@article{ART003266076},
author={Yonghun Jang and Jung Min Lim and Seong-Guk Nam and Minhyung Ryu and Eunjin Yoo and Myung-Sub Lee and Jong Wook Kwak},
title={MLM-based Misrecognized Word Correction for Speech Recognition},
journal={Journal of The Korea Society of Computer and Information},
issn={1598-849X},
year={2025},
volume={30},
number={11},
pages={79-89}
TY - JOUR
AU - Yonghun Jang
AU - Jung Min Lim
AU - Seong-Guk Nam
AU - Minhyung Ryu
AU - Eunjin Yoo
AU - Myung-Sub Lee
AU - Jong Wook Kwak
TI - MLM-based Misrecognized Word Correction for Speech Recognition
JO - Journal of The Korea Society of Computer and Information
PY - 2025
VL - 30
IS - 11
PB - The Korean Society Of Computer And Information
SP - 79
EP - 89
SN - 1598-849X
AB - In this study, we propose an integrated approach to improving the accuracy of Korean speech recognition by addressing phonetic similarity-induced misrecognitions. The proposed system combines three key components: (1) enhancing the signal-to-noise ratio through frequency-domain noise reduction using Minimum Mean Square Error (MMSE)-based log-spectral estimation and a high-pass emphasis filter, (2) detecting contextually inappropriate words using KoBERT-based Masked Language Modeling (MLM), and (3) selecting the final correction word using Jamo-level Levenshtein Distance, which reflects the phonetic characteristics of the Korean language.
In an experiment conducted on 1,000 Korean sentences containing misrecognized words, the proposed method reduced the Word Error Rate (WER) from 9.2% to 4.7% compared to the baseline. In addition, the proposed method achieved a maximum detection accuracy of 96.4% for misrecognized words. In conclusion, the proposed method was verified to significantly improve the performance of real-world speech recognition systems.
KW - Speech Recognition;Speech to Text;Error Correction;Language Model;Korean NLP
DO -
UR -
ER -
Yonghun Jang, Jung Min Lim, Seong-Guk Nam, Minhyung Ryu, Eunjin Yoo, Myung-Sub Lee and Jong Wook Kwak. (2025). MLM-based Misrecognized Word Correction for Speech Recognition. Journal of The Korea Society of Computer and Information, 30(11), 79-89.
Yonghun Jang, Jung Min Lim, Seong-Guk Nam, Minhyung Ryu, Eunjin Yoo, Myung-Sub Lee and Jong Wook Kwak. 2025, "MLM-based Misrecognized Word Correction for Speech Recognition", Journal of The Korea Society of Computer and Information, vol.30, no.11 pp.79-89.
Yonghun Jang, Jung Min Lim, Seong-Guk Nam, Minhyung Ryu, Eunjin Yoo, Myung-Sub Lee, Jong Wook Kwak "MLM-based Misrecognized Word Correction for Speech Recognition" Journal of The Korea Society of Computer and Information 30.11 pp.79-89 (2025) : 79.
Yonghun Jang, Jung Min Lim, Seong-Guk Nam, Minhyung Ryu, Eunjin Yoo, Myung-Sub Lee, Jong Wook Kwak. MLM-based Misrecognized Word Correction for Speech Recognition. 2025; 30(11), 79-89.
Yonghun Jang, Jung Min Lim, Seong-Guk Nam, Minhyung Ryu, Eunjin Yoo, Myung-Sub Lee and Jong Wook Kwak. "MLM-based Misrecognized Word Correction for Speech Recognition" Journal of The Korea Society of Computer and Information 30, no.11 (2025) : 79-89.
Yonghun Jang; Jung Min Lim; Seong-Guk Nam; Minhyung Ryu; Eunjin Yoo; Myung-Sub Lee; Jong Wook Kwak. MLM-based Misrecognized Word Correction for Speech Recognition. Journal of The Korea Society of Computer and Information, 30(11), 79-89.
Yonghun Jang; Jung Min Lim; Seong-Guk Nam; Minhyung Ryu; Eunjin Yoo; Myung-Sub Lee; Jong Wook Kwak. MLM-based Misrecognized Word Correction for Speech Recognition. Journal of The Korea Society of Computer and Information. 2025; 30(11) 79-89.
Yonghun Jang, Jung Min Lim, Seong-Guk Nam, Minhyung Ryu, Eunjin Yoo, Myung-Sub Lee, Jong Wook Kwak. MLM-based Misrecognized Word Correction for Speech Recognition. 2025; 30(11), 79-89.
Yonghun Jang, Jung Min Lim, Seong-Guk Nam, Minhyung Ryu, Eunjin Yoo, Myung-Sub Lee and Jong Wook Kwak. "MLM-based Misrecognized Word Correction for Speech Recognition" Journal of The Korea Society of Computer and Information 30, no.11 (2025) : 79-89.
