lacking depth, and so i wonder how much of a role cognitive dissonance in the face of "odd" reflections plays in our perception of a sound as "synthetic" or not?  my hunch is that Pianoteq is already processing waveforms essentially indistinguishable from that of physical instruments, but that it's the speaker setup that makes or breaks the sound as seeming authentic... 'would be very curious to hear (perhaps in another thread) about the experience of others with multiple speaker configurations!

edit/PS: further brainstorming about speaker setup here... i wonder what sort of effect housing such a speaker arrangement as described above in a dedicated open cabinet of quality spruce would yield?  at the very least 'should be able to dispense with using IR solutions for cabinet resonance?  'would also make for a fun joinery project and so i guess, with apologies to my wife in mind, 'only going to be one way to find out...

Last edited by _DJ_ (22-09-2017 04:31)
Matthieu 7:6

Re: Very high-end desktop and Piano modelling

I actually like the idea of using GPU as a supplemental computation node for more advanced simulation. I know that a lot of people benefit of low memory signature of Ptq by running it on different hand-held devices and light netbooks etc, but I'm among those who run it on my main desktop PC rig. And from all that recent fuss around cryptocurrency mining I believe everybody should already know that GPUs is capable to do some calculations times faster than even modern multicore CPUs. Utilizing them for additional calculations would allow to see much better fidelity in Ptq's models already today. So some sort of checkbox which would fire up an additional simulation and computation engine would be a blast for those who used to run it on some gaming PC, for example. Would be rock solid feature, still allowing a regular sound quality for those with an older/slower devices.

Re: Very high-end desktop and Piano modelling

I like that, AlexS.

Re: Very high-end desktop and Piano modelling

AlexS wrote:

I actually like the idea of using GPU as a supplemental computation node for more advanced simulation. I know that a lot of people benefit of low memory signature of Ptq by running it on different hand-held devices and light netbooks etc, but I'm among those who run it on my main desktop PC rig. And from all that recent fuss around cryptocurrency mining I believe everybody should already know that GPUs is capable to do some calculations times faster than even modern multicore CPUs. Utilizing them for additional calculations would allow to see much better fidelity in Ptq's models already today. So some sort of checkbox which would fire up an additional simulation and computation engine would be a blast for those who used to run it on some gaming PC, for example. Would be rock solid feature, still allowing a regular sound quality for those with an older/slower devices.

Again, the problem with GPU and audio computations is LATENCY. It would at least add 8-10 milliseconds to what's already there in the system with ASIO etc. It is just not worth it for realtime instruments.

Last edited by EvilDragon (23-09-2017 00:09)
Hard work and guts!

Re: Very high-end desktop and Piano modelling

EvilDragon is it possible to pass processing power from the CPU to the GPU so that the CPU generates the initial sound and then passes the processing on the the GPU so that the latency is seamlessly dovetailed?

Re: Very high-end desktop and Piano modelling

Sure it's theoretically possible but the latency would still be there, because data has to be sent to GPU, processed, merged, then returned back to CPU (and this doesn't even make sense in a modelled instrument, either, because you cannot send the later part of the sound that depends on some excitations and even new input during its expected computation on the GPU, it just won't work), which it then needs to buffer and output. In serial processes like instruments this is just not feasible. GPU is made for heavily parallel processing, not serial.

Last edited by EvilDragon (23-09-2017 00:15)
Hard work and guts!

Re: Very high-end desktop and Piano modelling

+EvilDragon Interesting, thank you!

Re: Very high-end desktop and Piano modelling

i hope to live long enough to see general purpose quantum processors become commercially & readily available (probably around the same time ITER-DEMO comes on-line, humans' first steps on mars, and teleportation of whole objects is a thing)... then you'll be cookn' with gas! 

Matthieu 7:6

Re: Very high-end desktop and Piano modelling

Speaker cabinets built with good "sound wood" woodn't(pun) be good.
You want the speaker cone to vibrate, not the sides of the box that it is in.
I can imagine few worse designs than a speaker cabinet with sides that are resonant in the audible range.

Re: Very high-end desktop and Piano modelling

Bagend's ?

