Xorijiy lingvistika va lingvodidaktika
–
Зарубежная
лингвистика
и
лингводидактика
–
Foreign
Linguistics and Linguodidactics
Journal home page:
https://inscience.uz/index.php/foreign-linguistics
Corpus-based discourse analysis: Automatic Speech
Recognition (ASR) technologies and spoken corpus
collection
Nargiza ASROROVA
1
Uzbekistan State World Languages University,
Университет TIFT
ARTICLE INFO
ABSTRACT
Article history:
Received May 2025
Received in revised form
15 May 2025
Accepted 25 June 2025
Available online
15 July 2025
This study investigates the integration of Automatic Speech
Recognition (ASR) technologies in the collection and analysis of
spoken learner corpora, with a focus on L2 contexts. Employing
a mixed-methods design, the research evaluates the
effectiveness of ASR systems, specifically, Whisper and BERT-
based models in producing accurate transcriptions and
facilitating
language
acquisition.
Quantitative
results
demonstrate high transcription accuracy, while qualitative data
reveal that ASR-supported feedback significantly enhances
learner engagement, pronunciation, and speaking proficiency.
Technological limitations and ethical concerns related to data
privacy and feedback mechanisms are also taken into account.
Overall, the findings highlight the transformative potential of
ASR technologies in language education by enabling scalable,
real-time assessment and personalized feedback, while
underscoring the need for continued refinement and equitable
implementation across diverse learner populations.
2181-3701
/©
2025 in Science LLC.
https://doi.org/10.47689/2181-3701-vol3-iss4-pp1
This is an open-access article under the Attribution 4.0 International
(CC BY 4.0) license (
https://creativecommons.org/licenses/by/4.0/deed.ru
Keywords:
Automatic Speech
Recognition (ASR),
speech transcription,
corpus,
spoken discourse analysis,
second language acquisition.
Korpusga asoslangan diskurs tahlili: Nutqni avtomatik
aniqlash
(ASR) dasturlari va og‘zaki nutq korpusini
to‘plash
ANNOTATSIYA
Kalit so‘zlar
:
Avtomatik nutqni aniqlash
(ASR),
nutq transkripsiyasi,
Ushbu tadqiqot Avtomatik nutqni aniqlash (ASR)
texnologiyalarining ingliz tilini ikkinchi til sifatida (L2)
o‘rganuvchilarning og‘zaki korpuslarini yig‘ish va tahlil qilish
1
PhD student, Uzbekistan State World Languages University, Teacher, TIFT University
Xorijiy lingvistika va lingvodidaktika
–
Зарубежная лингвистика
и лингводидактика
–
Foreign Linguistics and Linguodidactics
Issue
–3 № 4 (2025)
/ ISSN 2181-3701
2
korpus,
og‘zaki diskurs tahlili,
ikkinchi tilni o‘zlashtirish
.
jarayoniga integratsiyasini o‘rganadi. Turli metodlarni
aralashtirish usulida olib borilgan mazkur tadqiqot ishida
Whisper va BERT asosidagi modellar misolida ASR tizimlarining
aniq transkripsiyalar yaratish va ikkinchi til o‘zlashtirish
jarayonini oshirishdagi samaradorligi baholandi. Miqdoriy
natijalar transkripsiya yuqori aniqlikda ek
anligini ko‘rsatdi,
tavsifiy jihatdan esa ASR ga asoslangan fikr-mulohazalar
(feedback) talabalar faolligining, hamda talaffuz va og‘zaki nutq
ko‘nikmalarining sezilarli darajada yaxshilanishiga olib kelishini
aniqladi. Texnologiyaning kamchilik tomonlari,
ma’lumotlar
maxfiyligi hamda fikr-
mulohaza mexanizmlari bilan bog‘liq
axloqiy masalalar ham tadqiqot doirasida o‘rganib chiqildi.
