Topic: Pianoteq source for its simulation

I am almost sure no one from Modartt reads this forum but in case they do…

To Modartt team:
Dear team members, we as customers typically compare Pianoteq to other (competing) products such as Roland Earth Piano and also to some very good sampled libraries. Price-wise they are also comparable. But like 10 years ago in the guitar world, the simulations (amp-sims here represented by 'Pianoteq 9' by some analogy) were recognizable instantly from the real (sampled) recordings so is the Pianoteq even upon a hearing (3rd person) test, let alone playing it in real time.

With 'Pianoteq 9' out now, I personally do not see (hear) any much difference from 'Pianoteq 7', or maybe a bit but it is negligible or even placebo. The efforts put into Pianoteq are admirable but compared to sample libraries it is only space (gigabytes) saving tool for demo-practicing purposes (at least for pianist I talked to).

Here is my own speculation and of course I might be completely wrong indeed:
I can hear that for some notes you have modeled them after some static samples. That is ok as intention to simulate 'real world', but in the grand picture of schemes it is wrong. I have my own view on how to 'capture' the real thing (albeit there are more than one real pianos in this world and we can not equip them all with 'capturing' devices) and you have to forget about microphones! Sampled libraries also fall in this microphone christmas tree trap just to increase size of the library to justify price.

Take a look a glimpse at the guitar world. Just a hint. I really hope I am wrong in all this but something tells me I am not.

Re: Pianoteq source for its simulation

Pashkuli wrote:

I am almost sure no one from Modartt reads this forum but in case they do…

To Modartt team:
Dear team members, we as customers typically compare Pianoteq to other (competing) products such as Roland Earth Piano and also to some very good sampled libraries. Price-wise they are also comparable. But like 10 years ago in the guitar world, the simulations (amp-sims here represented by 'Pianoteq 9' by some analogy) were recognizable instantly from the real (sampled) recordings so is the Pianoteq even upon a hearing (3rd person) test, let alone playing it in real time.

With 'Pianoteq 9' out now, I personally do not see (hear) any much difference from 'Pianoteq 7', or maybe a bit but it is negligible or even placebo. The efforts put into Pianoteq are admirable but compared to sample libraries it is only space (gigabytes) saving tool for demo-practicing purposes (at least for pianist I talked to).

Here is my own speculation and of course I might be completely wrong indeed:
I can hear that for some notes you have modeled them after some static samples. That is ok as intention to simulate 'real world', but in the grand picture of schemes it is wrong. I have my own view on how to 'capture' the real thing (albeit there are more than one real pianos in this world and we can not equip them all with 'capturing' devices) and you have to forget about microphones! Sampled libraries also fall in this microphone christmas tree trap just to increase size of the library to justify price.

Take a look a glimpse at the guitar world. Just a hint. I really hope I am wrong in all this but something tells me I am not.

1. They do read the forum.
2. There is a difference between v7 and v9. Its not as big as v1 - v9 obviously, but like any technology it evolves gradually, and faster in the beginning. Can you spot the difference between the 2 latest iphone models?
3. Its not only a gigabyte save tool. Modelling generates the sound based on input while, samples selects a prerecorded sound closest to the user input. Modelling allows us to change properties of the piano which are fixed in sampled piano libraries. This includes string length and soundboard properties and hammer hardness etc. Pianoteq also allows for morphing of pianos to create hybrid instruments. Low footprint and low hardware requirements have its competitive edge in many cases, but not all.
4. Take your own view of capturing "the real thing" and start a competitor. I will cheer you on.

Personally, I think generative AI will be pianoteqs competitor in the future. When AI can control synthesis, which I guess it already can, its only a matter of time before it can react to user input and generate a sound based on the users preferences. We are not there yet, but its only a matter of time before someone has built it. Before then, pianoteq is a good option for ppl who likes small footprints, highly customizable, responsive piano.

