"Physical Modeling Has Gotten REALLY Good" (YouTube video)

[bxbrkrz]

[huggermugger]

What he has to say about Reason...

[Jackjackdaw]

I can’t bear it. Those modelled violins and saxophones sound awful. I’ve spent a lot of time around acoustic instruments and to my ears it’s not even close. I can see the value for arranging and demoing and they sound fine buried in a mix but that bit when he captions ‘don’t lie bruh, you would 100% believe this is recorded music’ I was like , bruh, you really like the smell of your own farts.

[DaveyG]

That guy talks too much, just like almost everyone else on YouTube.
Out of all the synths he used, Objekt was the only one that occasionally made sounds that made him flinch and scowl. Maybe that's a good thing!

But he's right about there needing to be a cheaper way for non-Reason users to get access to Reason synths.

[integerpoet]

I can’t bear it. Those modelled violins and saxophones sound awful. I’ve spent a lot of time around acoustic instruments and to my ears it’s not even close. I can see the value for arranging and demoing and they sound fine buried in a mix...

I suspect the real strength of something like Objekt is coming up with sounds from fictional physical objects. In other words, what if you randomize until you like it regardless of what real instrument it does or does not resemble? It'll be convincing enough to make the listener imagine what object could possibly have been involved. No two people will imagine exactly the same object because there was no real object. If you don't like its ability — or lack thereof — to output truth, then stop asking for truth and see where that takes you. Evaluating it as if it were a simulation engine seems like a mistake.

I'm reminded of people complaining that Large Language Models like ChatGPT are often wrong. Yeah, if you expect them to tell you about the real world, that'll happen. So don't! Ask for bullshit — i.e. fiction — and it will never lie to you. I've had a lot of fun with ChatGPT that way, but I would never ask it for, say, health advice. Or to be a saxophone.


As an aside, in one of my current projects, I've been playing with the Rhodes preset (one of the Rhodes presets?). If the goal were for Objekt to be a simulator for real-world instruments, they'd want to really nail this one, yes? It's kind of a perfect case.

By itself, it doesn't sound like the real thing because the real thing has "flaws" — deviations from math — which the preset doesn't have. I imagine the preset is trying to model the hammers and tines inside the box, not the entire signal chain you'd find in the real thing. And I like that because I didn't want a lie in the first place.

If I wanted a lie, I'd buy some super-specialized VST which lets me choose the model year and whatever else a historical obsessive would care about. And then, honestly, I would worry that I'd be caught out by someone who has the actual model (and year) of the simulated Rhodes piano and notices some flaw in the simulation that I would never catch.

That entire simulation exercise is, for me, beside the point, but if I did want something closer to the real thing, I could set up a similar-enough signal chain myself. All audio recordings are lies anyway, so I'd bet folding money whatever "flaws" I introduce atop the preset could convincingly mask the un-real-ness in what would be assumed by the non-nerd listener to be artifacts of the recording process.

Ultimately, though, I actually do want the overly perfect idealized essence that the preset is trying to be. I like that it's just a starting point — or even a mere sound design tutorial.

[huggermugger]

I really enjoy working with modelled instruments, but the appeal is not to magically sound like the real thing. It's simply the rich, organic quality of the sounds that physical modelling typically produces. They have a kind of substance, a realism (by which I don't mean realistic), that other synthesized sounds do not. FM and spectral synthesis come close to the same qualities, but they have a sterile quality to me.

[integerpoet]

I really enjoy working with modelled instruments, but the appeal is not to magically sound like the real thing. It's simply the rich, organic quality of the sounds that physical modelling typically produces. They have a kind of substance, a realism (by which I don't mean realistic), that other synthesized sounds do not. FM and spectral synthesis come close to the same qualities, but they have a sterile quality to me.

This forum needs a Like button.

[Jackjackdaw]

I totally get the creative aspect . But this video and thread in particular concerns itself with the synthesis of real life instruments and the accuracy of which p.m. is achieving. I’m still getting the same vibes as when people were dialling in string patches on a Juno and saying how real they sounded, if not worse because we are in uncanny valley territory now.

[integerpoet]

I totally get the creative aspect . But this video and thread in particular concerns itself with the synthesis of real life instruments and the accuracy of which p.m. is achieving. I’m still getting the same vibes as when people were dialling in string patches on a Juno and saying how real they sounded, if not worse because we are in uncanny valley territory now.

Right. What people say about a thing and the thing itself are always separate.

And I kinda like the uncanny valley phase of the evolution of this stuff because (as mentioned above) I never craved the lie in the first place. The uncanny valley might be enough for things like "simulations" of objects that never really existed but nevertheless sound kinda like they might have — and pleasingly mess with your head that way. Nobody can pedantically perform outrage that a synth botched a detail of a fictional object.

I like Benn's channel and I definitely don't mean to make him out to be a hapless rube, but I'm also reminded of early cinema in which the first audiences were terrified a shaky flickering series of low-fidelity black-and-white pictures of an oncoming train was an oncoming train. To them it must have felt like VR goggles feel to us.

It's probably also important to remember that what we think a sax sounds involves the player as well as the sax itself. A player does things with the reed and probably valve manipulations that are difficult or impossible to do with any MIDI controller (even a breath controller). This is not to excuse a synth which markets itself as a simulator and falls short, but maybe part of what we hear is that a sax played with a keyboard sounds like an unfeeling and/or crippled sax player. It might be an interesting experiment to hook a MIDI controller up to a bunch of servos or whatever mounted on a real sax and compare that to a physical model of a sax.

