With a generic KEMAR dummy-head HRTF, listeners in a 2026 VR study confused front and back on 17.2% of trials, more than double the 8.1% they scored with their own measured HRTFs (Pirard and Poole, AES Conference on Audio for Virtual and Augmented Reality 2026). The same study found that left-right judgments barely moved across five HRTF types, which matches decades of earlier work. Head tracking is the strongest single fix in the data: weighted front-back confusions fell from 16.5% in static headphone playback to 6.3% with head-tracked rendering (Oberem et al., bioRxiv 2020). Training helps too, with fitted spherical error dropping from 43.4 to 25.5 degrees in under two hours of practice (Steadman et al., Scientific Reports 2019). We aggregated data from the Journal of the Acoustical Society of America, the Journal of the Audio Engineering Society, Scientific Reports, Trends in Hearing, PLOS Computational Biology, Applied Sciences, and AES conference papers. For market and adoption figures, see our separate spatial audio statistics roundup; this page covers only how accurately people hear direction.
A quick glossary, because these terms get reused below. An HRTF (head-related transfer function) is the direction-dependent filtering your head, torso and outer ears apply to sound before it reaches the eardrum; virtual 7.1 and spatial audio work by applying an HRTF over headphones. A generic HRTF comes from a dummy head (such as KEMAR or the Neumann KU100) or from another person. Lateral angle means how far left or right a sound is; polar angle means where it sits on the circle around the ear axis, which covers up-down and front-back.
TL;DR
- Polar-angle error was 48.8° with a generic KEMAR HRTF vs 30.1° with the listener’s own measured HRTF (Pirard and Poole, AES AVAR 2026).
- A generic HRTF produced 21 front-back confusions in 78 responses (27%) vs 0% with individualized HRTFs (Jenny and Reuter, JMIR Serious Games 2020).
- Head-tracked playback cut weighted front-back confusions from 16.5% to 6.3% (Oberem et al., bioRxiv 2020).
- One 10° head turn lowered front-back confusions from 10.7% to 1.5% with real loudspeakers (McLachlan et al., PLOS Computational Biology 2025).
- Head movements through about 32° of azimuth almost eliminated front-back confusions (McAnally and Martin, Frontiers in Neuroscience 2014).
- With non-individual HRTFs, front-back confusions started at 40.7% and localization error halved after about 19 minutes of training (Steadman et al., Scientific Reports 2019).
- Quadrant errors fell from 27.5% to 18.3% within the first 400 training trials (Majdak et al., Attention, Perception and Psychophysics 2010).
- Frontal externalization rose from 25.56 to 65.15 out of 100 after level and reflection adjustment (Shen and Wang, Trends in Hearing 2026).
- 81% of SADIE II participants preferred dummy-head HRTFs to their own (University of York, Applied Sciences 2018).
- Choosing among clustered non-individual HRTFs reduced localization error by 22% (So et al., Ergonomics 2010).
1. Angular Error: Generic vs Individualized HRTFs
The headline finding is lopsided. Swapping a listener’s own HRTF for a generic one hurts the up-down and front-back dimension badly, while left-right accuracy hardly changes. In the 2026 five-HRTF study, polar-angle error grew from 30.1° with measured HRTFs to 48.8° with KEMAR (p < 0.001), while lateral metrics showed no significant effect (p > 0.05) (Pirard and Poole, arXiv 2026).
