Moving from monaural sound to head-tracked binaural audio with room acoustics raised VR immersion ratings from 3.88 to 7.91 on a 9-point scale in a study of 17 trained listeners. That lab gain sits next to a much soberer pooled figure: sound’s effect on presence across 13 effect sizes is r = .260, below user tracking at r = .408. Across 243 published studies, the median Igroup Presence Questionnaire (IPQ) score is only 0.28 on a scale running from -3 to 3, and the evidence that any of this carries over to commercial games is thin. We aggregated data from ACM TOCHI, ACM CHI, PLOS ONE, the Frontiers journals, Presence: Teleoperators and Virtual Environments, and Meta Reality Labs. For adoption and device data, see our separate spatial audio statistics roundup; this page covers what sound does to measured presence (“the feeling of being there”) and immersion scores.
TL;DR
- Sound’s pooled effect on presence is r = .260, 95% CI [.203, .317], N = 757, while user tracking reaches r = .408 (Media Psychology, Cummings and Bailenson 2016).
- The median IPQ score across 243 studies is 0.28 on a -3 to 3 scale, and head-mounted display studies average M = 0.5 (ACM TOCHI, Tran et al. 2024).
- Full audio rendering at 20% video scored 5.79, matching 5.76 for binaural audio without reverberation at 100% video, p = 0.891 (Frontiers in Signal Processing, Potter et al. 2022).
- A soundscape moved IPQ single-item presence with η² = 0.386, against η² = 0.130 for step sounds, yet IPQ Spatial Presence showed no significant effect (Frontiers in Robotics and AI, Kern and Ellermeier 2020).
- Enhanced audio raised Job Simulator audio quality from 7.11 to 7.43, F(1,63) = 24.49, p<0.001 (Frontiers in Virtual Reality, Mavridou et al. 2025, supported by the tool’s developer).
- A 3D audio vs stereo horror game test found no significant questionnaire difference in immersion or presence (Aalborg University, Larsen and Pilgaard 2015).
- The IPQ showed r = 0.82 for place illusion in a four-questionnaire comparison, N = 50 (ACM CHI, Degenhard et al. 2025).
- The IPQ reached α = .87 and took a mean of 2.449 minutes to complete (ACM CHI, Schwind et al. 2019).
- Wearing a head-mounted display raised sound elevation error by about 1.8° in a blindfolded test (PLOS ONE, Ahrens et al. 2019).
- Of 21 presence and cybersickness results, 11 are negative, 6 positive and 4 null (Frontiers in Psychology, Weech et al. 2019).
1. IPQ Benchmarks: What Presence Scores Look Like Across 243 Studies
The median published IPQ score is 0.28 on a -3 to 3 scale. A score that sounds modest is therefore normal, and the top 5% of conditions start at only 1.30. Any audio team quoting a raw IPQ number needs this context before calling a result good or bad, which is the purpose of the two-decade IPQ analysis in ACM TOCHI.
The subscale split is the useful part. Experienced Realism is the only subscale with a negative mean, so whether the virtual world feels real is the weakest dimension in the benchmark, and section 3 shows it is one that sound can move. The analysis gives no breakdown by sound condition, so nobody can yet say what audio contributes to these benchmarks.
| Metric | Value | Source |
|---|---|---|
| Corpus and overall IPQ score, rescaled to the -3 to 3 range | 243 studies from 229 papers, 9,354 participants; M = 0.4, SD = 0.5, Mdn = 0.3 | Tran et al., ACM TOCHI 2024 |
| Overall score percentile cut-offs, all studies | P50 = 0.28, P75 = 0.73, P90 = 1.07, P95 = 1.30 | Tran et al., ACM TOCHI 2024 |
| IPQ score by display type: head-mounted display vs projection vs monoscopic desktop or mobile | HMD M = 0.5 (P75 = 0.83, 180 studies); projection M = 0.2 (34 studies); monoscopic M = -0.2 (39 studies); Welch’s F = 23.37, p < .001 | Tran et al., ACM TOCHI 2024 |
| IPQ subscale means, 162 studies | General Presence M = 0.9, SD = 0.7; Spatial Presence M = 0.8, SD = 0.7; Involvement M = 0.3, SD = 0.6; Experienced Realism M = -0.3, SD = 0.6 | Tran et al., ACM TOCHI 2024 |
The class labels are percentile-based: 121 of 243 studies (49.8%) fall in the band named Low, because that band is defined as everything under the median.
