Research And Implementation Of Speech Enhancement Algorithm

In general, speech is often corrupted acoustically by ambient noise. The result is the degradation of the performance of digital voice processor, such as speech coder, speech recognition. So a system for speech enhancement is needed strongly to solve the problem. The objective of speech enhancement may be to improve the quality, to increase the intelligibility, to ensure the reliability of digital voice system. Therefore, the study on speech enhancement technology has important value in reality.A complete enhancement system has three parts: a noise estimator which is used to trace noise estimation, a noise reduction algorithm which is the crucial part in the system and a post-processor which is necessary if specific design is required.Generally speaking, a noise estimation algorithm is a noise variance estimator. A noise estimation based on minimum statistics is deeply studied in this thesis. The speech enhancement system of this thesis will adopt the noise estimator based on minimum statistics.This paper focuses on speech enhancement algorithms in the short-term spectral estimation methods. To compare the spectral subtraction, improved spectral subtraction and minimum mean square error methods, the Optimally-Modified Log-Spectral Amplitude estimator (OM-LSA) method is the best one of the three methods after making a comprehensive survey of increasing SNR and auditory perception. The OM-LSA is chosen in the speech enhancement system of this thesis.In this paper, the authors study the speech enhancement algorithm based on Minimum Mean Square Error short time Log-Spectral Amplitude estimation (MMSE-LSA) under the single channel condition, and simplify the algorithm in order to improve the real-time processing .We introduce an Optimally-Modified Log-Spectral Amplitude estimator, which minimizes the mean-square error of the log-spectra for speech signals under signal presence uncertainty. Experiments show that the algorithm enhances the speech very well, especially in the conditions of low SNR. The algorithm is implemented on DSP TMS320VC33, and can efficiently decrease background noise.