Research On Algebraic Codebook Transcoding Between AMR-NB And G.729AB

Research on parameter transcoding algorithms between AMR-NB and G.729ab has been a focused topic in speech coding field recently. Transcoding techniques of codebook are related with speech quality and computational complexity. In term of it, some transcoding techniques of algebraic codebook between AMR-NB and G.729AB are proposed in this thesis.Firstly, the codebook structures, searching algorithms, bits allocation,encoding methods in both of the codecs are compared and analyzed in detail. According to different transcoding rates, various transcoding techniques based on parameter domain, excitation domain or speech domain are adopted to satisfy both the requirement of improving speech quality and reducing computational complexity.Secondly, after analyzing the distribution of codebook pulse of AMR-NB at 12.2kb/s and G.729ab as well as observing the matching rate of codebooks between voice frames and unvoice frames, random number mapping and permutation & combination mapping codebook parameter transcoding algorithms are presented in this thesis. It is shown by experiment results that both algorithms are able to reduce transcoding computational complexity by 98%, with acceptable speech quanlity.Thirdly, the algebraic codebook transcoding based on pulse position reconstruction is proposed in this thesis through the research on depth-first tree algorithms of codebook. It is shown by results that, comparing with the method in which the search range is not shorten, the proposed algorithm can reduce the codebook search computational complexity by 30% or above, as well as guarantee the transcoded speech quality.Finally, by using the probability statistics methods and taking advantage of the effectiveness of description of stochastic processes of HMM, a codebook parameter transcoding algorithm based on HMM is proposed and realized in this thesis. Comparing with the traditional speech domain codebook transcoding, this algorithm can reduce computational complexity by 34.7% or above, with the guarantee of transcoded speech performance.