Search the Community

I’m a recent fan of Linus Tech Tips. I stumbled across LTT while researching sound cards a few months ago, and I’ve since watched much of the back catalogue of entertaining and informative videos. I’m especially fond of the videos that investigate the real-world impact of theoretical features like memory speed and many CPU cores. Unfortunately, my main question has not been answered in any LTT video, or anywhere else that I could find: is there an integrated or USB sound device that has good headphone virtual surround, that works well in games. I have seen the LTT/Techquicky video entitled Virtual Surround as Fast As Possible, but the focus is on home theatre equipment, and headphone virtual surround is not mentioned. By headphone virtual surround, I mean accurate simulation of positional surround sound on headphones. I wish to make a clear distinction between accurate virtual surround and accurate audio reproduction. I don’t advocate the use of virtual surround when listening to music. Virtual surround is a trade-off between sound quality and immersion, best suited for gaming. It may also suit movies with surround sound tracks but I have not investigated this. Despite many hours of searching I haven’t found a reputable source of substantive information about virtual surround functionality in audio hardware. I found discussions on forums, but opinions tend towards the consensus that virtual surround has little or no value. Opinions on virtual surround seem to be supported by anecdotal evidence alone. In this absence of evidence, I’ve done own experimentation, including purchasing more sound hardware for comparison. My main motivation for posting this is to stimulate constructive discussion about virtual surround for gaming, and maybe even encourage the creation of expert and professional produced information. At the very least, my research and experimental results may help others. Virtual Surround Works Firstly, proof that headphone virtual surround is real and works. With the assistance of a friend I conducted a blind trial and can provide the following results. The test protocol and configuration is detailed later in this post. Hardware Positional Accuracy1 Avg. Angle Error2 Front/Rear Confusions3 Via HD Audio (control) 50% 45 degrees 4 Sound Blaster X-Fi 81% 25 degrees 2 Sound BlasterX G1 USB 75% 11 degrees 0 ASUS Xonar DGX 81% 8 degrees 0 Positional Accuracy measures how well I could identify the direction of a test sound without seeing the screen. Avg. Angle Error is an average of the angle of difference between my perception and where the sound actually was. Front/Rear Confusion count is the number of times I thought a sound originated from the front when it originated from behind, and vice versa. All the non-control results are better than chance, with only 2 cases where I was not able to correctly identify the front/rear origin. Although there are caveats, which I will get into after detailing the testing protocol, I believe this conclusively proves that I can better perceive the direction of a sound with virtual surround. The X-Fi average angle error was inflated by the two 180-degree front/rear confusions. The Via HD Audio had fewer front/rear confusions than would be expected for pure chance (six expected) due to my guessing strategy. Test Protocol Testing virtual surround is difficult because it’s a subjective experience that’s significantly influenced by visual cues. The goal is to measure the sound hardware’s ability to generate sounds from arbitrary directions that can be accurately perceived. The test protocol should not focus on front/rear position, as virtual surround improves localisation around the sides as well. A test needs to be blind and easily repeatable. While experimenting, I found that it’s important that the environment have only a single sound source, or you will inadvertently train yourself to use other sound sources to “triangulate” the testing source. Other ambient noises and environmental effects must be minimised. After trying several options, I picked the game TOXIKK because it’s free, first person, uses the well-established Unreal Engine with the XAudio2 sound API, and quickly loads into a suitable environment where the player can me moved to a predetermined position. The sound source selected was a waterfall in the Ganesha map. For each test run I stood facing away from the screen with my headphones off. My assistant then oriented the view so that test sound lay in one of eight predetermined directions: front, front right, right, rear right, rear, rear left, left and front left. Once the test source was positioned, I put my headphones on and immediately called out the direction I thought the test sound originates, which was recorded by my assistant. We did this 16 times, twice for each direction. The sequence was randomised so that I had no idea of the test order. I used a pair of Audio Technica ATH-AD700 headphones. After all the results were