Leveraging Web Audio in the Design of an Experiment to Measure the Effect of Authentic Spatial Sound on Verbal Working Memory in Online Virtual Reality Learning Environments

Leveraging Web Audio in the Design of an Experiment to Measure the Effect of Authentic Spatial Sound on Verbal Working Memory in Online Virtual Reality Learning Environments Vincent Russell School of Computer Science & Information Technology University College Cork Cork, Ireland v[email protected] David Murphy School of Computer Science & Information Technology University College Cork Cork, Ireland d.murph[email protected] ABSTRACT This paper presents the planned experimental design of an online virtual reality learning environment (VRLE) to investigate the effect of authentic spatial sound on verbal working memory. Despite the increasing integration of virtual reality into education, the cognitive impact of spatial audio remains underexplored, particularly in relation to working memory processes. This study addresses that gap by embedding an adapted version of the Automated Operation Span (AOSPAN) task into a dimensionally accurate VR simulation of a real-world laboratory, constructed using Three.js on a browser-based WebXR platform and exploiting the advanced spatial sound processing capabilities of Web Audio. Authentic spatial audio is captured using a Sennheiser AMBEO VR microphone and rendered in first-order Ambisonics via Resonance Audio powered by Omnitone and delivered through the Web Audio API. The AOSPAN task is administered twice per participant, once with authentic ambient spatial sound and once without it, in a counterbalanced design. Performance is measured through standard AOSPAN scoring metrics, allowing for within-subject comparison to determine the influence of authentic spatial sound on verbal working memory, with particular focus on the articulatory rehearsal process of the phonological loop. By employing a fully web-based delivery model, this framework supports wide accessibility and future scalability while maintaining experimental control. The paper contributes a novel methodology for cognitive testing in immersive environments and offers insights into the role of spatial sound in enhancing or modulating memory performance in VR education. The findings of the proposed experiment have implications for both audio interface design and the development of effective and cognitively supportive virtual learning environments. Licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0). Attribution: owner/author(s). Web Audio Conference WAC-2025, November 19–21, 2025, Paris, France. © 2025 Copyright held by the owner/author(s). Keywords Virtual Reality, VR, WebXR, Web Audio, Spatial Sound, 3D Audio, Ambisonics, Working Memory, Working Memory Capacity, Memory Recall, Learning Environments, VRLE, Three.js. 1. INTRODUCTION The widespread availability of affordable consumer headmounted displays (HMDs) has made virtual reality (VR) one of the most cutting-edge platforms for educators to engage in interactive teaching and learning. Researchers, teachers, and instructional designers are increasingly recognising the benefits of incorporating VR technology into educational and training scenarios. The Internet, supported by the current WebXR specification, now provides a means to deliver VR experiences on various devices and platforms, enabling remote VR education that is accessible to all, whether through an immersive system, such as a HMD, or a nonimmersive system like that of a standard desktop browser [34]. VR presents countless potential educational advantages, yet numerous aspects of this immersive technology remain that require additional scrutiny. While the visual modality has received much of the focus from researchers, the aural modality has received comparatively less attention and is often an afterthought when designing VR experiences. One auditory area of research, that requires further investigation, is the impact of authentic spatial sound on working memory, and ultimately learning, when used in educational VR applications. In relation to spatial sound, authenticity has the quality of being true and accurate to its ‘real-world’ sound source. When considered in sound design for VR, audible authenticity influences a user’s conviction that the virtual experience they encounter possesses a tangible existence in the real world [40]. This belief in the authenticity presented impacts the sense of presence and immersion within the virtual space. 1.1 Working Memory Working memory (WM) is a cognitive system within the brain that offers temporary storage and manipulation of information essential for complex cognitive activities such as language comprehension, learning, and reasoning [2]. The original WM model, introduced by Baddeley and Hitch in 1974 [4] outlined short-term memory as a complex three-part system consisting of the central executive, and its slave systems; the phonological loop and the visuo-spatial sketchpad. The central executive controls the information flow to and from the slave systems, the phonological loop is responsible for processing auditory information, while the visuo-spatial sketchpad handles visual and spatial information. Baddeley later revised the model [3] to include a fourth component; the episodic buffer, which integrates information from the phonological loop, the visuo-spatial sketchpad, and longterm memory to form episodic representations. The phonological loop can be further divided into two subcomponents; the phonological store and the articulatory rehearsal process. The phonological store temporarily holds verbal information (e.g. words or digits), while the articulatory rehearsal process refreshes the information in the phonological store using subvocal rehearsal (i.e. silent repetition with the inner voice). The focus of the experiment outlined in this paper is the phonological loop, with particular attention to the articulatory rehearsal process. 