 |
|
Someone you want us to interview? Let us know at news@3dsoundsurge.com |
|
|
|
|
|
| Sound News | Press
Releases | Archives | Week In
Review | Editorials
| Articles | |
|
|
|
||||
|
|||||
|
|
Toni Schneider From Aureal Semiconductor
Toni Schneider, of Aureal Semiconductor, has been kind enough to answer a few questions that we had on wavetracing technology and the Vortex 2. To get the context of some of the questions and answers we suggest that you also read the recent MaximumPC interview with Dave Rossum, Chief Scientist of Creative Labs/ Emu and and a recent in-house interview with some Creative Labs scientists. On top of the answers the questions we asked Toni, he also supplied us with an introductory preamble, so with no further adieu, we will get into Toni's preamble and then our interview. In recent PR pieces, Creative seems to be trying to make a case against Wavetracing. This is nothing new for us. The competition is faced with a technology that they don’t yet have, and they try to make it go away by saying it’s no good, or that no one will use it. Then, when they see it catch on with developers, they jump on the bandwagon. This has happened with several Aureal innovations. Two years ago we brought positional 3D audio and HRTFs to gaming, last year all of our competitors embraced it. It’s happening again this year with occlusions. It’ll happen in the future with Wavetracing. The competition’s case against Wavetracing seems to go like this: Wavetracing is actually the correct way of doing acoustics, but we don't like it because it's too hard to do (I say: we did it), it uses too much CPU (I say: we don't), and we can fake it using reverb which will sound just as good (reviews say: wrong, Wavetracing clearly sounds better). Needless to say, I find this argument pretty non-sensical, since we’ve already proven it wrong with our products. Moreover, Wavetracing is just at the beginning of its evolution, i.e. it’ll get better and better sounding over the next few years. Yet it already sounds better for real-time games than reverb, which has been around for many decades and won’t evolve much further. Most importantly, our Vortex 2 product doesn’t force you to choose between technologies. It features A3D 2.0, a great next generation technology that does reverb for breakfast and adds a bunch of revolutionary new geometry based acoustics methods on top. Creative has only reverb, and they're trying to make a convoluted case for why their solution is better than ours. Bottom line: An Aureal Vortex 2 product will support both Wavetracing (the geometry approach) and reverb (the preset approach) for 3D acoustics. Plenty of developers and reviewers seem to agree that Wavetracing is clearly superior for 3D gaming applications, but in the end that doesn't matter because both methods are supported in leading games for now. That’s why we support both in our products. We want our customers to be able to enjoy all the latest 3D features and content. If our competition wants to fixate on just reverb that’s their choice. In my opinion, Wavetracing, or geometry based acoustics, offers an even more significant enhancement to gaming than positional 3D audio did. I therefore remain convinced that over time, the majority of real-time 3D content will be written to a geometry based method. You can fake acoustics for a while by using static, pre-rendered reverb, but sooner or later, you have to offer a real-time interactive acoustics solution like Wavetracing, or your products will be out of date. Let me draw a graphics analogy: If you want to visually render a 3D scene, you have two basic choices: Pre-rendering or real-time rendering. Pre-rendering gives you more detail and less interactivity. Real-time rendering gives you more interactivity and less detail. Since the exciting part of gaming is the ability to move around in a world and have your environment react to you interactively in real-time, pretty much all 3D games use real-time rendering for 3D graphics. The same will happen for acoustics (Wavetracing = real-time rendering, reverb = pre-rendering). For those of you who want to learn the details of how Wavetracing works and form an opinion for yourself, I'd recommend our A3D white paper at http://www.aureal.com/tech/te_whit.htm. It summarizes over 10 years of 3D audio research and development that Aureal and Crystal River Engineering (a company that merged with Aureal 2 1/2 years ago) have done in this field. This 3D audio work was pioneering work, a lot of it done for and used by NASA, and all of it started and invented long before the current battle over PCI audio. In other words, we didn't just cobble together A3D in response to someone else's product. We invented it from the ground up, we did it for no-compromise research customers who demanded the best possible algorithms and systems, we proved that it worked every step along the way and openly published the research results, we productized it for ultra-high end applications (if A3D wasn't incredibly accurate and powerful, would Navy researchers have put it in fighter jets?), and finally we were the first to bring it to the PC market 2 years ago with the original Monster Sound. By the way, the Wavetracing portion of A3D wasn't invented overnight either. The first Crystal River sound server that featured this technology (a $60,000 bank of 4 Convolvotrons, each of which featured 128 parallel DSP processors, the whole system produced 4 3D sources with 6 wall reflections each) was delivered in 1992. We've been working on the technology since then, and today, you can own a Vortex 2 board for $79 that is 4 times as powerful as that original product. A3D is the real thing, everybody else is playing catch up! 3DSS: Thanks for that introduction Toni. Moving right into the questions, Creative's recent EAX Q&A makes the following statement: "These techniques assume that sound waves are reflected in a "specular" (mirror-like) fashion, which is an approximation neglecting sound diffraction and diffusion phenomena" My understanding of