Spatial Audio in Conferencing and Group Calls

Why this matters

If you run a video conferencing product, a telemedicine platform, a virtual classroom, or a contact-centre app, your users spend hours a day listening to several voices stacked into one flat channel, and they get tired faster than anyone admits. Spatial audio is one of the few audio upgrades that improves comprehension rather than just fidelity — it makes overlapping talk easier to untangle and makes "who just said that?" obvious. This article is for a product manager, founder, or operations lead deciding whether to build it, and for the engineer who will have to wire it up. By the end you will know what spatial audio actually computes, why it only pays off under specific listening conditions, how Teams, Zoom, Google Meet, FaceTime, and LiveKit each implement it, and the three mistakes that turn the feature into a support ticket instead of a selling point.

What "spatial audio in a call" actually means

Start with the normal case, because the upgrade only makes sense against it. In an ordinary group call every participant's voice is mono — a single channel — and the app plays that one channel into both of your ears at the same level. The technical name for this is a head-locked or centred mix: every voice sits at the exact centre of your head, on top of every other voice. Your brain has no spatial cue to separate them, so when two people talk at once the words pile up and you lose both.

Spatial audio changes one thing: it gives each voice a position. The app decides where each talker should sit relative to you — usually matching the spot where their video tile appears on screen — and then processes their mono voice into a stereo pair (a left-ear signal and a right-ear signal) that your brain reads as "coming from over there." Microsoft describes its own implementation in exactly these terms: when someone speaks, "you'll hear their voices coming from their relative positions on the meeting screen," so a person on the far left of your screen sounds like they are on your far left, and the person beside them sounds slightly closer to centre.

The simplest way to do this is panning: make a left-positioned voice louder in the left ear and quieter in the right. That alone gives a usable left-to-right spread. A more convincing method uses a head-related transfer function, the measurement of how your own head and ears reshape a sound depending on which direction it comes from. We covered the head-related transfer function — almost always written HRTF — in depth in ambisonics, HRTF, and binaural rendering; here you only need the result: applying an HRTF to a voice makes a flat pair of earbuds sound like the voice is genuinely out in the room, above, below, or behind you, not just left or right.

Figure 1. A flat call collapses every voice to the centre of your head; a spatial call maps each tile to a position and renders it to the matching ear, so overlapping talkers stay separable.

Why your brain cares: spatial release from masking

The reason spatial audio helps is older than any software — it is how human hearing evolved to work in a crowd. The classic name is the cocktail party effect: your ability to lock onto one voice in a noisy room and tune out the rest. You do it mostly with two ears. Having a separate signal at each ear lets you process a target voice far more efficiently than a single channel can.

The specific benefit has a name: spatial release from masking. "Masking" is when one sound drowns out another. When a target voice and a competing voice come from the same direction, the competing voice masks the target heavily and you strain to follow it. Pull the two apart in angle, and intelligibility climbs — that is the "release." Researchers even define a minimum angular separation, the smallest angle between a target and competing talkers needed to improve speech intelligibility by twenty percent. Below that angle the voices blur together; above it they separate.

There is a second, related payoff that matters for long meetings: the spatial release of cognitive load. A 2017 study in the Journal of the Association for Research in Otolaryngology measured prefrontal brain activity while listeners followed a target talker against a competing one, and found that separating the talkers in space reduced the mental effort of listening. The same study added an important caveat the marketing pages skip: the benefit was strongest at an intermediate difference in loudness between the talkers and shrank when one talker was much louder or much quieter than the other. In plain terms, spatial separation helps most when the competing voices are roughly comparable in level — which is exactly the typical group-call situation where everyone's microphone is normalized to a similar volume.

Pitfall — selling "immersion" instead of comprehension. Spatial audio in a call is not about making meetings feel cinematic. The measurable, defensible benefit is reduced listening effort and easier separation of overlapping talkers. Pitch it as "follow crosstalk without fatigue," not "be transported into the room" — the second claim is the kind a sceptical buyer disproves in one demo on laptop speakers.

The one condition that makes or breaks it: the listening device

Here is the rule that catches every team building this for the first time. Spatial audio needs two independent channels reaching your two ears. If the device cannot deliver a distinct left and right signal, there is no spatial effect — at best, no benefit; at worst, the voices smear.

That single requirement explains every platform limitation you will read about next. Microsoft is explicit: Teams spatial audio works on USB-wired stereo headphones, wired stereo speakers, or the device's built-in stereo speakers — and not on Bluetooth audio devices, because classic Bluetooth headsets switch to a low-quality mono profile the moment a microphone is active. Microsoft notes that the next-generation LE Audio Bluetooth standard, which can carry stereo while the mic is live, will be supported. We explain why classic Bluetooth collapses to mono on a call in echo cancellation on speakerphones, Bluetooth, and AirPods and why LE Audio changes that in LC3 and LC3plus, the new Bluetooth audio default.