1.2 Related Research Although considerable research has been conducted in the area of WM in VR, it is more often used as a novel platform to conduct and evaluate memory tests (e.g. [1, 10, 9]), or to investigate responses in VR experiences that impact WM (e.g. [26, 20, 30]). However, the relationship between aural modality and WM within VR remains significantly overlooked. Research in this domain has mainly focused on shortterm memory (STM), particularly with regard to immediate memory recall. In the academic literature, STM is frequently equated with WM, despite the conceptual distinctions between the two. The discourse surrounding this topic is marked by considerable debate; some scholars argue that STM and WM exhibit significant overlap [27], while others maintain that they are fundamentally different constructs [4], with the distinction often being semantic. Nevertheless, there is a general agreement that STM and WM are interconnected, with STM being regarded a component of the broader WM framework [12]. As part of their study, Ruotolo et al. [35] used VR to assess the short-term verbal memory response of individuals to the noise of a projected new motorway. Their results indicated a negative impact on short-term verbal memory, with the influence increasing with proximity to noise source. As part of a wider study, Cowen et al. [11] investigated how sound impacted the time it took to complete a task in a virtual environment. The experiment used four auditory conditions: no sound, white noise, operating room ambient sounds, and operating room ambient sounds combined with drill noise. The study found that tasks completed under ambient sound conditions (without distracting drill noise) had the shortest completion times and, therefore, the highest performance. In the VirSchool research project [19], the effect of background music on memory recall was tested using two different immersive displays. The findings showed that the recall of facts, from material presented with background music, differed between the display devices. However, the more immersive display showed a positive increase in recall of facts accompanied by background music. It should be noted that none of the aforementioned research directly measured the effect of sound on WM and also did not incorporate spatialised sound. During a study involving desktop conferences, Baldis [6] explored the impact of spatial audio on memory. The author believed that spatial audio enhanced clarity and that the spatialisation of sound served as an additional memory cue to aid in recall. The results indicated that spatial audio allowed for more efficient use of working memory by making it easier to differentiate and pinpoint sound sources, enabling participants to concentrate more effectively on the presented content. Although the study suggested improved memory retention with spatialised audio, it utilised a traditional speaker setup rather than fully immersive binaural spatial audio in a VR setting, and did not include background ambient sound. 2. DESIGN TECHNOLOGIES 2.1 WebXR Extended reality (XR) encompasses virtual reality (VR), augmented reality (AR), and mixed reality (MR). The WebXR Device application programming interface (API)1is a W3C2standard that facilitates the creation of web-based XR applications, exposing access to input and output functionalities of VR and AR devices. This API establishes a uniform abstraction layer for real-time rendering [24], supporting a cross-platform“build once, deploy everywhere”development approach that uses familiar web technologies (e.g. HTML, CSS, and JavaScript) to create XR content. WebXR applications are compatible with a range of devices, including wireless and wired VR HMDs, AR headsets, mobile phones, tablets, and desktop computers, without the need for additional plug-ins or software. Due to extensive support and availability of the unified API across affordable and premium devices running on differing operating systems, WebXR is streamlining the access to XR content for everyone, enabling broader adoption and promoting more innovative applications. The WebVR API3, developed by the Immersive Web Group4in 2014, served as the precursor to the WebXR API. It significantly contributed to the accessibility of immersive experiences through web browsers and consumerlevel HMDs. WebVR addressed two critical issues in the VR field; limited content growth and lack of cross-platform support [14], by leveraging the openness of the Web. This led to a surge in content creation for various platforms, ultimately paving the way for the development of the WebXR API. The WebXR API seeks to merge the realms of VR and AR, and to provide users with enhanced virtual experiences by extending the range of input options to include the likes of hand gestures and voice commands [24]. The widespread acceptance of WebXR across platforms and browsers makes it the perfect technology to create educational applications to share learning materials [34, 16, 18]. Therefore, it is selected as the framework for constructing 1https://developer.mozilla.org/en-US/docs/Web/API/ WebXR Device API 2https://www.w3.org 3https://developer.mozilla.org/en-US/docs/Web/API/ WebVR API 4https://www.w3.org/immersive-web the experimental VR environment utilising Three.js. 2.2 Three.js Three.js5, an open-source JavaScript library and API, stands out as a highly recognised software framework designed to develop WebXR applications. It uses WebGL6 (Web Graphics Library), also an open-source and crossplatform compatible JavaScript API, to render 3D graphics in web browsers. WebGL, created by the Khronos Group, is built on OpenGL ES 2.07and exploits OpenGL’s shading language GLSL to access the additional processing power needed for 3D interactive websites by executing code directly on the graphics card [31]. WebGL offers direct access to hardware through the exposure of low-level Document Object Model interfaces of HTML5 webpages. Due to its low-level nature, creating basic 3D objects involves a significant amount of coding and is therefore prone to errors. Three.js streamlines the creation of interactive 3D web applications by offering a user-friendly JavaScript API built around the capabilities of WebGL [15], eliminating the need to manually code and compile intricate WebGL commands as all complex operations are managed in the backend. The decision to use Three.js for creating the VR platform of this experiment is also influenced by its integrated spatial audio functionality and its seamless integration with other JavaScript audio processing libraries, such as Resonance Audio, which are developed on the Web Audio JavaScript API. The inclusion of spatial audio is a critical factor in this research. 