2. How Much Sound Moves Presence: The Meta-Analytic Picture
Sound’s pooled effect on presence is r = .260, 95% CI [.203, .317]. That is a real effect, and the Stanford meta-analysis in Media Psychology calls it small to medium. It also concludes that sound’s impact is relatively small compared with other features, and significantly weaker than tracking. It still outranks image quality, whose effect was significantly smaller than sound’s.
Sound here is one composite covering presence or absence of sound, ambient, diegetic or spatialized sound, and channel count. The meta-analysis gives no pooled estimate for spatial audio alone, so a spatial audio r value cannot be quoted from it.
| Metric | Value | Source |
|---|---|---|
| All immersive features combined, effect on presence | r = .316, 95% CI [.295, .338], K = 115 effect sizes from 83 studies, N = 6,998 | Cummings and Bailenson, Media Psychology 2016 |
| Sound as a system feature, effect on presence | r = .260, 95% CI [.203, .317], K = 13, N = 757 | Cummings and Bailenson, Media Psychology 2016 |
| User tracking level, effect on presence | r = .408, 95% CI [.360, .456], K = 22, N = 1,566 | Cummings and Bailenson, Media Psychology 2016 |
| Stereoscopic visuals, field of view and image quality, each effect on presence | Stereoscopy r = .320, K = 18; field of view r = .304, K = 14; image quality r = .150, 95% CI [.086, .214], K = 10, N = 855 | Cummings and Bailenson, Media Psychology 2016 |
| Scoping review of audio in head-mounted display VR | n = 121 studies; results described as mixed for presence | Bosman et al., Behaviour and Information Technology 2024 |
| Earliest controlled test in this set: spatialized sound added to a stereoscopic display | Presence significantly higher, apparent realism not increased | Hendrix and Barfield, IEEE Virtual Reality Annual International Symposium 1995 |
Most recent available data: Cummings and Bailenson, 2016. The newer review of 121 studies is a scoping review, not a pooled analysis, and we found no audio-specific meta-analysis. The 1995 IEEE symposium paper is a useful historical marker because it already split presence from realism. A 2022 study in Presence: Teleoperators and Virtual Environments with 51 participants (28 stereoscopic HMD, 23 monitor) also found a large effect on spatial presence for stereoscopic viewing, without a numeric effect size in the abstract.
3. Soundscape Versus Footsteps: Which Sound Layer Moves the Score
An ambient soundscape explained far more of the presence rating than step sounds: η² = 0.386 against η² = 0.130 on the IPQ single item. A soundscape here means the continuous background sound of a place. The result suggests the budget question has an order: lay the environment first, then interaction sounds. Most recent available data: Kern and Ellermeier, Frontiers in Robotics and AI 2020.
The authors put no-sound presence at between 40 and 54 percent of real-world presence, with step sounds plus soundscape adding some 15-20 percentage points and a ceiling near 70%. The participants walked a treadmill through a nature scene for up to 5 minutes per condition, which is not a game, so treat this as a sound-layer result, not a genre result.
