Virtual Reality Exposure Therapy for Foreign Language Speaking Anxiety: Evidence from Electroencephalogram Signals and Subjective Self-Report Data

Abstract

This study examines the efficacy of virtual reality exposure therapy (VRET) in alleviating foreign language anxiety (FLA) among university students. Although research exists on FLA, interventions have relied on self-reporting measures, leaving a gap in understanding physiological indicators and anxiety reduction. While previous research has explored either the therapeutic potential of virtual reality or the neurophysiological correlations of anxiety through electroencephalography (EEG), few have integrated these methodologies within a single experimental framework. This study combined the foreign language classroom anxiety scale (FLCAS) with (EEG) data to capture subjective and neural responses to anxiety in second language (L2) speaking. The participants (n = 20) completed language speaking tasks both before and after VR intervention, which exposed them to anxiety-inducing conditions replicating language challenges. During these tasks, brainwave signals were recorded, focusing on frontal alpha asymmetry (FAA) and alpha power (F3, F4), indicating neural activity associated with stress and emotional regulation. Results showed participants experienced a significant decrease (p = 0.017 < 0.05) in self-reported FLCAS scores after VRET. The reduction in FLA showed a negative correlation with increased alpha power at F3 (r = −0.55, p = 0.012), suggesting a link between left frontal neural regulation and anxiety reduction. These findings underscored VRET’s effectiveness in influencing emotional responses during L2-speaking tasks.

1. Introduction

Foreign language communication is affected by anxiety-related issues in educational and academic settings. Researchers have pointed out that anxiety is a challenging obstacle for foreign language learners because it could disrupt the process of learning foreign languages at various levels [1,2]. Consequently, the overall performance and achievement of individuals are negatively affected by stressful feelings [3]. Specifically, regarding second language acquisition (SLA), this anxiety-provoking phenomenon has been highlighted as a hurdle among university students [4,5]. Researchers in the field of language studies have investigated the reasons for situational anxiety and have expressed reasons such as fear of being negatively evaluated, unwillingness to communicate orally, low self-esteem, and classroom atmosphere [2,6,7]. Therefore, foreign language anxiety (FLA) has been examined from different perspectives by academic researchers and language experts to determine appropriate remedies and methods for its alleviation.

With the advent of virtual reality (VR) technology, many researchers have identified its potential for applications in educational settings and language learning [8]. In addition, VR has attracted increasing interest from researchers because of its potential in the treatment of various phobias and anxiety disorders [9,10,11]. Virtual reality exposure therapy (VRET) is an engaging, computer-driven treatment where individuals are progressively and systematically exposed to anxiety-provoking stimuli within a controlled virtual setting [12]. Unlike traditional exposure techniques that depend on real-life or imagined exposure, VRET allows for precise control over the intensity, duration, and context of the stimuli in a secure and repeatable manner [13].

Recent studies have employed physiological data as an objective metric to measure and evaluate anxiety and cognitive load in various contexts. This has allowed scholars to clearly understand the impact of new therapeutic approaches (i.e., VR-based therapy) by providing objective evidence of their effectiveness in reducing anxiety and mental disorders [14].

Although extensive studies have investigated the significance of reducing FLA and the potential of incorporating VR technology as a therapeutic tool for the treatment of phobias and anxiety, the existing literature lacks comprehensive research on how VR influences FLA at both the psychological and physiological levels. Specifically, in the context of language anxiety, the interplay between subjective self-perceived anxiety and objective neurophysiological markers of stress remains insufficiently explored [15,16].

Understanding the relationship between subjective, self-perceived anxiety and objective physiological markers is crucial, as self-reported measures alone might be subject to bias due to individual perception and social desirability effects, resulting in inaccurate anxiety [17]. The incorporation of physiological markers such as electroencephalogram (EEG) signals enables researchers to capture real-time stress responses, thereby providing a more comprehensive understanding of anxiety manifestations during foreign language use. As a noninvasive and cost-effective method for recording brain electrical activity, EEG enables real-time tracking of neural dynamics. In anxiety research, EEG is especially valuable for observing rapid cortical activation changes linked to emotional regulation and stress responses [18]. In a recent study by Kelsen et al. [19], EEG signals have been effectively employed to explore FLA in non-clinical student populations.

This study is innovative in its approach by combining VRET with both subjective self-assessment tools and objective EEG-based neural markers to examine FLA. Although previous research has looked at these components separately, our study is distinct in integrating them into a single framework. This approach provides a more thorough understanding of how immersive VR interventions influence both the psychological experience and neural control of anxiety related to speaking a foreign language.

While previous literature demonstrates that VRET can reduce FLA, most studies evaluate outcomes primarily via self-report; very few incorporate objective neurophysiological markers during L2 speaking. Our study addresses this gap by integrating EEG-based indicators—specifically frontal alpha asymmetry and alpha power—with FLCAS in a pre–post VRET design.

In light of this gap, the present study aims to examine the impact of VR technology on FLA levels among university students using both subjective self-report data and objective physiological metrics. To address this research gap, we sought to answer the following research questions:

RQ1:

To what extent does virtual reality exposure therapy (VRET) reduce self-reported FLA among university students?