Re: Very high-end desktop and Piano modelling

_DJ_, here's a thread from my experience at the multiple speaker setup:

http://www.forum-pianoteq.com/viewtopic.php?id=4625

I have my mains firing upwards, with smaller secondary speakers, potted down, aimed just to the outside of the player. With Pianoteq 6.0, appropriate width setting (0.40 with my speakers outboard of my piano, and appropriate velocity settings and volume, it's getting harder to tell the difference between real and synthetic when I sit down to play.

- David

Re: Very high-end desktop and Piano modelling

Lots of interesting discussion on the topic! I am really hoping someone from Pianoteq is listening in!

Osho

Re: Very high-end desktop and Piano modelling

Yes, one of the amazing things about Pianoteq is that it is hardware-neutral and can be used with any type of sound-producing equipment from headphones to speaker systems. This is much different than a brand-name arranger keyboard that has built-in speakers. When a company has its own speakers, not only do they get to pick the speakers, but they also get to tweak the equalization curves appropriately for them. I found this out when I used Pianoteq through a Yamaha Clavinova, both as the keyboard and as the playback speakers -  the Yamaha voices and sounds work great on the Yamaha, but Pianoteq  sounded horrible (I  remember now a forum user here who once said this and published a Clavinova-specific preset for Pianoteq) -  that let me see that all of the Yamaha piano sounds must have equalization built-in that are specific for the hardware that Yamaha has chosen.  Therefore, Pianoteq  itself is a double-edged sword in that it does not tell you that you need to use specific speakers, but then again,  you may need to do something to make it sound the best on your particular application.

Last edited by dklein (23-09-2017 21:39)
- David

Re: Very high-end desktop and Piano modelling

dklein wrote:

Therefore, Pianoteq  itself is a double-edged sword in that it does not tell you that you need to use specific speakers, but then again,  you may need to do something to make it sound the best on your particular application.

This is true. I have found Pianoteq 'final sound to ear' vary greatly - even with the same preset and midi. It sounds great on higher range headphones - but not so much on some of the lower end ones. Similarly, it sounded good on the 'studio monitors' - but not as great as it does in headphones. It sounded really great on 'full range' home theater speakers.

Also, a bad velocity curve or the mismatch of the 'default normal' velocity curve to the midi controller can result in sound where all the keys feel really heavy and missing action and hammer hitting strings.

IMHO this partly explains why there are those who love Pianoteq and there are those who hate it. There are just too many variables to know what someone else using Pianoteq is actually hearing.

Osho

Re: Very high-end desktop and Piano modelling

I imagined good studio monitors could get the full range of sound keeping the high quality for all frequencies.

Re: Very high-end desktop and Piano modelling

skip wrote:
dstorfer wrote:

Maybe sans reverb one could find some differences, but that wouldn't be a practical use for it, so if with reverb you could barely tell, then - why do we need a more high-rez algorithm today?

Is "without reverb" an anechoic chamber?

Greg.


Greg, I just meant that in the digital sense. So if we had Pianoteq in its current form and played it without any reverb or effects, and then also used the hypothetical super-upgraded-ultra-Pianoteq, also with no effects, could we hear a difference in the sounds?  I suspect that as soon as you add reverb when trying to A/B the two versions, it would soften the "edges" and blend the sound in a way that it might be much harder to hear the differences.

Re: Very high-end desktop and Piano modelling

I've spent quite a bit of time with Pianoteq 6, the new Steinway, the Ruckers hpschd, and several others. Wonderful modelings, perhaps better than the huge Ivory II sampled pianos (and certainly more fun for Philippe and team to build!)  The greatest limit to quality is THE SPEAKER (or earphone), as it has been for 50 or more years.  I've tried PTQ6 with M-audio monitors, Cambridge Soundworks computer speakers,  20-buck bought-in-the-airport earphones, Apple earphones, and a few others. Which is better?  The 40-year-old KLH17 speakers!!! and the cheap earplugs!!  Forget the special version of Pianoteq software for $4000+ computers with the latest 32-core video processor. Speakers better than the 50-year-old acoustic-suspension ones, at a reasonable price (the best studio monitors are for billionaires or bIg recording studios), would show what Pianoteq can do!

Re: Very high-end desktop and Piano modelling

True on the speakers being biggest up.. or down grade of sound... I have two separate setups for my Pianoteq and change speakers and headphones just to stir it up some... interesting how some of the pianos sound better or worse depending on the particular setup.