Tadqiqot natijalari ASR texnologiyalarining til ta’limini tubdan
o‘zgartirish salohiyatiga ega ekanligini tasdiqladi, ya’ni m
azkur
texnologiya
o‘quvchilarni
real
vaqtda
baholash
va
individuallashtirilgan fikr-mulohazalar (feedback) taqdim etish
imkonini berib, til o‘rganuvchilarning har qanday toifa
guruhlari uchun qo‘llanilishi mumkin deb topildi
.
Корпусный
анализ
дискурса:
технологии
автоматического распознавания речи (ASR) и сбор
корпус устной речи
АННОТАЦИЯ
Ключевые слова:
автоматическое
распознавание речи (ASR),
транскрипция речи,
корпус,
анализ устного дискурса,
усвоение второго языка
.
Данное исследование посвящено интеграции технологий
автоматического распознавания речи (ASR) в процессы сбора
и анализа корпусов устной речи обучающихся, с особым
вниманием к контексту изучения английского языка как
второго языка (L2). Используя смешанный метод
исследования, автор оценивает эффективность ASR
-
систем
—
в частности, моделей Whisper и BERT —
в обеспечении
точной транскрипции и содействии усвоению иностранного
языка. Количественные данные подтверждают высокую
точность транскрипций, тогда как качественные результаты
свидетельствуют о том, что обратная связь, реализованная с
помощью ASR, способствует повышению вовлечённости
учащихся, улучшению произношения и развитию навыков
устной речи. Также рассматриваются технологические
ограничения
и
этические
аспекты,
связанные
с
конфиденциальностью данных и способами предоставления
обратной связи. В целом, результаты подчёркивают
значительный потенциал технологий ASR в трансформации
языкового образования за счёт масштабируемой оценки в
реальном времени и персонализированной поддержки
обучения, при этом акцентируется необходимость их
дальнейшего
совершенствования
и
справедливого
внедрения в условиях разнообразия обучающихся
.
Xorijiy lingvistika va lingvodidaktika
–
Зарубежная лингвистика
и лингводидактика
–
Foreign Linguistics and Linguodidactics
Issue
–3 № 4 (2025)
/ ISSN 2181-3701
3
INTRODUCTION
Automatic Speech Recognition (ASR) Technologies for spoken learner corpus
collection are vital tools that significantly impact language acquisition, especially in the
context of second language (L2) learning. ASR technologies facilitate the real-time
transcription of spoken language, allowing educators and researchers to collect and
analyze learner corpora effectively. This capability enhances the assessment of spoken
proficiency and offers personalized feedback, thus contributing to improved
pronunciation and language skills among learners [1][2].
The evolution of ASR systems has resulted in various approaches tailored
for educational settings, including speaker-dependent, speaker-independent, and
speaker-adaptive models. Each type presents unique advantages and challenges,
particularly in diverse classroom environments where multiple learners interact.
Notably, ASR technologies are integrated into applications such as automatic reading
tutors and conversational agents, which have been shown to engage learners and foster
interactive language practice [3][4]. These tools not only improve language skills but also
enhance the accessibility of language learning resources.
Despite their benefits, ASR systems face notable challenges, including accuracy
discrepancies across different languages and dialects, as well as ethical concerns
surrounding privacy and data usage. Recent studies have highlighted issues such as the
potential bias against speakers from diverse linguistic backgrounds, which could impact
the fairness of language assessments [5][6]. Furthermore, the technology's performance
can be compromised by environmental factors like background noise and the individual
characteristics of learners' speech patterns, raising questions about its reliability in
varied settings [7].
ASR technologies hold immense potential for revolutionizing language learning
through enhanced feedback and engagement, but ongoing research is necessary to
address the current limitations and ethical concerns associated with their use. The future
of ASR in educational contexts hinges on advancements that improve accuracy, reduce
bias, and broaden accessibility, thereby maximizing the benefits for language learners
worldwide [5][8].