RQ2:

How does VRET influence both EEG-based physiological indicators of anxiety and self-reported FLA levels?

2. Literature Review and Related Work

2.1. Addressing Foreign Language Anxiety FLA

FLA is described by MacIntyre [20] as “worry and negative emotional reaction aroused when learning or using a second language.” It is considered under the category of situation-specific anxiety in the context of mental health disorders that emerge from the “uniqueness of learning a foreign language” [1,21,22]. In practice, this type of anxiety tends to occur repeatedly in specific situations. Notably, FLA negatively affects academic performance and effectiveness among university students [23,24,25].

This form of anxiety could manifest in various ways, including communication apprehension, fear of negative evaluation, and test anxiety, all of which could hinder language learning [21]. Empirical studies have demonstrated that FLA negatively affects the linguistic performance, cognitive processing, and overall academic success of students [26]. High levels of FLA could lead to avoidance behaviors, reduced participation in language classrooms, and increased difficulty in processing and retaining linguistic information [5].

Given the significance of FLA, several studies have explored its causes and underlying sources [27,28,29]. These studies identified key sources of FLA in a classroom setting, including “learner’s low self-esteem & communication apprehension” [7], “teacher’s behavior and questioning” [30], “fear of negative evaluation & making mistakes” [2], and “fear of being laughed at & worrying about being judged by others” [31]. From a broader perspective, these sources have been categorized into the internal (e.g., self-esteem and fear of failure) and external variables (e.g., teacher behavior and classroom environment) of learners [32].

Previous studies [33] incorporated the foreign language classroom anxiety scale (FLCAS), a self-report questionnaire developed by Horwitz et al. [1] to subjectively evaluate and measure foreign language anxiety. The FLCAS is a widely recognized and validated instrument designed to measure the anxiety levels of learners in a foreign language classroom setting [34,35]. It consists of 33 Likert-scale items that collectively contribute to the anxiety of students when learning or using a second language in an academic environment. Since its introduction, numerous studies have employed the FLCAS to explore the relationship between language anxiety and various learner-related variables (e.g., proficiency levels, motivation, self-efficacy, and academic performance) [23,36,37]. Since then, some studies have used a modified version of the original FLCAS to better align with their specific research objectives [34,35].

2.2. VR in Language Learning and Exposure Therapy

VR is a three-dimensional (3D) combination of images and voices, along with other immersive elements, in which individuals could interact in a manner that they feel a real-world presence [38]. Since its emergence, VR technology has been used by researchers and practitioners in various fields (e.g., education, healthcare, and gaming) [8]. This technology enables researchers to develop environments that closely resemble real-world situations, allowing users to immerse themselves and experience a heightened sense of presence and engagement [39]. The growing adoption and application of virtual immersion in language learning has been encouraged by the research of many scholars in recent years [8,40]. Full immersion in a 3D environment allows learners to engage in life-like settings and participate in experiences that could not be replicated in traditional language classrooms [39]. The benefits of VR for language learning have been emphasized; learners experience higher levels of presence in the absence of the real world, resulting in more focus and engagement in SLA [41]. In a systematic review, researchers highlighted that university students are the primary adopters of VR technology in the context of language learning, and noted improvements in their performance and positive perceptions of VR technology integration [40].

In clinical and healthcare settings, VR has been used to treat a diverse range of anxiety disorders and phobias [9,42,43]. This is evident in the VRET approach, in which individuals are exposed to conditions in VR environments that gradually introduce anxiety-provoking situations. This therapeutic method has been employed to treat a wide range of phobias, such as (but not limited to) fear of driving [11], fear of flying [44], and fear of public speaking [45]. For example, in a study by Kaplan-Rakowski and Gruber [41] on the effect of VR on public speaking anxiety among participants (n = 12), the authors found that VR positively contributed to anxiety reduction after practice sessions in simulated classroom environments with virtual audiences. In a subsequent investigation conducted by Thrasher [46], the use of VR technology to mitigate FLA and enhance oral comprehensibility among French language learners was examined. This study employed both self-reported anxiety measures and physiological indicators (salivary cortisol) to assess anxiety levels in VR environments compared to traditional classroom settings. These findings indicated that VR environments were associated with reduced anxiety and improved comprehensibility. Furthermore, this study revealed a positive correlation between self-reported anxiety and cortisol levels.

Ding [47] reported pre–post foreign language speaking anxiety (FLSA) reductions in a randomized design comparing HiVR and classroom conditions, further supporting VR-based exposure as a viable pathway to reduce anxiety in EFL learners. The majority of participants indicated that HiVR provided them with a realistic setting and enjoyable learning experiences, which kept them engaged in the learning process. Complementing these results, Park et al. [48] showed anxiety reduction across repeated VR public-speaking sessions with EFL learners, highlighting the benefits of repeated exposure. Also, their research underscored the potential of VR as a versatile tool for both language acquisition and the development of professional skills, enabling users to control anxiety and practice public speaking in a tailored and controlled environment.

