IntroductionEnglish-speaking proficiency has become essential for academic success, professional advancement, and cross-cultural communication in today’s interconnected global society (Crystal, 2003). Although it is imperative, English as a Foreign Language (EFL) learners worldwide face considerable challenges in developing oral proficiency, with many struggling to achieve communicative competence, even after years of formal instruction (Jamshidnejad, 2020). This global challenge is particularly evident in East Asian contexts, including Taiwan, where traditional teaching methods focus on grammar and test scores rather than on communicative competence (Chen et al., 2020). Consequently, students often have a firm grasp of grammar and vocabulary but cannot speak English confidently. This lack of confidence creates a negative cycle that impedes oral production. Recent research highlights the importance of self-efficacy in language learning, particularly for speaking proficiency. Students with high levels of self-efficacy are more likely to engage in speaking activities and perform better on oral tasks across various cultural and educational contexts (Hoesny et al., 2023; Wicaksono et al., 2023; Wijaya, 2024).To promote L2 speaking self-efficacy, which is grounded in Bandura’s (1997) social cognitive theory and its emphasis on mastery experiences and social modeling, alternative assessment methodsFootnote 1 such as self-assessment (SA) and peer assessment (PA) have gained popularity in EFL classrooms. These approaches align with Vygotsky’s (1978) sociocultural theory, where learning occurs through social interaction, and with Zimmerman’s (2000) self-regulated learning framework, which emphasizes the cyclical nature of planning, monitoring, and reflection (Ashraf and Mahdinezhad, 2015; Wicaksono et al., 2023). While these theoretical frameworks highlight the importance of learner autonomy and self-regulation, implementing learner-centred speaking assessments remains challenging in traditional classroom settings where the teacher usually dominates the assessment practice. Technology integration has emerged as a feasible solution through platforms such as VoiceThread, a web-based multimedia tool that addresses these challenges by enabling users to create, share, and discuss audio, video, and text content asynchronously. VoiceThread is especially suited for speaking practice and evaluation in language learning contexts because it creates a low-anxiety environment for oral production while facilitating meaningful peer interaction and self-reflection. Although research findings on alternative assessments have been inconsistent, with some learners still perceiving teacher evaluations as being more reliable (Kumar et al., 2023; Ritonga et al., 2021), the shift from traditional paper-and-pencil methods (Phongsirikul, 2018) to digital platforms necessitates an investigation into how SA and PA function in digital environments, particularly given their potential to provide more flexible, authentic, and engaging speaking assessment experiences than conventional methods.Despite the growing adoption of technology-mediated assessments in language learning, several critical gaps remain in understanding their effectiveness and implementation. First, while research has examined self- and peer assessments separately, few studies have systematically compared their relative effectiveness when mediated through online platforms. Second, despite the relationship between self-/peer assessments and speaking proficiency having been explored (Hoesny et al., 2023; Wicaksono et al., 2023; Wijaya, 2024), the mechanisms by which technology-mediated SA and PA influence self-efficacy development remain underexamined, especially in the Taiwanese EFL context, where students often exhibit high language anxiety and limited speaking confidence (Jamshidnejad, 2020). Third, there is insufficient empirical evidence regarding how social cognitive theory (Bandura, 1997) manifests in digital assessment environments (Phongsirikul, 2018), particularly in cultural contexts where traditional assessment practices are deeply entrenched (Chen et al., 2020).To address these gaps, this study aimed to: (1) compare the effectiveness of VoiceThread-mediated self- and peer assessment on Taiwanese EFL learners’ speaking proficiency and self-efficacy (Ashraf and Mahdinezhad, 2015; Wicaksono et al., 2023) and (2) explore students’ perceptions of using VoiceThread for assessment activities. The findings will shed new light on how online platforms can facilitate self- and peer assessments while accounting for cultural and contextual factors specific to Taiwan’s EFL environment. This research will provide practical insights for Taiwanese English teachers seeking to implement technology-mediated assessment approaches that address their students’ unique needs and challenges in an increasingly interconnected global society (Crystal, 2003). In addition, the results will help bridge the gap between theoretical frameworks of self-regulated learning, and their practical application in online assessment environments, potentially informing the development of more culturally responsive and theoretically grounded assessment practices in similar EFL contexts.Literature reviewTheoretical frameworkThis study is grounded in two theoretical frameworks that inform the application of online self- and peer assessments to influence English speaking and self-efficacy: Bandura’s (1986, 1997) self-efficacy theory and Vygotsky’s (1978) sociocultural theory. These frameworks have been extensively applied in L2 speaking research (Al-khresheh and Alkursheh, 2024; Kim, 2024; Li et al., 2024), providing a comprehensive lens for understanding how different assessment modalities influence language learning outcomes and self-efficacy development.Bandura’s self-efficacy theory posits that learners’ beliefs about their capabilities significantly influence their performance through four primary sources: mastery experiences, vicarious experiences, social persuasion, and emotional states. First, mastery experiences, the most crucial source of self-efficacy, involve accomplishing a task (Usher and Pajares, 2008). Mastery experiences are particularly influential in language learning contexts where repeated successes build confidence and competence (Li et al., 2024). In this study, VoiceThread-mediated SA provided learners with structured opportunities to review and evaluate their own speaking performances, which allowed them to track their progress over time, identifying both improvements and areas needing attention. The asynchronous nature of VoiceThread allowed learners to attempt multiple recordings until satisfied, potentially fostering positive mastery experiences.Second, vicarious experiences occur through observing others’ successes and failures, mainly when the observed individuals are considered similar to the observers themselves (Schunk and Usher, 2019; Wang and