A new hybrid vector quantizer of LSF parameters

This paper proposes a new type of hybrid vector quantizer (VQ) of line spectrum frequencies (LSF) parameters of a speech signal. First, the LSF vector is split into two or three subvectors. Then, a tree-search procedure is performed until a terminal node is reached. Finally, a full-search is performed among the code vectors associated with the terminal node in order to find the result of the encoding process. The method finds the nearest neighbor of each input subvector. This new hybrid VQ has a much lower complexity than current VQs like split VQ or switched split VQ. In the design process of this VQ, the Voronoi cells which intersect with the rectangular cells must be determined. We propose a new method to solve this issue requiring less encoding operations compared to previous work.