2.3. Measuring Anxiety Level with EEG Physiological Signals

Anxiety is a multifaceted emotional state often linked to various physiological responses. Numerous studies have examined the association between anxiety and physiological indicators, as understanding this relationship offers critical insights into the underlying mechanisms of anxiety disorders and potential treatment approaches [14,49]. For example, a study measuring the heart rate (HR) variability of participants during oral presentations found a positive correlation between self-reported FLA and increased HR [50]. Consequently, engaging in foreign language conversations might lead to physiological changes due to associated stress or anxiety. To measure stress and anxiety using physiological metrics, EEG signals (a noninvasive method for assessing brain activity) have been widely used by professionals and scholars in the field [14,49,51]. Among EEG-based metrics, alpha power and frontal alpha asymmetry (FAA) are commonly used indicators of anxiety-related neural activity [17,52]. Alpha power, particularly in the frontal and central regions, has been linked to stress regulation, and lower alpha power is often associated with heightened anxiety. FAA, which reflects the asymmetry in alpha activity between the left and right prefrontal cortices, has also been used as a biomarker of emotional and stress responses [53]. The relationship between FAA, a neural marker of anxiety, and FLA was examined [19]. These findings indicated that FAA served as a relevant neurophysiological marker associated with emotional processing and regulation during foreign-language-speaking tasks.

Previous studies have demonstrated that the alpha power frequency range (8–12 Hz) is sensitive to stress- and anxiety-provoking stimuli. This phenomenon is particularly evident in asymmetrical patterns of frontal alpha activity [45,46]. As elucidated in previous investigations [47], measures of stress biomarkers and disorders associated with mood and stress are also correlated with FAA.

Although physiological signals provide objective measures of anxiety, they have certain limitations. Interpreting EEG data could be challenging because they are affected by certain factors, including variations in brain anatomy, the nature of the tasks performed during signal recording, and technical issues such as noise and artifacts [49,54]. Despite these challenges, combining physiological data with self-reported measures provides a comprehensive understanding of anxiety. This has been outlined in research conducted by Akil et al. [17], who investigated the association between FAA and self-reported measures of depression and anxiety. The researchers analyzed data from the participants (n = 130) and examined correlations between FAA and scores obtained from established self-reporting instruments. These findings indicated a potential relationship between FAA and anxiety, particularly among individuals experiencing moderate-to-high levels of psychological distress. Thus, the combination of physiological and self-reported data in FLA research is of significant interest.

As shown in

Table 1. Summary of the recent related works on FLA, incorporating subjective and/or objective measures.

Author(s) & Year	Sample/Method	Subjective Scale	Objective Measures	Intervention	Remarks
Park et al. [48], (2025)	(n = 13, native East Asian English learners)/VR presentations	Self-report FLA (pre/post)	Gaze tracking	Customizable immersive VR	Significant reduction in FLA scores after six VR sessions; improved speech clarity and fluency.
Kelsen et al. [19], (2024)	(n = 40, native Mandarin Chinese speakers)/L1 vs. L2 speaking Task	Second language skill-specific anxiety scale (L2AS)	EEG Signals (FAA)	Word-chain speaking Task (No VR Intervention)	Right-lateralized FAA predicted higher FLA during L2 speaking.
Kaplan-Rakowski et al. [41], (2023)	(n = 20, English Learners)/VR vs. Zoom presentation	Self-report FLA (pre/post)	None reported	High-immersion VR public-speaking vs. videoconferencing (Zoom)	The VR group showed significantly lower FLA than the Zoom group
Ding [47], (2024)	(n = 140, Chinese EFL learners)/randomized 2 × 2 (HiVR/classroom)	Foreign Language Speaking Anxiety (FLSA)/ (pre/post)	None reported	High-immersion VR vs. classroom role-play speaking	Learners perceived HiVR helpful, speaking anxiety reduced
Thrasher [46], (2022)	(n = 25, L2 French learners)/VR vs. traditional classroom	Language-anxiety self-report (pre/post)	Salivary Cortisol	Immersive social-VR speaking tasks	Reduced FLA with improved oral comprehensibility; positive correlation observed between self-report and cortisol data.
Gu [55], (2025)	(n = 1086, Chinese ESL learners)/large-scale survey study	FLCAS Questionnaire	None reported	VR exposure to language-use scenarios	VR exposure indirectly lowered FLA via enhanced communicative confidence and perceived fluency.
Our Study	(n = 20, Korean university students) VR & in-person English-speaking tasks	FLCAS Questionnaire (pre/post)	EEG Signals (F3, F4, FAA)	Scenario-based VRET with graded anxiety-provoking stimuli	Combined VR-based exposure with EEG + self-report; significant pre-post reduction in FLA

, prior studies in the context of FLA have typically focused either on virtual reality-based interventions without physiological measurements or on EEG-based analyses conducted outside immersive environments. The simultaneous inclusion of both approaches in the present work provides a more comprehensive understanding of how immersive VR exposure influences both perceived and neurophysiological dimensions of foreign language speaking anxiety. While previous research has demonstrated that VR can effectively reduce FLA [41] or that EEG signals can capture anxiety-related cortical activity [19], few, if any, studies have combined these methodologies within a single experimental framework. Our work bridges this methodological gap by evaluating foreign language speaking anxiety reduction through synchronized subjective (FLCAS) and objective (EEG) data within a VRET setting, thereby offering a novel contribution to the growing intersection of affective neuroscience and technology-enhanced language learning.