Li, 2019). Students gain exposure to various speaking models in PA conditions by watching their peers’ recorded speeches. This exposure serves dual purposes: stronger performance provides aspirational models, while observing peers at similar proficiency levels helps normalize the learning process and demonstrates achievable progress. Third, social persuasion through verbal encouragement and feedback influences self-efficacy (Aben et al., 2022). PA operationalizes this source through structured peer feedback, while SA incorporates elements of self-persuasion through guided reflection protocols. VoiceThread’s multimedia commenting features enable detailed, specific feedback that can reinforce positive self-beliefs while providing constructive suggestions for improvement. Fourth, emotional and physiological states affect how individuals judge their capabilities (Phan and Ngu, 2016). VoiceThread’s asynchronous environment potentially reduces speaking anxiety by removing immediate audience pressure. Both SA and PA conditions are designed to create low-anxiety evaluation environments where L2 learners can develop comfort with speaking English over time.Lastly, complementing self-efficacy theory, Vygotsky’s sociocultural framework emphasizes the social nature of language learning and development. This framework is particularly relevant to understanding how PA facilitates L2 speaking development through social interaction and collaborative evaluation. Furthermore, both assessment conditions incorporate structured evaluation rubrics and guiding questions that scaffold learners’ developing ability to assess speaking performance. This scaffolding gradually develops students’ metacognitive awareness and self-regulated learning. In short, the differential engagement of self-efficacy sources and sociocultural learning mechanisms between SA and PA conditions may explain potential variations in speaking proficiency and self-efficacy outcomes.Self-efficacy and sociocultural theories offer a solid foundation for understanding how assessment practices impact the development of L2 speaking and self-efficacy. However, substantial pedagogical challenges remain in applying these concepts to online learning settings. It is still unclear how to optimize online assessments to activate various sources of self-efficacy while maintaining the social learning advantages highlighted by sociocultural theory. This study addresses this challenge by exploring how different assessment types (self- versus peer) delivered via VoiceThread may vary in their capacity to engage those theoretical mechanisms to facilitate L2 speaking improvement and boost self-efficacy.Self-efficacy and L2 speakingLearning to speak a second language well is far more complicated than just activating the language learner’s cognitive processing and memorizing the grammar. Based on Bandura’s social learning theory (1997), affective factors such as self-efficacy also play a crucial role in shaping EFL learners’ oral proficiency. EFL teachers have frequently observed in their classrooms that students with poor proficiency in English are mostly those who do not believe they can learn English well. Recent research has consistently shown a strong positive correlation between learners’ self-efficacy and their EFL achievements (Kim, 2024; Li et al., 2024).Bandura (1986) defined “self-efficacy” as learners' self-belief in learning particular tasks or skills. Learners with high self-efficacy tend to face challenges rather than avoid them, and are able to recover quickly from setbacks. Therefore, Bandura proposed that “the stronger the perceived self-efficacy, the higher the goal challenges people set for themselves” (1986, p. 362). Research in this area continues to validate the critical implications of self-efficacy theory in academic settings, showing that learners’ self-perceived efficacy affects their academic achievement (Al-khresheh and Alkursheh, 2024; Wicaksono et al., 2023). In L2 learning specifically, recent research has demonstrated that anxiety and self-efficacy influence students’ speaking skills (Wijaya, 2024), while public speaking self-efficacy has been found to predict achievement (McNatt, 2019).In addition, literature on the relationship between self-efficacy beliefs and performance implies that self-efficacy is susceptible to change via positive experiences. Studies investigating self-efficacy belief change have consistently argued that self-efficacy beliefs are flexible and subject to development, rather than being fixed (Al-khresheh and Alkursheh, 2024; Bandura and Schunk, 1981; Kim, 2024). The critical issue is developing instructional activities and novel applications to enable students to experience success and to reconstruct their self-efficacy perceptions accordingly. Goal setting and online technology use have been identified as effective mechanisms for cultivating self-efficacy beliefs (Basileo et al., 2024; Wang and Sun, 2024).In brief, prior studies suggested that the relationship between mastery experience and self-efficacy is cyclical and two-way: greater efficacy leads to more significant effort and persistence, resulting in better performance and enhanced efficacy. While much research has focused on learners’ perceived self-efficacy in foreign language learning (Li and Zhang, 2023; Sengul and Buyukkarci, 2023), there remains a gap in understanding how specific instructional strategies, such as VoiceThread-mediated self- and peer assessment, might affect L2 speaking self-efficacy. In EFL classrooms, adopting alternative assessment methods such as self- and peer evaluation has proven valuable for promoting goal setting and enhancing learner autonomy (Ashraf and Mahdinezhad, 2015; Liao, 2023; Wicaksono et al., 2023). These assessment practices may play a crucial role in developing students’ self-efficacy. More empirical studies are needed to understand how these assessment practices play a crucial role in developing students’ self-efficacy by activating multiple sources of efficacy information within a sociocultural framework of learning and development.Theoretical and empirical studies on self-efficacy in L2 speaking have pointed out a significant pedagogical challenge. Although self-efficacy beliefs are malleable and crucial for speaking development, less is known about how specific instructional approaches can effectively cultivate these beliefs in online environments. This study’s second research question intended to fill the gap by examining how different VoiceThread-mediated assessment types might affect L2 speaking self-efficacy. Understanding these relationships could help educators design more effective online speaking activities that simultaneously develop linguistic competence and strengthen self-efficacy beliefs.Self- and peer assessment in L2 learningRecent studies on self- and peer assessment in language learning have consistently demonstrated their positive impact on learners’ language