2.3 Web Audio API The W3C Web Audio API8is a high-level JavaScript API designed to process and synthesise audio in web applications. It has powerful audio handling capabilities that enable developers to create rich and interactive audio experiences directly in the browser. Before its release, Flash and HTML5 audio were commonly used for audio playback on the Web, but they had limitations that hindered the development of complex games and interactive applications. Flash needed a plug-in to function, and HTML5 audio lacked precise timing controls, had restrictions on the number of simultaneous sounds, lacked reliable pre-buffering, and did not support real-time effects. The Web Audio API overcomes the constraints of its forerunners by incorporating functionalities and processing tools commonly seen in contemporary game engines and desktop audio production software. Also notable is that it is compatible across various platforms and browsers without the need for extra plugins to execute. It offers a range of standard nodes, AudioNodes, for fundamental audio processing tasks such as volume control, filtering, adding time delay, creating reverberation, and applying dynamics processing. In 2014, the W3C Audio Working Group introduced the AudioWorklet to enhance the capabilities of the API, enabling developers to incorporate bespoke low-level audio processing [8]. 5https://threejs.org 6https://www.khronos.org/webgl 7https://www.khronos.org/opengles 8https://developer.mozilla.org/en-US/docs/Web/API/ Web Audio API In the context of VR, the Web Audio API includes integrated hardware-accelerated positional audio capabilities and offers a low-latency precise-timing system [39]. These functionalities are crucial for interactive VR applications that require rapid auditory feedback to user inputs and the ability to schedule audio events at precise times during the virtual experience. To accomplish this, the Web Audio modules convert C++ audio processing code to WebAssembly9to execute complex computational tasks directly in the browser’s native code. This significantly enhances performance speeds compared to slower JavaScript implementations of mathematically-intensive algorithms [22, 23]. Audio operations are conducted exclusively within an audio context, which provides a consistent timing model and frame of reference for time. Spatialisation of sound is controlled in the Web Audio API through a source-listener model. The AudioListener represents the position and orientation of a single listener (i.e. the user) in 3D space within the audio context. The PannerNode interface defines the position and orientation of a sound source relative to the listener, calculating azimuth and elevation angles for rendering spatialisation effects. The ConvolverNode is regularly used to create room effects with reverberation. It does so by applying impulse responses to “dry” sound in an AudioBuffer and executing a Linear Convolution to create a convolved, “wet” mix, sound. The GainNode interface is used to adjust volume levels and, together with the ConvolverNode, exploited extensively in Omnitone. 2.4 Omnitone Omnitone10 is a JavaScript library that is cross-browser compatible and intended to be optimised for interactive web applications. It is designed to enable binaural rendering of Ambisonic audio directly within web browsers. Ambisonics, developed by Michael Gerzon [21] in the 1970s, is a fullsphere surround sound technology that captures the sound pressure field of an auditory scene and reproduces the audio in a way that allows for 3D spatial representation of the sound. Unlike conventional surround sound systems, Ambisonics encompasses sound sources above and below the listener, offering a more immersive and realistic audio experience. Using encapsulated spatial cues, Ambisonics is able to produce a 3D sound field that allows listeners to accurately localise sound sources around them [25]. Omnitone optimises its performance by leveraging fast, native features from the Web Audio API, including GainNode and ConvolverNode. In addition, it takes advantage of head-related transfer functions (HRTF) from SADIE’s binaural filter library11. HRTFs are crucial in virtual systems, as they replicate the direction-dependent filtering of sound as it interacts with the head and torso in real life. Omnitone supports First Order Ambisonic (FOA) and Higher Order (2nd and 3rd) Ambisonic (HOA) streams, using virtual speakers and HRTF convolution to output a final binaural rendered audio stream with full-sphere surround sound effect. When experienced through headphones, this technology effectively creates a sense of space, making it ideal 9https://webassembly.org 10https://googlechrome.github.io/omnitone 11https://www.york.ac.uk/sadie-project/GoogleVRSADIE. html for immersive VR environments. Implementing it internally for Ambisonic decoding and binaural rendering, Omnitone is the driving force behind Resonance Audio. 