LITERATURE REVIEW
Automatic Speech Recognition (ASR) technology has evolved significantly since its
inception in the mid-20th century, marking a transformative journey in how spoken
language is processed and understood by machines. The early attempts at ASR faced
challenges with variations in accents, dialects, and speech nuances, which often resulted
in inaccuracies in transcription. Over the decades, advancements in acoustic modeling
and machine learning have enhanced the robustness and accuracy of ASR systems,
allowing for more effective communication between humans and technology [1][9].
In the context of language learning, ASR plays a crucial role in developing spoken
learner corpora, which are essential for evaluating and improving speaking proficiency
among learners of a second language (L2). These corpora can be categorized into two
main types: those focusing on native-speaking children, used primarily for developing
virtual tutors in non-language subjects, and those aimed at young language learners to
support language acquisition [10][11]. ASR systems provide real-time feedback on
learners' pronunciation and fluency, thereby offering personalized practice exercises that
are vital for language development [12].
Xorijiy lingvistika va lingvodidaktika
–
Зарубежная лингвистика
и лингводидактика
–
Foreign Linguistics and Linguodidactics
Issue
–3 № 4 (2025)
/ ISSN 2181-3701
4
Moreover, the incorporation of ASR in educational settings not only enhances
accessibility through automated captions for instructional content but also aids in the
efficient assessment of students' speaking abilities. However, despite the impressive
progress in ASR technologies, challenges such as the influence of learners' backgrounds,
including prior language exposure and educational experiences, must be addressed to
ensure a comprehensive understanding of speaking proficiency [3] [13-][14]. Future
studies could benefit from including measures to evaluate these factors, potentially
through pre-assessment surveys or interviews [3].
As ASR continues to advance, its applications extend beyond language learning into
various domains, including healthcare, customer support, and accessibility tools,
underscoring its growing significance in our daily lives [9][14]. The ongoing research
aims to tackle issues related to privacy, bias, and the computational efficiency of ASR
systems, which are crucial for their widespread adoption and effectiveness [9][15].
Automatic Speech Recognition (ASR) technologies have emerged as essential tools
for collecting spoken learner corpora, enabling educators and researchers to analyze
language acquisition in various contexts. These systems facilitate the assessment of
learner language by capturing spoken interactions in real-time and providing
transcriptions for further analysis [2].
Types of ASR Systems
ASR systems can be classified into three main types: speaker-dependent,
speaker-independent, and speaker-adaptive. Speaker-dependent ASR requires
training on an individual's voice, making it less suitable for diverse classroom settings
where multiple learners are involved. In contrast, speaker-independent ASR operates
without prior training, leveraging extensive speech databases to recognize various
speakers' voices, which is particularly beneficial in educational environments [3].
Speaker-adaptive ASR combines aspects of both, using customized databases to improve
recognition accuracy based on user input.
Impact on Language Acquisition
ASR technologies are employed in various educational applications aimed at
enhancing learner engagement and language development. For example, automatic
reading tutors utilize ASR to provide feedback on pronunciation errors by comparing
learners' speech with reference transcripts [13]. These systems are generally designed
for quiet environments, such as libraries, ensuring that only one learner reads at a time,
which facilitates focused feedback.
Conversational tutors represent another innovative use of ASR, combining speech
recognition with language generation to create interactive learning experiences. Tools
like My Science Tutor (MyST) support small-group learning by engaging learners in real-
time dialogues, which enhances their spoken language skills [13][4].
Recent studies indicate that ASR technology significantly enhances learner
engagement during language learning activities. By providing instant feedback and
interactive practice opportunities, ASR systems contribute to improved pronunciation,
vocabulary acquisition, and overall language proficiency [4]. The accessibility of free ASR
tools has also democratized language learning, allowing for quick deployment in various
educational contexts [3].
Despite their advantages, ASR systems face challenges, particularly in terms of
accuracy across different languages and dialects. While modern ASR engines have made
Xorijiy lingvistika va lingvodidaktika
–
Зарубежная лингвистика
и лингводидактика
–
Foreign Linguistics and Linguodidactics
Issue
–3 № 4 (2025)
/ ISSN 2181-3701
5
considerable advancements, achieving 100% accuracy remains elusive due to the
complexities of human speech [5]. Additionally, privacy concerns regarding data use and
storage present significant ethical considerations for educational institutions
implementing ASR technologies [6].
