@
Whistle Snap could you elaborate on that at all? I mean, specific examples of what sounds were shared with what cars and how they differed? I'd love to be able to hear this effect, even more to be able to get involved with the hybriding this time since the audio designation has been separated from the engine designation this time (which might be indicative of being more update-friendly, but not necessarily.)
It's not about listening past my prejudice, that's not what I said. It's that I legitimately don't know what I'm supposed to be listening for in order to identify a sound simulation. And even if I did know, I doubt my ability to hear it without practise.
Without being able to hear something that I can clearly identify as sound simulation, I don't think it's unreasonable for me to doubt that they're actually doing it. Russell's Teapot and all that.
That's fine, but you used the word bodge, did you not? And you were adamant that PD are wasting their time building things from scratch instead of using an actual bodge! Whilst technically an argument from ignorance, it's really neither here nor there. I'm fully with you on finding an objective means of describing the phenomena at hand. I just question whether it's within my capabilities.
Are people really doing all this by ear with no frequency analysis or numerical modelling? I assumed that there would be some way to analyse the sound to identify the jumps between samples even if you can't hear them.
No, there isn't, really - I mean you could record some sampler output and run it through a spectrogram generator and compare it to a similar situation in a real recording, but the results would be difficult to control and there will be a lot of visual noise, in the sense of what changes you're looking for will be obscured by changes that aren't relevant. Our ear-brains are by far the most sophisticated piece of audio equipment we have access to, figure out how to rely on them "objectively" (it's remarkably useful). Besides, the evaluation of any synthesis scheme always involves a psychologically controlled test (e.g. "double-blind") for the perception of things like "accuracy" and "quality", since most schemes inherently introduce approximations or deliberate omissions based on previous tests, and specifically to gain performance at no expense of "quality". Sound as we perceive it is a construct of our brains, after all, and analytical results don't always correlate to the experiential "reality".
What separates the "new breed" of synthesis (now over 50 years old...) from the analytical old-school, is that they're built forwards from first principles using interactive whole systems, instead of reverse engineered by isolated parts - learning the first principles will go a long way to helping anyone's understanding, and they're not really audio-specific (except the psychoacoustic stuff, which is worth a quick look also).
Chippy did a good job some time ago demonstrating what it sounds like as you use less and less samples to build a rev range. There was a clear and demonstrable difference, even to a layman.
What difference would you expect to hear between a good set of samples, and a synthesised/simulated rev range? And then how could you identify that, without using your ears?
Yes, that demo was great; I wish he'd shown the effect of sample length as well, though. People are forced to try that for themselves to hear the effect, still. I can't at this point put a demo together, as I'm relegated to my laptop for the time being (convenient it most certainly is not).
Samples have a typical pitch shifting sound that you just have to listen through. You can see how that's "wrong" with a spectrum plot: think about singing two separate notes, then singing one note and pitch shifting it to the second, better yet grab a mic and try it for yourself. Knowing what pitch shifting sounds like in isolation might make it easier to recognise it in a multi-sampler, but that's ear training again. If you want to do it analytically, you'd have to design the experimental method yourself, as it's making my head hurt trying to pin it all down.
Generated methods don't suffer from the same spectral compressing and stretching: look into the idea of formants in speech synthesis, because pitch shifting doesn't preserve formants when it should, and relatively small inaccuracies in the formants at constant perceived pitch result in the perception of different vowels - see
here. There are analogues to be found in engine sounds; indeed "source-filter model" is a speech synthesis paradigm (and is a
very broad category), and is an area T10 are potentially looking at for the future of its engine sounds.
I don't want to build sounds, I want to analyse sounds. I suspect you're approaching this from a holistic perspective, because you have the skills and you can. I can't (or can't spend six months learning the entire field from scratch), and I need to be able to isolate the differences that separates the old GT solution from the new. I'm starting from the opposite end to someone designing a sound simulation, I merely need to be able to pick their sound sim from other techniques.
Think of it as a sound Turing Test. I need to be able to identify a sound simulation without being able to see whether it's a sound simulation or not. That doesn't necessarily require knowing how to build a sound sim.
True, but building the test does. That's beyond my abilities. At least getting a feel for the way the sounds are put together should get your critical ear twitching nicely, I never intended to simulate sounds at first either, I just wanted to know how they work (because I was investigating the possibility space of engine configurations from a physics standpoint, and got to wondering precisely how it affects sound, and stumbled across the "secret" wholly accidentally).
All the things you describe about the sounds of the Red Bull are great, but I don't see how any of them lead inescapably to the conclusion that it's sound simulation. If we're really going to go down this path of assuming that they're simulating sound I don't want to be giving the benefit of the doubt, I want to be sure.
Sound simulation is a tricky one, and arguably samples are a form of that (if you take the word "simulate" literally). It is inarguable that their exhaust rasp is physically motivated, though (take the time to think about what "rasp" actually is). "Simulation" or otherwise, it behaves in a way that cannot be reproduced except by these aforementioned "source-filter"-like interactions.
The parallels between the exhaust pipe resonance on the RB cars and a resonant filter on a traditional synthesiser (e.g.
subtractive) should be obvious. It's how you construct that "filter", and more importantly in this case, the "source", that is the clever bit. As with subtractive synth, the overall "flavour" of the sound tends to be dominated more by the behaviour of the filter than it does the source, so the source cannot easily be scrutinised, but the filter can be in this instance. In speech synthesis, you can just about use that filter analysis to construct an inverse filter and hence discover the source, but engine sounds are highly non-linear in both the "source" and the "filter", so that is impossible (speech is marginally non-linear, mostly in the source: the vocal cords). However, the inverse-filter process is not totally useless for practical re-synthesis of engine sounds (just no good for proper analysis).
As for being sure, that's up to you. I don't expect anyone to be sure based on my observations, but if you want to understand it to the same degree, then I don't see how you can avoid the leg work. I'm not the Alan Turing of sound synthesis by any stretch of anyone's imagination, so I can't make it easier for you - maybe someone else can.
If you have access to a copy of Live For Speed, that has a fully adjustable "source-filter"-like synthesis method that was the inspiration for my initial attempts (it is not in itself a true simulation). Notice how smooth and expressive it is, despite the laughably low fidelity. I love it.
I'm still going to read through those procedural sound generation tutorials because I think they'll be interesting, but I doubt that it's going to give me what I'm looking for. They're making sounds, not analysing sounds. The knowledge overlaps to some extent, but it's not necessarily identical.
You'll be surprised, I think; you should see there is a lot of analysing, and tons of research, that goes into each of those patches - how do you know how to reproduce a sound if you don't actually know what the sound is? I think you're being simplistic again, and you should concentrate on that holism concept you correctly identified. Equally, you ought to have read it a bit and then commented, because prejudice!
To experts in sound synthesis (not me), a rich, natural and inherent level of interactivity in a synthetic version of a real-life sound's controls usually points to a model-based approach (physical or otherwise), although this is not apparent to any non-interactive listeners (the example usually given is that a virtual instrument is preferably model-based as far as the musician is concerned, but the audience is agnostic.) The very act of interacting with the new sounds and recognising that they feel different is proof alone that they are different.
The proof you personally require as to their nature can only really be gained by trying to get a sampler to exhibit the same expressivity - again, start with the resonant rasp effect: dynamic resonance of any kind is
immediately indicative of a model of some sort, the result achieved is indicative of the motivation for that model, i.e. physical "simulation" or otherwise (because of psychoacoustic considerations).
Just that you know who you talking to you know? Because of the Thing with open ended stuff... and knowledge
