search for


치매 환자를 포함한 한국 노인 음성 데이터 딥러닝 기반 음성인식
Deep learning-based speech recognition for Korean elderly speech data including dementia patients
Korean J Appl Stat 2023;36(1):33-48
Published online February 28, 2023
© 2023 The Korean Statistical Society.

문정현*a, 강준서* a, 김기웅bcd, 배종빈 b c, 이현준e, 임창원1, a
Jeonghyeon Mun*a, Joonseo Kang* a, Kiwoong Kimbcd, Jongbin Baeb c, Hyeonjun Leee, Changwon Lim1, a

a중앙대학교 응용통계학과; b서울대학교병원 정신건강의학과; c서울대학교 정신의학과; d서울대학교 뇌인지과학과; e세븐포인트원

aDepartment of Applied Statistics, Chung-Ang University;
bDepartment of Neuropsychiatry, Seoul National University Bundang Hospital;
cDepartment of Psychiatry, Seoul National University;
dDepartment of Brain and Cognitive Sciences, Seoul National University; eSevenpointone
1Department of Applied Statistics, Chung-Ang University, 84 Heukseok-ro, Dongjak-gu, Seoul 06974, Korea. E-mail:
*These authors are co-first authors.
Received October 10, 2022; Revised December 11, 2022; Accepted December 13, 2022.
본 연구에서는 발화자가 동물이나 채소와 같은 일련의 단어를 무작위로 일 분 동안 말하는 한국어 음성데이터에 대한 자동 음성 인식(ASR) 문제를 고려하였다. 발화자의 대부분은 60세 이상의 노인이며 치매 환자를 포함하고 있다. 우리의 목표는 이러한 데이터에 대한 딥러닝 기반 자동 음성 인식 모델을 비교하고 성능이 좋은 모델을 찾는 것이다. 자동 음성 인식은 컴퓨터가 사람이 말하는 말을 자동으로 인식하여 음성을 텍스트로 변환할 수 있는 기술이다. 최근 들어 자동 음성 인식 분야에서 성능이 좋은 딥러닝 모델들이 많이 개발되어 왔다. 이러한 딥러닝 모델을 학습시키기 위한 데이터는 대부분 대화나 문장 형식으로 이루어져 있다. 게다가, 발화자들 대부분은 어휘를 정확하게 발음할 수 있어야 한다. 반면에, 우리 데이터의 발화자 대부분은 60세 이상의 노인으로 발음이 부정확한 경우가 많다. 또한, 우리 데이터는 발화자가 1분 동안 문장이 아닌 일련의 단어를 무작위로 말하는 한국어 음성 데이터이다. 따라서 이러한 일반적인 훈련 데이터를 기반으로 한 사전훈련 모델은 본 논문에서 고려하는 우리 데이터에 적합하지 않을 수 있으므로, 우리는 우리의 데이터를 사용하여 딥러닝 기반 자동 음성 인식 모델을 처음부터 훈련한다. 또한 데이터 크기가 작기 때문에 일부 데이터 증강 방법도 적용한다.
In this paper we consider automatic speech recognition (ASR) for Korean speech data in which elderly persons randomly speak a sequence of words such as animals and vegetables for one minute. Most of the speakers are over 60 years old and some of them are dementia patients. The goal is to compare deep-learning based ASR models for such data and to find models with good performance. ASR is a technology that can recognize spoken words and convert them into written text by computers. Recently, many deep-learning models with good performance have been developed for ASR. Training data for such models are mostly composed of the form of sentences. Furthermore, the speakers in the data should be able to pronounce accurately in most cases. However, in our data, most of the speakers are over the age of 60 and often have incorrect pronunciation. Also, it is Korean speech data in which speakers randomly say series of words, not sentences, for one minute. Therefore, pre-trained models based on typical training data may not be suitable for our data, and hence we train deep-learning based ASR models from scratch using our data. We also apply some data augmentation methods due to small data size.
주요어 : 한국 노인 음성 데이터, 자동 음성 인식, 딥러닝, 데이터 증강
Keywords : Korean elderly speech data, automatic speech recognition, deep-learning, data augmentation
  1. Malik M, Malik MK, Mehmood K, and Makhdoom I (2020). Automatic speech recognition: A survey, Multimedia Tools and Applications, 80, 9411-9457.
  2. Kim YG and Lee JW (2020). Development of Korean automatic speech recognition model using transformer, Proceedings of Symposium of the Korean Institute of Communications and Information Sciences, 71, 659-660.
  3. Lee SG and Kwon SI (2014). Elderly speech analysis for improving elderly speech recognition, Communications of the Korean Institute of Information Scientists and Engineers, 32, 16-20.
  4. Young V and Mihailidis A (2010). Di culties in automatic speech recognition of dysarthric speakers and implications for speech-based applications used by the elderly: A literature review, Assistive Technology, 22, 99-112.
    Pubmed CrossRef
  5. Audacity Team, Audacity® (2020). Free audio editor and recorder [Computer application] Version 2.4.1. 2020.
  6. Chi YK, Han JW, Jeong HJ et al. (2014). Development of a screening algorithm for Alzheimer’s disease using categorical verbal fluency, PLOS ONE, 9, e84111.
    Pubmed KoreaMed CrossRef
  7. Chakraborty K, Talele A, and Upadhya S (2014). Voice recognition using MFCC algorithm, International Journal of Innovative Research in Advanced Engineering (IJIRAE), 1, 2349-2163.
  8. ChanW, Jaitly N, Quoc VL, and Vinyals O (2016). Listen, attend and spell: A neural network for large vocabulary conversational speech recognition, In proceedings of 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Shanghai, 4960-4964.
  9. Dong L, Xu S, and Xu B (2018). Speech-transformer: A no-recurrence sequence-to-sequence model for speech recognition, In proceedings of 2018 IEEE International Conference on Acoustics, Speech and Signal Processing(ICASSP), Calgary, AB, 5884-5888.
  10. Han W, Zhang Z, Zhang Y et al. (2020). Contextnet: Improving convolutional neural networks for automatic speech recognition with global context,
    Pubmed CrossRef
  11. Wu Z, Liu Z, Lin J, Lin Y, and Han S (2020). Lite transformer with long-short range attention,
  12. Lu Y, Li Z, He D, Sun Z, Dong B, Qin T,Wang L, and Liu TY (2019). Understanding and improving transformer from a multi-particle dynamic system point of view,
  13. Gulati A, Qin J, Chiu CC et al. (2020). Conformer: Convolution-augmented transformer for speech recognition,
  14. Graves A and Jaitly N (2013). Towards end-to-end speech recognition with recurrent neural networks, Proceedings of the 31st International Conference on International Conference on Machine Learning PMLR, 32, 1764-1772.
  15. Kim KW, Ji YG, and Han JW (2014). inventors. Method of diagnosing dementia based on verbal fluency and apparatus therefore. South Korea patent KR101437569B1, 2014 Sep 04.
  16. Panayotov V, Chen G, Povey D, and Khudanpur S (2015). Librispeech: An ASR corpus based on public domain audio books, In proceedings of 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Brisbane, 5206-5210.
  17. Park DS, Chan W, Zhang Y, Chiu C, Zoph B, Cubuk ED, and Quoc VL (2019). Specaugment: A simple data augmentation method for automatic speech recognition,

February 2024, 37 (1)