Jump to content

Sampling vs. Physical Modelling Synthesis


Mediterranean

Recommended Posts

  • Members

 

What is the difference? And what are the advantages/disadvantages of one over the other? Which one provides a better emulation of real acoustic instruments such as pianos, wind instruments, strings, drums, etc?


Thanks a lot:)

 

 

Sampling is pretending to be able to play an instrument you shouldn't touch.

 

That's why all the workstation boards use samples. They are boards for pretenders.

 

PM on the other hand is creative and you can actually make an instrument out of it instead of a rompler.

Link to comment
Share on other sites

  • Members

Sampling - a static picture of a sound. Physical modeling - a freaking holodeck from Star Trek in comparison.

 

 

PM can produce all the nuances of the instrument that are very hard or impossible to get right with just samples. Hopefully I've made a clear cut difference here. :)

Link to comment
Share on other sites

  • Members

Samples are recordings of an actual instrument. To imitate the changing number of overtones based on things like the force used to blow/strike the instrument, they can switch other samples in, apply a dynamic filter, or do other tricks to achieve accurate tonal variety.

 

Physical modelling seeks to describe the behavior of the instrument in mathematical models and compute the output in real time. Advantages are much more sensitive variation to performer input. Disadvantages are that the basic tone of a complex instrument like piano is harder to describe in a physical model than just recording a sample.

 

Modelling is the way of the future, samples are going to eventually be used only for recording/playback of phrases and other found sounds. Right now, modelling is used on all the best Hammond organ emulations (VB-3, KeyB), and for acoustic piano you have Pianoteq, which is completely modelled and arguably the most responsive piano out there. While something like Ivory sounds more like a particular Steinway or Yamaha piano, something like Pianoteq can sound like a Steinaha or even a Bosenway.

 

Modelling instruments require lower memory than sampled, and generally require higher CPU.

 

Many instruments from major manufacturers like Nord, Roland, and Yamaha use a hybrid approach - samples for the main sound and modelling to add extra components like soundboard resonance.

Link to comment
Share on other sites

  • Members

Sampling is for the simply sane. Physical modeling is for demented players and programmers. Mate_stubb pretty much nailed it. Mathematical models of everything from lips, tongues, larynxes, bows, etc ad infintum. That's a lot of math but the results can be astounding and far beyond imitative into the realm of that holodeck. Try bowing a bell with a sample.

Link to comment
Share on other sites

  • Members

Sampling captures, accurately, the tonal qualities of another instrument in its least dynamic stages. Some of that dynamism can be recreated with multisampling and, in recent years, the complexity of multisampling leaves very little of the original instrument's performative nuances without representation.

 

The most effective approach these days is a mix of sampling with, basically, whatever you call it, physical modelling of the microarticulations that occur primarily during the attack stage of a sound's envelope, where the most dynamic and timbral variation is occurring in the greatest range. That dynamism can be represented more effectively with modelling, and then that modelling can blend into the sampled part of the sound, so you get the best of both worlds.

 

This multiple approach is something that, say, Roland has been evolving since the advent of the D-50.

Link to comment
Share on other sites

  • Members

its easy to explain how physical modelling works but kind of hard to describe what it sounds like. i think the real beauty of pm is being able to take a normal sound and kinda warp it.

 

in this example he makes a flute sound turn into marbles around 2:00. the rest of the tutorial shows kinda how sculpture works but more importantly what it sounds like

 

[video=vimeo;14850214]

 

the downfall imo of many sample libraries is they just get insanely large like the symphonic cube at 550 gb. but for realistic sounding stuff i'd stick with them personally.

 

in the not-so-distant future and with a really powerful computer something like pianoteq run through something like acustica nebula3 on steroids might achieve the sweet tones associated with scarbee, goldbaby, etc

Link to comment
Share on other sites

  • Members

A massive synth pad was created from that Bach sequence. Damn! I love pads!

 

About Pianoteq: for something that is not sampled, it sounds extraordinary to my ears. I think it surpasses many, if not all of the ROMpler pianos I have heard.

 

Thank you to each and everyone for your replies:)

Link to comment
Share on other sites

  • Members

 

Pianoteq definitely surpasses just about all rompler pianos I've tried in projection, tonal and timbral detail, decay richness and length, sympathetic resonance, velocity response, and pedalling techniques.

 

 

Ah, we have the same qualitative taste in pianos:)

 

But how was that piano created? from scratch using very complex mathematical formulas and algorithms to create the sound of an acoustic piano or was there a basic audio source they started with?

 

Also, when one tweaks a given patch/waveform on a synth to try to make it sound like a certain instrument or to create a brand-new one, is that patch/waveform being physically modelled?