proficiency, self-efficacy, and autonomy. For instance, research by Aldosari and Alsager (2023) highlighted how SA improves language skills and fosters resilience, creativity, and learner autonomy among EFL learners. These findings align with earlier studies that emphasized the role of SA in promoting metacognitive skills and self-directed learning (Alibakhshi and Sarani, 2014; Chen, 2010). Research suggests that these assessment practices can significantly enhance oral language skills. Alibakhshi and Sarani (2014) studied how SA affects speaking fluency and accuracy. They worked with 60 EFL students at pre- and upper-intermediate levels, split between a control and an experimental group. Their pre- and post-test results confirmed that SA boosted speaking accuracy and fluency, suggesting its value for students and teachers. SA has proven particularly beneficial in speaking classes.Besides SA, PA has also been shown to contribute significantly to language development. A study by Esfandiari and Tavassoli (2019) demonstrated that PA enhances learners’ oral proficiency and boosts their confidence and social skills by fostering a sense of accountability. Joo (2016) examined students’ ability to evaluate both their oral and peers’ work. The findings highlighted that speaking improvement stems from the assessment and students’ active involvement in the evaluation process. Similarly, research by Ashraf and Mahdinezhad (2015) found that PA activities positively impacted learner autonomy in speaking tasks. The collaborative nature of PA allows students to reflect on their peers’ performances while improving their own. Moreover, Zarei and Usefli (2019) investigated how different assessment types affected Iranian EFL learners’ general and academic self-efficacy. After taking the Primary English Test (PET), 94 students were divided into three groups, each experiencing self-, peer, or teacher assessment. Students completed general and academic self-efficacy questionnaires before and after the intervention. ANCOVA analysis revealed that while the assessment types had similar impacts on general self-efficacy, self- and peer assessment proved more effective than teacher assessment for academic self-efficacy. Interestingly, self- and peer assessment showed comparable effectiveness in boosting academic self-efficacy.Studies comparing the effects of self- and peer assessment on second language speaking have revealed that both assessment methods can positively impact learners’ speaking skills, although their effects may differ across implementation contexts. Imani (2021) found that self- and peer assessment had comparable positive effects on both impulsive and reflective EFL learners’ speaking ability, with no significant interaction between assessment type and cognitive style. However, Maleki et al. (2024) observed that SA was more effective than PA in terms of improving ESP students’ speaking ability and reducing their second language speaking anxiety. This finding was echoed by Zheng et al. (2024), who pinpointed the optimal sequence for integrating SA first for high-anxiety learners versus introducing PA for low-anxiety learners. Zheng et al. (2024) studied AI-supported formative assessment in English public speaking with three key findings: peer assessment led to higher social engagement than automated assessment, students had mixed views on different assessment types, and both self- and peer assessment improved speaking skills, with varying impacts on anxiety, engagement, and specific language abilities. The study highlights the limitations of automated assessment in replacing peer interaction for language learning.However, challenges related to the reliability of self- and peer assessment persist. Studies have identified issues such as assessor bias or evaluation inconsistency (Singh, 2015; Yinjaroen and Chiramanee, 2011). For example, some learners overestimate or underestimate their abilities compared to teacher assessments (Patri, 2002). In light of these challenges, implementing structured guidelines is key to mitigating such issues. Falchikov (2007) suggested that providing clear criteria and training can improve the accuracy of both self- and peer assessments. For example, Chen’s (2010) study showed that Chinese university students’ ratings aligned more closely with teachers’ assessments after two rounds of practice and guidance.To explicitly compare SA and PA, Table 1 summarizes the language-learning-related benefits and the limitations of SA and PA.Table 1 Comparing self-assessment and peer assessment.Full size tableIn short, the literature indicates that both SA and PA offer distinct yet complementary benefits for L2 speaking development. SA appears particularly effective for improving speaking accuracy/fluency and reducing anxiety (Maleki et al., 2024), whereas PA contributes to social engagement and collaborative learning (Zheng et al., 2024). Both assessment types show promising effects on learners’ academic self-efficacy (Zarei and Usefli, 2019); however, their relative effectiveness may vary depending on learner characteristics, such as anxiety levels (Zheng et al., 2023). Most existing studies have examined these assessment types in traditional classroom settings, leaving their comparative effects largely underexplored in online environments, which leads to the first research question. Moreover, considering the implementation challenges identified by the previous research, including assessment reliability and peer feedback quality (Singh, 2015; Zheng et al., 2024), online platforms have emerged as a feasible medium to mitigate the drawbacks of alternative assessments with their affordance of providing clear guidance, convenient training, and constant feedback for optimal self- and peer assessment. This potential of online platforms for enhancing SA and PA practice necessitates investigating how tools such as VoiceThread might mediate these assessment types to optimize L2 speaking development.Online assessment in L2 speakingOnline assessment, or e-assessment, refers to using digital tools to deliver, store, and grade student performance in various forms, including text, audio, video, and images. It can be conducted synchronously or asynchronously for individuals or groups (Crisp, 2011). Recent studies have documented the positive impact of online assessment on L2 speaking skills. One effective tool for online assessment is VoiceThread, which harnesses unique affordances to enhance oral proficiency through self- and peer assessment. The platform’s asynchronous feedback system enables assessors to provide targeted commentary at specific moments in speakers’ oral productions. At the same time, its multimodal interface combines audio, visual, and text-based interactions to support comprehensive assessment.For example, in Pagkalinawan’s (2021) study, students who recorded and assessed their presentations via VoiceThread improved their vocabulary, grammar, fluency, and pronunciation. The