Some ppl like the sound of pianoteq, some don't. Just like any instrument. I personally prefer uprights to grand pianos in most cases. Does that mean now that grand pianos only purpose is for those who want a loader sound? No. Same goes for virtual instruments. Its a matter of what you value the most in an instrument.

I porsonally like samplebased pianos just as much as modelled but for different reasons.

Re: Pianoteq source for its simulation

snurrfint wrote:

• They do read the forum.
• Can you spot the difference between the 2 latest iphone models?
• string length and soundboard properties and hammer hardness etc. Pianoteq also allows for morphing of pianos to create hybrid instruments
• capturing "the real thing" and start a competitor.

…generative AI will be pianoteqs competitor in the future. When AI can control synthesis, which I guess it already can…

Good points, especially about AI generative synthesis.
My point with this first post in the forum is that typically we know the math formula for a vibrating string, incorporating timbre and those parameters are 'on paper' for sure. But how Modartt team gets their sources of actual physical strings off of a piano to 'match' their model to those? This is crucial for the whole intended development of the simulated piano sound!

• is it from a microphone recording in an anechoic or real studio room chamber? (if yes this is ok but it is wrong)
• is it from this or that piano brand and model (if yes, then why not buliding a sort of "neutral" or scientific setup as I would have done it? But I understand they are not a piano manufacturer and brand that makes pianos; they are scientists and piano and music lovers)
• is it from a collection of sources and sort of 'averaging' the output in order to fit the parameters?
• and the actual engine's piano parameters are not quite 'available to be tinkered' by the end user it seems otherwise how would they sell all those add-ons branded with the respective logo of a real piano manufacturers as those add-ons seems to do exactly that – preset tweaks not available from the UI?

How I would have done it:
1. Place linear pickups (such as Cycfi Nu modular pickups) under every string of a neutral build of a piano frame\soundboard and maybe a few hammers and test keys
2. Attach piezo-pickups (piezoelectric materials produce an electrical signal when a stress\vibration is applied to certain spots on the soundboard non-destructively (per test or capture)
3. Setup with one string per note (real 'una corda') and another setup with up to three two strings with two aliquot strings for the higher register (not like the current pianos!)
4. Possibly use of a few solenoids triggers to get a few dynamic pushes to the keys mentioned to get various loudness and dynamic captures form all the above 'sensors'
5. All that in an semi-anechoic or full anechoic environment with or without piano cabinet (probably without as it is terrible by current construction)

Then I can say I have a close to neutral source sounds to compare the models against them and to try and match them.
Wavetable synthesis eventually to be incorporated for exploring the morphing of corresponding source material samples.

Again, that is my own opinion and how I would have attempted such endeavour. Then Math and Physics, AI and ML  models are generated for the purpose.

Last edited by Pashkuli (Yesterday 16:28)

Re: Pianoteq source for its simulation

Pashkuli wrote:
snurrfint wrote:

• They do read the forum.
• Can you spot the difference between the 2 latest iphone models?
• string length and soundboard properties and hammer hardness etc. Pianoteq also allows for morphing of pianos to create hybrid instruments
• capturing "the real thing" and start a competitor.

…generative AI will be pianoteqs competitor in the future. When AI can control synthesis, which I guess it already can…

Good points, especially about AI generative synthesis.
My point with this first post in the forum is that typically we know the math formula for a vibrating string, incorporating timbre and those parameters are 'on paper' for sure. But how Modartt team gets their sources of actual physical strings off of a piano to 'match' their model to those? This is crucial for the whole intended development of the simulated piano sound!

• is it from a microphone recording in an anechoic or real studio room chamber? (if yes this is ok but it is wrong)
• is it from this or that piano brand and model (if yes, then why not buliding a sort of "neutral" or scientific setup as I would have done it? But I understand they are not a piano manufacturer and brand that makes pianos; they are scientists and piano and music lovers)
• is it from a collection of sources and sort of 'averaging' the output in order to fit the parameters?
• and the actual engine's piano parameters are not quite 'available to be tinkered' by the end user it seems otherwise how would they sell all those add-ons branded with the respective logo of a real piano manufacturers as those add-ons seems to do exactly that – preset tweaks not available from the UI?