Link to comment
Share on other sites

  • Members

Modartt analyses the samples at various pitch and velocity ranges, then boils that down to differential equations (=mathematical model of the sound creating process), then creates the algorithms which are supposed to solve the equations in fastest possible time (=realtime). Not the easiest job in the world, but the creator of Pianoteq has a Ph.D in applied mathematics, AND a history as a piano tuner of a vast amount of different pianos ;)

 

Regarding your second question, nope, that wouldn't be physical modelling, unless you're using noise and/or impulse waveforms through one or more comb filters. That's Karplus-Strong physical modelling, then.

Link to comment
Share on other sites

  • Members

At this point until PM developes more they both have their purposes in music. And lets not forget the importance of sampling machines.

We need sample playback romplers for the budget minded musician to gig.

 

I don't want to return to the 80's where keyboardist in cover bands were limited by their budget as to whether they could even cover an entire top 40 show.

 

$2000.00 US. was usually the entry fee for two limited analog boards or one DX7.

Link to comment
Share on other sites

  • Members

 

At this point until PM developes more they both have their purposes in music. And lets not forget the importance of sampling machines.

We need sample playback romplers for the budget minded musician to gig.


I don't want to return to the 80's where keyboardist in cover bands were limited by their budget as to whether they could even cover an entire top 40 show.


$2000.00 US. was usually the entry fee for two limited analog boards or one DX7.

 

 

I remember those days. I had a Prophet 5 ($3k) and a CP-70 ($3.5K). When the band stopped touring and started playing local, I then sold the Yamaha and bought a DX-7 ($1.8K). These were 1980 dollars. Neither were multi-timbral but my band at the time covered all those Top40 songs of the day and that was all I had to do it with. Had to switch patches fast!

Link to comment
Share on other sites

  • Members

Modartt analyses the samples at various pitch and velocity ranges, then boils that down to differential equations (=mathematical model of the sound creating process), then creates the algorithms which are supposed to solve the equations in fastest possible time (=realtime). Not the easiest job in the world, but the creator of Pianoteq has a Ph.D in applied mathematics, AND a history as a piano tuner of a vast amount of different pianos
;)

 

Apparently there is a patent (7,915,515 B2 according to the Pianoteq web site) on the Pianoteq model. Being on vacation this week, I am not going anywhere near a network related to work, where it would be easy to look this up and have a look at the details. That will make a nice little "back to work" activity on Monday :D

 

Some intro stuff on physical modeling that might be of interest:

 

Professor James Clark's web page

especially his

Advanced programming techniques for modular synthesizers

 

Professor Julius Smith's web page

especially

Physical Audio Signal Processing

 

The latter work by Professor Smith is really interesting and immediately accessible to a person with a good background in EE and systems theory (probably at the graduate level). The stuff linked is pretty much based on linear systems theory. I am curious whether and how non-linearity is incorporated in Pianoteq.

Link to comment
Share on other sites

  • Members

 

Modartt analyses the samples at various pitch and velocity ranges, then boils that down to differential equations (=mathematical model of the sound creating process)

which proves that "physical modeling" isn't really quite strictly modeling of physical parts. It's a combination of hueristics, sample analysis and reconstruction, and simulating effects caused by physical parts. But it is rather different from traditional sampling and deserves its own name. Someday when they actually start modeling individual physical components, I wonder what they'll call it!

 

Another aspect in the gray area is convolutions, which are often used to model the soundboard resonance. These can exactly duplicate any linear physical process, if you can get the exact step response from that physical process. Well, that last bit is sampling of a sort. But not the kind we normally call sampling for digital instruments.

 

My favorite piano is TruePianos, which is modeled. But I've never had a chance to try Ivory. I've played the PianoTeq demo, but only using the presets. To my ear, they all sounded artificial compared to TruePianos. Well into the list of presets was, IIRC, a Yamaha that I did like a lot, but TP still beat it by a fair margin.

 

In any case, I think that when used well, samples can be amazingly good. I haven't played a modeled Rhodes that's as good as Scarbee or even my own (free) Rhodes soundfont. That's a bit odd because I suspect that a Rhodes is far easier to imitate than a piano, but no doubt the market is much smaller.

 

One problem with samples and the currently available memory is that more memory alone does not make a good sample set, and many of the best sample sets don't require vast amounts of memory. The skill of the people creating the sample set matters far more than sheer memory. Yet scads of creators of sample sets just gobble up memory and use that as a marketing PLUS.

 

BTW, excellent summary Mate!

Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

×
×
  • Create New...