잡음 섞인 한국어 인식을 위한 ICA 비교 연구

Comparison of ICA Methods for the Recognition of Corrupted Korean Speech

  • 김선일 (거제대학 조선정보기술계열)
  • Kim, Seon-Il (Department of Information Technology for Shipbuilding, Koje College)
  • 발행 : 2008.09.25

초록

두 가지 Independent Component Analysis(ICA) 알고리즘을 적용하여 자동차 엔진 소음과 섞인 음성 신호의 인식을 시도하였다. 이를 이용하여 추정한 신호를 HMM을 이용하여 인식하였고 이 신호의 인식률을 소음이 섞이기 전의 음성 신호의 인식률과 비교하였다. 음성 신호를 추정하는데 두 가지 서로 다른 ICA를 사용하였으며 그 중의 하나는 negentropy를 최대화하는 FastICA 알고리즘이며 다른 하나는 출력 신호 사이의 독립성을 최대화하여서 입력과 출력 사이의 mutual information을 최대화하는 information-maximization approach 이다. 남성 앵커가 진행한 한국어 뉴스 문장에 대한 단어 인식률은 87.85%이며 다양한 신호 대 잡음비를 갖도록 소음을 섞어서 추정을 한 후 인식을 시도한 결과 FastICA를 이용해 추정한 음성 신호에 대한 인식률은 1.65%, information-maximization을 이용해 추정한 음성 신호에 대한 인식률은 2.02% 인식률 저하가 나타났다. 따라서 어느 방법을 적용하든지 의미 있는 차이가 없음을 확인하였다.

Two independent component analysis(ICA) algorithms were applied for the recognition of speech signals corrupted by a car engine noise. Speech recognition was performed by hidden markov model(HMM) for the estimated signals and recognition rates were compared with those of orginal speech signals which are not corrupted. Two different ICA methods were applied for the estimation of speech signals, one of which is FastICA algorithm that maximizes negentropy, the other is information-maximization approach that maximizes the mutual information between inputs and outputs to give maximum independence among outputs. Word recognition rate for the Korean news sentences spoken by a male anchor is 87.85%, while there is 1.65% drop of performance on the average for the estimated speech signals by FastICA and 2.02% by information-maximization for the various signal to noise ratio(SNR). There is little difference between the methods.

키워드

참고문헌

  1. J. P. LdBlanc and P. L. De Leon, "Speech Separation by Kurtosis Maximization," Proc. ICASSP, vol. 2, pp. 1029-1032, 1998
  2. A. Hyvarinen, J. Karhunen, and E. Oja, Independent Component Analysis, John Wiley and Sons, 2000
  3. A. J. Bell land Terrence J. Sejnowski, "An information-maximisation approach to blind separation and blind deconvolution," Neural Computation, vol. 7, no. 6, pp. 1129-1159, 1995 https://doi.org/10.1162/neco.1995.7.6.1129
  4. P. Comon, "Independent component analys, A new concept?," Signal Processing, vol. 36, pp. 287-314, 1994 https://doi.org/10.1016/0165-1684(94)90029-9
  5. S. Amari and A. Cichocki, "A New Learning Algorithm for Blind Signal Separation," Advances in Neural Information Processing System, vol. 8, pp. 757-763, MIT Press, 1996
  6. A. Hyvarinen, "Fast and Robust Fixed-Point Algorithms for Independent Component Analysis," IEEE Trans. On Neural Networks, vol. 10, no. 3, May, 1999
  7. J. F. Cardoso, "Blind signal separation: statistical principles," Prod. IEEE, vol. 9, no. 10, pp. 2009-2025, Oct., 1998 https://doi.org/10.1109/5.720250
  8. E. Bisser, T. W. Lee and M. Otsuka, "Speech Enhancement in a Noisy Car Environment," Proc. 3rd International Conference on Independent Component Analysis and Source Separation. pp. 272-277, 2001
  9. J. F. Cardoso, "Learning in manifolds: the case of source separation," Proc. IEEE SSAP '98, Portland, Oregon
  10. T. M. Cover, and J. A. Thomas, Elements of information theory, New York: Wiley
  11. A. Hyvarinen, and E. Oja, "Independent component analysis: algorithms and applications," Neural Networks, vol. 13, no. 4/5, pp. 411-430, 2000 https://doi.org/10.1016/S0893-6080(00)00026-5