3. Research Design and Method

3.1. Participants

In this study, we recruited a total of (n = 20) native South Korean students (13 males and seven females). The participants were 22–29 years old, with an average age of M = 24.50 ± 2.27 y, and the selection criteria were as follows:

(1)

Holding a Test of English for International Communication score of 600 or lower, or a College Scholastic Ability Test English grade 3 or lower (both are standardized English proficiency tests widely used in Korea);

(2)

Having taken or is taking an English class;

(3)

Having self-reported feelings of nervousness, anxiety, or avoidance during English classes or when having to speak in English.

The abovementioned English tests and thresholds were chosen to ensure that the study participants were likely to experience challenges and anxiety in English use, particularly in speaking and classroom participation, which are central to FLA research. In addition, a previous FLA study showed that anxiety tended to be greater among learners who perceived their English ability to be low or insufficient [56]. Only the individuals who fulfilled all three conditions were eligible to participate in the experiment. Of the 20 participants, 16 had prior experience using VR technology. All participants signed written and printed consent forms before the experiment. The participants were provided with financial compensation upon successful completion of all the experimental steps of the study. This study was approved by the Institutional Review Board of Kwangwoon University (Approval no. 7001546-202401031-HR(SB)-010-04).

3.2. Prototype Development

In the experimental phase, a virtual university classroom environment was developed using Unity (Unity Technologies: San Francisco, CA, USA, version 2022.3.36f1). The virtual setting encompassed avatars representing classmates and teachers. In addition, a dashboard (control panel) was designed to trigger verbal and nonverbal stimuli and avatar actions during the VR sessions. The VR prototype was designed to incorporate various anxiety-inducing scenarios using a combination of verbal stimuli, avatar movements, and environmental sound effects to create an immersive and psychologically engaging experience. These scenarios were designed so that participants could engage in structured language tasks, including self-introductions, delivering short speeches on a given topic, and responding to follow-up questions posed by the virtual teacher, simulating real-world foreign language communication challenges. In addition, virtual classmates dynamically responded to the speaking tasks of participants by exhibiting emotions such as admiration, interest, or disappointment through facial expressions and body language, including raising their hands to ask questions, laughter, and clapping, further enhancing the simulated classroom environment. Furthermore, to progressively intensify the level of stress, the instructor gradually increased the number of virtual classmates during the VR intervention [57]. Through these progressively challenging scenarios, the researchers aimed to gradually expose participants to foreign language-speaking situations in a controlled and immersive environment, facilitating the development of confidence and resilience in managing FLA. Consequently, researchers were able to control anxiety-inducing situations in VR, such as those encountered during real presentations or talking in front of the classroom. Figure 1a illustrates the perspectives of participants within the virtual classroom, depicting the scene when virtual classmates are in an idle state, and the classroom is at full capacity.

Figure 1b illustrates virtual classmates standing and clapping in response to the speaking task of participants, conveying a positive reaction, while Figure 1c shows participants selecting various topics to talk to virtual classmates. Throughout the VR intervention, the researcher triggered various reactions from avatar classmates at different points in the task sequence, providing dynamic and context-sensitive feedback to the participant.

3.3. Apparatus

In this study, a 14-channel wireless Emotiv EPOC X headset (Emotiv Systems Inc., San Francisco, CA, USA) was used to capture the brain activity and signals of participants. The sampling frequency was set at 128 Hz, and EEG electrodes were positioned on the scalps of participants in accordance with the 10–20 International System (Figure 2) [58]. Noninvasive and washable saline solutions were applied to the felt pads of the sensors to enhance conductivity and signal quality. The recorded raw EEG signals were transmitted to a computer via Bluetooth connection.

Furthermore, an Oculus Quest 2 headset (Meta Platforms Inc., Menlo Park, CA, USA) and its accompanying controllers were employed for the participants’ VR intervention, with the headset connected to a computer.

3.4. EEG Signal Recording

To ensure the quality and reliability of the EEG data, the raw signals were preprocessed using Emotiv’s proprietary built-in software (EmotivPRO v.2.0). This preprocessing pipeline encompassed automated artifact detection and removal in addition to noise reduction procedures. The preprocessing steps involved band-pass filtering, typically ranging from 0.5 to 45 Hz, to eliminate low-frequency drifts and high-frequency noise. Additionally, automatic artifact detection and rejection were performed based on amplitude and frequency thresholds, and signal re-referencing was applied to reduce noise specific to individual channels. In this study, the alpha power was analyzed as a key EEG-based metric for assessing anxiety. Alpha power, particularly in the frontal and central brain regions, has been widely studied in relation to stress and emotional regulation [59].