For gamers, this explains why virtual surround makes a sniper on your left easy to place but a footstep above or behind you hard to read. Left-right cues come mostly from timing and level differences between the two ears, which every head shares in rough form. Elevation and front-back cues live in the fine spectral notches of each person’s outer ear, and those differ from person to person.
| Metric | Value | Source |
|---|---|---|
| Polar-angle absolute error, generic KEMAR HRTF vs listener’s own measured HRTF | 48.8° vs 30.1° (p < 0.001) | Pirard and Poole, AES AVAR 2026 |
| Polar-angle absolute error, randomly chosen other person’s HRTF vs high-resolution scan-based synthetic HRTF | 41.8° vs 35.6° | Pirard and Poole, AES AVAR 2026 |
| Average local azimuth unsigned error: own measured, own modeled, generic KU100 dummy head | 17°, 21°, 20° (no significant main effect) | Rummukainen et al., Applied Sciences 2021 |
| Average local elevation unsigned error, own measured HRTF vs generic KU100 HRTF | 21° vs 21° | Rummukainen et al., Applied Sciences 2021 |
| Listeners whose horizontal accuracy with another person’s HRTF matched real free-field sources | 12 of 16 | Wenzel et al., JASA 1993 |
| Vertical error with measured or estimated HRTFs vs generic KEMAR, 24 participants, head-tracked playback | Significantly smaller; horizontal differences not significant | Frankenbach et al., JAES 2025 |
| Best human accuracy with real sound sources (free field): azimuth vs elevation | 1-2° vs 3-4° | Carlini et al., Frontiers in Psychology 2024 |
Context note: the Rummukainen study tested only 8 listeners in its static experiment, and its 3° to 4° azimuth gaps did not reach significance. The Wenzel result is the classic reference for this pattern (most recent available data for that design: Wenzel et al., JASA 1993), and the 2025 Frankenbach study (JAES) reproduced the same split with modern dynamic rendering.
2. Front-Back Confusion and Quadrant Errors
A front-back confusion is the error gamers complain about most: an enemy behind you sounds as if they are in front. A quadrant error is a broader count that also includes up-down flips. These rates swing widely between studies because of stimulus type, training, and how strictly a confusion is defined, so compare rows within a source before comparing sources.
In a VR rating study, a generic KEMAR HRTF produced 21 front-back confusions in 78 responses (27%), while individualized HRTFs produced none (χ²(1,77)=22.0, P<.001) (Jenny and Reuter, JMIR Serious Games 2020). The pattern holds in the 2026 data as well: a random other person’s HRTF was close to the listener’s own on front-back confusions but still worse on quadrant errors.
| Metric | Value | Source |
|---|---|---|
| Front-back confusion rate: KEMAR vs own measured HRTF vs random other person’s HRTF | 17.2% vs 8.1% vs 9.1% | Pirard and Poole, AES AVAR 2026 |
| Quadrant error rate, random other person’s HRTF vs own measured HRTF | 22.6% vs 15.4% (p = 0.039) | Pirard and Poole, AES AVAR 2026 |
| Front-back confusions, generic KEMAR HRTF vs individualized HRTF | 21 of 78 responses (27%) vs 0% | Jenny and Reuter, JMIR Serious Games 2020 |
| Front-back confusions, most-deviant vs least-deviant non-individual HRTF from the ARI database | 19 of 78 (24%) vs 5 of 78 (6%) | Jenny and Reuter, JMIR Serious Games 2020 |
| Front-back confusion with a non-individual HRTF over headphones vs real free-field sources | 31% vs 19% | Wenzel et al. 1993, as reported in Rummukainen et al., Applied Sciences 2021 |
| Quadrant error when switching from generic to individual HRTFs | 21% to 8% | Middlebrooks 1999, as reported in Rummukainen et al., Applied Sciences 2021 |
| Quadrant error for virtual speech, chance level about 50% | 59% | Begault et al. 2001, as reported in Majdak et al., Attention, Perception and Psychophysics 2010 |
| Quadrant error for trained listeners with real loudspeaker sources | 6% | Makous and Middlebrooks 1990, as reported in Majdak et al., Attention, Perception and Psychophysics 2010 |
Outlier note: the 59% figure for virtual speech is above chance, and two later papers (Majdak et al. 2010 and Rummukainen et al. 2021) report the same value. It shows how badly virtual speech can fail on this metric. Rows citing 1990 to 2001 studies are most recent available data for those specific designs.