2.5 Resonance Audio Resonance Audio12, an open source spatial audio JavaScript software development kit (SDK), which replicates the interaction of real sound waves with human ears and the surrounding environment. It was introduced by Google in 2017 and is optimised for mobile and desktop computing. To achieve accurate simulation of real-world sound wave interactions with human ears, Resonance Audio takes advantage of the SADIE HRTF library inherited from Omnitone. Resonance Audio employs a user-defined room model to create realistic spatial audio reflections and reverberation in the VR scene. The room model, delimited by dimensions in metres, establishes an audio space with six surfaces (four walls, ceiling, and floor). The absorption properties of the materials used for the room surfaces are chosen from a set of predefined, frequency-dependent absorption coefficients (e.g. ‘acoustic-ceiling-tiles’, ‘brick-painted’ and ‘glass-thin’). An accurately defined room model ensures that the audio aligns with the user’s visual experience. Although Google’s official support for the SDK ended in 2019, and little development has occurred since, Resonance Audio continues to rank highly compared to more recent alternatives (e.g. Atmoky13 and Superpowered14) and remains a relevant library for creating immersive sonic experiences in WebXR [41]. Resonance Audio was selected to manage the spatial sound of the VRLE due to its detailed documentation, ease of implementation, Ambisonic playback, configurable room model, and its ability to update the sound scene based on the head orientation of the listener. 3. METHODOLOGY 3.1 Virtual Reality Learning Environment The complete experiment process is carried out within the online virtual reality learning environment (VRLE) with Three.js being employed to create the virtual platform, import the 3D models, and handle all the user interactions. The VRLE is a dimensionally accurate replica of a pharmacology laboratory (Fig. 1) located in the Western Gateway building on the campus of University College Cork. This learning space was chosen because it translates well into an online VRLE that could be populated with multiple student users, but more importantly for this experiment it was selected because of its rich acoustic signature. With the windows open, students in the laboratory are enveloped in sounds emanating from natural and man-made sources (e.g. birds, people, laboratory equipment, air conditioning, and road traffic) from all directions. The inclusion of this authentic spatial sound into a web-based VRLE and the investigation of its effect on working memory is the motivation behind the study. As the VRLE is web-based, a number of design decisions must be made to reduce download times and minimise excessive processing demands once deployed via the Internet 12https://resonance-audio.github.io/resonance-audio 13https://atmoky.com 14https://superpowered.com Figure 1: The pharmacology laboratory used as the basis for the VRLE. Figure 2: The 3D model of the laboratory constructed in Blender. browser. The virtual laboratory is modelled in Blender15 (Fig. 2) with the room and all objects joined in one mesh and exported as a GLB file, a binary version of the glTF (Graphics Library Transmission Format) file format. All colours and images used in the room and its contents are also merged onto one texture map and exported as a single large image texture. Using the GLTFLoader in Three.js, the GLB model is loaded into the virtual scene, and the image texture is linked to its single material. This combination method drastically reduces the number of draw calls required to build the scene, thereby providing optimal performance and efficiency while alleviating pressure on GPUs without a significant loss in visual quality. Illuminating virtual environments can require substantial computational resources, therefore restricting the number of lights used and utilising pre-rendered lighting on 3D model surfaces, lessens the burden on processors. To achieve this, in Blender, the lighting of the 3D laboratory model is baked into the image texture used to colour the model, and only one ambient light is used in Three.js to illuminate the virtual scene. Upon commencement of the VRLE immersive experience, participants are placed in the middle of the virtual laboratory (Fig. 3). For the purposes of the experiment, there is 15https://www.blender.org Figure 3: The completed WebXR VRLE with introductory user interfaces as part of the ”onboarding” stage. no requirement to explore the room. Therefore, movement within the scene is restricted, with only head and hand controller interactions enabled for the participant. Before starting the practice phase of the experiment, a period of time is set aside for self-guided orientation in the VRLE environment, allowing participants to acclimate and become acquainted with permissible interactions. This “onboarding” stage not only allows participants to settle into the virtual environment and get comfortable with the hardware equipment, but is essential to ensure that users fully comprehend and embrace the VR application [7]. Due to the variation in VR experience across participants, the onboarding phase in not time restricted. The Three.js implementation of sound within a scene is limited to non-positional (global) and positional audio objects. Although adequate for most online VR applications, this narrow application of spatial audio does not support the playback of Ambisonic sound files, a prerequisite for the experiment design. To overcome this limitation, Resonance Audio manages the audio configuration of the VRLE. 3.2 Audio Configuration To test the effect of authentic spatial sound on verbal working memory, it is imperative to first capture authentic spatial sound. To achieve this, the Sennheiser AMBEO VR Mic16 is used to capture the signature 360o, threedimensional sound field of the laboratory. The AMBEO VR Mic employs four matched KE 14 cardioid capsules in tetrahedral arrangement to record FOA spatial sound in raw, four-channel, A-format arrangement (1: front-left-up [FLU], 2: front-right-down [FRD], 3: back-left-down [BLD], 4: back-right-up [BRU]). To make the recording compatible with popular delivery platforms, including Resonance Audio, it is necessary to convert it to B-format Ambisonics. This