Are Sample Libraries Stagnating?

Sampler and Sampling discussion (techniques, tips and tricks, etc.)
RELATED
PRODUCTS

Post

AngelCityOutlaw wrote: Sun Sep 08, 2019 10:32 am
anomandaris1 wrote: Sun Sep 08, 2019 9:13 amYou have complete control in most samplers over this sort of stuff and can do it by yourself better than any so called "AI". Anyway, I doubt you can come perfectly authentic with samples (but with synthesis anything is possible)/
You can come close enough to fool laymen and make an acceptable "recording" if you're using the right stuff and playing to its strengths, though.

But, for example, the first page of the Star Wars score up there. The brass chordal rhythms of repeated notes sound better to me than most conventional libraries by a landslide and it took the arranger but the time and effort of notating it to achieve that. Such passages sound like nails on a chalkboard with CineBrass Core and many other libraries to me, which still costs several times as much as NotePerformer itself and the former still receives updates to this day.

There are other sample libraries, and techniques with layering samples, etc. that can do that, but it generally requires a lot more effort and the results can be unpredictable. I think if a sample technology is able to offer those results with much less effort, and at a far cheaper price, it could be competitive.

But it seems to be happening more in the notation field. Which is interesting because most composers today are relying on DAWs and many can't even read notation now. You'd think it would be the other way around?

What I do know, is that while the DAW sample libraries sound quality and such has improved vastly, the play-ability or ease-of-use of them (while maintaining a reasonable semblance of realism) seems to be lagging by comparison.

:shrug:
Notation is pretty vague in terms of actual performance instructions, compared to what you have under control in a DAW/sampler. But it takes time and knowledge to program what you have in mind.
You want to input very basic midi data and expect virtuoso performance, sorry, but this won't happen.
Sample libraries are also not real instruments and will never be as flexible or good (in terms of intonation - many orchestras play in some kind of adaptive just tuning in the more emotional moments) as real instruments; you will also need some kind of controller that allows idiomatic performance for the instruments that won't sound like keyboard style repertoire.

Let's say in the future exists a software tool with library of performance styles. It can still make mistakes and sound horrible. Why? Because software is dumb. (If we ever achieve real AI, your composing job is the last thing you should be worried about, because our species will be obsolete...)

Post

There was a system created for Reaper called VI-Folio (which appears to have been abandoned a few years ago, unfortunately) that seems like a stepping stone toward automation in creating orchestral mockups. It's kind of complicated to set up, but (among other things) analyzes the entirety of the MIDI item to enable things such as automation of the selection of articulations based on note length, or choose the type of legato transition based on the length of the overlap (e.g. use portamento for longer overlaps or whatever).

I bring this up only to illustrate my previous point that this sort of innovation is perhaps more suited for changes at the DAW level rather than the sample library level. Granted, this system is not entirely automated, and requires fairly extensive setup by the user in advance, but it's kind of a proof-of-concept for the use of analysis/algorithms to aid in the composition and mockup processes.
AngelCityOutlaw wrote: This is an interesting point because the same could be said of human musicians playing it too. Despite the notated instructions, you're sure to get the player's own interpretation regardless — which I always thought was a desirable aspect, and part of the charm of a musician's performance?
Well that's an interesting perspective I didn't consider. I guess it's really subjective after all. It's especially interesting because a lot of the virtual "orchestral" music we hear today has aspects of it that might be considered impossible (or at least very impractical) for human players to perform; sample libraries have changed the way we write and perceive a lot of styles of music. Some composers write a piece with a particular performer or group in mind from the outset. Will we have different AI performers with different "personalities" and playing styles?

I'm particularly interested in developments regarding physical modeling instruments (like Audio Modeling SWAM stuff) where constant manipulation of several parameters (e.g. bow pressure/direction/position, vibrato speed/intensity, bite, growl, formant, breath noise, microtuning, overblow, etc) are essential in obtaining a convincing sound, as opposed to sample libraries where the human element is baked into the actual recordings used to build the library (so the composer generally only needs to worry about articulation selection and CC1/CC11). Modeled instruments seem an entirely separate beast, with a steeper learning curve than sampled ones, in my opinion.

It's one thing to teach an AI to predict which articulations you want based on MIDI. But it's another thing entirely to teach it how to actually perform a modeled virtual instrument like a human would.