platform’s multimodal features allowed students to receive feedback in multiple formats. The asynchronous nature of the tool boosted their confidence by enabling them to review and respond to feedback multiple times, leading to enhanced preparation and delivery skills. Similarly, Salehi and Rowan (2015) found that VoiceThread’s multimodal commenting features facilitated detailed peer assessment, allowing students to critically evaluate each other’s work through oral and written feedback, improving their speaking performance. Moreover, Liao’s (2023) study on online self-and peer assessment demonstrated how VoiceThread’s combination of visual, audio, and text-based interaction supported the development of learner autonomy and improved L2 English speaking skills. Participants in the experimental group who engaged with these multimodal assessment features outperformed the control group in speaking performance and demonstrated a stronger sense of learner autonomy.Studies examining students’ perceptions of online speaking assessments have revealed both platform-specific benefits and challenges. Students applauded VoiceThread for its multimodal feedback options and the flexibility of working on assessments at their convenience (Amalia, 2018; Komang, 2021; Wibowo and Novitasari, 2021). Similarly, Çetin Köroğlu’s (2021) study with 52 English-major university students in Turkey revealed that participants using digital formative assessment tools outperformed those taking traditional speaking tests, attributing this success to the opportunity for learner self-reflection and peer collaboration through task completion.However, it is important to note that VoiceThread, like any tool, has its challenges. For instance, the delayed feedback from the asynchronous nature of the platform could disrupt the assessment flow. Giving specific feedback on speaking skills in a digital format can be challenging for L2 speakers, and so may require modeling and training. While some challenges stem from technical aspects of oral recording and playback quality, others involve pedagogical considerations, such as providing adequate training, keeping assessment consistent across different review sessions, and ensuring effective use of multimodal feedback tools. In short, while online speaking assessment platforms offer enhanced opportunities for detailed feedback and engagement in second language acquisition, successful implementation requires careful attention to platform-specific affordances and limitationsThe literature on online speaking assessment highlights both opportunities and challenges in implementing digital tools for L2 speaking development. While platforms such as VoiceThread offer unique affordances for multimodal feedback and flexible assessment opportunities, successful implementation requires careful consideration of both technical and pedagogical factors. This connects directly to all three research questions in this study, as how VoiceThread-mediated assessment types affect both learning outcomes and learner perceptions is examined.The present studyDespite growing interest in online speaking assessment for second language learners, significant gaps exist in understanding how theoretical frameworks, assessment practices, and digital platforms intersect to support L2 speaking and self-efficacy development. First, while self-efficacy theory and sociocultural theory suggest different mechanisms through which assessment practices might influence learning outcomes, empirical evidence about how these mechanisms operate in online environments remains scarce. Second, although research demonstrates the potential of both self- and peer assessment for L2 speaking development, educators face continued challenges in implementing these practices effectively in online contexts. Third, while digital platforms such as VoiceThread offer promising features for speaking assessment, a better understanding of how to leverage these tools to optimize speaking development and self-efficacy growth is needed. This study intended to fill these gaps by addressing the following three research questions:1.What are the effects of VoiceThread-mediated self-assessment compared to peer assessment on the oral proficiency development of L2 English speakers?2.What are the effects of VoiceThread-mediated self-assessment compared to peer assessment on the speaking self-efficacy of L2 English speakers?3.What are L2 English speakers’ perceptions of VoiceThread-mediated self- and peer assessments after participating in them for 1 year?MethodsResearch designThis mixed-methods study investigated the differential effects of VoiceThread-mediated SA versus PA on English-speaking proficiency and self-efficacy, grounded in self-regulated learning and sociocultural theory. The design aimed to address critical challenges in EFL contexts, such as low sense of confidence among Taiwanese students, limited opportunities for authentic speaking practice, and lack of formative assessment. VoiceThread-mediated assessment was chosen to create a structured environment that promotes self-regulated learning through its asynchronous and multimodal features. VoiceThread’s asynchronous and multimodal nature helps reduce performance anxiety by allowing multiple recording attempts, enables deeper reflection through unlimited replay capabilities, facilitates formative assessment through standardized protocols, and creates a permanent record of progress for metacognitive monitoring. Participants were 37 adult English as a foreign language (EFL) students (power analysis indicated minimum required n = 35, α = 0.05, power = 0.80, Faul et al., 2009) enrolled in a required English Oral Training One course at a 4-year university in central Taiwan. Two intact classes, previously formed based on standard university enrollment procedures, were randomly assigned to SA (n = 19) and PA (n = 18) conditions.Drawing from Vygotsky’s sociocultural theory, the design leverages technology-mediated interaction to create zones of proximal development, where learners develop oral proficiency through structured self-reflection (SA group) or peer scaffolding (PA group). The assessment protocols were designed to facilitate the internalization of speaking strategies through systematic engagement with cultural tools (VoiceThread features) and metacognitive processes. The implementation protocol operationalized key theoretical principles while addressing practical challenges in EFL contexts. For the SA group, the three-step evaluation process exemplifies self-regulated learning theory through its emphasis on planning (pre-speech preparation), monitoring (performance review), and reflection (video-based commentary). Similarly, the PA group’s systematic review process creates opportunities for collaborative learning and peer scaffolding, consistent with sociocultural theory’s emphasis on social interaction in learning.The quasi-experimental design was used to examine assessment