Lower alpha power is generally associated with increased anxiety and a heightened cognitive load, whereas higher alpha power reflects a more relaxed state. The alpha power was calculated by averaging the spectral power within the alpha frequency range (8–12 Hz) across relevant electrode sites (F3 and F4). In addition to the alpha power, the FAA was calculated by determining the mean alpha power values at the F3 (left frontal) and F4 (right frontal) electrode sites. This approach was based on the literature [52,60,61], which suggested that greater relative right frontal activation (lower FAA values) was associated with heightened anxiety, whereas greater frontal activation (higher FAA values) was associated with lower anxiety levels. Prior studies have employed diverse methodological approaches to evaluate FAA more thoroughly [62], often incorporating variations in equations (

Table 2. FAA calculation methods.

FAA Metric	Formula	Transformation Type	Reference
FAA₍ln₎	$Ln\left(F4\right)-Ln\left(F3\right)$	Logarithmic Difference	[17]
FAA₍ratio₎	$\left(F4-F3\right)/\left(F4+F3\right)$	Linear Ratio	[63]
FAA₍lnratio₎	$(Ln\left(F4\right)-Ln\left(F3\right))/(Ln\left(F3\right)+Ln\left(F4\right))$	Logarithmic Ratio	[64]
FAA₍lnrel₎	$Ln\left(rel\left(F4\right)\right)-Ln\left(rel\left(F3\right)\right)$	Log-Difference of Relative Power	[65]

) that primarily involved alpha power measurements at the right frontal electrode (F4) and left frontal electrode (F3). By implementing these diverse approaches, we aimed to examine the consistency of FAA measurements and explore how the results from each method correlated with subjective self-reported data derived from FLCAS scores.

3.5. Procedure

The experimental phase of our study comprised three primary stages: (1) pre-VR intervention, (2) VR intervention, and (3) post-VR intervention. All participants selected for the experimental phase of our study adhered to the procedure depicted in (Figure 3). Prior to the commencement of the experimental phase, the participants were provided with detailed written instructions to ensure a comprehensive understanding of the entire process, thus familiarizing them with the procedures. The details of each stage are described in subsequent sections.

3.6. Pre-VR Intervention

Initially, the participants completed a web-based FLCAS questionnaire comprising 33 statements, each rated on a five-point Likert scale ranging from one (strongly disagree) to five (strongly agree). As the study was conducted with South Korean participants, the original FLCAS questionnaire developed by Horwitz et al. [1] was translated into Korean with no modifications to ensure comprehension and accuracy of self-reported responses. The translated version was carefully reviewed to ensure consistency with the original instrument.

Before this, the demographic data and previous experience of the participants with VR technology were collected. Subsequently, the EEG headset was securely positioned on the scalp of the participants, and the connectivity of the electrodes, along with the EEG signal quality, was verified according to the guidelines of the device manufacturer to ensure that all electrodes provided high-quality data. Following this, participants were instructed to relax their minds for 15 s and then engage in an in-person English-speaking conversation with the researcher. Concurrently, EEG data were collected as the participants responded to questions posed by the researcher. Note that the participants were initially asked simple self-introduction questions that increased in complexity and challenge. Upon completion of this stage, the researchers assisted the participants in safely removing the EEG headset before guiding them to the subsequent stages.

3.7. VR Intervention

After completing the pre-VR intervention phase, the researchers guided participants to join the VR session. In the VR intervention stage, the participants were instructed to wear a VR headset and hold both controllers (Figure 4). After immersion in the virtual classroom, a VRET scenario-based approach comprising three scenarios (A, B, and C) with increasing levels of anxiety-provoking tasks was implemented. Participants engaged in activities such as delivering self-introductions, speaking on assigned topics, and responding to questions. The graded exposure procedure, frequently used in VRET, was implemented to systematically escalate speaking-related stress and cognitive load [66]. Within each VR scenario, the participants engaged in two speaking tasks: Task 1 involved a brief, structured monologue calibrated to the scenario’s level of difficulty, and Task 2 consisted of a researcher-led follow-up related to Task 1, during which the virtual teacher or classmates asked follow-up questions and brief comments to prompt clarification and elaboration. Scenario A entailed a simple self-introduction followed by a few basic questions (e.g., hobbies and educational background). Scenario B required a short presentation on a moderately challenging and familiar topic (e.g., favorite computer games or sports). Scenario C required a brief discussion on a more complex topic (e.g., “What is AI technology?”) with follow-up questions designed to progressively increase the linguistic and cognitive demands.

The researcher triggered positive and negative verbal and nonverbal feedback as well as reactions from the virtual teacher and classmates at predetermined intervals throughout the VR experiences of the participants. Positive verbal feedback included phrases such as “Well done!” and “Very good,” while negative verbal feedback consisted of comments like “Your explanations were confusing and hard to follow.” In addition, nonverbal feedback was incorporated through body language and facial expressions of virtual classmates, such as nodding and clapping to indicate approval or crossing arms and shaking heads to express disapproval. To further enhance the anxiety-inducing environment, the researcher introduced auditory stressors, including mobile phone ringing, keyboard typing, and the sound of closing a door, to simulate real-world distractions and increase the cognitive load of the participants during the VR experiment. It is noteworthy that the participants were allocated a 2 min rest period between scenarios.

