David Lai And I wouldn’t call scripting/coding modeling, these are different terms for me. Modeling is basically like what Pianoteq (very bad), Kawai (also bad) and Roland (lesser of evils) are doing. I’m religious about samples, because they originate from the actual acoustic piano, so the sound starts out realistic. And for what VILabs is doing, I think they are among the best makers of such virtual instruments!
Yes I see you're mainly religious about things, good, bad, evil and such 😉 Modelling is basically when you want to make a good replacement of something with different means, and it was around in other fields before piano. Using samples directly is very similar to finite-inpulse-response/finite-step-response models, while for more compact linear representation one might use transfer functions or state-space models; but for more complex systems, nonlinear extensions are needed. Obviously you have close relation with VI Labs, personally I'd prefer if the competition got better so that we'd get more quality virtual pianos to choose from.
David Lai AI audio still has high frequency distortion noise
Not sure what exactly you're referring to, but sure AI isn't something exactly magical and can be used in very different ways, and many half-baked AI solutions have obvious drawbacks. It takes hard work and expert knowledge to do it right in new applications (for instance, Grok must have invested a whole lot in undressing women because it isn't nearly as good at other kinds of image manipulation one'd like it to conduct). I'm just saying that deep neural networks have good potential for this purpose (to make modelled virtual pianos better than Ptq, not quite as compact yet more compact than direct-sample-based ones).
