Efficient Fixed Codebook Search Method for ACELP Speech Codecs

Abstract

There are several sub-optimal search techniques for fast algebraic codebook search of ACELP speech codecs. Focused search method, depth-first tree search method and pulse replacement methods are used to reduce computational complexity of algebraic codebook search. In previous pulse replacement methods, the computational load is increased as the pulse replacement procedure is repeated. In this paper, we propose a fast algebraic codebook search method based on iteration-free pulse replacement. The proposed method is composed of two stages. At the first stage, an initial codevector is determined by the backward filtered target vector or the pulse-position likelihood-estimate vector. At the second stage, after computing pulse contributions for every track the pulse replacement is performed to maximize the search criterion Q _k over all combination replacing the pulses of the initial codevector with the most important pulses for every track. The performance of the proposed algebraic codebook search method is measured in terms of the segmental signal to noise ratio (SNRseg) and PESQ (Perceptual Evaluation of Speech Quality) using various speech data. Experimental results show that the proposed method is very efficient in computational complexity and speech quality comparing to previous pulse replacement methods.