3.8. Post-VR Intervention

After completing the VR intervention, we first asked the participants to complete a web-based simulator sickness questionnaire to assess any potential effects of the VR experience. The participants then responded to the same web-based FLCAS. After the EEG device was placed on their heads, they were asked to have another in-person English-speaking conversation with the researcher, albeit with questions different from the pre-VR intervention.

3.9. Statistical Analysis

To evaluate the effectiveness of the VR intervention in reducing FLA, a paired t-test was conducted to compare FLCAS scores pre- and post-VR intervention. This analysis was used to determine whether there was a statistically significant difference in the self-reported anxiety levels of the participants following exposure to VR-based therapy. Pearson’s correlation analyses were conducted to explore the relationships among FAA, alpha power, and FLCAS scores. We computed the mean alpha power differences at the F3 and F4 electrode sites along with the FAA values derived from the four equations. This analysis aimed to determine the extent to which physiological indicators were associated with self-reported anxiety levels, offering deeper insight into the neural functions underlying FLA. To explore the interrelationships among these metrics, we constructed a correlation matrix of seven variables: difference F3 mean, difference F4 mean, delta_FAA_(ln), delta_FAA_(ratio), delta_FAA_(lnratio), delta_FAA_(lnrel), and the FLCAS score difference. Statistical significance was determined using Pearson’s correlation coefficients and corresponding p-values. This approach allowed for the examination of variations in anxiety reduction among participants, providing a more nuanced understanding of how different individuals responded to the VR intervention. All statistical analyses were performed using RStudio for Windows (version 2024.12.0) to ensure accurate data processing.

4. Results

4.1. Comparison of Pre- and Post-VR FLCAS Scores

To assess the changes in FLA levels pre- and post-VR intervention, the FLCAS questionnaire was administered at two time points. To ensure the reliability of measurements [36], the Cronbach alpha values for reliability of the FLCAS questionnaires were calculated, which show a high level of internal consistency (α = 0.96 for pre-VR; α = 0.97 for post-VR).

Table 3. FLCAS score descriptive statistics.

Variable	Possible Range	Min–Max	Mean	SD
FLCAS Score (Pre)	33–165	65–156	108.45	27.83
FLCAS Score (Post)	33–165	43–155	99.15	28.84

presents the descriptive statistics of the FLCAS scores for pre- and post-VR experience.

The observed decrease in the minimum score from 65 to 43 suggests that some participants experienced a substantial reduction in their FLA. Furthermore, the mean score after the intervention decreased to 99.15, indicating a reduction in overall anxiety levels among participants (Figure 5). A paired t-test was used to determine whether the reduction was statistically significant. The results of the paired t-test (p = 0.017 < 0.05) indicated a statistically significant difference between pre-VR (M = 108, SD = 27.83) and post-VR (M = 99.15, SD = 28.84). This finding suggests that the VR intervention has a meaningful impact on reducing FLA. The results from the Shapiro–Wilk normality test (pre-VR data: W = 0.94611, p-value = 0.3118; post-VR data: W = 0.95753, p-value = 0.4958) demonstrated that the data were normally distributed, and the assumptions for the correlation test were successfully fulfilled. To strengthen our results, a Wilcoxon Signed-Rank test was conducted (p = 0.0216 < 0.05), which demonstrated a statistically significant difference between pre- and post-VR intervention data. The results showed that among the 20 participants, 14 (70%) exhibited a reduction in their FLCAS scores (decreased FLA), five (25%) displayed an increase (increased FLA), and one (5%) participant’s score remained constant.

4.2. Assessment of VRET Effectiveness Using EEG and Subjective Metrics

To evaluate the effects of the VR intervention, we analyzed the physiological indicators of anxiety (alpha power and FAA) derived from EEG signals and recorded the pre-VR and post-VR experiences (

Table 4. Descriptive statistics of variables pre- and post-VR intervention.

Variable	Pre-VR Mean (SD)	Post-VR Mean (SD)
F3 alpha power	0.0301 (0.005)	0.0328 (0.004)
F4 alpha power	0.0412 (0.006)	0.0321 (0.005)
FAA₍ln₎	0.279 (0.021)	0.2665 (0.019)
FAA₍ratio₎	0.1367 (0.017)	0.1305 (0.014)
FAA₍lnratio₎	0.0615 (0.005)	0.0549 (0.006)
FAA₍lnrel₎	0.279 (0.019)	0.2665 (0.018)
FLCAS score	27.83 (2.453)	28.84 (2.87)

). The mean value and standard deviation (SD) of the EEG alpha band (8 Hz–12 Hz) were calculated for the F3 and F4 electrode sites [67].

To further investigate the relationship between the subjective and physiological indicators of the FLA, a correlation analysis was performed (

Table 5. Correlation showing Pearson’s r for variables.