| Metric | Value | Source |
|---|---|---|
| Presence rating on a 1-100 scale, Experiment 1 (N = 36): soundscape added vs not added, and step sounds added vs not added | Soundscape F(1, 35) = 48.95, p < 0.001, η² = 0.583; steps F(1, 35) = 0.003, p = 0.96 | Kern and Ellermeier, Frontiers in Robotics and AI 2020 |
| IPQ single-item presence, Experiment 2 (N = 40): soundscape vs step sounds | Soundscape F(1, 39) = 24.512, p < 0.001, η² = 0.386; steps F(1, 39) = 5.821, p = 0.021, η² = 0.130 | Kern and Ellermeier, Frontiers in Robotics and AI 2020 |
| IPQ Realism mean: noise-cancelling only vs steps plus soundscape | -1.375 vs -0.306 | Kern and Ellermeier, Frontiers in Robotics and AI 2020 |
| IPQ Involvement mean: noise-cancelling only vs steps plus soundscape | -0.331 vs 0.256; soundscape F(1, 39) = 36.983, p < 0.001, η² = 0.487 | Kern and Ellermeier, Frontiers in Robotics and AI 2020 |
| IPQ Spatial Presence mean: noise-cancelling only vs steps plus soundscape | 0.17 vs 0.33; no significant effect of either factor | Kern and Ellermeier, Frontiers in Robotics and AI 2020 |
The gain on IPQ Realism between those two conditions is -0.306 minus -1.375 = 1.069 points. Spatial Presence is the surprise: sound changed how real and absorbing the scene felt, without a significant change in the subscale that shares its name with spatial audio.
4. Spatial Rendering Versus Video Resolution
With full audio rendering, video could drop to 20% resolution and still score 5.79, statistically level with 5.76 at 100% video without reverberation (U = 567.0, p = 0.891). Reverberation here means the room reflections that tell the ear how big a space is. This result puts audio rendering in the same budget discussion as resolution, though the authors note that naive subjects would be more representative than their trained listeners.
The test content was speech and music on an HTC Vive, not game content. For the physical side of reverberation, see our concert hall acoustics statistics. The percentages from older studies below cannot be stacked, because each one compared different conditions on different hardware.
| Metric | Value | Source |
|---|---|---|
| Immersion rating on a 9-point scale, 100% video: monaural vs head-tracked binaural with reverberation | 3.88 vs 7.91 (N = 17) | Potter et al., Frontiers in Signal Processing 2022 |
| Immersion rating: head-tracked binaural with reverberation at 20% video vs head-tracked binaural at 100% video | 5.79 (95% CI 5.24-6.35) vs 5.76 (95% CI 5.18-6.35); U = 567.0, p = 0.891 | Potter et al., Frontiers in Signal Processing 2022 |
| Kruskal-Wallis result for immersion by spatial audio method vs by video resolution | Audio H(2) = 145.917, p < 0.001; video H(2) = 34.709, p < 0.001 | Potter et al., Frontiers in Signal Processing 2022 |
| Change in mean immersion at 100% video: monaural to head-tracked binaural, and removing reverberation | Increase of 48%; decrease of 27.2% | Potter et al., Frontiers in Signal Processing 2022 |
| Mean presence, no sound vs spatialised sound with reverberation in a loudspeaker VR cube | Increase of 25.6% (p < 0.01) | Larsson et al. 2007 (secondary report, see Data watch) |
| Mean presence, monaural vs binaural without reverberation, no head tracking | Increase of 20% | Hendrix and Barfield 1996 (secondary report, see Data watch) |
| Mean presence, no sound vs simple stereo sounds | Increase of 9.0% (p < 0.01) | Dinh et al. 1999 (secondary report, see Data watch) |
The monaural to full rendering gain in the first row is 7.91 / 3.88 = 2.04 times. Most recent available data: Potter et al., Frontiers in Signal Processing 2022.
5. Inside Actual VR Games: Where the Effect Gets Harder to Find
Enhanced audio lifted Job Simulator’s audio quality rating from 7.11 to 7.43 (F(1,63) = 24.49, p<0.001), while the same tool produced no significant change in self-reported presence in a non-commercial simulation. The 2025 Frontiers in Virtual Reality study is the closest thing to a modern test on a commercial game, and the authors say the stronger commercial response is likely due to sound architecture. The reading for developers: post-processing can amplify sound design that exists, and it cannot supply one that does not.