Aureal's Wavetracing technology is that it does take into account diffraction and diffusion by identifying materials. Thus a carpeted surface will absorb more of the wave resulting in a duller sound that the wave reflected off a steel surface. Could you please comment on both my understanding and Creative’s statement? TONI: You are correct in that we take into effect how a wall material affects the spectral shape of a sound's reflection off that wall. I think the statement is referring to the fact that sound waves don't exactly behave like a straight line as they bounce around a room. Wavetracing uses various models of sound wave behavior to account for that. I'm not sure what Creative is trying to say with the statement. If the implication is supposed to be that our reflections are not accurate enough, that would a. be wrong, and b. very funny, since they would be criticizing a feature that don't have in the first place. It's like saying, we offer a $99 Honda, and Aureal offers a $99 Ferrari, and you should buy our Honda, because Aureal’s Ferrari isn’t a Porsche! 3DSS: Again from Creative's Q&A we have the statement that: "The exhaustive calculation of sound reflections for multiple sound sources off many objects is complex. Although the underlying physical principles of the geometrical model are simple (and provide only an approximation of real reflections), it requires serious computing resources." Based on my experience and the experience of others, this is quite correct. Wavetracing calculations are currently resulting in anywhere between a 10 and 20 percent hit on the overall game performance (typically measured by FPS). While acknowledging that these statements as they pertain to A3D 2.0 are stemming from one title (Half Life) can you comment on your progress in bringing Wavetracing CPU usage down to the 5 percent level that has previously been stated as your goal? TONI: Yes, Wavetracing requires a lot of computation. Our first working prototype of A3D 2.0 (spring 98) used most of a PII233 to do a single 3D source with 6 reflections. Today, we do 16 sources with 60 reflections using just a few percent of a PII233. In other words, brute force Wavetracing takes lots of CPU, but we have found lots and lots of ways to speed things up dramatically. Half-Life was the first shipping example of Wavetracing 3 months ago. It sounds awesome and uses up to 15% CPU on certain machines. We have since improved the efficiency by another factor of 10. With all the latest drivers, we have reached our target of sub 5% with the latest A3D 2.0 title, Heretic II. On today’s average gaming rig (300-400Mhz CPU), all you’ll ever notice is that your audio just got amazingly better. And we're not done yet... 3DSS: Still on reflections...Can you explain how you came to the conclusion that 60 reflections could general a compelling audio experience? TONI: We did a lot of listening tests and modeling. We found that 30-60 reflections start to sound really good. More reflections are always better to make things sound even richer, but you can actually get great results with as few as 8 or 16. That's because the most important aspect of reflections is how dynamic they are - the fact that they are computed on the fly based on your position and the geometry and wall materials around you. This makes them sound incredibly interactive, and makes spaces really come alive when you move around in them. You can get those results even with just a handful of reflections. Incidentally, you get none of that interactivity with just reverb =).: 3DSS: A question that seems to come up again and again is that centers around the the statement that an acoustic ray tracing technique can only handle a limited number of reflections and cannot be used to simulate the later reverberation decay. From our Unofficial Vortex2 Faq I quote your previous response on this statement: "Fixed early reflections as part of a reverb are pretty much standard in any reverb. Our reverb engine will model fixed early reflections just like Creative's when a game uses reverb only (via something like EAX). As we've described, the disadvantage of that approach is that things sound somewhat static and flat, because the reverb itself doesn’t change as you move around (the only thing that changes is the wet/dry mix between direct path and reverb). If the game uses A3D 2.0, the idea is to turn the reverb's fixed early reflections off, and substitute dynamically computed and individually rendered early reflections. The reverb can be left in there to do the late field rendering. That way you get it all, the interactivity of early reflections, and the long echoes of the reverb" Now some of the limitations discussed with EAX in the old response would be gone with the enhancements of EAX 2 so what I would like to do is again focus back to how you propose to handle later order reflections (reverb). TONI: I haven’t seen the full EAX 2.0 spec, but from what I've seen in press releases it seems to actually do nothing to address the above shortcomings. EAX 2.0 introduces some form of occlusions, but still doesn't seem to handle reflections or geometry based inputs. 