recorded, the number of correct guesses divided by the total guesses gives the positional accuracy. As there is equal chance a test will be front or rear, anything higher than 50% is better than pure chance. Acclimatisation Period Because of my self-experimentation, I was surprised that my assistant could not localise sounds in virtual surround very accurately, even though her ability to localise real-world sound is easily demonstrable. This suggests the modelling is not accurate enough to simulate reality, or that the experience is unnatural enough to interfere with natural ability. My leading theory is that listening to virtual surround on the X-Fi for the last 9 years has trained me to hear its method of positional audio. I don’t know how long it would take for a listener to acclimatise to virtual surround, and how transferrable the training is to other hardware or software approaches. Final Results My original intention was not to review sound hardware, but in the course of my research I formed some views, which may be useful to others. Subject Sound Blaster X-Fi ASUS Xonar DGX Sound BlasterX G1 USB Via HD Audio (control) Configuration Tricky, but OK once you’re used to it. Initially confusing, but notably flexible. Confusing, with numerous worthless modes. Easy Simulation Immersion focused 7.1 simulation focused Immersion focused None Speaker Model 135-degree side speakers 135-degree side speakers 135-degree side speakers Stereo Accuracy 81% 8% front/rear errors 81% 75% 50% 25% front/rear errors Reverb Low Uncomfortably high at the lowest setting. Medium at the best Surround setting. None Comments Weak in direct front/direct rear positioning, but very clear at the sides. The 7.1 Virtual Speaker option surprisingly useful. Shame about the reverb. Strong front/rear positioning, but somewhat weak around the sides. Very pure sound. No front rear positioning, and weak around the sides. Preference Best Second Equal Second Equal Worst Improving This Protocol Because I could not find a way to precisely orient the view, I decided to use the eight predetermined directions in an attempt to limit the inaccuracy of manually orienting the sound source. This means that exaggerates accuracy when correct, and exaggerates error when wrong. It also makes it possible to use a guessing strategy. Finding a suitably quiet environment and stable loud sound source was also challenging. I chose a 16 sample-count because it meant that each direction could be tested twice. A higher sample count resulted in sound-fatigue where I started to concentrate too hard. This is counterproductive, when localising sound sources should be subconscious. Chosen at random, 16 is an insufficient number of samples to ensure each direction is tested at least once, so a randomised list of two times eight directions was used. Any type of randomisation will occasionally choose sequential tests from the same direction, which was corrected by switching one test with another. Knowing that there was a fixed distribution and that no two tests in order could be the same, I was compelled to make the occasional logical guess, defeating the purpose of the testing. This most benefited the Via HD Audio, since even though all sound seemed like it originated from in front, and it was very hard to distinguish sound sources within 45 degrees of right of left, I used a guessing strategy that artificially inflated the result. The average angle of error would be informative, but the test protocol did not produce meaningful data. Given the fixed orientation of tests, a correct guess is considered 0 degrees of error, which artificially inflates accuracy. Each error is in units of 45 degrees, which will exaggerate error. My ideal solution is a simple custom map for a game that uses a popular engine and sound API. The inner space of the map is large cylinder with a centre mark to stand on and a single clear looping sound source. The wall texture marks 360 degrees anti-clockwise around the inner wall of the cylinder, with the sound positioned at zero. Orienting the test sound at a random angle could then be achieved by moving the crosshair to the correct mark on the wall. The test subject can then indicate the direction with a protractor wheel. This way test angles could be properly randomised, and angle of error measured. Testing multiple engines, sound APIs, different frequencies of sound would provide a more conclusive analysis. I discovered in testing that Unreal Engine and ID Tech 6 use different speaker models, which I have detailed further on in this post. Configuration and Product Details With virtual surround you need time experimenting with settings to get it working properly. Typically it’s about finding the options in the vendor-specific configuration applet that will set Windows Speaker Setup to 5.1/7.1, with the device output to stereo/headphones, and enable positional audio rather than simple down-mixing. The options and behaviour of the configuration program