Treat the size of that result with care. The study was supported in part by Bongiovi Acoustic Labs, the developer of the audio tool, several co-authors are affiliated with it, and no effect sizes are reported. The two other game-context sources below are a master’s thesis from Aalborg University and a paper on VR game space acoustics, and neither one gives a presence score we could quote.
| Metric | Value | Source |
|---|---|---|
| Sample and design of the enhanced vs normal audio test | 68 recruited, within-subjects; 58 kept for physiological analysis; HTC VIVE Pro Eye | Mavridou et al., Frontiers in Virtual Reality 2025 (supported by the tool’s developer) |
| Job Simulator audio quality rating on a 9-point item, enhanced vs normal audio | 7.43 ± 0.35 vs 7.11 ± 0.36; F(1,63) = 24.49, p<0.001 | Mavridou et al., Frontiers in Virtual Reality 2025 (supported by the tool’s developer) |
| Job Simulator involvement and sound localization ratings, enhanced vs normal audio | Involvement F(1,63) = 6.79, p = 0.011; localization F(1,63) = 5.95, p = 0.018 | Mavridou et al., Frontiers in Virtual Reality 2025 (supported by the tool’s developer) |
| Job Simulator emotional valence from facial muscle sensors, enhanced vs normal audio | 5.2 ± 1.1 vs 4.7 ± 0.86; Wilcoxon t = 569.0, p = 0.0265 | Mavridou et al., Frontiers in Virtual Reality 2025 (supported by the tool’s developer) |
| Escape VR self-reported presence, enhanced vs normal audio, whole experience | No significant difference | Mavridou et al., Frontiers in Virtual Reality 2025 (supported by the tool’s developer) |
| Horror game, 3D audio vs stereo | No significant questionnaire difference in immersion or presence; electrodermal (skin conductance) response differed significantly | Larsen and Pilgaard, Aalborg University master’s thesis 2015 |
| VR game spaces: sound-designed reverberation vs reverberation matching the visible room | Users preferred the altered version only when the original was very dry or very ringing | Popp and Murphy, Virtual Worlds 2024 |
The thesis abstract adds its own warning: the 3D audio condition also had higher sound intensity, which could account for the skin-conductance result. Most recent available data for that test: Larsen and Pilgaard, 2015.
6. Which Questionnaire Actually Detects an Audio Change
The IPQ showed r = 0.82 for place illusion, the highest of four questionnaires, in a CHI 2025 comparison with N = 50. Place illusion means the sense of being in the virtual location, which is the dimension spatial audio aims at. If the instrument cannot see a change, an audio team will read no effect as no value.
The same study warns that presence tools are blunt: participants generally judged score differences as too small, and no questionnaire picked up the embodiment changes. A CHI 2019 paper adds a practical result: filling in the form inside the headset did not change mean scores, which avoids a break in the experience.
| Metric | Value | Source |
|---|---|---|
| Mean score gap, high minus low intensity condition, by questionnaire (with item count) | SIP 0.57 (1 item), IPQ 0.43 (14 items), PQ 0.38 (19 items), SUS 0.36 (6 items) | Degenhard et al., ACM CHI 2025 |
| Effect size r for place illusion, Wilcoxon signed-rank test | IPQ 0.82, SIP 0.75, SUS 0.72, PQ 0.57 | Degenhard et al., ACM CHI 2025 |
| Effect size r for plausibility, meaning how believable events feel | PQ 0.8, IPQ 0.67, SUS 0.38, SIP 0.36 | Degenhard et al., ACM CHI 2025 |
| Participants agreeing with the score difference | IPQ 33.6%, PQ 25.6% | Degenhard et al., ACM CHI 2025 |
| Internal consistency, Cronbach’s α (N = 36) | IPQ α = .87, WS α = .82, SUS α = .75 | Schwind et al., ACM CHI 2019 |
| Mean completion time in minutes | SUS 1.107, IPQ 2.449, WS 5.982 | Schwind et al., ACM CHI 2019 |
| IPQ mean, completed inside VR vs outside VR | F(1, 34) = .013, p = .909 | Schwind et al., ACM CHI 2019 |
SIP is a single-item presence scale, SUS is the Slater-Usoh-Steed questionnaire, PQ and WS are forms of the Witmer and Singer Presence Questionnaire. Most recent available data for reliability and timing: Schwind et al., 2019.
