Here are some random one-offs and thoughts:
Pianoteq is the brainchild of Philippe Guillaume who is one of the brightest math professors (if I'm not mistaken). And despite that, Pianoteq seems to have hit a plateau. It has come a long way indeed and is now quite piano-like in timbre, compared to earlier versions that sounded like a cymbalom. Yet, there are some traits of Pianoteq that are always there. Which is why I think it's not about better hardware or better algorithms. We have one of the smartest guys and we have extremely powerful computers, yet Pianoteq is not able to produce high-quality sound like the sample-based libraries. That makes me think additive synthesis is not the way. We should instead use real "modeling", i.e. finite-element analysis, or simply speaking, have virtual piano with all its mechanical parts moving inside the computer, rather than attempting to recreate the sound. However, that would require a quantum leap in computing power, no pun intended. Hopefully, we can have that in our lifetime but I'm rather skeptical. Modeling is the future but not this modeling. We need entirely different type.
I'm wondering, if Pianoteq uses a variation of additive synthesis optimized for piano sound, how do they make the attack? IMO, the piano attack is where there's most chaos and unpredictability. Do they use sampled attack portions, in the same way they use sampled noises? If not, how are they actually synthesizing attacks? Maybe that's where the weakness of the algorithm is. Maybe it's what makes repeated strikes to be so similar?
To me it seems Pianoteq is mostly a clever way to synthesize piano sound without taking much space. The added bonus is lack of noise and better resonance behavior. Well, who cares about disk space nowadays? So, it's only noise and resonances. Digital piano manufacturers already have excellent resonance modeled engines. Some software ones also have resonance engines, I believe Ivory. Hopefully more software piano manufacturers will start adding resonance modeling. Then to me it seems, the future is in hybrid software piano engines: samples + resonance modeling.