can be confusing and challenging. These are the settings I used in my testing. I’ve included extended notes on each product below. Sound Blaster X-Fi ASUS Xonar DGX Sound BlasterX G1 USB Config Applet Creative Console Audio Center BlasterX Acoustic Engine Pro Speaker Setting in Config Program Headphones Analog Out set to Headphone, Audio Channel set to 8 Channels Headphone (Virtual 7.1 Surround) Virtual Surround Mode X-Fi CMSS-3D Dolby Headphone, DH-1 reference room BlatserX Acoustic Engine Surround on 75 Windows Speaker Setup After choosing Headphones in Creative Console, Windows needs to be set to 7.1 Correctly synchronises with Audio Channel option. Correctly synchronises with Speakers / Headphones Configuration profiles. Setting the Windows Speaker Setup to 7.1 or 5.1 is important because TOXIKK, like most modern games, uses it to set the mixing mode. This can be an issue, which I have detailed further on. Sound Blaster X-Fi I’ve had the Creative Sound Blaster X-Fi ExtremeGamer for nearly 10 years and have been very happy with it. The Creative Console configuration software is a bit silly with its mode switcher but I’m used to it now. The Virtual Surround feature is called CMSS-3D. The only nuisance is that choosing Headphones in Creative Console also sets the Windows Speaker Setup to stereo, which has to be changed back to 7.1 for virtual surround to work. There are some extra options under the Headphone… button, but I have not been able to verify they do anything, so I have left MacroFX and ElevationFilter on Auto. Creative Sound BlasterX Pro-Gaming G1 The Speakers / Headphones options in the configuration applet for the G1 are confusing. There is “Virtual 7.1 Surround Sound”, and “Headphone (Virtual 7.1 Surround)”. Through experimentation I discovered these do quite different things. When selected, Headphone (Virtual 7.1 Surround) will set Windows Speaker Setup to 7.1 and enable a simple down-mixer until you enable Surround in the BlasterX Acoustic Engine. The Surround option can be set from zero which does nothing, to 100 which does a bit too much. Around 20 there’s a noticeable difference between front and rear, but not much side spatialisation. It seems to perform best between 50 and 75, offering good positioning without too much reverb. Over 75 I found the reverb overwhelming. All other Acoustic Engine features were disabled. The similarly named but remarkably different “Virtual 7.1 Surround Sound” mode is not suitable for gaming. This mode seems to map front left and right channels directly into the headphone stereo channels, while adding specialisation effects to the other five channels. This does preserve stereo sources which is an area of weakness for virtual surround, but at the expense of positional accuracy. I do not believe the diagram of the speaker positions in the Speakers/Headphones section of the config applet is accurate. This is part of a wider issue I have discussed further on. ASUS Xonar DGX The Xonar’s Audio Center is something out of the early noughties. The cramped controls are not ordered or labelled very sensibly, but once you figure out its counterintuitive logic it’s easy enough to use. “Audio Channel” is the Windows Speaker Setup, Analog Out is the sound card output, and Dolby Headphone is the Virtual Surround option. There are three room settings that just seem to apply different levels of aggressive reverb. I found HD-1 to be the least offensive. The Audio Center also has a “7.1 Virtual Speaker” option to change the orientation. You must enable Dolby Headphone first, which I found confusing since the checkboxes are in reverse order and have an inconsistent indenting behaviour. This option lets you move virtual speakers around to change the way the virtual surround mix is created. Moving a speaker further away will make sounds from that direction seem quieter. Change the angle of the speaker will change the direction you perceive sounds originate. I found rear sounds a little loud, so I moved the rear speakers further away. I also found that ID Tech 6 engine games use a different speaker model and could correct for that using this option. I believe the GX mode enables an Creative ALchemy-like EAX wrapper, so I left this off. I’m not entirely sure what the HF/Hi-Fi mode does apart from turn off Dobly Headphone. The default positions of the speakers in the 7.1 Virtual Speaker diagram initially look incorrect, but this is one of the rarely found accurate representations of the virtual speakers modelling that is used. Game Configuration Like most modern games, TOXIKK detects the Windows Speaker Setup during launch to set the mixing mode. This can be confirmed in the launch log. Following is Stereo set in Windows Speaker Setup: [0001.12] Init: XAudio2 using 'Speakers (2- Creative SB X-Fi)' : 2 channels at 48 kHz using 32 bits per sample (channel mask 0x3) This is with 7.1 Surround Windows Speaker Setup: [0001.10] Init: XAudio2 using 'Speakers (2- Creative SB X-Fi)' : 8 channels at 48 kHz using 32 bits per sample (channel mask 0x63f) Some games, especially older games, allow you to choose your speaker layout. I have not done any conclusive testing on how this interacts with virtual surround. As long as the game is mixing in more than 2 channels, and the sound card is producing a virtual surround stereo mix, you should see benefits. I found positional accuracy of virtual surround in TOXIKK to be best in 7.1. 