An Extended Speech De-noising Method Using GGM-Based ICA Feature Extraction
ICA (independent component analysis) feature extraction is an efficient sparse coding method for noise reduction. In many ICA-based de-noising processing, however, they need noise-free source data to train the basis vectors as a priori knowledge. The noise-free data is always not acquirable in practice. In this paper, the generalized Gaussian model (GGM) is proposed as the p.d.f. estimator in ICA to extract the basis vectors directly from the noisy observation, since GGM can easily characterize a wide class of non-Gaussian statistical distributions. Simultaneously, the distribution of the coefficients learned by GGM is benefit for obtaining the shrinkage functions. The de-nosing experiments of noisy speech signals show that the proposed method is more efficient than conventional methods in the environment of additive white Gaussian noise. It demonstrates that the proposed method offer an efficient approach for detecting weak signals from the noise environment.
Unable to display preview. Download preview PDF.
- 1.Hyvärinen, A.: Sparse code shrinkage: Denoising of nongaussian data by maximum likelihood estimation. Technical Report A51, Helsinki University of Technology, Laboratory of Computer and Information Science (1998)Google Scholar
- 2.Lee, T.-W., Jang, G.-J.: The Statistical Structures of Male and Female Speech Signals. In: Proc. ICASSP (Salt Lack City, Utah) (May 2001)Google Scholar
- 3.Lee, J.-H., Jung, H.-Y.: Speech Feature Extraction Using Independent Component Analysis. In: Proc. ICASP, Istanbul, Turkey, June 2000, vol. 3, pp. 1631–1634 (2000)Google Scholar
- 7.Lee, T.-W., Lewicki, M.S.: The Generalized Gaussian Mixture Model Using ICA. In: international workshop on Independent Component Analysis (ICA 2000), Helsinki, Finland, June 2000, pp. 239–244 (2000)Google Scholar