process is simplified by the specifically designed Sennheiser AMBEO A-B format converter plugin (Fig. 4), which can be seamlessly embedded in post-production software, in this case the Reaper17 digital audio workstation. Reaper is capable of handling up to 64 audio channels per track, which is equivalent to seventh-order Ambisonics, exceeding the 4-channel FOA requirements of the recording. The B-format represents the sound field around the microphone in a W, X, Y, Z configuration, with W being the omnidirectional sum of all four capsules, and X, Y and Z 16https://www.sennheiser.com/en-gb/catalog/products/ microphones/ambeo-vr-mic/ambeo-vr-mic-507195 17https://www.reaper.fm Figure 4: The AMBEO A-B Format Converter plugin for Reaper. being the three virtual bi-directional microphone patterns for front/back, left/right and up/down. It is important to highlight that the AMBEO VR Mic is limited to capturing FOA audio, and is chosen due to limited hardware availability. Although there are other microphones available that can record HOA, which provide superior audio resolution, they also result in significantly larger file sizes. This presents a trade-off, as the lower resolution granularity offered by FOA allows for smaller file sizes, facilitating quicker delivery over the Internet. Despite being restricted to FOA, the AMBEO VR Mic effectively captures the authentic sound signature of the laboratory environment, making it suitable for the objectives of this experiment. The decision to utilise this microphone balances the need for manageable file sizes with the requirement for adequate audio fidelity, ensuring that the recorded sound meets the experimental standards. The AMBEO VR Mic is placed in the exact centre of the laboratory, in upright orientation and facing the front of the room, at a height of 1.6m (Fig. 5). This matches the positioning of the default camera in 3D space coordinates (0, 1.6, 0) within the VRLE. The recording is captured on a Tascam DR-680 MkII18 multitrack field recorder in highresolution WAV format at 48kHz sampling rate, with 24 bits per sample. The four mono WAV files, of the 4-channel raw A-format, are subsequently loaded into Reaper where they are converted to B-format and edited to loop seamlessly. Ultimately, the recording is rendered as a 4-channel WAV at 48kHz with 16-bit PCM bit depth in AmbiX (ACN-SN3D) format [28], which is supported by Resonance Audio. To prevent the introduction of unwanted audio artefacts as an effect of upmixing from FOA to HOA, the Ambisonics order output in Resonance Audio is configured to 1, aligning with the FOA of the recorded audio file. The dimensions of the Resonance Audio room are configured to match the measurements of the virtual room, which replicates the real-world laboratory. With Resonance Audio units in metres, this corresponds to an audio room of 9.25m in width, 11.5m in length, with a height of 2.88m. The predefined room material coefficients also match the materials in the virtual and real space (i.e. ’acoustic-ceiling18https://tascam.jp/int/product/dr-680mkii/spec Figure 5: The AMBEO VR Mic positioned at the centre of the laboratory, at a height of 1.6m. This corresponds to the location of the virtual camera in VR, i.e. the viewpoint of the participant. tiles’, ’glass-thick’, ’linoleum-on-concrete’, and ’concreteblock-painted’). To enhance the experiment experience, instructional voiceover messages are included during the process to reinforce visual text instructions. Artificial intelligence (AI) speech audio is created using Micmonster19, a text-to-speech tool that generates natural sounding voices with emotions. ‘Emily’ is the chosen AI voice, as it has a soft Irish accent, is moderately paced, and is intelligible. With the purpose of enhancing the experience, these voiceovers are not critical to the research and are not used in any audio loops. Therefore, to reduce download sizes while maintaining respectable quality, the speech audio is saved in mono MP3 format at 48kHz with a bitrate of 192kbps. The mono audio files are then positioned in 3D space, like other virtual objects, close to speakers on the front wall of the VRLE. Convolving mono audio objects with HRTFs spatialises the audio sources according to their position coordinates and is a conventional technique in VR experience design [44]. Resonance Audio dynamically convolves spatialisation and room effects into the “dry” voiceovers in real-time. As a requirement of the experiment, participants must wear headphones so that they are fully immersed in the HRTF binaural rendering of spatialised audio within the VRLE. 3.3 Working Memory Test To measure working memory capacity (WMC) the experiment implements an adapted version of the automated operation span (AOSPAN) task developed by Unsworth et al. [43]. The AOSPAN task is a widely used cognitive psychology tool designed to assess an individual’s ability to store and process information simultaneously, reflecting their WMC. Originally, the OSPAN task was administered by an experimenter and required participants to remember items and recall them in the correct order, while also solv19https://micmonster.com Figure 6: The virtual keypad user interface used for letter recall. ing math problems that serve as a distraction [42, 17]. The AOSPAN task was created as an easy-to-administer, less time-consuming, self-scoring, reliable, and valid version of the original that required little intervention from the experimenter. Although acceptable variations (e.g. number of practice trials and presentation of maths equations) exist in computer-based implementations of the AOSPAN task [29, 33, 5, 13], the arrangement planned for this experiment closely follows the structure and specifications outlined in the Unsworth et al. version [43]. In this experiment, participants are tasked with memorising a sequence of letters that are shown to them, while simultaneously engaging in mental calculations of mathematical problems. As this is a VR