• Register
  • Help
Page 1 of 3 123 LastLast
Results 1 to 10 of 21

Topic: Sampling >> Modelling [CONCEPT/IDEA]

Share/Bookmark
  1. #1

    Sampling >> Modelling [CONCEPT/IDEA]

    This being a SAMPLE LIBRARY forum, I may make some enemies with this post; but I'm REALLY curious. I posted this in a thread on the Cakewalk Sonar forum. Then it occurred to me that this forum might have some great feedback.

    ...just a curiosity question: Has anybody explored the concept of an adaptive modelling engine that morphs to match a sample-set? Instead of developing a specific set of physical modelling algorithms, say you create a modelling engine that analyzes an enormous sample library and then "learns" how to model (i.e. creates its own algorithms) the SAMPLE WAVEFORMS -- instead of the instrument/physics/system itself.

    ...anybody ever hear of such a thing? It seems like it would be fairly easy to model all of the possible outputs of a finite set of sample waveforms. Granted, you wouldn't have the detail and flexibility of a true physical model....but it would be interesting for several reasons:

    1) Sample-set >> Modelling conversion [i.e. for those who've got thousands invested in samples]
    2) It would allow me to model anything I can sample [anything from a short one-shot to a 20GB multi-sample/layer/velocity/etc. monster]
    3) Enormous potential sound palette
    4) Tremendous sound-sculpting possibilities available to those who wish to 'tweak' the auto-created algorithms/models
    5) New 'models' (algorithms) would take minutes/hours to develop -- instead of months/years

    With the colossal CPU power that AMD & Intel have promised is on the horizon, it makes sense in my mind to start using modelling VIs instead of sampling VIs. They eliminate LONG load times, and/or multiple-PC setups, disk streaming issues, etc. [Plus...I wouldn't have to continuously backup my sample library as it expands.] However, a product/system like this would still allow folks with superior sampling equipment/skills to produce efficient and innovative products.


    EDIT: I guess I should clarify here. This idea might not work so well for diverse libraries (like percussion, ethnic instruments, etc.) that have a wide variety of waveforms and sound types. But for something like a piano, strings, choirs, etc., I imagine it should work well.

  2. #2

    Re: Sampling >> Modelling [CONCEPT/IDEA]

    Quote Originally Posted by Infinite5ths
    ...just a curiosity question: Has anybody explored the concept of an adaptive modelling engine that morphs to match a sample-set? Instead of developing a specific set of physical modelling algorithms, say you create a modelling engine that analyzes an enormous sample library and then "learns" how to model (i.e. creates its own algorithms) the SAMPLE WAVEFORMS -- instead of the instrument/physics/system itself.
    I worked on it at university for about a year and a half - then gave up. It's somewhat of a granular model, but thing fall apart pretty quickly simply because of correlating instrument behavior to acoustic response - too many variations - too many non-linearities. That's why the concept exploited by Synful is a dead-end.
    Houston Haynes - Titan Line Music

  3. #3

    Re: Sampling >> Modelling [CONCEPT/IDEA]

    Hey Houston! I just read your IMDB listing and found out that you're from Raleigh -- my home town since age 4.

    Thanks for your feedback. I've read some about granular synthesis....never about a granular MODEL. ...sounds fascinating. Where were you in school, what was your field of study, and how did you end up doing what you do now? [Note: This is all just pure curiosity, especially since you're from around here.]

  4. #4

    Re: Sampling >> Modelling [CONCEPT/IDEA]

    Lots of questions spawned out from a simple retort - I'll need to be more careful.

    I was born at Rex Hospital, if you can believe that (though I spent most of my formative years in the Hendersonville/Asheville area). My folks live in Cameron Park, and one of my sisters lives on the edge of town and is the Assistant Secretary of State for NC, so we definitely have ties to the area even though my wife and I live in Los Angeles now.

    As you might have read in my bio - I spent a few years in a conservatory atmosphere - Petrie School of Music at Converse College, down in Spartanburg, South Carolina. Part of my scholarship/work study involved assisting in the recording of all of the performances on-campus. From there I got bit by the recording bug, and grew tired waiting for them to come up with an electronic music major, so I left and went to UNC-Asheville where Bob Moog had just come on as the Research Professor in Music. It was there that I studied electronic music from the ground up. I started in Bob's first synthesis class with nothing more than a WWII era tone generator and pair of reel-to-reel decks - graduated to modular synths that next semester - and eventually on to my own computer setup (a NeXT machine) where I conducted research on advanced synthesis and automated composition. I received a grant to research the Schillinger System of Music Composition for the possibility of turning it into an application for automated composition/orchestration. After realizing how much I had bitten off (I was young then - things like that came to me slowly) I decided to simply make the research about de-constructing a Bach two-part invention and re-constructing it into a new piece in the new style. I ran out of time (and grant funds) before I created anything of real use, but that's what research is all about, no? I also did some research into CPU-driven synthesis, which was in its infancy then (I used MusicV and a few other research applications to generate audio through the built-in DSP chip on the NeXT - a Motorola 560002 - pretty hefty for its day). I got interested in cascading a series of all-pass filters to re-construct an acoustic sound in real time, and modify it in real time with MIDI parameters. The best I could come up with is reading from a table of wave snippets, based on acoustic sounds I had cut up by hand using the built-in wave editor on the NeXT. What I got out of it was *interesting* but not immediately musically useful. It seems that Synful uses something of the same approach, which is why I haven't even given it a first look. A process like that requires deep and wide tables of data - and *really* good predictive coding to know how and when to "pick up" the right snippet, given the input data.

