Speech processing systems are inevitably interfered by various noise. The noise not only degrades the quality and the intelligibility of the processing systems, seriously the system couldn抰 work well . To minimizing the effects of the noise on the performance of the processing systems, speech enhancement technology is applied in the various speech processing systems. Consequently the study of speech enhancement technology is very significant.in this thesis, we study some speech enhancement methods based on short-time spectral magnitude of noisy speech, such as spectral magnitude subtraction~ power spectrum subtraction~ Wiener filtering-, maximum likelihood envelope estimation. We study the fundamental and the implementation of these methods and their improved forms, and their performance under the wide SNR(signal-to-noise ratio) range. Meanwhile we reveal the differences and the relations among these speech enhancement methods.Under this condition that the noise (white or color) is additive and stationary, the experiment results indicates that the methods can greatly improve the quality and intelligibility of noisy speech, and have other advantages such as the widely applicable SNR range, less computation load. Our works in this thesis can be the basis of further development in the practical speech enhancement systems. |