Study And Implementation On Speech Enhancement Based On Subband

Speech is an important way for information exchange nowadays. With the development of communication and multimedia techniques, speech signals with higher quality in communication are required. The existence of background noise can influence the effects of listening and turn down the performance of many speech processing systems. As a preprocessing scheme of the speech processing system, speech enhancement not only gains pure speech from noisy speech signals and improve the quality of speech, but also offers comfortable listening. As a result, speech enhancement has been applied more and more broadly and become indispensable in speech communication.Recently, the theory of subband is mature and applied to some novel speech enhancement methods, but the research of related techniques in single channel systems is very limited. In this paper, several single channel speech enhancement methods based on subband are studied and implemented. The contents of this thesis are discussed as follows:First, the basic theory of speech processing is summarized. Besides, traditional speech enhancement methods and the development of speech enhancement based on subband are briefly reviewed.Second, traditional speech enhancement methods based on short-time spectrum estimation are simulated and the delay of existing algorithms is analyzed.Third, the theory of cosine modulated filter banks is carefully studied. A method for the design of the lowest-delay cosine modulated perfect reconstruction is deducted. Moreover, the feasibility of single channel system with subband processing which can obtain lower algorithm delay is discussed.Fourth, by combining filter banks theory and the speech enhancement method based on short time spectrum estimation, two speech enhancement methods with lower algorithm delay are proposed. Simulation results show that the enhancement effects are satisfying.Finally, fixed point program processing of the proposed algorithm is implemented and the codes are transplanted to Samsung S3C2410A microcontroller.