implementation, all letters and equations are displayed on flat planar interfaces positioned in 3D space. The letters to be recalled are shown one at a time for 800 milliseconds and each is preceded by a mathematical equation (e.g. (2 x 4) – 1 = 7) that requires a true or false response. When the participant is asked to recall the sequence of letters, in the correct order of presentation, a virtual keypad (Fig. 6) appears with twelve letters (F, H, J, K, L, N, P, Q, R, S, T, and Y), an ‘Undo’ key to delete an incorrect entry, and an ‘Enter’ button for submission. The recall phase is not time limited, but the time to solve the mathematical equations is calculated based on the mean response time, plus 2.5 standard deviations, from a preceding practice session. The AOSPAN task’s practice session consists of three sections before the real trials begin. The first practice section tests a participant’s ability to recall a series of letters. This simple letter span, of letter sets ranging from two to four letters, is presented four times and has no time limitation. The second practice provides an opportunity to familiarise with the equation-solving portion of the task. Participants complete fifteen maths equations and are asked to do so as quickly as possible. This is required to calculate the time (mean response time plus 2.5SD) that is allowed to solve the equations in the real trial, so that letter rehearsal time is minimised. It also factors in the pace of the participant operating in the virtual environment based on their VR experience. The final practice section is a combination of the two previous letter-recall and equation-solving practices in which series of math problems are presented alongside a series of letters, exactly as they would appear in the real trials. Participants receive three equation-letter combinations (of 2-letter size set) before advancing to the real trials. As standard with other versions of the AOSPAN, main trials consist of 15 sets made up of three administrations of each letter set ranging from 3 to 7 letters. This produces a total of 75 equations and 75 letters, that is, 75 operationstorage pairs. It should be noted that the order of letter set sizes is randomised and that participants who score less than 85% in equation-solving are excluded from the data. This criterion is imposed so that participants who are genuinely trying to solve the equations and remember the letter sequences are included in the final results analysis. The AOSPAN task reports five scores on completion: 1. OSPAN score (the sum of all perfectly recalled sets) 2. Total number correct (the total number of letters recalled in the correct position) 3. Maths errors (the total number of maths equation errors) 4. Speed errors (the number of unsolved maths equations due to elapsed time) 5. Accuracy errors (the number of incorrectly solved maths equations) Unlike standard implementations of the AOSPAN, participants undertake the main test twice but under two different audio conditions; once with the presence of authentic ambient sounds and another without, the order of which is randomised. The goal of the experiment is to determine whether the inclusion of authentic spatial sound has an effect on WMC. 3.4 Participants As a WebXR-based experience, the entire experimental procedure and administration can be performed online, making it accessible to a worldwide audience of participants on various devices and platforms. However, to maintain consistency, manage experimental conditions, and ultimately ensure the integrity of the data collected, the experiment is carried out under traditional laboratory conditions, adhering to standard participant recruitment protocols. Individuals of any gender, technical proficiency, and level of VR experience are welcome to participate; however, only those aged 18 and above are considered eligible for the experiment. All participants are required to use VR headsets and are obligated to use the same headphones set at an invariable volume level. All participants experience similar experimental conditions, each completing the onboarding and practice phases under active ambient spatial sound configuration. However, the sequence of the audio conditions is varied during the two main tests. Whether individuals complete the first test with ambient sound on and the second test in silence, or vice versa, is randomised. All participants receive 15 trials in each of the main tests, consisting of three administrations for each set size ranging from 3 to 7 letters. The order of the trials, and the content of the letter sets, is randomised for each individual. To preclude any potential compromise to the results, the experiment is carried out in a controlled setting within an anechoic chamber, which provides an audio-neutral environment for testing. All participants are monitored and utilise the same model HMD with identical audio configuration (i.e. the Meta Quest 3 HMD using its built-in headphones set at 90% volume level). 3.5 Data Collection & Analysis Before starting the immersive VRLE session, participants receive a unique 6-digit alphanumeric ID randomly generated by the system, known only to them, and used should they wish to withdraw their data. The responses to the consent declaration and the demographic questionnaire are collected using web forms in the HMD Internet browser. This anonymised data is transferred to the VRLE for storage until the completion of the experiment, upon which it is combined with the OSPAN test scores and emailed to the researcher via a server-side PHP script. Participants respond to the OSPAN trails and the posttest questionnaire while in the VRLE immersive session. This design decision was made as it is crucial to avoid any break in presence (BIP) [38] that would result from exiting the VRLE. This could cause negative reactions within the participant and, consequently, responses that are less reliable [37]. Research has shown that the use of in-VR questionnaires (inVRQs) is