5.1 surround/6 channel mode was notably less accurate. Current State Headphone virtual surround suffers from the following problems: It’s hard to configure, and hard to test The terms and concepts are not clearly defined with an overuse of complicated scientific-sounding technical explanations in place of intuitive concepts It’s easily confused with up-mixing and stereo expansion which offer nothing in terms of positional accuracy Some implementations sound unpleasant There hasn’t been any significant innovation in 20 years There are no commonly agreed methods to objectively measure virtual surround accuracy Linus’ previous advice has been buy the most expensive sound card you can afford, or just use on-board because it’s good enough. There isn’t much information to qualify this. Do you Need a Sound Card? Mostly discusses marketing concepts pitting sound cards against external DACs, with no mention of virtual surround features. As previously mentioned, the Techquicky video entitled Virtual Surround as Fast As Possible focuses on 2.1 home theatre equipment, and the idea that virtual surround works by bouncing sound off walls. I admit I didn’t even know this was a thing. I want to stress this isn’t an LTT problem, it’s apparently an industry problem. It reminds me of the old days of graphics cards reviews exclusively concentrating on average FPS for real world games with rarely any measurement of minimums, and no mention of frame pacing. To supplement this you’d get entire articles dedicates to analysing texture filtering boundaries. It seems a bit ridiculous looking back, but the parallels are clear. Dmitry Novoselov of HardwareCanuckes recently posted several videos covering the Sennheiser GSX 1000. The fact sheet on Sennheiser’s website seem to state it supports eight channel mixing; a necessity for accurate virtual surround. However, this comparison using in-game footage of The Witcher III sounds like nothing more than stereo expansion. Virtual surround may be working correctly and accurately, but the testing method doesn’t prove it, and details of the system configuration are not disclosed. This HardwareCanucks video briefly discusses virtual surround in gaming headsets, but the explanation of how it works is simplified to such a degree that it is effectively incorrect. Dmitry then goes on to suggest open backed headphones. Then there are videos like Gaming Audio Myths: Avoid The BS & Save Your Audio Life which are filled with jargon and pseudo-science, contain very little genuine truth, and miss the entire point of “gaming audio”. You only have two ears You only have two ears, and yet you can generally tell where a sound is coming from with astonishing accuracy even with your eyes closed. There are many theories and factors that contribute to this. The Virtual Barber Shop created by QSound Labs is an often cited proof that you don’t need a discrete satellite speaker sound system to have an immersive experience. The virtual barbershop is a binaural recording. As explained by QSound, this is achieved by recording with a pair of microphones placed approximately where your ears are, and to get this level of quality, presumably attached to a model of head and shoulders with realistic ears. This is an entirely fixed perspective experience. You can’t move your head and nothing new can ever happen. Comparing any kind of surround sound to a binaural recording to is like comparing real-time in-game graphics to a pre-rendered Pixar movie. Virtual surround is something different. True Surround A game knows the exact point of origin of any sound. With a true surround sound system (i.e., at least four satellite speakers disturbed around you) and the right software architecture the angle of origin of each sound can be calculated, with respect to your head, and interpolated across the fixed angles of your satellite speakers to achieve positional audio with nothing more than simple panning. The sounds can be attenuated for distance or occlusion, and additional effects such as reverb can be added on top like post processing effects in graphics. This is all relatively straight forward and takes advantage of the two ears, head, shoulders, and brain that you have. Virtual 7.1 Surround The concept of virtual 7.1 surround might initially sound like nonsense. By definition, 7.1 surround means having 7 physical satellite speakers (plus a sub) that should be placed in the prescribed optimal angles so that the position of a sound can be easily produced through panning. What is a “virtual