Variable	Difference F3 Mean	Difference F4 Mean	Delta_FAA(ln)
FLCAS_difference	−0.55 *	−0.07	0.34

* p < 0.05, Correlation strength was interpreted based on Evans [68], where absolute values of Pearson’s r are classified as follows: 0.00–0.19 (very weak), 0.20–0.39 (weak), 0.40–0.59 (moderate), 0.60–0.79 (strong), and 0.80–1.00 (very strong). All statistical analyses followed the standards.

). This analysis aimed to evaluate the strength and direction of linear associations among key variables, including the mean alpha power differences at the F3 and F4 electrodes, changes in log-transformed FAA (delta_FAA_(ln)), and differences in FLCAS scores pre-VR and post-VR intervention. A moderate negative correlation was observed between the change in alpha power at F3 and the change in the FLCAS score (r = −0.55, p = 0.012 < 0.05), indicating that participants who showed a larger decrease in F3 alpha power following the VR intervention also tended to exhibit a larger reduction in self-reported FLA. As alpha power is inversely related to cortical activity, this finding suggests that increased left frontal cortical engagement (i.e., decreased alpha power) is linked to reduced anxiety symptoms in response to the VR exposure task. In contrast, the correlation between differences in mean F4 and FLCAS was weak and not statistically significant (r = −0.07). In addition, FAA_(ln) was selected as the representative FAA metric for primary correlation analyses with FLCAS scores, which exhibited a positive but non-significant correlation with the FLCAS difference (r = 0.34). The inclusion of multiple FAA formulations served only as a cross-validation check to assess the robustness of FAA-related patterns using various computational methods.

Although these results did not reach statistical significance, they suggested a possible association between FAA changes and reductions in subjective FLA.

5. Discussion

This study investigated the effectiveness of VR Exposure Therapy in mitigating FLA among university students by assessing both subjective and objective indicators of anxiety. This research was guided by fundamental questions regarding whether VRET could reduce self-reported FLA and how physiological indicators, such as EEG-derived metrics, correlate with subjective perceptions of anxiety pre- and post-VR intervention. Through a carefully designed experimental framework involving scenario-based VR speaking tasks, this study uniquely integrated self-report measures (FLCAS) with EEG signal recordings, specifically focusing on alpha power in the F3 and F4 regions and four different calculation methods of FAA.

5.1. Effectiveness of VRET in Reducing Self-Reported FLA

The results of the subjective self-report FLCAS comparison demonstrated a statistically significant decrease in the FLA of the participants following the VR-based intervention, thereby affirming the efficacy of VR exposure therapy in language learning environments. The mean FLCAS score declined from 108.45 to 99.15, with 70% of participants indicating reduced post-intervention anxiety levels. This reduction is consistent with prior research that highlights the effectiveness of immersive technologies in alleviating phobias and social anxiety [10,69], public speaking anxiety [45], and affective barriers in SLA [41,46]. In our study, the VR intervention phase employed a progressive exposure technique wherein participants were incrementally introduced to speaking tasks that escalated anxiety within a controlled virtual classroom setting. As expected, this strategy seems to have been effective in desensitizing participants to typical foreign language use stressors such as peer evaluation, speaking in front of audiences, and other key triggers of FLA.

5.2. Correspondence Between FLCAS Scores and EEG-Based Changes

The rationale for integrating both subjective and objective data is the recognition that anxiety, particularly FLA, is a multidimensional construct. Self-reported instruments such as the FLCAS provide valuable insights into the conscious perceptions and emotional states of learners. However, they are inherently prone to cognitive biases and social desirability. Objective measures such as EEG-derived metrics capture real-time neurophysiological responses, allowing for the identification of underlying affective and attentional processes that are not easily accessed through self-reporting. One of the key findings supporting the efficacy of VRET in reducing FLA was the observed convergence between a statistically significant reduction in FLCAS scores and directional changes in EEG-derived metrics. Analysis of the relationship between EEG-derived metrics and self-reported FLA revealed noteworthy patterns, particularly with respect to left frontal alpha activity. The moderate negative correlation between changes in the FLCAS scores and F3 alpha power suggests that increased alpha power in the left frontal region is associated with decreased subjective anxiety. This pattern might reflect cognitive effort or task-specific dynamics in FLA [19], warranting further investigation. In contrast, right frontal alpha activity (F4) did not show a meaningful correlation with changes in self-reported FLA. The correlation matrix further supported this interpretation. Notably, there was a moderate negative correlation between the FLCAS difference and the F3 alpha power difference (r = −0.55), indicating that participants who showed greater increases in left frontal alpha activity tended to report greater reductions in subjective anxiety. In addition, changes in FAA_(ln), FAA_(ratio), and FAA_(lnrel) were all moderately correlated (but nonsignificant) with FLCAS reduction (r ≈ 0.34), suggesting that shifts in asymmetrical patterns were at least partially reflective of the psychological relief reported by participants. Beyond group-level trends, an important observation in this study was the variability in synchrony between subjective and objective measures at the individual level. While 70% of the participants showed a reduction in the FLCAS scores, the degree of change varied substantially, and not all reductions were matched by proportionate shifts in the EEG metrics. This suggests that individual differences—such as trait anxiety, neurocognitive profiles, and prior language experiences—might influence how strongly the subjective and physiological states of a participant align [70,71].