types as the independent variable. English oral proficiency (measured using the validated in-house speaking assessment, inter-rater reliability α = 0.89) and English-speaking self-efficacy (measured using the ESSEQ scale, Cronbach’s α = 0.92) were dependent variables. The random assignment of intact classes, rather than individual students, was chosen to maintain ecological validity and prevent treatment diffusion while still providing reasonable control over potential confounds. Baseline equivalences in dependent variables between groups were established via independent samples t-tests. Two independent samples t-tests were conducted to establish baseline equivalence between the SA and PA groups based on their pre-test scores. For English oral proficiency, there was no significant difference between the SA group (M = 13.84, SD = 1.97) and the PA group (M = 14.56, SD = 1.55), t(35) = 1.221, p = 0.230. Similarly, speaking self-efficacy scores did not differ significantly between the SA group (M = 3.46, SD = 0.48) and the PA group (M = 3.54, SD = 0.62), t(35) = −0.416, p = 0.680. These results indicate that both groups began the study with comparable levels of English oral proficiency and speaking self-efficacy, thereby supporting the assumption of baseline equivalence between the groups. Both quantitative and qualitative data were collected over the academic year. Quantitative data included pre- and post-intervention measures of oral proficiency and self-efficacy. Qualitative data comprised a post-project questionnaire, open-ended survey, and semi-structured group interview.VoiceThread-mediated assessments were conducted monthly following a standardized protocol. ANCOVA was employed to analyse quantitative data after confirming assumptions of normality and homogeneity of variance. The selection of ANCOVA over repeated measures ANOVA was guided by both statistical and methodological considerations. Given the quasi-experimental design with two intact classes (SA = 19, PA = 18), ANCOVA offered superior statistical control by treating pre-test scores as covariates, thus accounting for initial between-group differences in English proficiency and speaking self-efficacy. This approach enhanced statistical power through reduced error variance, and provided more precise estimates of the intervention effects while controlling for pre-existing differences that could influence the outcomes. Additionally, the primary research focus was comparing the effectiveness of two assessment approaches rather than the temporal pattern of change; ANCOVA aligned more closely with the analytical objectives. Qualitative data underwent thematic analysis following a systematic approach. Integration of quantitative and qualitative findings followed a convergent parallel design.Participants and pedagogical contextUsing convenience sampling, 37 first-year, English-major students from the two intact Oral Training One classes taught by the researcher were recruited with their informed consent. Convenience sampling provided practical and easy access to the participants. However, the researcher acknowledges its potential limitations, such as reduced generalizability, potential bias, and insufficient representation of the broader population. Participants’ English proficiency levels, based on the institution’s placement examination scores calibrated to the Common European Framework of Reference for Languages, ranged from B1 to B2. Two intact classes were randomly assigned to the SA or PA condition. Prior to data collection, institutional review board approval was obtained, and all participants provided written informed consent.The English Oral Training One course, a year-long required component of the undergraduate English curriculum at a comprehensive university in Taiwan, was designed to develop students’ public speaking competence. The course curriculum comprised eight systematically sequenced speech tasks of increasing complexity, ranging from a 3-minute descriptive speech about a meaningful object to a 7-min persuasive presentation. Speech assignments were spaced approximately 2–3 weeks apart to allow adequate preparation time. Both SA and PA groups received identical instructional content, speech assignments, and evaluation criteria to maintain treatment fidelity. During non-presentation weeks, instruction focused on genre-specific rhetorical strategies, organizational patterns, and delivery techniques. Model speeches demonstrating varying levels of performance were analysed to establish clear quality benchmarks. Prior to each speech task, detailed scoring rubrics were provided and explained.All speeches were video-recorded and uploaded to VoiceThread for subsequent assessment activities. The instructor provided standardized 2–3 min oral feedback to each student immediately following their in-class presentations, using consistent evaluation criteria across both groups.Implementation of VoiceThread-mediated self- and peer assessmentFollowing the in-class speech delivery, students in both groups were required to complete their retrospective, VoiceThread-mediated assessment by the end of the following week. VoiceThread, a web-based multimodal collaboration platform, was selected for its multimedia capabilities and accessibility features. The platform enables asynchronous video-based interaction, and supports both individual reflection and peer feedback through integrated multimedia commenting tools.The assessment protocol was implemented consistently across all eight speech tasks throughout the academic year. The VoiceThread platform facilitated the systematic scaffolding of assessment skills through several structured supports. Initially, the instructor provided students with detailed assessment guidelines and exemplar videos demonstrating effective evaluation techniques. The platform’s multimedia interface allowed the instructor to embed time-stamped comments at specific points in sample speeches, modeling how to provide focused feedback on various aspects of oral performance (e.g., pronunciation, organization, and delivery). The following forms of scaffolding were carried out to prepare participants for actual assessments.For the SA group, scaffolding was implemented through a three-tiered process. First, students used guided SA forms that prompted reflection on specific speech components. Second, they analysed instructor-annotated model speeches that highlighted key assessment criteria. Third, they practiced self-evaluation with shorter speech segments before posting a finalized one. The platform’s playback features enabled students to revisit specific portions of their speeches repeatedly, supporting detailed analysis and reflection.The PA group followed a similar form of scaffolding but emphasized peer interaction features. Students first practiced peer assessment using pre-selected speech samples with instructor guidance. The platform’s threaded commenting feature allowed collaborative dialogue about