Is physically modeled stuff actually worth the hype?

[tapper mike]

We have a perfectly good instrument forum here at KVR.

Not that I'm trying to dissuade anyone from advocating their opinion.

[mystran]

SWAM strings can be quite convincing, but it takes a lot of work from the user. The new Expressive E Soliste also sounds quite good.

This is getting a bit off-topic, so please excuse me...

...but I used to actually play violin to some level of proficiency and let me tell you the Paganini in the video posted sounds very stiff and unnatural, because it kinds sounds like individual notes slapped together and that's not really how you play violin. The bow is more like a paint brush and you'd constantly varying both the pressure and speed to shape the articulation even within the notes, so the music breathes.

Here is how Caprice n.24 should sound:


Now, to get back to topic, there's some of this for guitar as well: there's so many ways the player can intentionally or unintentionally change the way in which the strings vibrate that there is both a huge control problem with expression and also with all the "practical limitations" and little mistakes that even if the "physical" model is theoretically capable of producing the perfect sound, it'll be next to impossible to really match the real thing.

On the other hand, with samples you're obviously stuck with finite choice, so even if it is more difficult for physical modelling to get the overall sound right, it definitely has better potential for varied articulation... but for really expressive instruments, I just don't see it every sounding completely like the real thing unless you also do a physical model of the player.

[mystran]

I mean... with violin beyond a certain point of proficiency, the way you "express" the music become perhaps the main thing to practice. You certainly have to practice your technique as well, but that's really just the prerequisite for music.

I think it's the exact same thing with guitars as well. Pretty much all Paul Gilbert (quite proficient technically) ever talks about is how to articulate notes and phrase things and somewhere in this video Satriani (also quite proficient technically) casually says something along the lines of: you'll get as fast a you'll ever get in a few years, then it's all about expression:


With something like a piano, things are a bit easier, 'cos there's not that many variables you can apply to either hitting or releasing a key (or pedal). There's some, but it's not like you can physically touch the strings.

So... I don't know... if the choice is between samples vs. physical modelling, then modelling could be worth it. If the choice is between physical modelling and someone (skilled) playing the thing for real, I don't think physical modelling has much of a chance.

[jamcat]

...but I used to actually play violin to some level of proficiency and let me tell you the Paganini in the video posted sounds very stiff and unnatural, because it kinds sounds like individual notes slapped together and that's not really how you play violin. The bow is more like a paint brush and you'd constantly varying both the pressure and speed to shape the articulation even within the notes, so the music breathes.

As I said, you really have to put the work in with SWAM instruments. You absolutely could program a Paganini performance to have the qualities you describe. That’s the greatest strength of SWAM over sampled instruments. I’m assuming the Paganini piece in the video was a real-time performance, which is why it has some distinct keyboard performance qualities to it. But a sample-based instrument would not and could not perform any better, though SWAM easily could with some deliberate programming choices.

I chose the SWAM videos more to show off the basic tonal quality of the instruments, as they are 100% physically modeled and do not rely on any sampled content whatsoever. Both of these videos show off the latest SWAM engine, which saw some significant upgrades late last year with version 3.10. I mainly just wanted to make sure the videos I selected showcased the latest engine.

There are some 3rd party videos done with older versions of SWAM that painstakingly recreate specific performances of pieces and present both. These performances are nearly indistinguishable.

Here is one such piece programmed by Luigi Zaccheo:


Original performance:


And here is a different SWAM performance of the same Paganini piece, this time programmed by Luigi Zaccheo:

[mystran]

There are some 3rd party videos done with older versions of SWAM that painstakingly recreate specific performances of pieces and present both. These performances are nearly indistinguishable.

I think these still sound quite stiff and unnatural. Maybe they would be fine to the ears of a casual listener, but at least if you know how to play yourself, you start to notice these little details that just seem off and would be physically impossible to perform like that.

[tapper mike]

I've seen my fair share of violin performances where the violinist plays stiff because they've played to hard for too long. Wrist and finger cramping happens to many musicians and can affect performance as well.

[tapper mike]

Here is what I'm seeing from my favorite YT classical film scoring guys.

They aren't using swam.

They are using lots of key switches and fader boxes.

[mystran]

I've seen my fair share of violin performances where the violinist plays stiff because they've played to hard for too long. Wrist and finger cramping happens to many musicians and can affect performance as well.

Well, it's not really about playing too hard ... it's just about not keeping your hand relaxed enough.

Exact same thing as picking at higher speeds on a guitar. Pretty much the same muscles and you can even use pretty much the same motions (well, if you play with what Troy Grady calls "reverse dart-thrower" then that'd be closed to the contrabass bow-grip, I guess, but that and the "violin motion" are pretty much the good ones), just the details and posture and a bit different ... and exact same problems with the natural instinct to try to cramp the entire hand from fingers to wrist to elbow to shoulder in order to force speed... with sort of "works" (in the short term), but isn't really necessary.

ps. Finger-style guitar is obviously bit of a different beast, but flat-picking and bowing have a lot of similarities and ... lot of the same challenges with keeping things relaxed. You use your elbow a lot more (or at least in a much wider range) for the bow, for sure, but other than that.