beneficial due to their ability to minimize the impact of a BIP, their non-intrusive nature, and their ability to generate more reliable responses [32]. Although in W3C working draft20, WebXR currently does not support the display of HTML overlays within immersive VR sessions. Due to the lack of compatibility, user response interfaces are instead drawn on canvases, simple 2D planar objects positioned within the 3D space of the VRLE. This approach has shown greater advantages for users when answering questionnaires in VR, as it is faster and less frustrating compared to impractical 3D object-based interfaces [36]. During the experiment, qualitative and quantitative data are collected. The qualitative data consists of nominal and ordinal data types from the demographic and post-test questionnaire, while the quantitative data consist of the discrete OSPAN test scores. The parametric paired t-test is used to interpret the results within subjects (repeated measures) should the underlying scoring data satisfy certain statistical criteria and assumptions (e.g. normality). Alternatively, non-parametric statistical tools, such as the Wilcoxon signed-rank test, are to be applied to interpret the test scoring data. The spatial sound condition (ambient sound present or not) is the independent variable, with the OSPAN test scores being dependent variables. 4. CONCLUSIONS This paper presents the design of an innovative, browserbased experimental framework that uses WebXR and Web Audio technologies to investigate the effect of authentic spatial sound on verbal working memory within online VRLEs. By integrating advanced spatial audio tools, namely the Web Audio API, Omnitone, and Resonance Audio, into a WebXR-based VRLE, the study addresses a significant gap in current cognitive and auditory research by examining the underexplored influence of spatial sound on working memory performance. The methodological rigor applied in the experiment design ensures that the collected data will contribute meaningfully to both the audio research community and broader discussions around immersive educational technologies. In addition, the dual audio conditions of the experiment (i.e. presence and absence of authentic ambient spatial sound) 20https://www.w3.org/TR/webxr-dom-overlays-1 facilitate accurate within-subject comparisons of working memory performance, thereby allowing a robust analysis of how auditory authenticity may enhance or hinder cognitive functioning in immersive learning contexts. Ultimately, this work not only introduces a replicable model for assessing cognitive performance in VR using webnative audio tools but also lays the groundwork for future studies exploring the psychoacoustic and educational implications of spatial sound. As immersive online learning platforms continue to grow in prevalence, understanding how sound design influences cognition will be crucial when creating effective and inclusive VR educational experiences. 5. ETHICS This study has received institutional ethics approval. Social Research and Ethics Committee (SREC) 2024-083. 6. ACKNOWLEDGMENTS The authors wish to thank Billy O’Mahony and Eoghan O’Riain of the MAVRIC research lab, University College Cork, Ireland, and Dr. David Murphy, Broadmoor Hospital, UK. Special thanks to Fionnuala McAdam O’Connell for assistance with the development of the automated operation span (AOSPAN) task in VR. 7. REFERENCES [1] A. Armougum, E. Orriols, A. Gaston-Bellegarde, C. Joie-La Marle, and P. Piolino. Virtual Reality: A New Method to Investigate Cognitive Load During Navigation. Journal of Environmental Psychology, 65:101338, 2019. Publisher: Elsevier. [2] A. Baddeley. Working Memory. Science, 255(5044):556–559, Jan. 1992. [3] A. Baddeley. The Episodic Buffer: A New Component of Working Memory? Trends in Cognitive Sciences, 4(11):417–423, Nov. 2000. Publisher: Elsevier. [4] A. D. Baddeley and G. Hitch. Working Memory. In G. H. Bower, editor, Psychology of Learning and Motivation, volume 8, pages 47–89. Academic Press, Jan. 1974. [5] H. Bailey. Computer-Paced Versus Experimenter-Paced Working Memory Span Tasks: Are They Equally Reliable and Valid? Learning and Individual Differences, 22(6):875–881, Dec. 2012. [6] J. J. Baldis. Effects of Spatial Audio on Memory, Comprehension, and Preference During Desktop Conferences. In Proceedings of the SIGCHI conference on Human factors in computing systems, pages 166–173, 2001. [7] E. Chauvergne, M. Hachet, and A. Prouzeau. User Onboarding in Virtual Reality: An Investigation of Current Practices. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems, pages 1–15, Hamburg Germany, Apr. 2023. ACM. [8] H. Choi. Audioworklet: The Future of Web Audio. In ICMC, 2018. [9] G. Climent, C. Rodr´ıguez, T. Garc´ıa, D. Areces, M. Mej´ıas, A. Aierbe, M. Moreno, E. Cueto, J. Castell´a, and M. Feli Gonz´alez. New Virtual Reality Tool (nesplora Aquarium) for Assessing Attention and Working Memory in Adults: A Normative Study. Applied Neuropsychology: Adult, 28(4):403–415, July 2021. [10] B. Coleman, S. Marion, A. Rizzo, J. Turnbull, and A. Nolty. Virtual Reality Assessment of Classroom–Related Attention: An Ecologically Relevant Approach to Evaluating the Effectiveness of Working Memory Training. Frontiers in psychology, 10:1851, 2019. Publisher: Frontiers Media SA. [11] B. Cowan, D. Rojas, B. Kapralos, F. Moussa, and A. Dubrowski. Effects of Sound on Visual Realism Perception and Task Performance. The Visual Computer, 31(9):1207–1216, Sept. 2015. [12] N. Cowan. What Are the Differences Between Long-Term, Short-Term, and Working Memory? Progress in brain research, 169:323–338, 2008. Publisher: Elsevier. [13] W. De Neys, G. d Ydewalle, W. Schaeken, and G. Vos. A Dutch, Computerized, and Group Administrable Adaptation of the Operation Span Test. Psychologica Belgica, 42(3):177–190, 2002. Publisher: The Belgian Psychological Society. [14] C. Dibbern, M. Uhr, D. Krupke, and F. Steinicke. Can WebVR Further the Adoption of Virtual