The practical consequence is that a large share of your users — anyone on common Bluetooth earbuds during a call, or anyone on a laptop with the lid angled so the stereo speakers fire sideways — gets no spatial effect at all. A well-built feature detects this and falls back to the normal mono mix silently rather than shipping a broken-sounding stereo image. Teams does exactly that, and also disables spatial audio when network bandwidth or device memory runs low.

How the major platforms do it in 2026

Four consumer-grade and one developer-grade implementation cover the field. They differ less in the underlying idea than in how far they take it and what they require of the listener.

Microsoft Teams

Teams maps each participant's voice to the position of their tile on your screen and renders the result to stereo. It is a meeting-room metaphor: voices spread left to right across the gallery. The constraints are the ones above — stereo wired output only, no Bluetooth, and it is not available in one-to-one calls or in meetings larger than 100 attendees, the two cases where a spatial spread either adds nothing (one person) or cannot be laid out cleanly (a hundred tiles). You enable it under Settings → Devices.

Zoom

Zoom positions participants' voices in the stereo field based on where they sit in Gallery or Immersive View. Its current limitation is the reverse of a consumer feature: spatial audio is tied to Zoom Rooms and the Zoom Workplace desktop app, in gallery or immersive layouts, and likewise does not support wireless headphones. The design target is the conference room with a proper stereo speaker bar, not the individual on earbuds.

Google Meet and Google Beam

Standard Google Meet has moved more cautiously: in late 2025 it added stereo sound for shared presentations, widening the audio image for screen-shared content rather than positioning each participant. The more ambitious spatial work lives in Google Beam, the 3D telepresence system, which in 2026 extended to group meetings on Zoom and Meet and anchors each voice to the speaker rendered at life-size on the Beam display — the strongest position-to-person binding of any mainstream system, because the "screen position" is a true 3D rendering of the person.

Apple FaceTime

Apple takes the binaural route furthest. FaceTime spatial audio places voices to match each person's position on screen and renders them with an HRTF and head tracking through AirPods, so the sound field stays fixed as you turn your head. Apple also offers Personalized Spatial Audio: you scan your head and ears with an iPhone camera to build a custom HRTF, replacing the generic one for a sharper, more individual effect, and that profile syncs across your Apple devices. FaceTime pairs this with Voice Isolation, a microphone mode that suppresses background noise so the positioned voices stay clean — the kind of noise suppression we cover in noise suppression: RNNoise, Krisp, NVIDIA RTX Voice.

Winner cell for "rendering" depends on use case: HRTF + head tracking (FaceTime) is the most convincing on headphones; a stereo pan (Teams/Zoom) is the most portable.

Building it yourself: the WebRTC path

If you run your own real-time service on WebRTC, you do not get spatial audio for free — but the browser already ships the tools. The pipeline that delivers each participant's voice to the browser is the one we walk through in the WebRTC audio pipeline end-to-end; spatial audio is a processing stage you add after each remote voice arrives and before it reaches the speakers.

The engine is the Web Audio API, a W3C Recommendation since 17 June 2021, and specifically its PannerNode. A PannerNode takes an (x, y, z) position for a sound source and an (x, y, z) position and orientation for the listener, and computes the stereo output. It offers two panningModel settings. The equalpower model is a cheap left-right pan — fast, low memory, good enough for a flat gallery. The HRTF model convolves the voice with measured human ear-impulse responses for a true binaural image; it sounds far better on headphones but uses more memory and CPU, which the spec itself flags as a concern on low-end mobile devices.

The architecture in practice, drawn from LiveKit's published WebRTC tutorial, is straightforward. Each remote participant's audio track becomes a MediaStreamAudioSourceNode. You give each one its own PannerNode. Because the panner accepts only one position, you compute each remote voice's position relative to you with one subtraction, then update the panner as people move or as the gallery re-lays-out. Here is the core, reduced to the essential calls:

// Turn one remote WebRTC voice into a positioned stereo source.
const ctx = new AudioContext();
const source = ctx.createMediaStreamSource(remoteStream);
const panner = ctx.createPanner();

panner.panningModel = "HRTF";      // true binaural; use "equalpower" to save CPU
panner.distanceModel = "exponential";
panner.refDistance = 100;          // distance at which volume is unattenuated
panner.maxDistance = 500;          // beyond this, no further attenuation
panner.rolloffFactor = 2;          // how fast volume drops with distance

source.connect(panner).connect(ctx.destination);

// Position = remote tile minus my own position, so it is relative to me.
const relX = remote.x - me.x;
const relY = remote.y - me.y;
panner.positionX.setTargetAtTime(relX, ctx.currentTime, 0.02); // 20 ms glide
panner.positionZ.setTargetAtTime(relY, ctx.currentTime, 0.02); // map 2D y → z

Two details from that snippet are worth the words. First, the distance model (exponential, with refDistance, maxDistance, and rolloffFactor) lets a far-away participant in a virtual room sound quieter — useful in a spatial "office" layout, irrelevant in a fixed gallery where everyone is the same distance. Second, the setTargetAtTime call with a small time constant (here twenty milliseconds) glides the position instead of snapping it, so a voice that jumps tiles does not click. Snapping positions is a common cause of artefacts.

