The atom got too bif today...  I men...  we reach a point that a processor have a limit of zise dua atom size, from this point we just add cores and don't get a new faster processor (core) itself.

Light speed also became too slow today... you can't get a transcontinental duet em real time without latency problems due light speed.

Moddart will need to change the laws of physic.   :-)

I guess we're all on the same page.
Though just one thing on my mind now - go go ptq 4!!

Kind regards,
Eran

My point is this. Yes, you can take it to the absurd and say that every molecule should be considered in an absolutely perfect piano model, but for the most part, the components of a piano are sufficiently consistent to fit into a set of equations.

The strings are probably quite uniform in their construction from end to end. Of course they differ by thickness and tension, but those can be accounted for. The soundboard wood is, on decent pianos, probably quite consistent. The hammer felt ought to be consistent, if a skilled tuner has voiced it properly.

If you assume these consistencies, then the behavior of the strings and other parts can be perfectly modeled. I wasn't claiming that a perfect molecular model is feasible, just that once you have made assumptions about your model, a perfect mathematical model is possible. This model is what has to then be quantified and simplified into something that can be computed live.

I'm not asking for the absolute perfect piano model, simply one that has not been reduced so far as to be a live model.

Think about it.

The secrete is to use the esential, sellect what is more important in a model.  That's very difficult.

etalmor:...

I don't think it's naïve to think this is the case. In fact, I think it's naïve to think that Modartt simply hacked their code together without an underlying set of equations that tell how strings and other materials vibrate and interact. They certainly do not have their "best model currently possible" in Pianoteq now, because they had to greatly simplify and approximate their ideal equations in order to make them computable in a reasonably "live" fashion. (I won't say "realtime" because that really means something different.)

I agree Modartt definitely did not hack their code, and they do have an underlying set of equations. However, consider this: what is the equation of a string? there is no one answer to that.
- On one end, it could be something rather simplistic that describes a "perfect" string that has no harmonic distortion. We know Modartt's model is better than that - as it takes into account the length of the string which affects harmonic distortion.

- On the other end, it could be something impossibly complex that describes each and every molecule in the string. This takes into account the edges of the string around the tuning pins, the flexibility and temperature changes in every tiny bit of the string, etc, etc...
And theoretically, even that is not a perfect model - it doesn't describe chemical reactions such as corrosion of the metal, and i'm not even talking about sub-atomic phenomena. Perhaps practically these details are not interesting from an acoustic experience perspective. After all, one just needs to model what can be perceived.

My point is that there is no such thing as a perfect model - every model makes compromises. The question is (and I think I'm quoting Guillaume from an old post) *what* to model, which is the same question as what not to model. Certainly you can't model *everything* (you'd need a theoretical quantum computer for that!)

Having said this - I deeply respect and admire the work being done by Modartt.

Regards,
Eran.

I think pianoteq it's set to old computers for too long.    Every notbook selled today have at least dual core, and most have i3.

Probably many of us have tried to do the basic midi tracks with Pteq, and then render using a sampling library. So far I've never been satisfied with my results, because of the major differences between the two approaches.

If both segments were done using two versions of Pteq- 'live' and 'rendering' versions, this might work very nicely.

I don't think it's naïve to think this is the case. In fact, I think it's naïve to think that Modartt simply hacked their code together without an underlying set of equations that tell how strings and other materials vibrate and interact. They certainly do not have their "best model currently possible" in Pianoteq now, because they had to greatly simplify and approximate their ideal equations in order to make them computable in a reasonably "live" fashion. (I won't say "realtime" because that really means something different.)

If this is the case, I think they should release a more accurate model that doesn't try to be live. It wouldn't necessarily require 8 cores or tons of memory, but it would concentrate more on quality than on performance. I'd pay extra for such a product.

I am not even remotely a computational acoustics expert, but my computer science background tells me that you simply must have your ideal equations figured out in order to accomplish something like what Modartt has done. Maybe they've moved on from their mathematics and work entirely with their current approximated models. Maybe it would be a lot more work to go back to the math and make a new version. Or, maybe they already have a prototype they could turn into an extreme-quality non-live rendering program. Either way, it's just an idea that I think would be good.

I ask to anyone who can to trace a parallel.  The evolution of CPU performance along the last 6 years, and the CPU demand of each pianoteq version along this 6 yeras.

I have a feeling that pianoteq changed too little in terms of CPU demand, comparing to the processors evolution.

Second, regarding sampling and modeling. Well I think there is a whole spectrum of possible combinations between the two.
For example, instead of sampling the final sound, you can sample an "intermediate" reaction, and feed that into the model. e.g. if you have a modelled trumpet, you can sample the air flow through the mouth piece. Or in a piano, the intricate bounce of the hammer against the string.
Another example, is that you can sample instead of a whole sound, just some partials of it. Think of a convolution reverb, it's actually a non trivial blend of sampling and modelling...

Finally - there is much love for Modartt in this forum, well deserved, and they set the expectations pretty high. I say - let's give them the credit for their passion, creativity and ingenuity, that they continue to produce superb products. I think there is reason to be optimistic about the technology - it has took sampling 30 years to get to it's current state, and maybe start platueing. I don't believe physical acoustic modelling has run out of juice yet, or anytime soon.
The other side of the coin is: let's be realistic - this technology has yet to run it's course. There is still enough room for improvement - and hopefully we'll continue to see gradual improvements over the coming years. Modartt is doing a fine job leading the way, and I wish them the good fortune to continue to do so.

-- E

I bet you are pretty happy with the results of pianoteq, even in beta stages of V4.

This is a good signal, as you already can "taste the cheese".

I'm in agreement with the crowd that believes a blend of samples and modelling would produce the most realistic and importantly- EVEN sounding and playing- results.

Not sure about what you refer.

There are already hybrid version, like truepainaos, that uses small sample data bank and agorithms to create harmonic and new velocities,and the result it's nothing great. 
Ivory use modeled (if we think in symphatethic resonance algoritym as moddeled) to try add nice harmonics and a very large data bank of sounds, but the result it's not perfect, and bny listening we still feel the touch of sampler it have.

If you refer about mix the sound of a sampled note with the sound of moddeled note, you would get something like a chorus, since, for exemple, one people singing it's diferent than 2 people singing, even the the two people have very similar voices and sing extremelly in fine sincronization.