Development of an Open-Source–Based AI Speech-to-Text System and Performance Analysis Using Presidential Speech (오픈소스 기반의 AI 음성·텍스트 변환 기능 개발 및 대통령 음성을 통한 성능 분석)

Minsoo Bae (배민수); Young-Moon Yu (유영문)

doi:10.14404/JKSARM.2025.25.3.243

Development of an Open-Source–Based AI Speech-to-Text System and Performance Analysis Using Presidential Speech

Journal of Korean Society of Archives and Records Management
Abbr : JRMASK
2025, 25(3), pp.243~258
DOI : 10.14404/JKSARM.2025.25.3.243
Publisher : Korean Society of Archives and Records Management
Research Area : Interdisciplinary Studies > Library and Information Science > Archival Studies / Conservation
Received : July 15, 2025
Accepted : August 22, 2025
Published : August 31, 2025

Minsoo Bae ¹, Young-Moon Yu ²

¹대통령기록관 공업연구사
²대통령기록관 공업연구관

Accredited

ABSTRACT

This study developed an open-source–based AI Speech-to-Text (STT) system and analyzed its performance by applying it to presidential speech. While various high-performance STT services are currently commercialized, most are provided online for a fee. However, because of the nature of presidential records, using online services can raise security concerns, and incurring continuous costs for processing accumulating records is inefficient. To address this, the Presidential Archives has developed an offline STT system based on open-source AI models, which is currently under testing and operation. In this study, approximately three hours of presidential audiovisual records were transcribed into text using this function, and the error rate was measured by comparing with the actual text. The results showed that the overall performance is comparable to the latest commercial online services. Additionally, speech rate and recording quality were extracted and analyzed for their correlation with the error rate. Finally, this research highlights the feasibility of applying open-source AI technologies for the utilization of records.

KEYWORDS

Artificial intelligence, Speech-to-text, Open-source software, Presidential Archives, audiovisual records

Journal Copyright Policy

This license enables reusers to copy and distribute the material in any medium or format in unadapted form only, for noncommercial purposes only, and only if attribution is given to the creator.

Citation status

* References for papers published after 2025 are currently being built.

[report] 권영환 / 2021 / 오픈소스 활성화를 위한 오픈소스 연구개발 생태계 연구(RE-102) / 소프트웨어정책연구소

[report] 김현숙 / 2019 / 디지털기반 대통령기록관리체계 모델 재설계 : 1~17

[journal] 나미선 / 2016 / 대통령기록관 신청사 보존·복원 인프라 구축 / 기록인 34 : 22~31

[journal] 남태우 / 2007 / A Study of President Records Management Law / 한국기록관리학회지 7(2) : 165~188

[journal] 안대진 / 2017 / 지능형 기록정보서비스 방안 / 기록인 41 : 38~45

[report] 이해수 / 2025 / 2024년 국내외 인공지능 산업 동향 연구 / 소프트웨어정책연구소

[web] FFmpeg / FFmpeg Website / https://ffmpeg.org

[web] GitHub / GitHub Website / https://github.com

[web] Hugging Face / Hugging Face Website / https://huggingface.co

[journal] Malik, M. / 2021 / Automatic speech recognition: a survey / Multimedia Tools and Applications 80(6) : 9411~9457

[web] McFee B. / Librosa / https://librosa.org/doc/latest/index.html

[confproc] Mikolov T. / 2010 / Recurrent neural network based language model / Interspeech : 1045~1048

[web] OpenNMT / CTranslate2 / OpenNMT / https://opennmt.net/CTranslate2

[journal] Rabiner, L. / 1989 / A tutorial on hidden Markov models and selected applications in speech recognition / Proceedings of the IEEE 77(2) : 257~286

[confproc] Radford, A. / 2023 / Robust speech recognition via large-scale weak supervision / International Conference on Machine Learning 202 : 28492~28518

[confproc] Vaswani A. / 2017 / Attention is all you need / Advances in Neural Information Processing Systems 30 : 6000~6010

KJCKorea
Journal Central

Journal of Korean Society of Archives and Records Management 2025 KCI Impact Factor : 1.23