7. Localization, Personalization, and Side Effects
Wearing the headset alone raised sound elevation error by about 1.8° (F(1,2049) = 35.29, p<0.0001) in a blindfolded test. Part of the audio problem in VR is the hardware on the head, before any rendering runs. Visuals repay some of that debt: in the PLOS ONE study, showing the room and hand position cut elevation error. For headset-side numbers, see our gaming headset localization statistics.
Personal head-related transfer functions (HRTFs, the filters that describe how a person’s own ears and head shape sound) showed a presence benefit in a 2004 Chalmers paper, but the abstract gives no number and Meta’s measurement setup shows why it does not scale. Sickness is the other open variable: a review in Frontiers in Psychology cites Viaud-Delmon et al. (2006), where adding auditory cues raised presence but also raised cybersickness.
| Metric | Value | Source |
|---|---|---|
| Sound elevation error, with vs without HMD, blindfolded (N = 10) | Increase of about 1.8°; F(1,2049) = 35.29, p<0.0001 | Ahrens et al., PLOS ONE 2019 |
| Elevation error, blindfolded with HMD vs virtual room with hand location | 10.2° vs 8.1° | Ahrens et al., PLOS ONE 2019 |
| Published presence and cybersickness results, by direction of the relationship | 21 entries from 20 publications: 11 negative, 6 positive, 4 null | Weech et al., Frontiers in Psychology 2019 |
| Spread of reported correlations between presence and cybersickness | From r(6) = -0.63 (Cooper et al., 2016) to r(58) = 0.67 (Liu and Uang, 2011) | Weech et al., Frontiers in Psychology 2019 |
| Individualized vs generic HRTF, presence ratings for binaural stimuli | Significant increase; no figure in the abstract | Väljamäe et al., Proceedings of the 7th Annual Workshop Presence 2004 |
| HRTF capture rig at Facebook Reality Labs Research | 54 loudspeakers on a rotating arm; the process takes about half an hour (vendor’s own description) | Meta, Inside Facebook Reality Labs Research 2020 |
Most recent available data: Weech et al., 2019, Ahrens et al., 2019, Meta, 2020, and Väljamäe et al., 2004. Only 10 subjects took part in the PLOS ONE test, on one headset.
Summary: VR Game Audio Presence and Immersion by the Numbers
| Metric | Value | Source |
|---|---|---|
| Sound as a system feature, effect on presence | r = .260, 95% CI [.203, .317], N = 757 | Media Psychology 2016 |
| Median overall IPQ score, -3 to 3 range | 0.28 (P50) | ACM TOCHI 2024 |
| Immersion, monaural vs head-tracked binaural with reverberation, 100% video | 3.88 vs 7.91 | Frontiers in Signal Processing 2022 |
| Soundscape vs step sounds, IPQ single-item presence | η² = 0.386 vs η² = 0.130 | Frontiers in Robotics and AI 2020 |
| Mean presence, no sound vs spatialised sound with reverberation | Increase of 25.6% (p < 0.01) | Larsson et al. 2007 (secondary report) |
| Job Simulator audio quality, enhanced vs normal audio | 7.43 vs 7.11, p<0.001 | Frontiers in Virtual Reality 2025 (supported by the tool’s developer) |
| Escape VR self-reported presence, enhanced vs normal audio | No significant difference | Frontiers in Virtual Reality 2025 (supported by the tool’s developer) |
| Horror game, 3D audio vs stereo, questionnaire immersion and presence | No significant difference | Aalborg University master’s thesis 2015 |
| Audio in head-mounted display VR, studies scoped | n = 121, mixed for presence | Behaviour and Information Technology 2024 |
| IPQ effect size for place illusion | r = 0.82 | ACM CHI 2025 |
| IPQ reliability and mean completion time | α = .87; 2.449 minutes | ACM CHI 2019 |
| Headset effect on sound elevation error | About 1.8° higher | PLOS ONE 2019 |
| Elevation error, blindfolded with HMD vs virtual room with hand location | 10.2° vs 8.1° | PLOS ONE 2019 |
| Presence and cybersickness results by direction | 11 negative, 6 positive, 4 null | Frontiers in Psychology 2019 |
| HRTF capture rig | 54 loudspeakers, about half an hour (vendor’s own description) | Meta 2020 |
Methodology and Sources
We read each study below directly, in its full text or abstract, and report questionnaire names, sample sizes and statistical notation as the study prints them. Figures that appeared only in search snippets were dropped.
