An economical binary tree structure for vector quantization

The authors present a new way to organize the codebook for vector quantization of speech. The new structure can reduce the storage requirement almost by half without any increase in the computational cost. The key is to store only the final code vectors in a tree structure. To facilitate the search at each node, one stores the indices that point to the code vectors that lie nearest to the other branches instead of the centroids of branches. The performance of the new method is compared with that of the conventional method. The new method offers a better compromise for applications that have limited computational resources and storage room and have less stringent distortion rate requirements