The two-step search strategy in the vector codebooks for speech compression

Abstract

There has been developed the approach, which combines the advantages of the fast nearest neighbor search in the codebooks using weighted Euclidean metric. The proposed two-step search strategy is that: at first on the basis of the binary tree a few of codebook vectors (candidates) are chosen, using Euclidean metrics; among which then really the nearest vector is chosen. It is deduced the analytical dependence of the candidates and the number of operations needed to find the nearest vector. The given results of experiments confirm the performance of the proposed approach.