[Charlie Firpo]

...but I used to actually play violin to some level of proficiency and let me tell you the Paganini in the video posted sounds very stiff and unnatural, because it kinds sounds like individual notes slapped together and that's not really how you play violin. The bow is more like a paint brush and you'd constantly varying both the pressure and speed to shape the articulation even within the notes, so the music breathes.

As I said, you really have to put the work in with SWAM instruments. You absolutely could program a Paganini performance to have the qualities you describe. That’s the greatest strength of SWAM over sampled instruments. I’m assuming the Paganini piece in the video was a real-time performance, which is why it has some distinct keyboard performance qualities to it. But a sample-based instrument would not and could not perform any better, though SWAM easily could with some deliberate programming choices.

I chose the SWAM videos more to show off the basic tonal quality of the instruments, as they are 100% physically modeled and do not rely on any sampled content whatsoever. Both of these videos show off the latest SWAM engine, which saw some significant upgrades late last year with version 3.10. I mainly just wanted to make sure the videos I selected showcased the latest engine.

There are some 3rd party videos done with older versions of SWAM that painstakingly recreate specific performances of pieces and present both. These performances are nearly indistinguishable.

Here is one such piece programmed by Luigi Zaccheo:


Original performance:


And here is a different SWAM performance of the same Paganini piece, this time programmed by Luigi Zaccheo:

I thought they sounded beautiful. The mix of the swam was a little bit to loud. But i really liked the expression.

[tapper mike]

A few observations.

Many in the instrument forum seem to have animosity regarding expressive technique in performance for recording classical performance especially when key switching and cc or mpe capable controllers are involved. Thankfully that appears not to be the case. I actually am quite a fan of mpe and live usage of expression rather than post processing even though I can see how the opinions I've expressed about swam may lead one to the contrary.

Those of us linnstrumentalists may not be aware that the linnstrument has a strum mode
viewtopic.php?t=569371
I do not use that function, I get around pretty well without it.

The things that work for me may not work for you. I love the linnstrument.

I own two 200 models. One has the standard surface and the second has a bumpy surface


Roger Linn may love swam. I think it's over priced and I think I do well enough with sample libraries. I have sufficient storage, ram and processing power. I don't need to save any of that for recording/performing and if I did I could do what everyone else does freeze the tracks I'm not recording to save in that area.

I founded and am the admin of the Linnstrument forum on FB. I have fellow admins who check in and make sure those that seek membership get a quick response. A large amount of members joining want to try before they buy. And some join simply to sell their linnstruments.

It's work to learn the linnstrument regardless of how many instruments you may already play. If you aren't willing to go down to brass tacks practice scales and chords on it then you'll generally muck about have fun for a little bit get bored and move on.

For individual orchestral instruments I'm more impressed with LUX than I am with SWAM (ymmv)

[CapnLockheed]

A lot of new topics from a new member...hmmm

Says the guy that's only been a member since January and only has 271 posts...hrmnn...LOL!!

Not quite

Oh, did you have make a new profile after getting banned?

[Seafire Mk2]

A lot of new topics from a new member...hmmm

Says the guy that's only been a member since January and only has 271 posts...hrmnn...LOL!!

Not quite

Oh, did you have make a new profile after getting banned?

Mk1 lost access due to new security measures and old email being unavailable

[Uncle E]

I think these still sound quite stiff and unnatural. Maybe they would be fine to the ears of a casual listener, but at least if you know how to play yourself, you start to notice these little details that just seem off and would be physically impossible to perform like that.

Yamaha arrangers add random little slides and string noises to invoke realism. Like you said, if you're a player, you'll hear the incongruities and be put off by them, but I find them to be surprisingly convincing in the context of a full mix.

[jamcat]

Arranging prerecorded single hit samples is going to produce a more incongruous performance than automating a physically modeled violin performance 100% of the time.

So it’s kind of a dumb debate, because SWAM is absolutely going to win every time, unless you’re recording a real violinist in your studio.

Mystran’s argument seems to be that SWAM is good, but still not good enough to fool trained violin virtuosos, without offering a viable alternative to SWAM that would actually be better.

[mystran]

Mystran’s argument seems to be that SWAM is good, but still not good enough to fool trained violin virtuosos, without offering a viable alternative to SWAM that would actually be better.

Right. I think the core models themselves are pretty convincing and it's mostly a control problem with the whole physics system that is the bow ... and I'd absolutely love to have a good solution to that, but sadly I don't.

That said, I don't really see samples doing any better (quite the contrary) unless you happen to have samples of the exact phrases you want ... and then then splicing them together convincingly is less than trivial. I just think that in general we should not completely write off the value of real musicians yet.

ps. For context, while I've not done much physical modelling specifically, I spend my days listening to my own algorithms and being equally unhappy with them, so I'm probably hyper critical, but it's because I just want the world to have the best sounding music it possibly can have.