speaker”? Virtual 7.1 is a bit of a misnomer, but essentially it means taking the eight discrete channels mixed by the game, and using signal processing algorithms to create a two-channel mix that that is suitable for your two ears (see “you only have two ears” above). Detail on how this is achieved is easy to find. A graphics analogy is ambient occlusion, which works by analysing the scene and adding additional detail by making an educated guess. The above diagram omits the sub/LFE channel. It’s also worth considering that even if the virtual surround is not very accurate, you could learn to localise sounds using it, like learning to read another language. Up-mixing and Expansion Stereo can be up-mixed to 7.1 or virtual 7.1, but there won’t be any front/rear positional effects because that information is missing from the source. Stereo can be expanded with additional signal processing to give a sense of physical depth, but it does nothing for positional accuracy. If you’re using headphones, up-mixing a source that has been carefully mastered and mixed for stereo, such as music, will decrease its accuracy. Obviously, up-mixing a binaural source to virtual 7.1 would ruin it. Using a graphics analogy, music is comparable to pre-computed light maps; offline lighting calculations using a more sophisticated model than is practical in real-time. Applying ambient occlusion on top would be an unnecessary waste of time and will just make everything darker. Native Virtual Surround Occasionally a game has built-in native virtual surround, implemented by the game developer through a middleware API. In theory, Native Virtual Surround could be a lot better because positional algorithms can be applied to each sound before mixing to the target output channels. If the output is headphones, virtual surround would not need to go via a 7.1 mix because the positional information would already be mixed into the output. If Virtual 7.1 is like ambient occlusion, native virtual surround would be like global illumination; a more accurate simulation with fewer restrictions and shortcuts. I haven’t had many good experiences with this, and I think it probably comes down to the cost-benefit-analysis on licensing, development, and system resource utilisation budget. Marketing Marketing for gaming products can be hyperbolic and misleading. The box for the Creative Sound BlasterX Pro-Gaming G1 is confusing and misleading at best, nonsense at worst. On the back it advertises “True 7.1 HD Surround Sound” which is a liberal use of the word “True” since there is only a single audio jack. It then goes on to state that it has “Native 7.1 Channel input signals for realistic virtualization of gaming effects and audio cues with perfect accuracy and position”, which in my results-backed opinion is an exaggeration. In the GSX 1000 product details, Sennheiser repeatedly misuses the term “binaural”, probably to make an association with the popular virtual barber shop. 3D Positional Audio Virtual surround is not inherently 3D sound. Information about the height of the sound that is included in the 7.1 channel model. 3D sound is theoretically possible, and could even be achieved with virtual surround if the elevation was already mixed into the 7.1 source, but I do not know of any implementations. Fake 7.1 Surround Virtual surround is not “fake” surround. As the creators of the virtual barber shop experience, QSound puts it: “Stereo or surround audio reproduction relies on illusion to create a believable acoustic image or sound stage, in which instruments, voices and sound effects appear to originate from physical locations other than the actual speakers or headphones.” You can take this to mean it’s all fake, or fake is not a subtle enough concept in this case. Hardware vs Software I don’t know how much hardware contributes to virtual surround these days. I imagine that a convincing simulation would involve complex signal processing, however the accuracy of the devices I tested does not imply the calculations are very sophisticated. Before Windows Vista the windows sound stack supported hardware acceleration, A3D and EAX were hardware based. Even if the hardware is irrelevant, the software that comes with the hardware seems to make all the difference. Decent Headphones and a Decent DAC Using a decent pair of well-regarded fit for purpose headphones is always a good idea. No matter how sophisticated the virtual surround algorithms are, if your headphones cannot accurately reproduce the signal, it won’t sound right. The difference between cheap and reputable sound products is instantly noticeable and testable. Regardless of the quality of the headphones, and the price and reputation of the DAC, if virtual surround is not being mixed, you will not be able to tell if a sound originates from in front or behind, under controlled testing conditions. The setup may sound lively, and have a very wide and open sound stage, but the positional information needed for sound