assessment criteria and standards. Students could build upon each other’s observations while receiving peer feedback. The asynchronous nature of VoiceThread allowed students to carefully construct and revise their feedback before posting it on VoiceThread.To facilitate peer-assessment, all participants were placed in groups of three or four on VoiceThread. Participants in the SA group completed a three-step evaluation process: reviewing their recorded speech performance on VoiceThread, completing a structured SA form, and recording video-based reflective commentary using specified assessment criteria (see Appendix A). For the PA group, participants followed a systematic review process that involved viewing their group members’ recorded speeches, analysing performance using standardized assessment criteria, and recording video-based evaluative feedback.To ensure meaningful engagement with the feedback, students were required to submit their 3-min-long video-recorded comments on VoiceThread, guided by structured assessment sheets. This standardized length was chosen to ensure consistency in feedback depth across all assessments. The students were encouraged to address all the evaluation criteria on the assessment sheets rather than just meeting the minimal length of the video-recorded assessment. The teaching assistant monitored both submission of assessments and the quality of the submission. All students had a week after their speech to complete their assessments (self- or peer), with grade penalties for late submissions. In groups of three to four students, each PA participant evaluated two group members’ speeches to ensure a balanced distribution of peer feedback within the group, and to avoid always receiving feedback from the same group member. English was used for all recordings as it is the primary language of instruction in this English department. While this might have affected some students’ ability to express complex thoughts, it aligned with the course’s language learning objectives and provided additional speaking practice.In sum, assessment guidelines and rubrics were standardized across both conditions, with only the focus of evaluation (self- vs. peer) differing between groups. Regular monitoring of assessment completion and feedback quality was conducted to maintain implementation fidelity. The structured nature of both assessment protocols ensured consistency in feedback provision while accommodating the distinct characteristics of the self- and peer evaluation processes.Data collectionQuantitative data were collected from the English proficiency evaluation, English-speaking self-efficacy questionnaire (ESSEQ), and VoiceThread-mediated assessment survey, while qualitative data were collected from a post-project open-ended question survey and group interview. English proficiency evaluation and ESSEQ were administered to the participants before and after the year-long engagement in VoiceThread-mediated assessments. The perception questionnaire survey, open-ended survey, and group interview were conducted toward the end of the project. Four data collection instruments were utilized to answer the three research questions.English speaking proficiency test: 37 students recorded one 3-min speech before and after the intervention. The two topics, “My most memorable vacation” and “My most unforgettable achievement”, were counterbalanced between tests through random assignment within each condition (SA and PA). Within each condition, approximately half the students were randomly assigned to speak about their vacation in the pre-test and about their achievement in the post-test. At the same time, the other half did the reverse. This between-test counterbalancing aimed to reduce potential differences in topic difficulty or complexity. Both topics were chosen to be similar, encouraging comparable narrative skills and complexity, thus alleviating the effect of topic familiarity on performance outcomes.Three trained raters, blind to participant condition and assessment timing, evaluated the recordings using an adapted TOEFL Independent Speaking Rubric (interrater reliability = 0.89, Appendix B). The rubric assessed delivery (fluency and pronunciation) and language use (accuracy and complexity) on a five-point scale (1 = very poor to 5 = very good). Three experienced lecturers from the English language centre, each with at least 5 years of teaching experience, served as raters. The raters underwent a 3-h training session conducted by the researcher. It involved carefully listening to five sample speech videos, studying the scoring rubrics, and comparing scores among the three raters until they established 89% inter-rater reliability. Every recorded speech was evaluated independently by two raters, with the average of their scores serving as the pre-test or post-test English oral proficiency scores. Each rater scored 24 to 25 recorded speeches for the pre-test at the beginning and another 24 or 25 for the post-test at the end of the academic year.The English-speaking self-efficacy questionnaire (ESSEQ, Appendix C): This questionnaire was adapted from Alavi et al. (2004) validated questionnaire through a systematic process. The adaptation involved reviewing the items to align with the study context. Two experts in computer-assisted language learning and educational psychology reviewed the adapted items using a standardized evaluation rubric focusing on content relevance, linguistic clarity, and cultural appropriateness. Their feedback led to the finalized version of the 24-item ESSEQ, which was pilot-tested with 15 students matching the target population’s characteristics. The final instrument measured three dimensions:Accuracy (8 items, α = 0.92): measures participants’ confidence in grammatical and phonological precision (e.g., “I can use accurate grammar while speaking English” and “I am able to speak with correct pronunciation”).Fluency (9 items, α = 0.91): assesses confidence in maintaining natural, smooth speech flow (e.g., “I am capable of speaking for a few minutes in such a way as to clearly express my ideas” and “I am able to give a well-organized speech”).Interlocutory self-efficacy (7 items, α = 0.93): Evaluates confidence in various interactive speaking contexts (e.g., “I am comfortable talking with foreigners” and “I am able to discuss subjects of general interest with my classmates in English”). Respondents used a five-point Likert scale (1 = strongly disagree to 5 = strongly agree).The VoiceThread-mediated self-/peer assessment questionnaire (VTSQ): This questionnaire was administered post-intervention to assess participants’ perceptions of the assessment process. The 19-item instrument was adapted from Chiang’s (2018) validated questionnaire to address the specific context of VoiceThread-mediated assessment. The questionnaire encompassed four key aspects: (a) the usefulness of VoiceThread-mediated assessment for enhancing speaking fluency, (b) the