METHODOLOGY
This study employed an explanatory sequential mixed methods design, which
involved the systematic collection and analysis of both quantitative and qualitative data
to comprehensively address the effectiveness of Automatic Speech Recognition (ASR)
technologies in collecting spoken learner corpora. The methodology was divided into
distinct phases to provide a holistic view of the impact of ASR technologies on language
learning outcomes, particularly in the context of English L2 learners [3].
Participants
The research focused on 20 intermediate-level L2 learners from Uzbekistan,
differentiating it from previous studies by its emphasis on integrating ASR technology
with peer correction techniques during pronunciation and speaking instruction. An
experienced IELTS teacher collaborated closely with the researcher to implement the
intervention and offer feedback to participants [3]. This structured approach not only
ensured consistency across groups but also enriched the data collected through both
objective measurements and subjective learner perceptions [3].
Data Collection
The initial quantitative phase assessed the effectiveness of various ASR systems by
measuring their precision, recall, and F1 score in detecting keywords in learner speech. A
specific focus was placed on evaluating the classification accuracy of a BERT-based model
in predicting students' oral assessment grades from ASR transcripts. The results
indicated a notable 2% absolute drop in classification accuracy for African American
English (AAE) speaking students compared to their non-AAE counterparts [16]. ASR
systems, particularly Whisper, achieved a high scoring accuracy of 95.6% in comparison
to human-labeled transcripts, which scored at 96.3% [16].
Following the quantitative analysis, the qualitative phase utilized interviews to
delve deeper into learners' perceptions of ASR technology. This phase involved a
thematic analysis of the interview data, guided by Grbich's (2012) methodologies. Initial
reviews identified broad categories that were further dissected into subthemes,
capturing nuanced learner experiences and attitudes towards ASR technology's role in
their learning process [3].
Procedure
The intervention protocol was meticulously designed to align with the research
objectives, and regular meetings between researchers and the instructor ensured a
consistent approach to data interpretation and theme development. The study's design
allowed for a rigorous examination of how ASR technologies influence pronunciation and
speaking skills, filling a significant gap in existing research by highlighting the
perspectives of intermediate L2 learners [3][16].
The methodology incorporated comprehensive testing of ASR systems across
various configurations, yielding valuable insights into the strengths and weaknesses of
each technology in practical classroom settings. Future studies may expand upon this
framework by utilizing larger samples and varying procedures to further validate these
findings [3].
Xorijiy lingvistika va lingvodidaktika
–
Зарубежная лингвистика
и лингводидактика
–
Foreign Linguistics and Linguodidactics
Issue
–3 № 4 (2025)
/ ISSN 2181-3701
6
RESULTS AND DISCUSSION
Quantitative results demonstrate that the ASR systems exhibit varying levels of
precision and recall in detecting keywords, with an F1 score reflecting the balance
between these metrics. For instance, the BERT classification model for automatic
response scoring achieved its highest accuracy at 95.6% when using Whisper ASR
transcripts, in contrast to a 96.3% accuracy with human-labeled transcripts [16]. This
suggests that while ASR technologies can effectively transcribe learner speech, the
quality of the transcription can significantly affect the assessment outcomes. The
integrated approach tested on the UASpeech dataset resulted in an overall word
recognition accuracy of 69.4% across multiple speakers, indicating that while ASR
systems can achieve reasonable performance, there remains room for improvement,
particularly in diverse speaking styles [11][16].
The ASR systems provide a variety of feedback mechanisms, which are crucial for
language learners. Explicit corrective feedback is enabled through graphical
representations of speech, such as waveforms and spectrograms, allowing learners to
understand discrepancies between their pronunciation and that of native speakers [17].