[MuttReason]

I totally get the creative aspect . But this video and thread in particular concerns itself with the synthesis of real life instruments and the accuracy of which p.m. is achieving. I’m still getting the same vibes as when people were dialling in string patches on a Juno and saying how real they sounded, if not worse because we are in uncanny valley territory now.

Right. What people say about a thing and the thing itself are always separate.

And I kinda like the uncanny valley phase of the evolution of this stuff because (as mentioned above) I never craved the lie in the first place. The uncanny valley might be enough for things like "simulations" of objects that never really existed but nevertheless sound kinda like they might have — and pleasingly mess with your head that way. Nobody can pedantically perform outrage that a synth botched a detail of a fictional object.

I like Benn's channel and I definitely don't mean to make him out to be a hapless rube, but I'm also reminded of early cinema in which the first audiences were terrified a shaky flickering series of low-fidelity black-and-white pictures of an oncoming train was an oncoming train. To them it must have felt like VR goggles feel to us.

It's probably also important to remember that what we think a sax sounds involves the player as well as the sax itself. A player does things with the reed and probably valve manipulations that are difficult or impossible to do with any MIDI controller (even a breath controller). This is not to excuse a synth which markets itself as a simulator and falls short, but maybe part of what we hear is that a sax played with a keyboard sounds like an unfeeling and/or crippled sax player. It might be an interesting experiment to hook a MIDI controller up to a bunch of servos or whatever mounted on a real sax and compare that to a physical model of a sax.

I’m a sax player. Been playing for more than 40 years. So I’m always listening carefully when any plugin designer say they’ve nailed the sound of a sax. The most recent modelled versions like SWAM are way better than the older multisampled approaches but compared with the real thing? Not bad, but isolated, solo? Still pretty cheesy. Might be able to get away with short and simple lines but Michael Brecker in software form this is not. In a busy mix, tucked away within a virtual horn section playing horn riffs? Yes, convincing, probably (but I still think Tower of Power can sleep easy at night).

The problem is a sax is unusually complex sonically. There’s a weird combination of wood (the reed), different metals (the alloy composition used hugely affects the sound) and mouthpiece materials, and tiny (like millimetre tiny) shifts in the player’s embouchure fundamentally change the timbre of each note as it’s played. That’s phenomenally hard to capture in software, and there is no MIDI controller of any kind that can replicate the sheer range of inputs from tongue, mouth shape, bite pressure, breath speed, subtle amounts of finger pressure on the keys (they aren’t simply binary switches) and a bunch of other weird and microscopic physical tricks that sax players learn.

Where real world instruments do seem to be modelled brilliantly in software is where the sound source is an impact like a felt hammer on a piano string rather than a sustained note like a sax or violin (which also always sounds less convincing to my ears). I have Modartt Pianoteq, I had 12 years of classical piano lessons as a kid, and honestly the Pianoteq Steinway sounds about as close to an actual Steinway being recorded live as I could ever imagine. It’s pretty damn near indistinguishable.

[integerpoet]

It's probably also important to remember that what we think a sax sounds involves the player as well as the sax itself. A player does things with the reed and probably valve manipulations that are difficult or impossible to do with any MIDI controller (even a breath controller). This is not to excuse a synth which markets itself as a simulator and falls short, but maybe part of what we hear is that a sax played with a keyboard sounds like an unfeeling and/or crippled sax player. It might be an interesting experiment to hook a MIDI controller up to a bunch of servos or whatever mounted on a real sax and compare that to a physical model of a sax.

I’m a sax player. Been playing for more than 40 years. So I’m always listening carefully when any plugin designer say they’ve nailed the sound of a sax. The most recent modelled versions like SWAM are way better than the older multisampled approaches but compared with the real thing? Not bad, but isolated, solo? Still pretty cheesy. Might be able to get away with short and simple lines but Michael Brecker in software form this is not. In a busy mix, tucked away within a virtual horn section playing horn riffs? Yes, convincing, probably (but I still think Tower of Power can sleep easy at night).

The problem is a sax is unusually complex sonically. There’s a weird combination of wood (the reed), different metals (the alloy composition used hugely affects the sound) and mouthpiece materials, and tiny (like millimetre tiny) shifts in the player’s embouchure fundamentally change the timbre of each note as it’s played. That’s phenomenally hard to capture in software, and there is no MIDI controller of any kind that can replicate the sheer range of inputs from tongue, mouth shape, bite pressure, breath speed, subtle amounts of finger pressure on the keys (they aren’t simply binary switches) and a bunch of other weird and microscopic physical tricks that sax players learn.

Where real world instruments do seem to be modelled brilliantly in software is where the sound source is an impact like a felt hammer on a piano string rather than a sustained note like a sax or violin (which also always sounds less convincing to my ears). I have Modartt Pianoteq, I had 12 years of classical piano lessons as a kid, and honestly the Pianoteq Steinway sounds about as close to an actual Steinway being recorded live as I could ever imagine. It’s pretty damn near indistinguishable.

I played sax for about six months in the fifth grade. And yeah: What you said. But I suspect it's even worse.

There are just too many variables for a MIDI controller which isn't, basically, a sax. And at that point why would we use MIDI to pilot a sax simulator when we could just mic a sax? And would anyone really want to fake a performance with that many variables by nudging around little boxes on a screen?

The only way out of this problem, I suspect, is an AI assistant that pilots the variables you don't, and if that's what you want… Well, I suppose it would have a purpose. For TV commercials, where self-expression is kind of the opposite of the point. This is what I mean when I say the absence of a sax player from any physical model is probably an under-appreciated deal-breaker. Because it might even be that a human can't pilot a sax simulator.

Also, I apparently need to check out Pianoteq Steinway.

[avasopht]

Expression is always a key component, and when it's not, it probably isn't a big deal if it's not modelled