(1-2. School of Electronic and Information Engineering, South China University of Technology, Guangzhou, 510641)
Abstract: To effectively extract inherent information from measured speech signals, it is important to preprocess data to reduce noise. In this letter, we propose an algorithm---Bayesian multi-solution shrinker (BMS) for speech enhancement. The basic idea of BMS is to utilize empirical Bayesian model in the wavelet coefficients shrinkage step. Using speech data and calculating signal-to-noise ratio (SNR) and segmental signal-to-noise ratio (SSNR), we show that shown that the proposed approach outperforms the benchmark methods based on log-spectral amplitude (LSA), spectral subtraction and Stein's Unbiased Risk Estimate (SURE) wavelet denoising, respectively.
Keywords: Speech enhancement; denoise; Bayesian multi-solution shrinker.
REFERENCE
(1)Boll S.F. Suppression of acoustic noise in speech using spectral subtraction. IEEE Trans. Acoustics Speech Signal Process., 1979, 27: 113-120.
(2)I. Cohen, B. Berdugo. Speech enhancement for non-stationary noise environments. Signal Processing, 2001, 81: 2403-2418.
(3)Donoho D.L., Johnstone I.M. Ideal spatial adaptation by wavelet shrinkage. Biometrika, 1994, 81: 425–455.
(4)Donoho D.L., Johnstone I.M. Ideal Spatial Adaptation by Wavelet Shrinkage. Department of Statistics, Stanford University, USA. April 1993.
(5)Vidakovic B., Ruggeri F. BAMS method: Theory and simulations. The Indian Journal of Statistics, Series B, 2001, 63(2): 234-249.
(6)Vidakovic B. Statistical Modeling by Wavelets. Wiley, New York, 1999.
(7)Yang R., Berger J. A catalog of noninformative priors. Discussion Paper, 97042, ISDS, Duke University, NC.
(8)Garofolo J., Lamel L., Fisher W., Fiscus J., Pallett D., Dahlgren N., et al. TIMIT Acoustic–Phonetic Continuous Speech Corpus: Linguistic Data Consortium, 1993.
(9)Johnson M.T., Yuan X., Ren Y.. Speech signal enhancement through adaptive wavelet thresholding. Speech Communication, 2007, 49: 123-133.
(10)The MathWorks Inc. Wavelet Toolbox. Matlab: The Language of the Technical Computing. Version 7.14, Release 2012a, 2012.
作者简介:
谢宗伯,博士,华南理工大学副研究员,IEEE/IEICE会员,主要研究方向为信号处理与机器学习。先后主持国家和省部级项目多项,在国际高水平期刊发表论文多篇。
冯久超,教授,博导,华南理工大学教授,IEEE会员,广东省“珠江学者”特聘教授,主要研究方向为非线性系统理论。