Tree-structured delta codebook for an efficient implementation of CELP

Abstract

In this paper, we present a new structured codebook for reducing the complexity and memory of a CELP stochastic codebook search. The tree-structured delta codebook, whose code vectors are generated from a small number of delta vectors, can be searched efficiently by calculating the vector correlations recursively. The complexity was reduced to 1/70 that of a conventional full-gaussian codebook, and the memory for codebook storage was reduced to 1/100.Another excellent feature of this codebook is the feasibility of changing the distribution of code vectors adaptively. The codebook adaptation method (delta vector sorting) provides an SNRseg improvement of 0.6 dB, and consistent improvement in perceptual quality, by changing the order of delta vectors according to fit the input speech.plexity. As an extension of the delta codebook, we propose a tree-structured delta codebook which not only reduces the complexity but reduces the memory requirements of CELP. We will also investigate a method for adapting the distribution of the tree-structured delta codebook based on the input speech signal for improved CELP performance.In the following section, we will first introduce the treestructured delta codebook, and discuss how it can reduce the complexity of the CELP stochastic codebook search. In section 3, we will describe the codebook adaptation method which controls the distribution of code vectors adaptively based on the input speech, using the special nature of the treestructured delta codebook. Then, in section 4, we will analyze the performance of a CELP coder which uses the tree-structured delta codebook, and the codebook adaptation method. TREE-STRUCTURED DELTA CODEBOOK 1. INTRODUCTION Codebook structureSince it was first introduced in 1984, Code Excited Linear Prediction (CELP) [ 1,2] has been intensively investigated as a promising coding algorithm to provide good quality speech at low bit rates. CELP is a name for the class of coding algorithms which employ vector quantization (VQ) using a perceptually weighted error criterion measured in an Analysis-by-Synthesis loop. Although this process gives an efficient representation of the excitation signal, and exhibits better performance than conventional coding methods, the codebook search requires a huge computational load. This was a major drawback of CELP in its practical implementation. In particular, in the area of digital cellular communication, which is considered the biggest application area for lowbit-rate speech coding, reducing the complexity of CELP is important for small hardware size and low power consumption.In the last few years, several computational reduction methods have been studied [3], and some of them, using structured stochastic codebooks, have been reported to give a good compromise between complexity and performance [4-71. We have also proposed a hexagonal lattice codebook [8] and a sparse-delta codebook [91 effective for reducing the com-The tree-structured delta codebook is a variation on the delta codebook which we proposed in [9]. In the d...