3. Head Tracking: The Biggest Single Fix
Head tracking means the virtual scene stays fixed in the world while your head turns, the way real sound does. This is the “dynamic” binaural rendering found in VR headsets and some head-tracked earbuds, and it is absent from any virtual 7.1 mode that does not track the head. The data suggests it matters more than which HRTF you pick.
Switching from static to head-tracked playback cut unsigned horizontal error from 38.7° to 21.2° and weighted front-back confusions from 16.5% to 6.3% (Oberem et al., bioRxiv 2020). Even more telling, the gap between individual and non-individual HRTFs shrank from 6.8 percentage points in static playback to 1.7 points with head tracking. Movement does much of the work that a personal HRTF would otherwise do.
| Metric | Value | Source |
|---|---|---|
| Weighted front-back confusions, static playback vs head-tracked playback (5° HRTF resolution) | 16.5% vs 6.3% | Oberem et al., bioRxiv 2020 |
| Unsigned horizontal error, static vs head-tracked playback | 38.7° vs 21.2° | Oberem et al., bioRxiv 2020 |
| Front-back confusion gap between individual and non-individual HRTFs, static vs head-tracked | 6.8 vs 1.7 percentage points | Oberem et al., bioRxiv 2020 |
| Front-back confusions without vs with a single 10° head turn (real loudspeakers, 8 listeners) | 10.7 ± 6.3% vs 1.5 ± 1.2% | McLachlan et al., PLOS Computational Biology 2025 |
| Polar RMS error without vs with the 10° head turn | 25.1 ± 3.2° vs 20.5 ± 5.2° | McLachlan et al., PLOS Computational Biology 2025 |
| Head rotation window that significantly reduced vs almost eliminated front-back confusions | 4° vs 32° of azimuth | McAnally and Martin, Frontiers in Neuroscience 2014 |
| Head rotation window needed to significantly reduce elevation error | 16° of azimuth | McAnally and Martin, Frontiers in Neuroscience 2014 |
Context note: the McLachlan and McAnally studies used real loudspeakers, not headphones, so they show what head movement can do in principle (McLachlan et al. 2025; McAnally and Martin 2014). Lateral error did not change with head movement in either study (lateral RMS error 8.1° vs 8.2° in McLachlan).
4. Externalization: Out-of-Head vs In-Head Sound
Externalization is whether a sound seems to come from a point in space outside your head, rather than from inside it. Poor externalization is why a virtual surround mix can feel like “sound between the ears” even when directions are correct. Frontal sources are the hardest case, and the screen, where much of a game’s action plays out, sits in front of the player.
For a frontal source at 0° with a non-individualized room response, externalization scored 25.56 out of 100, rising to 65.15 after level and reflection adjustment (Cohen’s d = 2.32) (Shen and Wang, Trends in Hearing 2026). That room response is a BRIR (binaural room impulse response: an HRTF that also contains a room’s reflections). Reflections are a double-edged tool: they help sounds leave the head but can blur direction.
| Metric | Value | Source |
|---|---|---|
| In-head localization rate, static vs head-tracked playback | 6.4% vs 0.5% | Oberem et al., bioRxiv 2020 |
| Externalization rating (0-100) for a frontal source at 0°, original vs adjusted non-individualized room response | 25.56 vs 65.15 | Shen and Wang, Trends in Hearing 2026 |
| Externalization rating at 15° and 30° azimuth, original vs adjusted room response | 42.33 vs 67.66 (15°); 60.04 vs 85.67 (30°) | Shen and Wang, Trends in Hearing 2026 |
| Listeners choosing the further-modified room response in forced-choice tests at 0° and 15° | 84 of 100 and 83 of 100 (p < .001) | Shen and Wang, Trends in Hearing 2026 |
| Externalization score, individual vs non-individual room responses | 53.4% vs 51.2% | Leclère et al. 2019, as reported in Shen and Wang, Trends in Hearing 2026 |
| Length of reverberant tail that improves externalization before gains stop | Up to about 80 ms | Best et al., Trends in Hearing 2020 (review) |
| Mean elevation error for virtual speech without vs with reverberation | 17.6° vs 28.7° | Begault et al. 2001, as reported in Carlini et al., Frontiers in Psychology 2024 |
| SADIE II listeners who rated their own measured HRTF as the most external | 1 of 16 | University of York, Applied Sciences 2018 |
Context note: externalization barely depends on whether the HRTF is personal (53.4% vs 51.2%), but it depends heavily on head tracking, reflections and source angle. The Best et al. review also notes that sources moving with the head tend to internalize, while world-fixed sources externalize better. Begault’s elevation figure is most recent available data from a 2001 study.