Where do the positions come from? In a gallery, you map each tile's column to an x between left and right. In a virtual-space app — a 2D office, a game-style room — each avatar's coordinates feed the panner directly, and you also send your own position to peers over a WebRTC data channel. Whether the mixing happens in each client or on the server depends on your topology, which we compare in audio in SFU vs MCU vs P2P: an SFU forwards each voice as a separate track, which is exactly what client-side spatialization needs, whereas a server-side mixer (MCU) would have to render the spatial mix per listener — far more expensive.

A worked layout example

Suppose a four-person gallery and you want voices spread across a 180-degree arc in front of the listener. Map the four columns to azimuth angles of −60°, −20°, +20°, and +60°. Convert each angle to a unit position on the horizontal circle: x = sin(angle), z = −cos(angle). For the +60° talker:

x = sin(60°)  = 0.87   (well to the right)
z = −cos(60°) = −0.50  (in front of the listener)

Feed (0.87, 0, −0.50) to that participant's panner and their voice lands up and to the right of centre — matching the top-right tile. Repeat for each, and the gallery's left-to-right layout becomes an audible left-to-right arc.

The coming change: spatial voice over the network itself

Everything above renders spatial audio on the listener's device from mono voices. A different approach is arriving from the telecoms standards world: carry the spatial information in the codec, across the network.

That codec is IVAS — Immersive Voice and Audio Services — standardized by 3GPP in Release 18 (frozen June 2023) as the first codec built for immersive communication on 5G networks. IVAS is backward-compatible with the EVS speech codec already in the speech-codec family, and it carries mono, stereo, multichannel, ambisonics, and object-based audio. Its conferencing mode, called Independent Streams with Metadata, transmits each participant as a separate voice stream with position metadata, and the receiving device renders them into a spatial scene — and can line them up with a video scene sent in parallel.

The numbers to remember: IVAS spans 13.2 to 512 kbit/s, and in its parametric object mode it can carry three or four freely-placed voices at just 24.4 or 32 kbit/s and reconstruct their positions at the receiver. That is spatial conferencing at near-ordinary-call bitrates. The catch is deployment: IVAS needs support in both the network and the handset, and that rollout is early in 2026 — so for a service you ship today, the device-side Web Audio approach is the one that works now, with IVAS as the path to watch for native mobile spatial calls.

When it is worth it, and when it is a gimmick

Spatial audio earns its keep when three things are true: people talk over each other often, listeners wear stereo headphones or sit at stereo speakers, and meetings run long enough that fatigue matters. Multi-party discussions, panels, classrooms, telehealth group sessions, and contact-centre supervisor monitoring all fit. It is a gimmick — cost without benefit — in the opposite cases: one-to-one calls (no crosstalk, nothing to separate, which is why Teams disables it there), large town-halls where one person presents to a silent audience, and any context where most users are on a single mono speaker.

Pitfall — ignoring the device mix. Before building, measure what your users actually listen on. If most of your traffic is mobile users on a single phone speaker, or on classic Bluetooth earbuds, spatial audio will be off for the majority and the engineering spend buys little. Instrument the output-device type first; build the feature for the segment that can hear it.

Where Fora Soft fits in

We have built real-time audio for video conferencing, telemedicine, e-learning, and live-shopping products since 2005, and spatial audio sits squarely in the part of the pipeline we work in daily: the WebRTC client, the SFU that forwards each voice as its own track, and the device-and-network conditions that decide whether a feature like this helps or quietly fails. In group-call products the recurring engineering questions are the unglamorous ones — detecting the listener's output device, falling back to mono cleanly, gliding positions without artefacts, and deciding whether to spatialize per client or per listener on the server. Those are exactly the decisions that separate a spatial-audio feature that reduces meeting fatigue from one that generates support tickets.