This level of detail can foster a more tailored learning experience, accommodate
individual learning styles, and address specific pronunciation challenges [3]. However,
there are concerns that some ASR systems may inadvertently limit learners' practice
opportunities by relying too heavily on pre-recorded samples for comparison, thus
hindering spontaneous speech production [3].
Research indicates that ASR programs providing explicit corrective feedback are
particularly beneficial for pronunciation development in language learners. For instance,
systems like MyET, which deliver immediate, targeted feedback, have been shown to
enhance learner outcomes significantly when compared to systems that offer indirect
feedback, such as Speechnotes, which often fail to accurately identify the nature and
location of pronunciation errors [16]. This distinction emphasizes the importance of
selecting ASR technologies that align with the pedagogical goals of language instruction.
Regarding the effectiveness of ASR in language classrooms, the research
highlighted that ASR technology can lead to improved pronunciation among English as a
Foreign Language (EFL) learners. ASR systems were found to facilitate instant feedback,
allowing learners to recognize and rectify errors more efficiently [3]. Furthermore, the
integration of ASR in pronunciation training has been shown to increase student
engagement and interaction, contributing positively to their overall learning experience
[4][3].
Despite the positive outcomes associated with ASR technologies, several
challenges have been identified. For example, some students reported frustrations with
ASR systems misinterpreting words, which hindered their learning experience [16].
Additionally, the effectiveness of ASR may be compromised in suboptimal audio
conditions, leading to increased inaccuracies in speech recognition [13]. These factors
underscore the need for continued research and development in ASR technologies to
enhance their reliability and effectiveness in educational contexts.
As ASR technologies continue to evolve, future case studies should explore their
application across diverse learner demographics and educational settings. Investigating
the long-term effects of ASR on language acquisition and exploring the integration of ASR
with other technological tools could provide valuable insights into optimizing language
Xorijiy lingvistika va lingvodidaktika
–
Зарубежная лингвистика
и лингводидактика
–
Foreign Linguistics and Linguodidactics
Issue
–3 № 4 (2025)
/ ISSN 2181-3701
7
instruction and learner engagement strategies [14][15]. By addressing current
limitations and refining system capabilities, ASR has the potential to transform language
learning experiences for students worldwide.
CONCLUSION
This study underscores the significant potential of Automatic Speech Recognition
(ASR) technologies to enhance language learning outcomes, particularly within the
context of English as a Second Language (L2) instruction. Employing a mixed-methods
design, the research demonstrates that ASR systems, specifically, Whisper and BERT-
based models, are capable of generating highly accurate transcriptions and delivering
real-time, individualized feedback th
at supports improvements in learners’
pronunciation, fluency, and overall oral proficiency. The integration of ASR into
pedagogical practice not only facilitates scalable and efficient language assessment but
also promotes learner autonomy and engagement.
Nevertheless, the findings also reveal persistent challenges, including reduced
accuracy in varied linguistic and acoustic conditions, potential algorithmic bias against
speakers of certain dialects, and ethical concerns related to data privacy and informed
consent. These limitations highlight the need for continued refinement of ASR
technologies and the development of equitable, ethically sound implementation
strategies in educational contexts. Addressing these dimensions is essential for
maximizing the pedagogical benefits of ASR and ensuring its effective and inclusive
integration into language learning environments.
REFERENCES:
1.
Nakamura, S., Spring, R., & Sakurai, S. (2024). The impact of ASR-based
interactive video activities on speaking skills: Japane
se EFL learners’ perceptions. The
Electronic
Journal
for
English
as
a
Second
Language,
27(4).
https://doi.org/10.55593/ej.27108a5
2.
Gladia. (2023, December 19). A review of the best ASR engines and the
models powering them in 2024. Gladia Blog. https://www.gladia.io/blog/a-review-of-
the-best-asr-engines-and-the-models-powering-them-in-2024
3.
Michot, J., Hürlimann, M., Deriu, J., Sauer, L., Mlynchyk, K., & Cieliebak, M.
