Trying to find out what PelPix means by granular sound, I used a couple of low notes using a modified U4 preset to bring out as much as possible the raw synthesized sound.
Here is the fxp and test for U4.
http://www.forum-pianoteq.com/uploads.php?id=1685
http://www.forum-pianoteq.com/uploads.p...und_U4.mp3
This example shows the first 20 seconds or so with a normal U4 Tall preset, sounding quite good to me, letting the last note decay, and then the "raw" preset playing the same notes, and made to decay much longer.
If you listen again to the first portion, you may now notice more of this raw, a bit granular sound far in the background.
Maybe PelPix can say if this is what he means.
Doing the same test with Black Grand samples or my keyboard's sampled piano sound, I don't hear such a texture. It's a bit hard to say in fact because my keyboard is a looped sample and playing the midi file to my GigaStudio Black Grand samples cuts the last note short...but I find a different tail to the sound, smoother and I hear only partials interacting in the decay.
I tried to minimize this partial interaction in the raw preset, using a flat temperament and no unison detuning to bring out the basic waveform.
My hypothesis, since PelPix hears it in all modeled pianos, is that it may be a side effect of producing a waveform with many limited resolution computations (even 64 bit floating point is limited)
Pianoteq's modeling is in final analysis a very, very, very, ... , VERY sophisticated form of additive synthesis where the final waveform is the result of a very high number of calculations.
Maybe this background texture is a result of a lot of very tiny errors accumulating, while a sample or a piano recording is one shot. No comparison also with a real instrument that doesn't go through amplification and produces a 3D soundfield directly by moving air.
By the way, in response to Jake, my soundcard can output 24 or 16 bits samples and I do hear a slight difference, 24 bits samples being slightly smoother by having finer steps. But since we usually output 16/44.1 wav or compressed mp3, this difference is only useful when listening live.
This textured effect has always been there I think, qualified maybe as the "pianoteq sound" by some, but the better the modeling of all the components of an instrument gets, the further this raw sound is pushed in the background, to the point where it doesn't bother me at all.
My hypothesis may be wrong of course, this is just a very limited experiment. But then Pianoteq's goal has always been real-time output and it does fine as such, so approximations are unavoidable in that case.