What to read next

• Ambisonics, HRTF, and binaural rendering
• The WebRTC audio pipeline end-to-end
• Audio in SFU vs MCU vs P2P

Call to action

• Talk to a audio engineer — book a 30-minute scoping call to talk through your spatial audio in conferencing plan.
• See our case studies — 250+ shipped projects across video streaming, WebRTC, OTT, telemedicine, e-learning, surveillance, and AR/VR.
• Download the Spatial Conferencing Decision Sheet — One page: when to build it, the device gate, how each platform does it, and the Web Audio recipe.

References

1. Microsoft Support — "Spatial audio in Microsoft Teams meetings." Voices map to on-screen tile positions; requires USB-wired stereo headphones, wired stereo speakers, or built-in stereo speakers; Bluetooth not supported (LE Audio future); not available in 1:1 calls or meetings > 100; auto-disables on low bandwidth/memory. Accessed 2026-06-07. https://support.microsoft.com/en-us/office/spatial-audio-in-microsoft-teams-meetings-547b5f81-1825-4ee1-a1cf-f02e12db4fdb
2. Microsoft Teams Blog — "Follow conversations with ease using Spatial Audio in Microsoft Teams." Spatial separation reduces cognitive load and meeting fatigue; cocktail-party framing. Accessed 2026-06-07. https://techcommunity.microsoft.com/blog/microsoftteamsblog/follow-conversations-with-ease-using-spatial-audio-in-microsoft-teams/3888524
3. W3C Recommendation — "Web Audio API," 17 June 2021. PannerNode, panningModel enum (equalpower vs HRTF), AudioListener, distance models. Official W3C Recommendation (final stage). https://www.w3.org/TR/2021/REC-webaudio-20210617/
4. MDN Web Docs — "PannerNode: panningModel property." equalpower is simple/efficient; HRTF convolves measured human impulse responses, higher quality, higher memory/CPU; concern on low-end mobile. Accessed 2026-06-07. https://developer.mozilla.org/en-US/docs/Web/API/PannerNode/panningModel
5. 3GPP — "IVAS: taking 3GPP voice and audio services to a new immersive level." IVAS standardized in Release 18 (June 2023); 13.2–512 kbit/s; object/ISM mode for multi-party conferencing; receiver-side spatial rendering. Standards-body source. https://www.3gpp.org/technologies/ivas-highlights
6. Fraunhofer IIS — "Immersive Voice and Audio Services (IVAS)." ISM multi-party conferencing combines participants into an audio scene; parametric object coding carries 3–4 placed objects at 24.4 / 32 kbit/s; can match a parallel video scene. Accessed 2026-06-07. https://www.iis.fraunhofer.de/en/ff/amm/communication/ivas.html
7. LiveKit Blog — "Using WebRTC + React + WebAudio to create spatial audio" (Neil Dwyer). Per-track PannerNode, relative-position math, distanceModel = exponential, refDistance 100 / maxDistance 500 / rolloffFactor 2, setTargetAtTime glide. Verbatim implementation pattern. Accessed 2026-06-07. https://livekit.com/blog/tutorial-using-webrtc-react-webaudio-to-create-spatial-audio
8. Zoom Support — "Using spatial audio in meetings and webinars." Positions voices by Gallery / Immersive View seat; available in Zoom Rooms and Zoom Workplace desktop; no wireless headphone support. Accessed 2026-06-07. https://support.zoom.com/hc/en/article?id=zm_kb&sysparm_article=KB0075693
9. Apple Support — "Change the FaceTime audio settings on iPhone" and "Control Spatial Audio and head tracking." FaceTime spatial audio with HRTF + head tracking via AirPods; Personalized Spatial Audio from a head/ear scan; Voice Isolation mic mode. Accessed 2026-06-07. https://support.apple.com/guide/iphone/change-the-facetime-audio-settings-iphb54d5dee2/ios
10. Wenzel et al. (2017), "The Spatial Release of Cognitive Load in Cocktail Party Is Determined by the Relative Levels of the Talkers," J. Assoc. Res. Otolaryngol. 18(2). Spatial separation reduces prefrontal listening effort, strongest at intermediate target-to-masker ratio. Peer-reviewed primary source. https://pmc.ncbi.nlm.nih.gov/articles/PMC5418156/
11. AV Magazine — "Google Beam adds 3D group support for Zoom and Meet" (2026) and Google Meet stereo-presentation update (2025). Beam anchors each voice to the life-size rendered person; Meet adds stereo for shared presentations. Accessed 2026-06-07. https://www.avinteractive.com/news/collaboration/google-beam-adds-3d-meeting-support-for-zoom-and-meet-26-05-2026/
12. IETF RFC 8825 — "Overview: Real-Time Protocols for Browser-Based Applications" (January 2021). Applicability statement defining the WebRTC protocol suite that carries each participant's audio between browsers — the transport layer beneath the device-side spatialization described here. Official IETF standard. https://www.rfc-editor.org/info/rfc8825/