localisation simply isn’t there. Problems with Virtual Surround There are problems with virtual surround, most of which stem from the fact that it’s basically a hack. Simulating 7.1 Some virtual surround solutions are putting too much emphasis on simulating 7.1, and room simulation. This approach is neglecting the fact that surround sound with seven physical satellite speakers and a sub is a simulation to begin with and has its own significant constraints, one of which being that the room dimensions interfere with the accuracy of sound reproduction. Clarity Sound clarity is often negatively affected by virtual surround processing, especially sound that is pre-mixed with stereo separation such as music, voice, menu sound effects. The game will typically map stereo sounds to the front left and front right channels, which virtual surround naively interprets as originating somewhere ahead and approximately 90 degrees apart. This effective doubling up of conflicting the signal processing negatively affects clarity and can sound undesirable. Windows After a lot of experimentation, I believe that Windows Sound Speaker Setup is not independent, but closely tied to the sound driver. An example is the optional speakers step after choosing 5.1 Surround or higher. In true surround output mode, this mixes the non-present speakers into the other channels. In virtual surround mode this does nothing. None of the devices I tested have matching functionality in their configuration applets. However, when forced in Speaker Setup, the Creative Console does indicate the speakers are missing. I confirmed that in this mode does reproduce virtual surround. The missing four channels are simply downmixed to the remaining two. Conflicting Speaker Models Through experimentation, I found that contrary to nearly every diagram of 7.1 speaker placement, the actual model used by the Sound Blaster X-Fi, Sound BlasterX G1, and Xonar DGX position the side speakers at 135 degrees; approximately opposite to the front. The rear speakers are then evenly distributed between the side speakers. Only the ASUS Xonar DXG 7.1 Virtual Speaker diagram displays this accurately. Games vary significantly in how they model surround sound. If the speaker layout assumed by the game’s surround mix does not match the virtual speaker layout of the sound driver, positional accuracy is compromised. TOXIKK, which is based on the Unreal Engine and XAudio2 API uses the same 135-degree as the sound drivers. I found that Doom 2016 and Dishonored 2, both of which are based on ID Tech 6 and the WWise API, use a conventional 7.1 layout with side speakers positioned at 90 degrees. In both ID Tech 6 games sounds to the left side are perceived as oriented towards the rear left, while sounds directly I confirmed this by using the Xonar and the 7.1 Virtual Speaker option, by placing the left side satellite speaker directly in front, and moving the others as far away as possible. After loading up a game and finding a suitable sound source, I rotated the view so the sound was loudest, and observed the orientation of the sound. In Doom and Dishonored 2 the sound was loudest when oriented directly to my left. In TOXIKK the sound was oriented approximately at approximately 135-degrees. My assistant was able to easily verify my findings. This means that without correction, virtual surround in ID Tech 6 games is inaccurate on my tested hardware. I assume that this issue affects true surround setups as well. Conclusion I think LTT could help here by including some extra content on sound hardware videos. There is even an opportunity for LTT to make a video explaining what headphone virtual surround is, and dispel some of the myths and misconceptions, while providing some unbiased product reviews. Maybe, like when AMD started putting serious effort into rectifying their frame pacing issues because reputable sites were reporting the problem, this may even lead to better quality products. The relationship between innovation and consumer demand can be complicated. The maximum refresh rate for LCD monitors was 75 Hz for a decade, until manufacturers started pushing 3D. Consumers were not interested in 3D, but the 120 Hz refresh rates that came with it proved popular with gamers. Now high refresh rates are a selling point. The fact that the devices I tested range from less than a year old to 11 years old is evidence of a lack of innovation, and in my opinion is the reason consumers largely lost interest. To put this in context, Intel’s 7th generation Core CPU is a 5-10% performance improvement but discussion centres around Intel’s stagnation. I would love to see a virtual surround reviews of devices like Sennheiser GSX 1000, The Creative Sound BlasterX G5, the choice gaming headsets, even how they compare to Razer Surround. I can’t afford to buy all of these products to test them, but I would buy one of them if I knew it worked. Besides, at this moment I have entirely too many sound devices.