In this paper, we proposed method of Compression perception theory for Parameters quantify base on MELP model. In order to reduce the speed of speech compression encoding, The super frame used in LSP parameters for the coding part. In order to vector quantization for the super frame, compressed perception theory is used of the super frame to the theory of multidimensional vector for dimension reduction LSP. If the multidimensional LSP parameter are mapped to low dimensional space, we will get a low dimensional quantitative space. The further more, We can give a division of vector quantization .In the decoding part, We can reconstruct the original high dimensional LSP of vector through using the orthogonal matching algorithm. Thus, We can synthesis the speech signal to the original .The experiment shows that the algorithm improves the synthesized speech intelligibility and naturalness, reduce the coding required number of bits and yards of the storage complexity. Low rate of speech coding research has the vital significance. http://www.sjie.org/paperInfo.aspx?ID=177
