Research On Transcoding Techniques Of LSP And Pitch Parameters Based On GMM Between AMR-NB And G.729A

Nowadays research on transcoding techniques between AMR-NB and G.729A is popular in the field of speech coding. There are severl problems in traditional transcoding algorithms of LSP and pitch parameters based on DTE such as high complexity and high speech distortion. In terms of it, an effective solution is proposed in this paper.Firstly, in this thesis, it is detailedly contrasted for the algorithms of linear predictive analysis, searching techniques of pitch, searching methods and structure of algebraic codebook, quantization methods of gain as well as speech performance of both the codecs.Secondly, the traditional transcoding techniques of LSP and pitch parameters are analyzed. By means of the analysis of non-linear and linear interpolation methods in transcoding of LSP parameters, a calculation method of the linear interpolating parameter vector of LSP parameters based on least square method is proposed in this thesis.Finally, transcoding algorithms of LSP and pitch parameters basd on GMM are proposed and realized through the mathematical analysis of GMM in this thesis. Meanwhile, several practical issues in details are analysed under experiments. It is shown by the results of the experiments that: by employing the proposed LSP transcoding algorithm to the transcoders, the complexity is reduced by 78.9% or above, with the speech quality not worse than DTE method; by employing the proposed pitch transcoding algorithm to the transcoders, the complexity is reduced by 60.74% or above, with the speech quality approaching to DTE method.