The Mazda is using samples that have a distinct intake texture to the exhaust note. They've been in service since GT2. At least they're trying to plug the gaps in authenticity, but it is odd that it's still with older assets and techniques.
...
That will be the benefit of AES when if finally starts to sound realistic, you won't need to plough so much time and money in to recording vehicles, but at this time... for me... it simply isn't working. Mike is excellent at his job, but I imagine he's banging his head against a brick wall right now in frustration.
Or is that just what Sony want you to think?
Ha ha, nope not even close
Imagine it. "Hi, I'm from Codemasters. We don't have a lot of money like Sony and Microsoft but we'd like to record your ultra rare, ultra expensive car. Not only that but we want to put it on track and put additional stress on the engine by red-lining it several times."
...
Just pretend to be Alain de Cadenet... I'm sure nothing could go wrong!
So is AES a thrid party middleware? I wondered if it was simply the Roberson system.
It's no so much that, but with granular if you are accelerating much slower than the source recording, the system has to insert repeating samples to make up for it, so it all gets a tad robotic sounding.
As for 'sample every 1000 revs', don't forget that each doubling of the RPM is an octave (doubling) in pitch so 1000-2000 is an octave pitch change, 2000-3000 is half an octave... etc. etc.
AES is the label we give to the method to distinguish it from sampling, based on a few strings and magic numbers ("illegally") extracted from GT6's encrypted game files by skilled individuals. It appears to be proprietary, and (by ear as much as anything else) similar to a method I came up with independently after being inspired by the example set by Scawen Roberts' approach in LFS.
They are all similar to the Roberson technique (listen
here), in terms of how the different engine configuration sounds are extracted from the generation method, but I suspect they all use distinct generation techniques (if only perhaps a couple of different classes of technique, in respect of the nature of engine sounds). PD's generation technique is far more sophisticated than anything I could produce.
The Roberson method stops short of exhaust tract modeling, AES has a rudimentary implementation at present, most of which is turned off, and I have personally tried a complete "waveguide" approach, which has surprising properties even with a basic generation step.
A major problem I found with wavelets, remembering that I only tried to resynthesise existing samples to sound like different engines, as opposed to a more direct synthesis path, is that it doesn't really do interpolation between samples any better. Not unless you can account for the apparently random changes between sampled points ("engine sounds are stochastic"). In which case, other generation methods are more desirable from a computational efficiency standpoint.
I think PD don't use Wwise or FMOD for the same reasons iRacing abandoned FMOD implementation. And what use is it to AES anyway?