    It's like looking at one of those old 3-D photo viewers that were so popular in the 19th and early 20th century - amazing for their time but not exactly satisfying if you're looking for IMAX 3-D.
    Houston Haynes - Titan Line Music

  5. #5

    Re: Sampling >> Modelling [CONCEPT/IDEA]

    Wow! Thanks for all the info. That is VERY inspiring. I'm still young...and anxious/happy to bite off more than I can chew. The grant money is a little evasive at present.

    I love to hear stories/bios like that. If you're the type who wants to be involved in EVERYTHING all the time and learn about EVERY topic under the sun (i.e. you think George Lucas has the world's greatest job ever), it's fascinating to hear about others who've successfully combined a wide variety of skills.

    I wish you the very best! And thanks for the feedback, info & life story.

  6. #6

    Re: Sampling >> Modelling [CONCEPT/IDEA]

    Have a look at this thread:

    http://northernsounds.com/forum/showthread.php?t=45424

    I don't know if it's exactly what you mean, but I think it's quite impressive, and he (arwa02) says he's going to release a public beta soon (january/february), so I'd keep an eye on it.

    He says he only needs high quality samples to analyze, and make the "Models", and he could make strings too (in the demo there are only brass and woodwinds). It's all synthesis, and it seems to be CPU friendly (up to 128 instruments running at once)
    -Dani-

  7. #7

    Re: Sampling >> Modelling [CONCEPT/IDEA]

    That demo sounds pretty good - except for one bit on the oboe I think anyone would be fooled in a double-blind test. He mentions that it's pure synthesis, so my guess would be that they've done some sort of analysis/resynthesis, which is not the same as the method I describe above. If their homework was done right, it should go a far piece further than anything in the vein of Synful.
    Houston Haynes - Titan Line Music

  8. #8

    Re: Sampling >> Modelling [CONCEPT/IDEA]

    I hope this project continues to develop, and gets mature. I don't want to have 6 different computers to have a complete orchestra running, as some pro's have.

    I think the new era of sampling is starting from now. I'm sure Gary is going to give us all a nice surprise with its upcoming GOS and GPOA. I don't like VSL approach to sampling, where you need lots of HD space and CPU power to run the whole thing. We need virtual instruments, not terabytes of samples, and Gary is leading the way towards human-sound interaction. I recently recieved my copy of the Strad, and still can't believe the level of playability Gary and his team have obtained.

    Only a few days for the opening of Pandora's box... (NAMM) Let's see what is going on there!
    -Dani-

  9. #9
    Senior Member
    Join Date
    Apr 2005
    Location
    St. John's NL
    Posts
    1,040

    Re: Sampling >> Modelling [CONCEPT/IDEA]

    Quote Originally Posted by ZareOne
    I think the new era of sampling is starting from now. I'm sure Gary is going to give us all a nice surprise with its upcoming GOS and GPOA. I don't like VSL approach to sampling, where you need lots of HD space and CPU power to run the whole thing. We need virtual instruments, not terabytes of samples,....
    Not sure I agree with the entirety of your statement. I haven't followed the modelling side of things very closely, but I do own the strad, and I agree that this is the way things are going. I also agree to you prediction that Gary's gonna blow some minds at NAMM. What I do not agree with is the opinion that terrabytes and terrabytes of samples aren't needed. Everyone's RAM, CPU power, and HD space is growing, and every year people say, "I never thought we'd ever be using Gazziga-Bytes!" My viewpoint on it is that (great programming assumed) a virtual instrument or library is only as good as the number of articulations sampled, their consistency, and the number of alternate samples blah blah. Somewhere in the discussion on articulations one must tangent into prograaming, as the next problem is the kind of thiing Gary has excelled at doing;wth the Strad, he's given us something that plays like a lib with more (infinite) velocity layers samples by means of various dynamic filters, and has used convolution on top of simultaneous realtime speed/depth control of an LFO to give us ultra-exression without have ing to sample a kinds of vibrato. Other great programming features abound, but they all hing on having enough articulations recorded to assemble sounds with real nuance. It is great to use the tools available in a resourceful manner, but I don't think we can (in most cases) justify making smaller libs yet. You can pull some programming tricks to reduce samples needed, but you can't fake the funk. A violin where the long-note samples cut off at the end of the first bow-length is one thing to fit on a DVD (not a complaint, indeed it adds to the realism), but a Sampletekk piano is an entirely different matter!

    Belbin

  10. #10

    Re: Sampling >> Modelling [CONCEPT/IDEA]

    Physical, Spectral, Sample Modeling, or Modulation, or Covulation?

    What's really behind the Stradivari, the Gothriller, and their mysterious Italian creators?


    The unanswered question.

Go Back to forum

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •