"I think this is highly inappropriate and your nasty comments are unfair to women and to men, not to mention tiresome to read, and one cannot assume gender based on what was said, and the vast majority of people would agree with me, and I will have to complain to the moderators because I've been a member of this forum since 1986 and I have the most posts here."
Wasn’t it nice whence Sofia used to stop by ‘round here right after school?
I remember her saying ‘I just got home from school, and no sooner had I dropped my backpack, whence I felt giddy with anticipation, and could not wait to meet, once again, with my dear PC friends….’
Yet how did we repay her?
Well, at first we loved her, but then it came to our attention that Sofia, 22, wasn’t perfect after all, and our attitude towards her took a turn for the worse. A reflection, I say, on our idealistic relationship with perfection, or the lack thereof.
“Wait, she’s not perfect?”
“No! She has some flaws.”
“Screw her, then.”
“Isn’t that what you wanted to do?”
“How dare you! I’m married!”
Don’t be a stranger, Sofia……….
To me, Sofia will always be a woman, a shy and mysterious italian beauty, geeking out over the most delightful piano samples, wearing the skimpiest outfits in the guarded secrecy of her remote neapolitan villa.
QuasiUnaFantasia there is no modelling without previous sampling; the modelling algorithm does not know what to aim for unless it has a specific goal. That goal is provided by samples
So technically… what you are saying is:
All modeling is likely to involve plagiarism / copyright infringement?!
burkey No, not necessarily. And I assume Modartt has done their own sampling, in which case they of course own the samples.
Can anyone explain to me (in layman terms) why they can’t use an actual acoustic instrument as a point of reference?
Pete14 Because no one trusts their ears as much as they trust digital measurements. Theoretically, you could create a completely theoretical model (hmm, that's a lot of theory), and then have it generate sound which you then compare, through listening, with sound from an acoustic instrument. But that comparison involves not only the model itself, but also the loudspeakers over which it is heard, so it would conflate model and loudspeakers.
In order to compare only the model with the acoustic, you would have the model generate sound, and record similar sound from the acoustic, and then compare these two. Of course, that way you conflate the sound of the piano with the recording device, but it is so much simpler to do, and highly consistent (because no human hearing is involved), that it will probably always be the favoured approach.
Okay, so I ask, could Modartt take an existing sampled library (let’s say VSL) as their source instead of sampling an acoustic from scratch?
The former seems more practical/cheaper to me (hack the heck out of VSL).
Incidentally, this brings me back to Sofia, and why it is that we don’t embrace the act of re-sampling?
Keep in mind that most sampled libraries have been “stealing” from the acoustic world for years, or do you really believe that the “American Grand,” or the “Huge D,” are anything but Steinway(s) in disguise; surely, Steinway hasn’t seen a penny from these samples, so, if VI-Labs can have a “Big D,” why can’t Sofia release the “BallsandDorfer”? for you see, the latter sounds like “Bosendorfer,” but is it really? (Especially once it goes thru Sofia’s magical touch).
Pete14 Okay, so I ask, could Modartt take an existing sampled library (let’s say VSL) as their source instead of sampling an acoustic from scratch?
Yes, they could. And they did exactly that for their "C. Bechstein DG" instrument. But they clearly obtained permission from the creator of the samples (which was Bechstein themselves, I believe).
Pete14 ... and why it is that we don’t embrace the act of re-sampling?
There will probably be more of that in the future. I, myself am working with this as a goal (although I've been detoured quite a bit in recent times, and I don't have any plans to distribute whatever virtual instruments I create). But it seems obvious to me that statistically based models of piano's will gain ground in the not-too-distant-future.
QuasiUnaFantasia I, myself am working with this as a goal (although I've been detoured quite a bit in recent times, and I don't have any plans to distribute whatever virtual instruments I create)….
Why won’t you show us your “Huge D”?
Surely, we won’t tell anyone, and if you show me your “Huge D,” I’ll show you my “BallsandDorfer.”
Granted, I’ve never sampled anything, so I’ll simply be the face of BallsandDorfer, but Sofia would have to get down on her dirty knees to re-sample that there Dorfer!
Pete14 I think the world will do just fine without whatever I create, and additionally there would be a lot of hassle exactly about the samples which would be the foundation of such models (the one who creates the samples, owns them, I suppose,, and I would also not want to take anything away from them by using their samples to create competing instruments).
Oh, and I would probably never do a "D", since I have had thoroughly enough of that "D" sound already. 😆
Pete14 Granted, I’ve never sampled anything, so I’ll simply be the face of BallsandDorfer, but Sofia would have to get down on her dirty knees to re-sample that there Dorfer!
You've really outdone yourself this time, Pete. 😆
😁
QuasiUnaFantasia There will probably be more of that in the future. I, myself am working with this as a goal (although I've been detoured quite a bit in recent times, and I don't have any plans to distribute whatever virtual instruments I create). But it seems obvious to me that statistically based models of piano's will gain ground in the not-too-distant-future.
Yes, Resampling is what I'm doing nowadays. I have my own recordings, programming and velocity transitions already in Kontakt. I now simply re-sample them into 24 layer instruments for SFZ. Much easier than re-programming.
Also, I wonder if anyone is thinking of taking sample libraries as training material and using machine learning to create modeled pianos.
dore_m I have done a few preliminary experiments, and they revealed that building a useful machine learning model is not going to be a walk in the park. Particularly my initial desire, to create a model with high fidelity to the original samples using much less memory than the samples themselves. is going to be a tough nut to crack.
Come back, Sofia! Tell us how you did it.
While I might think there's something to be said about producing samples from Pianoteq, might be seen as unethical, I don't see any real consequences for doing so. Legal action would be a waste of time and money, especially if Sofia is 16. And, I doubt Modartt really gives a damn. After all, Pianoteq is more than just a piano sound player. People who cry fraud are a bit over the top; fraud is misrepresentation for financial gains, which didn't happen here. And stop whining about the buy me a coffee thing.
I think Sofia should put everything back, the website, the sample sets, the bio, and bring her back to life; embrace the alter ego, the creativity in making a character, like a writer does in a novel or a play.
But what I really want is the details of how she did it - the method, the software used. Even if one might say it's wrong to make samples from Pianoteq, I see no problem with Sofia sharing the technical details of how it was done. I call her she just because it's easier, and a bit of fun, too. So, come back, Sofia. And let the bons temps roule!
incredibly accurate chart.
It’s not such a big deal to make a MIDI file playing all the keys in specified velocities and with a certain duration that makes sure it can cover the entire sustain at the particular range. You export a WAV and split it into samples and voila.
Pretty dumb solution. Not sure what sort of people and how many people fell for it but apparently those don’t have good ears. I needed just 30 seconds to realize it was Pianoteq. I would have loved to be the one that debunked it but I was not interested in primitive pianos like these, so never bothered to even listen to demos, let alone download and test.