Speech Enhancement Based On Dual Wiener Filter

In real life,people's learning,work and entertainments are based on the transmission of speech.However,in the process of transmission,the clean speech will be affected by the external noise inevitably so that the speech signal we received has been degraded in the quality and intelligibility.And the received signal seriously affects people's work and quality of life.In order to improve the quality and intelligibility of the received signal which is conformable to the subjective feeling of the ear,it is necessary to deal with the degraded signal.The aim of speech enhancement is to deal with noisy speech signal so that the background noise is eliminated as much as possible and the original speech is restored.At present,according to the ways of speech processing,speech enhancement algorithms can be divided into two categories as time domain and transform domain algorithms.The clean speech and noise energy are more concentrated and the characteristic is easier to deal with in the transform domain.Therefore,researchers who are at home and abroad have been concentrated on the algorithm in the transform domain.In this paper,the main work focuses on the speech enhancement technology based on Wiener filtering in the discrete cosine transform(DCT)domain.By analyzing the deficiencies of the gain function of the existing speech enhancement algorithm,a new speech enhancement algorithm is proposed by using the dual Wiener filtering technique based on the statistical model,and the performance of the proposed algorithm is verified from the theory and experiment.In this paper,the innovative research is as follows: Firstly,the problem of the gain function in the transform domain is studied and the actual speech and noise data are used to analyze the clean speech and noise speech coefficients in the DCT domain.And the shortcomings of the traditional Wiener filtering algorithms are found.Secondly,the dual Wiener filtering is used to improve the original algorithm according to the previous experimental analysis results.At the same time,the speech and noise signals are modeled by Gaussian and Laplacian models which commonly used in DCT domain.The dual Wiener filtering algorithms are designed based on three statistical models.Finally,the validity and superiority of the proposed algorithm are verified through analyzing theory and experiment.Summarizing and prospecting the future research and development of speech enhancement technology.