Pianoteq 7, all the pianos , a  Casio:  Px-560M, PX 3000, (2) PX350's, Mac i27 and MacBook Pro, Focusrite, Scarlette 18/20 and a bunch of speakers and headphones

Re: Very high-end desktop and Piano modelling

I agree - the weakest link in sound is the last link, speakers. But it is about how we place them, too. We have to remember, that usually, or as a rule, speakers i set, in relation to listener, their angle has to be in 60 degree, equilateral triangle, to get out the most of the sound, sitting symmetrically in relation with speakers. But, because left speakers sound is for left ear and right speakers sound is for right ear, and some sound from left speaker still goes to right ear (and vice vers), it distract the listening a bit (with headphones you hear only left channel in left ear and right channel in right ear) That always make the difference between listening with headphones or speakers, and why headphones often sound better. Not to mention room, acustic a.s.o. Sorry for a bit bad english. Anyway, this is what I have experienced. I try to sit in equilateral triangle with my speakers when playing Ptq. It sounds best that way. And one get the best sound out of not so good speakers. Well, thats what I think about it.

Re: Very high-end desktop and Piano modelling

But when playing an actual piano which ear gets what? And will it make me cross eyed?

Pianoteq 7, all the pianos , a  Casio:  Px-560M, PX 3000, (2) PX350's, Mac i27 and MacBook Pro, Focusrite, Scarlette 18/20 and a bunch of speakers and headphones

Re: Very high-end desktop and Piano modelling

Yeah...like I know that Polk Audio sells 6-, 8-, and 10-element speakers, but the piano is an 88-element speaker (as long as each unison is tuned, otherwise more like a 200-element speaker), yet worse (or better), as each element resonates against a large planar Magnepan- or Quad ESL-style electrostatic speaker element that lies behind each and all of the 88 elements.  And then there's that crazy asymmetric case resonance, as well as that big metal vibrating harp that everything ties in to.

So, as Kramster1 asked, "which ear gets what?"

I have no idea, which is why it's been such a struggle to make my Pianoteq virtual piano playing through my Steinway piano's keyboard sound like the acoustic of the Steinway itself when Pianoteq is turned off and the stop bar is kept away from the hammers.

Kudos to all who go before us and figure out the answer to Kramster1's question!

- David

Re: Very high-end desktop and Piano modelling

@Kramster1 @dklein   ”Which ear get what?”
Figure it out or not - but this is how I see it, my simple commentary. Both ears get the sound, but not at the same time. Because we have two ears, our brain can decide the direction where the sound is coming from. When we play, we always move our head more or less. If looking a bit to the right, left ear get sound first and right ear a bit later (30ms for ex.), left ear sound more loudly. This give us an acustic image/reverberation. And in addition we have the sound reflexes in the room, not coming at the same time to both ears either. As I se it, many of the audio products are still technical limited, in a way, that they cant yet beat human brain understanding/experience of the nature of the sound in the real world (regardless of if there are 11 or 88 elements in a  speaker). But the development process is never-ending, fortunately, and Pianoteq team, for example, have come so far with their product, almost reached their goal. I am glad that this forum give them a gentle push every now and then and hope they have the strength to go on.
- And the other question (probably a joke), cross eyed? No, I dont see why we should get cross eyed, although we move our eyes and head when we are playing. Well, thats what I think about it.

Re: Very high-end desktop and Piano modelling

Maybe a solution to the problem?

https://www.waves.com/plugins/nx

Pianoteq 7 Pro with all pianos

Re: Very high-end desktop and Piano modelling

https://www.theoretica.us/bacch4mac/

another and the opposite solution .. The price of it is a little bit biting)

Re: Very high-end desktop and Piano modelling

Urs Zimmermann wrote:

Maybe a solution to the problem?

https://www.waves.com/plugins/nx

Not sure that official Waves marketing video was so convincing ("Introducing NX - Virtual Mix Room").

This video from Waves marketing provides a view on what the plugin does:

https://www.youtube.com/watch?v=U3GMg5CdBoI

The PianoTeq built-in binaural settings attempt to address similar challenges; regardless, they should provide better latency performance.