Reality? Mensch und computer 2018-usability professionals, 2018. Publisher: Gesellschaft f¨ ur Informatik eV Und German UPA eV. [15] J. Dirksen. Learning Three.js: The JavaScript 3D Library for WebGL. Packt Publishing Ltd, 2013. [16] M. Engberg, J. D. Bolter, and B. Maclntyre. RealityMedia: An Experimental Digital Book in WebXR. In 2018 IEEE international symposium on mixed and augmented reality adjunct (ISMAR-Adjunct), pages 324–327. IEEE, 2018. [17] R. W. Engle. Working Memory Capacity as Executive Attention. Current Directions in Psychological Science, 11(1):19–23, Feb. 2002. [18] B. Fanini, D. Ferdani, and E. Demetrescu. Temporal Lensing: An Interactive and Scalable Technique for Web3D/WebXR Applications in Cultural Heritage. Heritage, 4(2):710–724, 2021. Publisher: Multidisciplinary Digital Publishing Institute. [19] E. Fassbender, D. Richards, A. Bilgin, W. F. Thompson, and W. Heiden. VirSchool: The Effect of Background Music and Immersive Display Systems on Memory for Facts Learned in an Educational Virtual Environment. Computers & Education, 58(1):490–500, 2012. Publisher: Elsevier. [20] D. Gabana, L. Tokarchuk, E. Hannon, and H. Gunes. Effects of Valence and Arousal on Working Memory Performance in Virtual Reality Gaming. In 2017 Seventh International Conference on Affective Computing and Intelligent Interaction (ACII), pages 36–41. IEEE, 2017. [21] M. A. Gerzon. Periphony: With-Height Sound Reproduction. Journal of the audio engineering society, 21(1):2–10, 1973. Publisher: Audio Engineering Society. [22] J. Kleimola and O. Larkin. Web Audio Modules. In Proc. 12th Sound and Music Computing Conference, 2015. [23] O. Larkin, A. Harker, and J. Kleimola. iPlug 2: Desktop Plug-in Framework Meets Web Audio Modules. In Proceedings of the 4th Web Audio Conference, 2018. [24] B. Maclntyre and T. F. Smith. Thoughts on the Future of WebXR and the Immersive Web. In 2018 IEEE international symposium on mixed and augmented reality adjunct (ISMAR-Adjunct), pages 338–342. IEEE, 2018. [25] D. G. Malham and A. Myatt. 3-D Sound Spatialization Using Ambisonic Techniques. Computer Music Journal, 19(4):58–70, 1995. Publisher: The MIT Press. [26] M. A. Martens, A. Antley, D. Freeman, M. Slater, P. J. Harrison, and E. M. Tunbridge. It Feels Real: Physiological Responses to a Stressful Virtual Reality Environment and Its Impact on Working Memory. Journal of Psychopharmacology, 33(10):1264–1273, Oct. 2019. [27] G. A. Miller, E. Galanter, and K. H. Pribram. Plans and the Structure of Behavior. 1960. Publisher: Henry Holt and Co. [28] C. Nachbar, F. Zotter, E. Deleflie, and A. Sontacchi. Ambix - A Suggested Ambisonics Format. In Ambisonics Symposium, volume 2011, 2011. [29] F. L. Oswald, S. T. McAbee, T. S. Redick, and D. Z. Hambrick. The Development of a Short Domain-General Measure of Working Memory Capacity. Behavior Research Methods, 47(4):1343–1355, Dec. 2015. [30] D. Pan, Q. Xu, S. Ma, and K. Zhang. The Impact of Fear of the Sea on Working Memory Performance: A Research Based on Virtual Reality. In Proceedings of the 24th ACM Symposium on Virtual Reality Software and Technology, pages 1–9, Tokyo Japan, Nov. 2018. ACM. [31] T. Parisi. WebGL: Up and Running. ” O’Reilly Media, Inc.”, 2012. [32] S. Putze, D. Alexandrovsky, F. Putze, S. H¨ offner, J. D. Smeddinck, and R. Malaka. Breaking The Experience: Effects of Questionnaires in VR User Studies. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, pages 1–15. Association for Computing Machinery, New York, NY, USA, Apr. 2020. [33] T. S. Redick, J. M. Broadway, M. E. Meier, P. S. Kuriakose, N. Unsworth, M. J. Kane, and R. W. Engle. Measuring Working Memory Capacity With Automated Complex Span Tasks. European Journal of Psychological Assessment, 28(3):164–171, Jan. 2012. [34] F. C. Rodr´ıguez, M. Dal Peraro, and L. A. Abriata. Democratizing Interactive, Immersive Experiences for Science Education with WebXR. Nature Computational Science, 1(10):631–632, 2021. Publisher: Nature Publishing Group. [35] F. Ruotolo, L. Maffei, M. Di Gabriele, T. Iachini, M. Masullo, G. Ruggiero, and V. P. Senese. Immersive Virtual Reality and Environmental Noise Assessment: An Innovative Audio–Visual Approach. Environmental impact assessment review, 41:10–20, 2013. Publisher: Elsevier. [36] S. Safikhani, M. Holly, A. Kainz, and J. Pirker. The Influence of in-VR Questionnaire Design on the User Experience. In Proceedings of the 27th ACM Symposium on Virtual Reality Software and Technology, VRST ’21, pages 1–8, New York, NY, USA, Dec. 2021. Association for Computing Machinery. [37] M. Slater. How Colorful Was Your Day? Why Questionnaires Cannot Assess Presence in Virtual Environments. Presence, 13(4):484–493, 2004. Publisher: MIT Press. [38] M. Slater, A. Brogni, and A. Steed. Physiological Responses to Breaks in Presence: A Pilot Study. In Presence 2003: The 6th annual international workshop on presence, volume 157. Citeseer, 2003. [39] B. Smus. Web Audio API: Advanced Sound for Games and Interactive Apps. ” O’Reilly Media, Inc.”, 2013. [40] S. Tatlow. Authenticity in Sound Design for Virtual Reality. In History as Fantasy in Music, Sound, Image, and Media, pages 161–184. Routledge, 2024. [41] M. Tomasetti, A. Boem, and L. Turchet. How to Spatial Audio with the WebXR API: A Comparison of the Tools and Techniques for Creating Immersive Sonic Experiences on the Browser. In 2023 Immersive and 3D Audio: from Architecture to Automotive (I3DA), pages 1–9. IEEE, 2023. [42] M. L. Turner and R. W. Engle. Is Working Memory Capacity Task Dependent? Journal of memory and language, 28(2):127–154, 1989. Publisher: Elsevier. [43] N. Unsworth, R. P. Heitz, J. C. Schrock, and R. W. Engle. An Automated Version of the Operation Span Task. Behavior Research Methods, 37(3):498–505, Aug. 2005. [44] W. Zhang, P. N. Samarasinghe, H. Chen, and T. D. Abhayapala. Surround by Sound: A Review of Spatial Audio Recording and Reproduction. Applied Sciences, 7(5):532, 2017. Publisher: Multidisciplinary Digital Publishing Institute.