5. Training and Adaptation to a Generic HRTF
Brains are plastic, and a mismatched HRTF can be partly learned. This is good news for players who will never get their ears measured: the question shifts from “which HRTF” to “how much deliberate practice with feedback.”
In a 36-participant study, fitted spherical localization error fell from 43.4° to 25.5° with fewer than two hours of training (nine 12-minute blocks), and errors halved after about 19 minutes (Steadman et al., Scientific Reports 2019). Gamification alone did not beat the untrained control group (p = 0.940); the version that added head-tracked active listening did. Adaptation also varies by person, as the Stitt results below show.
| Metric | Value | Source |
|---|---|---|
| Fitted spherical error before vs after training with non-individual HRTFs (108 minutes in total) | 43.4° vs 25.5° | Steadman et al., Scientific Reports 2019 |
| Initial front-back confusion rate with non-individual HRTFs, and average reduction after training | 40.7%; about 15% | Steadman et al., Scientific Reports 2019 |
| Gamified training without active listening vs untrained control, spherical error | Not significant (p = 0.940) | Steadman et al., Scientific Reports 2019 |
| Median polar error improvement over the first four weekly sessions with a poorly matched HRTF | 11.9° | Stitt et al., Scientific Reports 2019 |
| Total polar error improvement over 10 sessions, poorly matched HRTF group vs own-HRTF control group | About 25° vs about 10° | Stitt et al., Scientific Reports 2019 |
| Poorly matched listeners who reached control-group polar error after 10 sessions | 3 of 8 | Stitt et al., Scientific Reports 2019 |
| Quadrant error before vs after the first 400 training trials with listeners’ own measured HRTFs, head pointing vs manual pointing | 27.5% to 18.3%; 28.8% to 18.4% | Majdak et al., Attention, Perception and Psychophysics 2010 |
| Error reduction after 60 seconds of audio-visual exposure with generic HRTFs vs audio-only exposure | t(16) = 2.87, p = 0.011 vs p = 0.53 (n.s.) | Berger et al., Frontiers in Neuroscience 2018 |
Context note: Stitt et al. attribute about 15° of that group’s 25° gain to HRTF adaptation, since own-HRTF controls also improved by about 10°. A real-ear experiment with physical ear molds (not headphones) found full adaptation to the new ears within 20 minutes, with about 10 minutes of audio-motor training (Parise et al., iScience 2024). Majdak’s training data dates from 2010 and is most recent available data for that protocol; because it used each listener’s own HRTFs, it shows how much of any gain is plain task learning, which any generic-HRTF study also contains.
6. Personalization Shortcuts and HRTF Databases
Full HRTF measurement is slow and lab-bound: the ARI-style protocol measured 1550 HRTFs per listener in about 20 minutes inside a specialized setup (Majdak et al. 2010). That is why research has moved to shortcuts: picking the best match from a database, reshaping a generic HRTF, or synthesizing one from a scan.