How I would have done it:
1. Place linear pickups (such as Cycfi Nu modular pickups) under every string of a neutral build of a piano frame\soundboard and maybe a few hammers and test keys
2. Attach piezo-pickups (piezoelectric materials produce an electrical signal when a stress\vibration is applied to certain spots on the soundboard non-destructively (per test or capture)
3. Setup with one string per note (real 'una corda') and another setup with up to three two strings with two aliquot strings for the higher register (not like the current pianos!)
4. Possibly use of a few solenoids triggers to get a few dynamic pushes to the keys mentioned to get various loudness and dynamic captures form all the above 'sensors'
5. All that in an semi-anechoic or full anechoic environment with or without piano cabinet (probably without as it is terrible by current construction)

Then I can say I have a close to neutral source sounds to compare the models against them and to try and match them.
Wavetable synthesis eventually to be incorporated for exploring the morphing of corresponding source material samples.

Again, that is my own opinion and how I would have attempted such endeavour. Then Math and Physics, AI and ML  models are generated for the purpose.

I think they are modelling all parameters of a piano, string vibrations (string resonance etc), soundboard vibrations (completely new model for just for v9), hammer hardness etc. They use recordings of pianos to match and test the modelling against real physical pianos and the frequencies they produce. Its quite delicate, to the point where you can modify specific properties of a certain part of the piano. Its a scientific research project turned into a product. I would bet that everything you know, or thing you know about modelling of a piano, they know aswell or better. And that is not meant as a insult to you at all, I would gladly play a piano modelled after your approach. This is where my knowlede ends in this matter. Hope you get answers, in some way or another.

Re: Pianoteq source for its simulation

Pashkuli wrote:
snurrfint wrote:

• They do read the forum.
• Can you spot the difference between the 2 latest iphone models?
• string length and soundboard properties and hammer hardness etc. Pianoteq also allows for morphing of pianos to create hybrid instruments
• capturing "the real thing" and start a competitor.

…generative AI will be pianoteqs competitor in the future. When AI can control synthesis, which I guess it already can…

Good points, especially about AI generative synthesis.
My point with this first post in the forum is that typically we know the math formula for a vibrating string, incorporating timbre and those parameters are 'on paper' for sure. But how Modartt team gets their sources of actual physical strings off of a piano to 'match' their model to those? This is crucial for the whole intended development of the simulated piano sound!

• is it from a microphone recording in an anechoic or real studio room chamber? (if yes this is ok but it is wrong)
• is it from this or that piano brand and model (if yes, then why not buliding a sort of "neutral" or scientific setup as I would have done it? But I understand they are not a piano manufacturer and brand that makes pianos; they are scientists and piano and music lovers)
• is it from a collection of sources and sort of 'averaging' the output in order to fit the parameters?
• and the actual engine's piano parameters are not quite 'available to be tinkered' by the end user it seems otherwise how would they sell all those add-ons branded with the respective logo of a real piano manufacturers as those add-ons seems to do exactly that – preset tweaks not available from the UI?

How I would have done it:
1. Place linear pickups (such as Cycfi Nu modular pickups) under every string of a neutral build of a piano frame\soundboard and maybe a few hammers and test keys
2. Attach piezo-pickups (piezoelectric materials produce an electrical signal when a stress\vibration is applied to certain spots on the soundboard non-destructively (per test or capture)
3. Setup with one string per note (real 'una corda') and another setup with up to three two strings with two aliquot strings for the higher register (not like the current pianos!)
4. Possibly use of a few solenoids triggers to get a few dynamic pushes to the keys mentioned to get various loudness and dynamic captures form all the above 'sensors'
5. All that in an semi-anechoic or full anechoic environment with or without piano cabinet (probably without as it is terrible by current construction)

Then I can say I have a close to neutral source sounds to compare the models against them and to try and match them.
Wavetable synthesis eventually to be incorporated for exploring the morphing of corresponding source material samples.

