pianow agreed their steinway is ten times better then the vsl's….
Once again, let’s keep things civil, but I do wonder, how did you come to this conclusion, “ten times better”?
Was this number chosen arbitrarily, or did you do the proper measurements to arrive at this number (10)?
It could very well be that your calculations are correct, but we simply need verification, and it’s not that I don’t trust you, but you know how the saying goes…..”trust but verify.”
Have a great day!
Pete14
alright I'll play but i warn you without having experience with at least one of them, you wouldn't fully understand what I mean.
1 it's rounder (which is the exact opposite of the vsl steinway, that one is sharp and cold as a knife)
2 it's darker (makes it easier to listen to without being fatigued)
3 it has excellent timbre (unlike vsl i doesn't sound thin and weak in the ppp range and doesn't sound hard and metallic in the fortes)
4 it's more cohesive (vsl steinway samples are all over the place)
5 it's more enjoyable overall
6 the resonances make it much fuller
7 the voicing is good for most genres (vsl is only good for classical)
8 its tuning is perfect
9 it's a lot dryer so you can put it in any arrangement
10 it sounds like a real steinway d
and i think this combination worth more then 10 times.
David Lai The 280 VC is too bright, like the Steinway
if i may, the 280vc sounds nothing like their steinway and that's why i hate their demos and presets, they make everything sound harsh and annoying
the 280vc is bright but also has excellent weight and impact, the fundamentals are heavy and warm, rhe basic tone is quite rounded unless you reach the FF range
the Vsl steinway on the other hand sounds empty, thin, bright and cold throughout all the velocities
it's in my opinion not fair to even compare these two
Not only modelled pianos, there are modelled guitars from MusicLab, modelled drums and basses from IK, modelled eps from AAS, modelled strings, brasses, woodwinds from Audio Modeling. They share PTQ's strong points and weak points, surely have many fans but haven't convinced many producers yet. Most producers still prefer sampled libraries or real singers, real band, real orchestra.
pianow The first time I heard VSL 280VC, I almost mistook it for PTQ. LOL. I thought I would rather had a PTQ version instead. At that time, PTQ had not released 280VC, and VSL had not compressed data yet. Later more demos came out. I heard more variations of the tone. It is not bad but I prefer their Imperial.
pipAT woodwinds from Audio Modelling
Those saxophones are quite cool.
Funny thing is that Audio Modeling and Sample Modeling were two brother companies separated from the original company. Their opinions became different towards virtual instruments against each other. Thus the original company split.
pipAT
completely agreed
To bring back the discussion to something technical. Most (all) sampled pianos are more or less modeled in the same sense as Pianoteq: they “model” the behavior of acoustic pianos through various means:
It’s basically behaving like a piano. That’s modeling. Because Pianoteq doesn’t use finite element modeling (what many people erroneously imagine modeling means: having the entire piano and its moving parts working in a virtual simulation. Nope). They just re-synthesize sound that approximates pre-recorded samples and by using pre-calculated sine-wave envelopes, curves, tables and other realtime (additive/subtractive) synth-like audio parameters, so that the synth sounds and behaves similar to the recording of the target piano.
Interesting how you listed “10” things; coincidence?
Now, on the notion that we’re not getting “true” modeling, and according to Quasi, ‘it’s not just a matter of having computers that can handle massive calculations per second, but also the immense complexity involved in capturing these elements,’ which I believe he deemed all but impossible, I wonder, could AI aid in this “gathering” process, and, therefore, allow for “true” physical modeling?
Compared with early funny GM midi like PTQ, 4Front Truepianos, Acoustica Pianissimo etc., PTQ 8 has improved in piano modeling sphere with many more nuanced funtions. It's not totally bad for playing for fun or practice. Some even only use PTQ for practice, sampled vis for production. And talking about pure synthersizers, Xfer, Uhe, Spire, or real hardware synth, nothing comes close PTQ in emulating acoustic pianos. I also treat PTQ as synth. To me, kind of , other synth pianos are like regular racing games like NFS, while PTQ like Asseto Corsa, detailed emulations. Of course, different from driving a real car. Sampled vis, modelled vi, digital pianos are all different from a real piano, sound and touch when playing.
Pete14 AI (or Machine Learning, to be strict) can help in creating better coefficients/curves/tables (or whatever they call these internal tuning parameters) to try to approximate the original recorded samples as close as possible but it can’t help in making faster computers 😃 I don’t believe we will ever have computers that are fast enough to run a detailed finite element model of a piano in real time.