Post

anomandaris1 wrote: Mon Sep 09, 2019 10:59 amNotation is pretty vague in terms of actual performance instructions
I disagree, it's as precise as you make it.
anomandaris1 wrote: Mon Sep 09, 2019 10:59 amYou want to input very basic midi data and expect virtuoso performance, sorry, but this won't happen.
I'm not sure, not just because of Noteperformer, but because someone told me about this, today.

http://orchestraltools.com/soloists_ser ... violin.php

An expressive solo violin that is controlled by velocity switch, many different note lengths sampled, no CC data to screw around with for dynamics, etc.

This is actually the kind of development I'm talking about. You feed the library basic MIDI data and instructions, and you get a performance back akin to a musician reading those instructions off notation. Clearly, this is something that could be done.

What I want, is to make the kinds of mockups I can now, but with tools more autonomous and intelligent, so that I don't have to input everything by hand/MIDI controller all the time. As I said in the OP, the way I see it is that samples are supposed to able to give us an idea of what our music might sound like were it to be recorded. While making it sound "good" or "convincing" has come a long way and is still an area of focus, making them intuitive is much less so.
anomandaris1 wrote: Mon Sep 09, 2019 10:59 am(If we ever achieve real AI, your composing job is the last thing you should be worried about, because our species will be obsolete...)
The thing is that we control the machines, not the other way around. We have the power to say "no" and that's all it takes. I'm not going to go off-topic into politics about AI, but as far as music is concerned if an AI ever can compose music like a human, that doesn't mean we stop composing music, or that we must listen to AI-composed music, or that we have to use AI-assisted tools in their entirety or even at all.
funky lime wrote: Mon Sep 09, 2019 2:54 pm I'm particularly interested in developments regarding physical modeling instruments (like Audio Modeling SWAM stuff) where constant manipulation of several parameters (e.g. bow pressure/direction/position, vibrato speed/intensity, bite, growl, formant, breath noise, microtuning, overblow, etc) are essential in obtaining a convincing sound, as opposed to sample libraries where the human element is baked into the actual recordings used to build the library (so the composer generally only needs to worry about articulation selection and CC1/CC11). Modeled instruments seem an entirely separate beast, with a steeper learning curve than sampled ones, in my opinion.

It's one thing to teach an AI to predict which articulations you want based on MIDI. But it's another thing entirely to teach it how to actually perform a modeled virtual instrument like a human would.
All great points, once again.

Post

I think frankly, the only way is if a better sampler makes itself available. Kontakt is too expensive. People just aren't paying. If they did it would be a different game. But I look to wavetable synths, basically doing the same thing.

I'm working on a sfz sampler with which I can make it easy for people to make patches. Hopefully is good.

Post

funky lime wrote: Sun Sep 08, 2019 1:52 pm

Part of this probably has to do with the ubiquity of Kontakt in the orchestral sample library market. KSP as a scripting language is not well suited for AI or machine learning.
Absolutely - KSP is a nightmare to get things done in, if its not trivially simple that is.

funky lime wrote: Sun Sep 08, 2019 1:52 pm
As far as innovation/stagnation, there is possibly a trend toward proprietary playback engines (Spitfire, Orchestral Tools) or more open environments like HISE. This would, theoretically, open up some possibilities that weren't available in Kontakt. But it remains to be seen which possibilities they might explore.
Exploring as I speak...HISE looks like it might be a rich vein of capabilities going forwards...
VST/AU Developer for Hire

Post

kingozrecords wrote: Wed Sep 11, 2019 4:58 pmI think frankly, the only way is if a better sampler makes itself available. Kontakt is too expensive. People just aren't paying.
Not to go on too much of a tangent here, but this raises a good point. Don't know about you, but I always roll my eyes whenever someone argues how "cheap" all these Kontakt libraries are just because "a real orchestra is way more expensive." It's like, a Ferrari is more expensive than a brand new Hyundai; that doesn't mean a new Hyundai is therefore "cheap". :?

Ilya Efimov's Tin Whistle costs like $80, but it only works with full Kontakt which is around $500 regular price. That's like $600 for a freakin' Tin Whistle library. A real tin whistle by the brand Illya sampled costs about $15 at your local music store. :ud:

Not to mention you see a lot of these libraries that have been around for 20 years still cost hundreds of dollars, or they drop prices by as much as 75%, 4 times a year. That indicates that the product is over-priced if you are having that many huge sales a year.

Really, this whole thread is kind of showing me that Kontakt is an obstacle. :lol:

Post

edit: nevermind, not worth

Post Reply

Return to “Samplers, Sampling & Sample Libraries”