Again, that is my own opinion and how I would have attempted such endeavour. Then Math and Physics, AI and ML  models are generated for the purpose.

I think you are making a lot of incorrect assumptions. I recommend the reading of the following article written by Juliette Chabassier , PHD, who was involved in the development of pianoteq.
The document is public domain .

https://inria.hal.science/hal-01894219/file/hal.pdf

Re: Pianoteq source for its simulation

Pianistically wrote:

I think you are making a lot of incorrect assumptions. I recommend the reading of the following article written by Juliette Chabassier , PHD, who was involved in the development of pianoteq.
The document is public domain.

Fabulous! I will read it right now. And I really thank you for the link to that document.

Re: Pianoteq source for its simulation

yeah, I think the degree of difficulty here is magnitudes above (and I'm a keys guy with at least 10 guitars) guitar amp modeling.

multiple strings per note, and they interact (and this interaction are also parameters in the models).
the soundboard is > than an amp
the micing is > room simulation.

let alone that for guitar (amp) modeling, or even stomp and effects modeling, there's still the core instrument itself - okay Variax is an actual model of various guitars, as is a few Fender models, but I doubt anybody really commits those "naked" as solo instruments yet folks (including some A-list pros) actually do that with Pianoteq.

I've been with Pianoteq since v1 - and I would say "it was an idea" then, but I could use it for pop pianos as a synth. It *still* has all of the synth capabilities. But...  Today? I'll also use it as a naked solo piano in a typical piano audio chain, meaning compression, EQ, verbs are all external to what is in native Pianoteq - because I love the models, can make the micing work for me, but just prefer better EFX.  I own more than enough premium (priced!) piano libraries and Pianoteq really is for me "looking at the challenge from the other side", and most times it's the better side.

I will *def* give the new micing setup in v9 a serious workout.

I just think this is quite a bit more modeling than an entire audio path on a guitar, it'd be closer if we model the guitar itself too.  If you have any doubt just how deep these models are?  switch down to a single mic, and move it what would approximate to just a few centimeters and hear the effect.  It's like the everywhere, not only on the virtual model of the instrument, but in every spot in the space shown.  It's stunning.  Now take that by 8x stereo. Voila, not just a model of the piano, but the piano in the space it's in.  That's huge. :-)

Re: Pianoteq source for its simulation

michaelhuisman wrote:

yeah, I think the degree of difficulty here is magnitudes above…

I just think this is quite a bit more modeling than an entire audio path on a guitar, it'd be closer if we model the guitar itself too. Voila, not just a model of the piano, but the piano in the space it's in.  That's huge. :-)

Modeling a "body" is resonance reverb – you can hear it through piezo in any acoustic guitar performance. That is nothing to worry about. Modelling strings for a guitar and how fingers (not just hammer hits) interact with those is not magnitudes but parallel universes more complex than any piano and the way pianos are played (triggering their strings via hammers).
The number of strings is also not a problem but phasing is. It is a chorus like effect but very short.

The main question I am asking is what the models have been matched to and compared to? What sounds could those be and how they have been captured? If with mics, then this is a huge mistake!

Re: Pianoteq source for its simulation

Accurately simulating the behavior of a piano's sound production is one thing, but running the simulation in real-time for playing is another. In industrial simulations, it is common for large servers to run for dozens of hours to produce results that last just a few seconds. If we take every detail into account as much as possible, real-time playing would become impossible. Therefore, approximations must have been used here. I believe the Modartt team has put in a great deal of effort to balance accuracy and real-time performance. However, it would be desirable to have an "accuracy" mode for offline rendering.

Re: Pianoteq source for its simulation

Pashkuli wrote:
Pianistically wrote:

I think you are making a lot of incorrect assumptions. I recommend the reading of the following article written by Juliette Chabassier , PHD, who was involved in the development of pianoteq.
The document is public domain.

Fabulous! I will read it right now. And I really thank you for the link to that document.

you are welcome .

Last edited by Pianistically (Today 12:18)