Letting listeners pick from six clusters built from 196 non-individual HRTFs reduced sound localization error by 22%, and 65% of listeners reduced their front-back confusion rates (So et al., Ergonomics 2010). Larger databases make such matching more realistic. Note the twist from the SADIE II data: people often prefer the tonal balance of a dummy head over their own measurements, so “sounds best” and “localizes best” are different targets (our headphone target curve preference roundup covers the tonal side).
| Metric | Value | Source |
|---|---|---|
| Localization error reduction from choosing among 6 clustered non-individual HRTFs | 22%; 65% of listeners also reduced front-back confusion | So et al., Ergonomics 2010 |
| Best results from boosting or cutting selected HRTF frequency bands by 12 or 18 dB, 32 participants | Up to 70% less localization error and 66% less front-back confusion | So et al., Human Factors 2011 |
| Median elevation error, best single HRTF vs best HRTF chosen per direction (8 listeners) | 27.8° vs 22.2° (p = 0.056) | Goldring et al., AES AVAR 2024 |
| Median azimuth error, best single HRTF vs best HRTF per direction | 18.13° vs 15.74° (p = 0.12) | Goldring et al., AES AVAR 2024 |
| Effect of frequency-scaling another person’s ear transfer functions | More than halved the penalty for using another subject’s ear transfer functions | Middlebrooks, JASA 1999 |
| SONICOM HRTF dataset, initial release vs extended release (measured subjects) | 120 vs 300 | Engel et al., JAES 2023; Poole et al., arXiv 2025 |
| SONICOM extended release: subjects with synthesised HRTFs, and measurement grid | 200 of 300; 793 positions | Poole et al., arXiv 2025 |
| SADIE II listening-test participants (16) who preferred KU100 dummy-head HRTFs to their own | 81% (the database itself holds 20 subjects) | University of York, Applied Sciences 2018 |
Context note: the So et al. and Middlebrooks findings are most recent available data for those methods (2010, 2011 and 1999). The SONICOM extended dataset and the SADIE II evaluation are the public resources most likely to feed the next generation of HRTF matching tools.
7. What the Evidence Says for Virtual 7.1 and Game Audio
Studies run inside actual games are still few and small, but they point the same way as the lab data. Personalization helps most where generic HRTFs fail most (elevation and front-back), while left-right placement is already good enough with almost any HRTF. In a desktop first-person shooter test, the 22 participants performed significantly better in the localization task with their individualized HRTF than with a generic one, in both azimuth and elevation (Andersen et al., I3DA 2021).
Gains do not always require a full HRTF. A KAIST team tuned only the interaural time difference (ITD: the tiny arrival-time gap between the two ears) and still found significantly faster response times than with two generic HRTFs (Kim et al., AES Show 2024). For hardware context on the headsets these features ship in, see our gaming headset statistics.
| Metric | Value | Source |
|---|---|---|
| Rendering conditions compared in a desktop shooter localization task, 22 participants | 3 (stereo panning, generic binaural, individualized binaural); individualized significantly better | Andersen et al., I3DA 2021 |
| Response time with an ITD-optimized HRTF vs MIT KEMAR and a default game-engine HRTF | Significantly reduced (figure not given in abstract) | Kim et al., AES Show 2024 |
| Lateral (left-right) accuracy across five HRTF types, from measured to dummy head | No significant effect (p > 0.05) | Pirard and Poole, AES AVAR 2026 |
| Pairwise differences in rated localization between measured, modeled and generic HRTFs in 6-degrees-of-freedom VR, 12 listeners | None significant in post-hoc tests | Rummukainen et al., Applied Sciences 2021 |
| Responses rating tone color as natural, individualized HRTF vs generic KEMAR HRTF | 62 of 78 (79%) vs 1 of 78 (1%) | Jenny and Reuter, JMIR Serious Games 2020 |
| Horizontal localization error, measured or estimated HRTFs vs KEMAR with head-tracked playback | No significant difference | Frankenbach et al., JAES 2025 |
Context note: none of these game studies published per-condition error figures in their abstracts, so treat them as direction-of-effect evidence. The practical reading is consistent across sections: head tracking (section 3) and under two hours of feedback training (section 5) close much of the gap that a personal HRTF would close.