I remember my work as a beta tester on one of the Pianoteq models. (My NDA has expired long ago). Philippe would prepare these internal calibrations and would give it to us. People would comment, record, discuss. Personally I used to create comparative recordings between Pianoteq and Vintage D (that’s what I had at the time). I would say something like: see how this phrase accentuates metallic overtones in PT and how nicely it blends in Vintage D. Philippe would then do another internal re-configuration based on beta testers feedback. Indeed, sometimes he addressed my comments but then something else got wrong. And in his words (not literal quote, but along the lines of): yes, it’s tricky because adjusting one of these (internal) params changes many things at a time. It’s also when I lost faith in PT. The entire beta process was just moving in circles and it never gradually went better, only changed certain aspects at the expense of others. This might be a subjective understanding and I may be biased but my participation was valuable and they invited me for a few other models but I declined politely.
ML can help in such a process by first learning when a piano sound and behavior is close enough to another one and then apply combinatorial tweaking of all the internal params iteratively until they approach the target sound, rather than relying on beta testers and one man who tweaks some knobs (figuratively) to address comments.
CyberGene I remember my work as a beta tester on one of the Pianoteq models..
Wait, you once “worked” for Pianoteq, and then decided to become a whistleblower? 😉
Perhaps it might’ve been better if you’d let us believe that it was indeed a real physical model, but I suppose you felt it your responsibility to blow the whistle on Philippe!
BTW, wouldn’t it be great if he stopped by? but come to think of it, he might not be too motivated considering the amount of “love” Pianoteq gets ‘round here….
P.S.
Has anyone tried a totally different approach to capture “the piano” other than a bunch of recordings, and/or a bunch of knobs? It would be great if we had a third option, and no, the “hybrid” approach (sampling + modeling) doesn’t count as a third option!
Pete14 Wait, you once “worked” for Pianoteq, and then decided to become a whistleblower? 😉
I thought you knew that. Many people know I’ve been on PT beta test team long ago, I’ve mentioned it numerous times on PW 😃
One great thing about VILBAS pianos is that their "modelling" or programming doesn't include noticeable sample distortion or artifacts, while I saw here, pw and vc, people report artifacts in the tone of Ivory 3 probably because of the hybrid modelling.
Honestly, CG, I didn’t know!
And, since we’re being honest, I’ll admit it, I was one of those who thought it a “physical model” until you came along and ruined that for us.
You know, CG, sometimes it’s okay to be a little bit oblivious to what’s happening behind the curtains, and perhaps you could’ve been a bit more gentle in breaking the news to us; because there’s nothing wrong with believing that Houdini was a “real” magician, and not just a trickster, but now you’ve ruined that for us, and on top of it, you won’t even give us a little hope that some day the magic will be real…….”It’ll never happen,” you say!
AI could train a piano library, but there hasn't been one yet I know. I only know Dreamtonics trained AI voice packs. They have standard voice packs but sounds like Vocaloid you know. Creepy unnatural voice singers, however many fans still like that.
BTW, beta testing for PT is volunteering. They found me on PW I guess (I was hating PT as always) and contacted me. We are not paid (I think I got free license for the version at the time) and we are not told how it works, nor any internal knowledge 🙂 I think Philippe once replied to me that my interpretation of their patent is not completely right (but he didn’t say it was completely wrong either) and he also said the patent is old and they have added many new things. He never explained any details though. He’s a very nice and knowledgeable guy as can be expected. What I am afraid though is, as the sole mastermind behind all of it, it’s only him who can make things better, yet he doesn’t have the ear of a recording engineer (hope he’s not insulted by my opinion) and is probably why he can’t approach the piano sound the way some of us expect it and there’s no one else who can do something. I hope I’m wrong 🙂
CyberGene What I am afraid though is, as the sole mastermind behind all of it, it’s only him who can make things better, yet he has no the ear of a recording engineer (hope he’s not insulted by my opinion) and is probably why he can’t approach the piano sound the way some of us expect it and there’s no one else who can do something. I hope I’m wrong 🙂
But I’m assuming he has a “team” giving him constant feedback and helping in the process; it can’t just be Philippe all on his own, even though he most certainly has the last word….
pipAT AI could train a piano library…
Are you saying there’s hope? 🤔