usefulness of VoiceThread-mediated assessment for enhancing accuracy, (c) the usefulness of VoiceThread-mediated assessment for cultivating speaking self-efficacy, and (d) the features and functions of the VoiceThread platform for assessment purposes. Two parallel versions were developed: one for the SA group and one for the PA group, with items appropriately worded to reflect the respective assessment type (e.g., “Reviewing my own speech…” vs. “Reviewing my peers’ speech…”). The equivalence between versions was established through expert review. Participants rated items on a five-point Likert scale (1 = strongly disagree to 5 = strongly agree). The overall instrument demonstrated strong internal consistency (α = 0.94) and content validity through expert review.Qualitative data: The study employed a sequential qualitative data collection approach, beginning with open-ended surveys followed by semi-structured group interviews. Both the surveys and interviews were conducted in English, the primary language of instruction in the program, to maintain consistency with the learning environment. This sequential design was deliberately chosen to allow participants to first reflect on and document their experiences thoughtfully through written responses, without the immediate pressure of face-to-face interaction. The open-ended survey explored participants’ experiences with VoiceThread-mediated assessment, including perceived benefits, challenges, linguistic gains, and future implementation suggestions. The SA group was asked about reviewing their recorded speech, whereas the PA group was asked about giving comments to and receiving comments from their peers. The written format provided participants with time to process their year-long learning journey and to articulate their thoughts in detail. Building on these initial reflections, follow-up semi-structured group interviews (45–60 min) were conducted to probe more deeply into themes emerging from the survey responses. The sample interview questions included inquiries about changes in video-recording comments with VoiceThread (for both), suggestions for platform improvements, self-reflection on recording review processes and personal growth in self-observation (for SA), and perspectives on balancing supportive and constructive peer feedback (for PA). Twelve participants (six from the SA and six from the PA group) were purposively selected for the interviews based on the following criteria: participants provided detailed examples of their VoiceThread experiences that aligned with the study’s focus on assessment practices; participants raised unique challenges or benefits that needed elaboration; and participants whose initial responses suggested distinctive patterns in their use of the platform. This sequential approach enabled participants to engage in both individual written reflection and collaborative dialogue, where they could elaborate on their written responses and react to peers’ experiences.Data analysisQuantitative analyses were conducted using IBM SPSS 27.0. Analysis of covariance (ANCOVA) was employed to examine the differential effects of VoiceThread-mediated SA versus peer assessment on participants’ English-speaking proficiency and self-efficacy. Pre-test scores served as covariates to control for initial group differences. Prior to conducting ANCOVAs, assumptions of normality (Shapiro-Wilk test), homogeneity of variance (Levene’s test), and homogeneity of regression slopes were tested. Effect sizes were reported using partial eta squared (η2p). Descriptive statistics and reliability coefficients were calculated for the VoiceThread-mediated Self-/Peer Assessment Questionnaire (VTSQ) to examine participants’ perceptions of the assessment process.A systematic approach was employed for qualitative data analysis to analyse responses from open-ended surveys and semi-structured group interviews (2 groups, 45–60 min each). Initial thematic analysis of survey responses informed the subsequent interview protocols. Using an iterative coding process, two researchers independently coded interview transcripts to triangulate with the tentative themes identified from the open-ended survey. When coding disagreements occurred, the researchers discussed until a consensus was reached. Emerging themes were categorized into main themes and sub-themes using constant comparative analysis. Chi-square tests were conducted to compare the occurrence of themes between the SA and PA groups. Thus, the potential group differences in their perceived effects of self- versus peer assessment were identified. Following mixed-methods best practices, qualitative findings were integrated with quantitative results through methodological triangulation to enhance the validity and comprehensiveness of interpretations. To ensure trustworthiness of the qualitative analysis, member checking was conducted with voluntary participants (n = 12) by sharing the identified themes and interpretations for their feedback.ResultsEffects of VoiceThread-mediated self-assessment versus peer assessment on English oral proficiencyIndependent oral assessments were administered to the SA (n = 19) and PA (n = 18) groups at the beginning and toward the end of the school year. The PA had slightly higher pre-test scores (M = 14.56, SD = 1.55) compared to the SA (M = 13.84, SD = 1.96) in terms of overall oral proficiency.Fig. 1 shows how English oral sub-skills developed differently between the groups. After the interventions, both groups demonstrated improvement in their post-test scores, with the SA showing a considerably higher overall mean score (M = 17.50, SD = 0.50) compared to the PA (M = 16.42, SD = 1.05). This pattern of improvement was consistent across most subcategories, including accuracy (PA: M = 4.08, SD = 0.31; SA: M = 4.32, SD = 0.25), fluency (PA: M = 4.22, SD = 0.43; SA: M = 4.34, SD = 0.29), pronunciation (PA: M = 4.11, SD = 0.27; SA: M = 4.21, SD = 0.25), and complexity (PA: M = 4.00, SD = 0.34; SA: M = 4.34, SD = 0.24).ANCOVAs were conducted with assessment type (peer vs. self-assessment) as the independent variable on four dimensions of English oral proficiency, using post-test scores as dependent variables and pre-test scores as covariates. The assumptions of normality and homogeneity of variances were met for all analyses (Levene’s tests: oral proficiency F(1, 35) = 3.757, p = 0.061; accuracy F(1, 35) = 0.142, p = 0.708; complexity F(1, 35) = 0.059, p = 0.809; fluency F(1, 35) = 0.404, p = 0.529; pronunciation F(1, 35) = 0.032, p = 0.860).Fig. 1Comparing the English oral proficiency sub-skills of the PA and SA groups.Full size imageTable 2 shows significant effects of assessment type on oral proficiency (F(1, 34) = 15.603, p 0.14), ANCOVAs were conducted with assessment type (self- vs. peer) as the independent variable, using post-test scores of four dimensions of English-speaking self-efficacy as dependent variables, controlling for the pre-test scores. SA