Summary: Gaming Headset HRTF Localization Accuracy by the Numbers
| Metric | Value | Source |
|---|---|---|
| Front-back confusion rate, generic KEMAR vs own measured HRTF | 17.2% vs 8.1% | Pirard and Poole, AES AVAR 2026 |
| Front-back confusions, generic KEMAR vs individualized HRTF in VR | 27% vs 0% | Jenny and Reuter, JMIR Serious Games 2020 |
| Average local azimuth error, own measured vs generic KU100 | 17° vs 20° | Rummukainen et al., Applied Sciences 2021 |
| Weighted front-back confusions, static vs head-tracked playback | 16.5% vs 6.3% | Oberem et al., bioRxiv 2020 |
| Unsigned horizontal error, static vs head-tracked playback | 38.7° vs 21.2° | Oberem et al., bioRxiv 2020 |
| Front-back confusions without vs with a 10° head turn | 10.7% vs 1.5% | McLachlan et al., PLOS Computational Biology 2025 |
| Head rotation that almost eliminated front-back confusions | About 32° of azimuth | McAnally and Martin, Frontiers in Neuroscience 2014 |
| Frontal externalization rating, original vs adjusted room response | 25.56 vs 65.15 out of 100 | Shen and Wang, Trends in Hearing 2026 |
| Reverberant tail length that helps externalization | Up to about 80 ms | Best et al., Trends in Hearing 2020 |
| Fitted spherical error before vs after under two hours of training | 43.4° vs 25.5° | Steadman et al., Scientific Reports 2019 |
| Training time to halve localization error | About 19 minutes | Steadman et al., Scientific Reports 2019 |
| Poorly matched listeners reaching control-level polar error after 10 sessions | 3 of 8 | Stitt et al., Scientific Reports 2019 |
| Quadrant error before vs after 400 training trials (head pointing) | 27.5% vs 18.3% | Majdak et al., Attention, Perception and Psychophysics 2010 |
| Localization error reduction from choosing among clustered HRTFs | 22% | So et al., Ergonomics 2010 |
| Median elevation error, best single HRTF vs best HRTF per direction | 27.8° vs 22.2° | Goldring et al., AES AVAR 2024 |
| SONICOM measured subjects, extended release | 300 | Poole et al., arXiv 2025 |
| SADIE II participants preferring dummy-head HRTFs to their own | 81% | University of York, Applied Sciences 2018 |
Methodology and Sources
- Pirard and Poole, Evaluation of Head-Related Transfer Functions Across Five Levels of Individualisation in Virtual Reality, AES AVAR 2026 (arXiv preprint): 19 listeners, Meta Quest headsets, 297 test trials per experiment.
- Jenny and Reuter, Usability of Individualized HRTFs in Virtual Reality, JMIR Serious Games 2020: 39 subjects, ARI and MIT KEMAR HRTFs.
- Rummukainen, Robotham and Habets, Head-Related Transfer Functions for Dynamic Listeners in Virtual Reality, Applied Sciences 2021: also the intermediary for the Wenzel 1993 and Middlebrooks 1999 comparison figures.
- Wenzel, Arruda, Kistler and Wightman, Localization Using Nonindividualized Head-Related Transfer Functions, JASA 1993.
- Middlebrooks, Virtual Localization Improved by Scaling Nonindividualized External-Ear Transfer Functions in Frequency, JASA 1999.
- Frankenbach, Herzog, Seitz and Fels, Evaluating Localization Accuracy of Dynamic Binaural Synthesis, JAES 2025.
- Majdak, Goupell and Laback, 3-D Localization of Virtual Sound Sources, Attention, Perception and Psychophysics 2010: also the intermediary for Begault et al. 2001 and Makous and Middlebrooks 1990 quadrant-error figures.
- Oberem, Richter, Setzer, Seibold, Koch and Fels, Localization Accuracy with Non-Individual and Individual HRTFs Comparing Static and Dynamic Reproduction, bioRxiv 2020: 11 listeners, right hemisphere only.