The considerable variability across measured variables (Table 4) likely reflects the typical diversity of foreign language anxiety levels observed among university students. Given the inclusion criteria applied during participant recruitment, such variation is expected and reasonable, as it arises from natural psychological and experiential differences among individuals.

The integration of subjective and objective metrics in evaluating the effectiveness holds important implications for educational technology and the design of language learning interventions. First, it validates the utility of EEG monitoring as a noninvasive, real-time tool for assessing learner anxiety in communication scenarios. The ability to track neurophysiological responses during language tasks could provide educators and researchers with deeper insights into when and how anxiety manifests, thereby enabling the development of more responsive and individualized learning environments.

Second, the observed synchrony between FLCAS reduction and EEG asymmetry shifts supports the viability of VR-based interventions as scalable approaches for FLA reduction. Rather than relying solely on traditional classroom desensitization techniques (e.g., role-playing or peer presentations), immersive VR environments offer replicable and controllable platforms for exposure therapy. The findings of this study suggest that such environments not only alleviate anxiety at the conscious level but also induce measurable changes in the underlying neural substrates, particularly those associated with emotional regulation and executive functioning.

5.3. Limitations and Future Research

Although our study presents encouraging evidence supporting the effectiveness of VR exposure therapy in reducing FLA, there are a few limitations that need to be considered. First, the relatively small number of participants limits the generalizability of the results. Although statistically significant trends were observed in both subjective and physiological data, future research with a larger and more diverse participant pool would enhance the statistical power and facilitate subgroup analyses based on individual characteristics such as language proficiency, gender, or prior experience with VR. Second, the VR intervention was limited to a single session. While short-term effects have been successfully observed, it remains uncertain whether these reductions in anxiety are sustained over time or whether repeated exposure is required to maintain effectiveness. Prospective studies should consider implementing longitudinal designs that incorporate multiple VR sessions over several weeks in order to better understand the long-term efficacy of VRET for FLA. In addition, expanding the scope of physiological measurements to include HR, galvanic skin response, and eye-tracking data could provide a more comprehensive view of the stress and anxiety responses of the participants. Future studies might consider employing a multimodal fusion strategy that combines different physiological signals, such as EEG, ECG, and EDA, to strengthen the reliability of anxiety detection. By integrating these modalities using advanced fusion methods alongside machine learning classifiers, the precision and consistency in differentiating between anxious and non-anxious states in the context of FLA could be further enhanced. Moreover, future research directions should adopt a more systematic baseline measurement to control individual variability, improve reliability, and investigate physiological differences between high- and low-anxiety groups. Such an approach could reveal differences in the physiological signals between baseline measures and task conditions within this subgroup of participants. Exploring the personalization of VR content based on individual anxiety profiles might also enhance intervention outcomes.

Recent advances in EEG modeling suggest that dynamic, multichannel frameworks such as state-space models [72] and graph-based deep learning approaches for evolving brain networks [73] could significantly enhance the analysis of temporal and spatial neural dynamics. In future work, we propose applying such methods to the acquired EEG data, thereby capturing non-linear dependencies and cross-channel interactions to improve the sensitivity and interpretability of anxiety-related neural responses.

Another limitation of the present study is that the EEG analysis focused exclusively on alpha-band activity to derive FAA. Although alpha rhythms are well-established neural markers of anxiety regulation, other EEG frequency bands- such as beta activity associated with cognitive processing and arousal, or theta rhythms related to emotional regulation- may also provide valuable information. Future research should incorporate a broader range of EEG frequency bands to obtain a more comprehensive understanding of neural dynamics related to foreign language speaking anxiety.

6. Conclusions

This study examined the efficacy of VRET in alleviating FLA by integrating subjective self-reported measures with objective physiological indicators. Specifically, this study aimed to determine whether immersive VR experiences could reduce FLA among university students and whether alterations in EEG-based metrics, particularly alpha power and FAA, were associated with decreased self-reported anxiety. We believe that our study is one of the few investigations to incorporate both subjective self-reports and physiological indicators in the examination of FLA and immersive VR therapy. The results indicated a statistically significant reduction in the FLCAS scores of the participants after the VR intervention, with 70% of the participants experiencing decreased anxiety. This finding demonstrates the effectiveness of immersive VR environments in delivering controlled, scalable, and anxiety-inducing exposure tasks that are consistent with established desensitization techniques. Furthermore, the significance of this study lies in its potential to yield practical implications in academic and other educational contexts.

The importance of this research is two-fold: it validates VR as a promising tool for anxiety reduction in second-language learning contexts and introduces physiological indicators, such as EEG-based FAA, as objective markers for evaluating intervention outcomes. These contributions support the design of more personalized and data-driven language-learning interventions. Future research should explore larger and more diverse samples, extend physiological monitoring to real-time VR sessions, and compare different VR scenarios and anxiety-inducing conditions.