- Tran et al., Classifying Presence Scores: Insights and Analysis from Two Decades of the Igroup Presence Questionnaire, ACM TOCHI 2024
- Cummings and Bailenson, How Immersive Is Enough?, Media Psychology 2016
- Kern and Ellermeier, Audio in VR: Effects of a Soundscape and Movement-Triggered Step Sounds on Presence, Frontiers in Robotics and AI 2020
- Potter, Cvetković and De Sena, On the Relative Importance of Visual and Spatial Audio Rendering on VR Immersion, Frontiers in Signal Processing 2022
- Mavridou et al., Did You Hear That?, Frontiers in Virtual Reality 2025
- Degenhard et al., When Do We Feel Present in a Virtual Reality?, ACM CHI 2025
- Schwind et al., Using Presence Questionnaires in Virtual Reality, ACM CHI 2019
- Bosman et al., The Effect of Audio on the Experience in Virtual Reality: A Scoping Review, Behaviour and Information Technology 2024
- Ahrens et al., Sound Source Localization with Varying Amount of Visual Information in Virtual Reality, PLOS ONE 2019
- Weech, Kenny and Barnett-Cowan, Presence and Cybersickness in Virtual Reality Are Negatively Related: A Review, Frontiers in Psychology 2019
- Hendrix and Barfield, Presence in Virtual Environments as a Function of Visual and Auditory Cues, IEEE 1995
- Väljamäe et al., Auditory Presence, Individualized Head-Related Transfer Functions, and Illusory Ego-Motion in Virtual Environments, 2004
- Uz-Bilgin, Thompson and Klopfer, Stereoscopic Views Improve Spatial Presence but Not Spatial Learning in VR Games, Presence: Teleoperators and Virtual Environments 2022
- Popp and Murphy, Speech Intelligibility Versus Congruency, Virtual Worlds 2024
- Larsen and Pilgaard, The Effect of Spatial Audio on Immersion, Presence, and Physiological Response in Games, Aalborg University 2015
- Meta, Inside Facebook Reality Labs Research: The Future of Audio, 2020
- Data watch: Controlled evidence on commercial VR games is thin, and most figures here come from speech, music, walking or simulation scenes. Scales differ between studies (IPQ -3 to 3, 1-100 items, 9-point items), so values cannot be compared across rows unless a row says so. The display subgroups in section 1 sum to 253 studies against 237 reported as display-reporting, and the overall mean is M = 0.4 for all 243 studies but M = 0.3 in the 237-study display table, so we do not combine those counts. In the 2025 Frontiers in Virtual Reality paper we left out several figures that look inconsistent in the published text: a helicopter-event localization mean of 9.09 on a 9-point item, an Escape VR audio quality direction that differs between text and figure caption, and an involvement test printed as t(2); its funding note and its no-conflict statement also sit in tension. In the rendering study in section 4, one audio-visual quality p-value differs between text and table, and no correction for multiple comparisons was applied. The questionnaire comparison in section 6 says SUS and PQ had the lowest reflection of changes users agreed with, yet also ranks PQ second on agreement, so we quote only the two printed percentages. Weech et al. list Mania and Chalmers (2001) as null with r(16) = -0.4, an inconsistency we did not resolve. The two CHI papers in section 6 used different forms of the Witmer and Singer instrument (32 items as WS, 19 items as PQ). Figures from Larsson et al. (2007), Hendrix and Barfield (1996) and Dinh et al. (1999) come from the discussion of the Frontiers in Signal Processing paper, and those from Viaud-Delmon et al. (2006), Cooper et al. (2016) and Liu and Uang (2011) from the Frontiers in Psychology review; we did not read those originals. Paywalled or blocked work, including AES papers and CHI 2018 work on game audio, was excluded rather than quoted second-hand.
Last updated: October 31, 2026. We update this roundup quarterly, and the next refresh is expected when the IEEE VR 2027 conference proceedings are published.