3DSS: Have any of the plans on using reverb to simulate later order reflections while maintaining Wavetracing for first order reflections changed? TONI: No, same plan. Ideally, you want enough reflections to model early and late field on the fly. Some day we will be able to do that. Until then, the best results come from using the reflections you have for early field and fill in (or fake) the rest with reverb. When given a choice of only one (because not all developers have time to put in both), I'd recommend reflections, because I personally love interactivity, and I'd happily give up some late room echoes for a more dynamic gaming experience.3DSS: We know that the first order reflections are tweaked by the materials. What level of customization will a developer have available to tweak reverb that will accompany the first order reflections? TONI: This gets into some yet to be disclosed areas. Needless to say, the developer gets all the basic reverb controls you get with something like EAX (decay time, reverb type, etc.). We are also adding some more interesting controls on top of that. 3DSS: There is a lot of talk on the net about how difficult it is to use A3D 2.0 and how much additional sound work is necessary in order to end up with a compelling audio experience. Can you comment on the level of effort required to implement A3D 2.0 in a title compared to the work that would be necessary to implement both A3D 1.x and an API such as the proposed I3D2? I would appreciate a two part answer here. Part 1 would be for games that use a common engine such as Unreal or Quake and Part 2 would be starting A3D 2.0 implementation in a brand new title. TONI: I wonder where people get that information? Developers that have used A3D 2.0 say that it is an amazingly clean and easy to use API. It seems that it is mostly Creative's marketing department who finds A3D 2.0 very difficult. Of course, they've probably never used or even seen a copy of the API... Part 1: If you're looking at adding 3D audio to an existing game all bets are off. It completely depends on the game’s audio engine. We've had several titles that have added A3D 2.0 support within just a few days. Others take longer, because they weren't designed with this kind of functionality in mind. For example, Unreal is a little more difficult for us because it was architected around a host-based reverb engine. Part 2: A3D 2.0 was actually designed for ground-up use by games. Unlike other 3D audio standards, A3D 2.0 is a complete engine and can do the audio for an entire game. If it is used in that way, its benefits really start to pay off, including ease of implementation. Such A3D 2.0 exclusive applications will start to show up in a few months. 3DSS: What options does a developer have to tweak the sound if they do not like the results of the default geometry based reflections? TONI: There are many reflection controls, including wall material settings (metal, wood, carpet, etc.), reflection intensity, or reflection delay. So far, most developers have used default settings. I'd expect future titles to take more advantage of these controls. 3DSS: Are you planning on supporting I3D2? Please elaborate on the rationale of your response. TONI: We are planning on supporting I3D2. It is intended to be a simple reverb and occlusion API that the industry can standardize on. Occlusions have obviously been one of our big innovations, so we're happy to see it getting adopted more broadly. Reverb isn't quite as advanced as we'd like to see, but at least it's a start. It's a natural process in the PC space for new functionality to be introduced and proven by companies like Aureal whose business it is to innovate, and if the new ideas catch on they get adopted by the mainstream. I3D2 is such a step, and it'll help proliferate high quality audio in games. After I3D2, we hope there will be an I3D3 that will be incorporating even more features of today's A3D 2.0. 3DSS: Do you plan to make A3D 2.0 be fully backwards compatible with I3D2 supporting hardware. To more clearly explain, what I mean is will a game coded to A3D 2.0 when playing on I3D2 hardware support all of the features of I3D2 (namely reverb and occlusions)? Again please elaborate on the rationale behind the answer. TONI: Since I3D2 hasn't been finalized, I can't say for sure. Our first target is to be fully backward compatible with DS3D. If I3D2 has wide adoption, we'll seriously consider wrapping it as well sometime in the future. 3DSS: We have heard one side from Creative Labs scientists, both in a recent issue of MaximumPC magazine and in an in-house Q&A on the Official SBLive site. We would like to offer you this opportunity to tell us why you feel that the geometrical model is superior to the statistical models and preset occlusion that will be in EAX 2.0? TONI: I can't comment on EAX 2.0, I haven't heard it yet. In general, a geometrical model works better because it is much more dynamic. With reverb (or if you prefer the more fancy name for it: statistical model), if I walk into a room the reverb comes on, then nothing changes until I walk into the next room, at which point a different reverb comes on (if the game actually has a reverb for every room in the map, in many cases the reverb only changes in a few spots). With Wavetracing, if I walk into a room, reflections come on, and they constantly change every frame as I move around and duck and walk by pillars, and it just keeps going as you move through the map. It's simply more interactive, and that's what 3D games are all about. In the long run, it makes most sense to base all of a game’s rendering on a single geometry data base: 3D graphics, 3D audio, physics, force feedback, etc.Or let me answer your question in a different way: Creative has probably out-spent us 30:1 when it comes to marketing and PR, yet most reviews I've seen say the exact same thing: the SB Live is a good product, but the Vortex 2 is better, because it has better 3D audio that plain and simple makes your games sound amazing. One of the main reasons we sound better is that our geometrical model is superior. 3DSS: Thanks Toni for taking the time to do this interview with us. If you have any question's or comments on this interview please
direct them to mark@3dsoundsurge.com.
|
Interviews Aureal: Wavetracing and Vortex 2 Aureal : Vortex 2 Drakan - Order of the Flame Sim City 3000 Weapons Factory Sound Card Reviews Hurricane Extreme Initial Review SoundBlaster Live! reference review SoundBlaster Live! Platinum 5.1 European Edition SoundBlaster Live! Platinum 5.1 America's Edition Sound Blaster Live! Player 5.1 Sound Blaster Live! X-Gamer 5.1 Hoontech SoundTrack Digital-XG Boostaroo Headphone Amplifier / Splitter Sound Blaster Live!Drive I and II Sound Blaster Creative Digital I/O 2 |
|
|
|
All content, design and work is ©
2001 - 3D Sound Surge Please respect the copyrights of the articles and writers herein. All copyrights are enforced by 3DSS.