A Novel Algorithm of Vector Quantization Codebook Design Using Fuzzy Clustering Theory

The design of the codebook is of great importance in the vector quantization.This paper,pointing to the drawbacks of the classic algorithm LBG,proposes a novel efficient algorithm of designing codebook based on the fuzzy clustering theory.Through initialization of distance threshold and the following adaptation of reducing threshold for each level,the iteration time is reduced and the convergent speed is then improved.At the same time,the adapta-tion of reducing threshold for each level can induce the generation of new clusters,thus resulting in more typical codebook vector.At last,through optimization selection of vectors in the decreasing order,the designed codebook will achieve better performance and be closer to global optimization.The novel algorithm is applied to LSF vector quantization codebook generation in mobile speech/audio coding,and the experiment result shows that the algorithm can improve the objective and subjective performance of the audio codec.