- McLachlan, Majdak, Reijniers, Mihocic and Peremans, Bayesian Active Sound Localisation, PLOS Computational Biology 2025.
- McAnally and Martin, Sound Localization with Head Movement: Implications for 3-D Audio Displays, Frontiers in Neuroscience 2014.
- Shen and Wang, Enhancing Externalization of Frontal Virtual Sound Sources, Trends in Hearing 2026: also the intermediary for the Leclère et al. 2019 figures.
- Best, Baumgartner, Lavandier, Majdak and Kopčo, Sound Externalization: A Review of Recent Research, Trends in Hearing 2020.
- Carlini, Bordeau and Ambard, Auditory Localization: A Comprehensive Practical Review, Frontiers in Psychology 2024: free-field baselines and the Begault 2001 reverberation figures.
- Steadman, Kim, Lestang, Goodman and Picinali, Short-Term Effects of Sound Localization Training in Virtual Reality, Scientific Reports 2019.
- Stitt, Picinali and Katz, Auditory Accommodation to Poorly Matched Non-Individual Spectral Localization Cues Through Active Learning, Scientific Reports 2019.
- Berger et al., Generic HRTFs May Be Good Enough in Virtual Reality, Frontiers in Neuroscience 2018.
- Parise et al., Happy New Ears: Rapid Adaptation to Novel Spectral Cues in Vertical Sound Localization, iScience 2024.
- So et al., Toward Orthogonal Non-Individualised HRTFs for Forward and Backward Directional Sound, Ergonomics 2010.
- So et al., Effects of Spectral Manipulation on Nonindividualized HRTFs, Human Factors 2011.
- Goldring, Ben Hur, Alon and Rafaely, Assessing the Potential Impact of Direction-Dependent HRTF Selection on Sound Localization Accuracy, AES AVAR 2024 (arXiv).
- Engel et al., The SONICOM HRTF Dataset, JAES 2023.
- Poole et al., The Extended SONICOM HRTF Dataset and Spatial Audio Metrics Toolbox, arXiv 2025.
- University of York AudioLab, A Perceptual Evaluation of Individual and Non-Individual HRTFs: A Case Study of the SADIE II Database, Applied Sciences 2018.
- Andersen, Miccini, Serafin and Spagnol, Evaluation of Individualized HRTFs in a 3D Shooter Game, I3DA 2021.
- Kim et al., Investigating the Role of Customized Interaural Time Differences on First-Person Shooter Gaming Performance, AES Show 2024.
- Data watch: error metrics are not standardized. Studies report polar vs elevation error, mean vs median, front-back confusion vs quadrant error (which also counts up-down flips), and some weight confusions by angle, so compare values within one source before comparing across sources. Several key figures come from preprints or conference papers not yet in a journal (Pirard and Poole 2026, Oberem et al. 2020, Goldring et al. 2024, Poole et al. 2025). The Middlebrooks individual-HRTF quadrant error appears as 8% in Rummukainen et al. and as 7.7 ± 8.0% for trained subjects in Majdak et al., which is consistent within rounding. SONICOM sources differ on the size of the initial release (120 subjects in the 2023 JAES abstract vs 200 HRTFs described in the 2025 extended-dataset paper); we cite 120 for the original paper and 300 for the extended release. The Steadman et al. paper lists 36 participants by group but gives a sex breakdown that sums to 38, and a few of its p-values are garbled in the text, so we quote only group-level results with clear statistics. The McLachlan and McAnally head-movement figures come from loudspeaker experiments, not headphones. Several rows rely on 1990 to 2011 studies and are flagged as most recent available data. No manufacturer or vendor figures are used in this roundup.
Last updated: October 25, 2026. We update this roundup quarterly, and the next refresh is expected when the peer-reviewed proceedings of the AES 2026 International Conference on Audio for Virtual and Augmented Reality and Immersive Games and the next SONICOM dataset release are published.