(2024). Error-
preserving automatic speech recognition of young English learners’
language. Beta archives, https://arxiv.org/html/2406.03235v1
4.
Qian, Z., Xiao, K., & Yu, C. (2023). A survey of technologies for automatic
dysarthric speech recognition. Journal of Audio, Speech, and Music Processing, 2023(48).
https://doi.org/10.1186/s13636-023-00318-2
5.
Bhatnagar, N. (2024 Deep dive into ASR systems. Medium.
https://medium.com/@captnitinbhatnagar/deep-dive-into-asr-systems-c571a576ff26
6.
Sun, W. (2023). The impact of automatic speech recognition technology on
second language pronunciation and speaking skills of EFL learners: A mixed methods
investigation.
Frontiers
in
Psychology,
14,
Article
1210187.
https://doi.org/10.3389/fpsyg.2023.1210187
7.
Southwell, R., Pugh, S., Perkoff, E. M., Clevenger, C., Bush, J., Lieber, R., Ward,
W., Foltz, P., & D’Mello, S.
(2022). Challenges and feasibility of automatic speech
recognition for modeling student collaborative discourse in classrooms. In Proceedings of
the 15th International Conference on Educational Data Mining (EDM 2022) (pp. 342
–
Xorijiy lingvistika va lingvodidaktika
–
Зарубежная лингвистика
и лингводидактика
–
Foreign Linguistics and Linguodidactics
Issue
–3 № 4 (2025)
/ ISSN 2181-3701
8
353).
https://educationaldatamining.org/edm2022/proceedings/2022.EDM-long-
papers.26/index.html
8.
What is ASR? (2023) How does it work? Our in-depth 2023 guide.
SimonTech. https://simontech.com/asr-guide-2023
9.
Sciforce. (2021, July 7). Automatic speech recognition (ASR) systems
compared: Ultimate guide to Google, Azure, IBM, Amazon, SpeechMatics, Kaldi, and HTK
ASR systems. Medium. https://medium.com/@sciforce/automatic-speech-recognition-
asr-systems-compared
10.
Abe, M., & Kondo, Y. (2019). Constructing a longitudinal learner corpus to
track L2 spoken English. Journal of Modern Languages, 29(1), 21
–
42.
https://doi.org/10.22452/jml.vol29no1.2
11.
Xiao, Y. (2025). The impact of AI-driven speech recognition on EFL listening
comprehension, flow experience, and anxiety: A randomized controlled trial. Humanities
and Social Sciences Communications, 12, Article 425. https://doi.org/10.1057/s41599-
025-02626-3
12.
Torres, G. (2025, May 6). What is automatic speech recognition? An
overview of ASR technology. Voice Flow. https://www.voiceflow.com/blog/automatic-
speech-recognition
13.
Omniscien Technologies. (2025). Automatic speech recognition accuracy
and
customization.
Retrieved
[July
20,
2025],
from
https://omniscien.com/lsev6/features/asr/automatic-speech-recognition-accuracy-and-
customization/
14.
Johnson, A., Chance, C., Stiemke, K., Veeramani, H., Shankar, N. B., & Alwan, A.
(2023). An analysis of large language models for African American English speaking
children’s oral language assessment. Black Excellence in Engineering, Science &
Technology Manuscripts, 1. https://doi.org/10.36227/beestm.vol1.art1
15.
Ngo, T. T.-N., Chen, H. H.-J., & Lai, K. K.-W. (2023). The effectiveness of
automatic speech recognition in ESL/EFL pronunciation: A meta-analysis. ReCALL, 35(2),
169
–
185. Cambridge University Press. https://doi.org/10.1017/S0958344023000070
16.
Arriaga, C., Pozo, A., Conde, J., & Alonso, A. (2024, September 9). Evaluation
of real‑time transcriptions using end‑to‑end ASR models [Preprint]. arXiv.
https://doi.org/10.48550/arXiv.2409.05674