was found to be more effective than PA in terms of enhancing students’ overall English-speaking self-efficacy, particularly in their confidence during interactive speaking tasks. Table 3 shows significant differences in overall English-speaking self-efficacy between the groups, F(1, 34) = 5.07, p 0.05, η2 = 0.001), suggesting that the assessment method primarily influenced students’ confidence in their interactive speaking abilities rather than their perceived technical language skills.Table 3 ANCOVA results of English-speaking self-efficacy (n = 37).Full size tableIn short, the assessment type had a significant impact on students’ English-speaking self-efficacy, with SA being more beneficial than PA. The assessment type did not affect students’ accuracy-related or fluency-related speaking self-efficacy. VoiceThread-mediated SA had a greater positive effect on L2 learners’ interlocutory self-efficacy than PA, highlighting its potential benefits and encouraging further exploration.The SA and PA groups’ perceptions of VoiceThread-mediated assessmentThe questionnaire survey, open-ended survey, and group interview were carried out at the end of the project to gain an emic view of VoiceThread-mediated self- and peer assessment from the participants. The descriptive statistics of the questionnaire survey will be presented first, followed by the chi-square tests of different perceptions deducted from the qualitative analysis of the themes which emerged from the open-ended survey and group interview. The overall results showed that the SA and PA groups perceived VoiceThread-mediated assessment differently, despite some common retrospective thoughts after a year-long participation.The comparative analysis of the 19-item questionnaire survey (Appendix D) from the SA and PA groups highlights significant disparities in their perceptions of VoiceThread-mediated assessment. The SA group consistently demonstrated more positive perceptions across questionnaire items, with mean scores ranging from 3.28 to 4.33, compared to the PA group’s lower range of 2.39–4.11. This overarching trend suggests that students who engaged in SA generally found their experience to be more beneficial. A notable point of agreement between both groups was their highest ranking of in-class speech practices (Item 1), with means of 4.33 and 4.11 for the SA and PA groups, respectively. This shared emphasis indicates that, regardless of the assessment approach, students valued the confidence-building aspects of in-class speaking activities. However, the groups diverged significantly in their other top-ranked items. The SA group particularly valued personal growth through self-review (mean = 4.28) and instructor feedback (mean = 4.06), while the PA group prioritized instructor feedback (mean = 3.78) and benefits from reviewing peer speeches (mean = 3.72).The SA and PA groups perceived learning outcomes differently. SA participants reported substantially higher benefits in oral fluency improvement (SA mean = 3.83 vs. PA mean = 2.78), grammatical accuracy and complexity (SA mean = 3.83 vs. PA mean = 2.78), and lexical diversity (SA mean = 3.51 vs. PA mean = 2.72). This pattern suggests that self-reflection might be more effective than peer feedback in promoting perceived language development. The qualitative data revealed that SA participants engaged in systematic self-correction and targeted improvement, which might help explain the differently perceived language growth between the groups.Motivation and engagement levels also differed significantly between the groups, particularly in related measures of self-efficacy. The SA group demonstrated higher motivation for creating video comments (mean = 3.44 vs. 2.89), greater willingness to continue with online tasks (mean = 3.42 vs. 2.83), and more comfort with online video posting (mean = 3.78 vs. 3.44). The qualitative data help explain this growth in self-efficacy. For instance, SA participants described a progression from discomfort to confidence through their self-review process. As one SA student noted, “I became less shy of oral tests,” while another emphasized their growth journey: “I have grown more confident.” In contrast, PA participants’ comments revealed ongoing struggles with self-efficacy: “I’m not brave enough to make mistakes in public even though I’m still in the learning process. But I think I made great progress with it”. These differences in affective aspiration help explain why SA participants reported higher comfort levels with video tasks.A chi-square test of independence was conducted to examine potential differences in the SA and PA groups’ perceived benefits. The analysis shows a statistically significant association between the two groups and the perceived benefits (χ2 = 9.847, df = 3, p = 0.020). Table 4 presents the two most significant differences of the four analysed.Table 4 Categories and sub-categories of students’ perceived benefits of VoiceThread-mediated self- and peer assessment.Full size tableThe most notable difference emerged in the advantage of performance review, where the SA group exhibited a substantially higher frequency (52.0%, adjusted residual = 3.1) compared to the PA group (25.8%, adjusted residual = −3.1). For instance, one female student in the SA group commented, “If I don’t see my video, I won’t find out that my pronunciation is weird. My gestures are also weird in the video, which helped me see I have a lot to improve while giving a speech”. Another SA participant said, “I often write down that I failed to express my idea so that I can fix it and say it right next time”. These specific, detailed approaches to self-review help explain why SA participants reported higher motivation scores for video tasks (mean = 3.44 vs. 2.89).While both groups maintained relatively balanced distributions across observation opportunity (self: 28.0%, peer: 32.3%), their observation patterns revealed different learning approaches. The PA group’s comments showed more general observations, as exemplified by one student: “I saw some problems in other people’s speeches, and I try not to make the same mistakes in my speech. That’s a benefit to me. Also, I can learn from others’ public speaking skills while watching their speeches”. The PA group valued emotional support through peer feedback, as evidenced by one male student’s comment: “I always feel too nervous to speak at a normal speed, but one time I saw a comment from my teammate; she said I was so brave to challenge myself to do an impromptu speech. I was inspired a lot”. This focus on general observation and emotional support, rather than specific language improvements, helps explain why PA participants reported lower scores in specific language skills such as grammatical accuracy (mean = 2.78) and lexical diversity (mean = 2.72).A chi-square test of independence revealed significant differences in the types of difficulties experienced by the SA and PA groups (χ2 = 19.7847, df = 4, p