The Research Of Fixed Codebook Search Method In Wide-Band Speech Coding

The development of times promotes the communication mode diversification and high quality. The traditional phone signal is narrow band, sometimes it may limit the intact expression of the speech. So, people brought up wide-band speech codec, for the expanded part of low frequency can make the speech heard become more natural and let the listeners feel more comfortable, expanding part of high frequency can raise understood degree of the speech. making it convenient to distinguish the person who speaks, and realize "recognizing persons when listening the sound"But the systems which support wide-band speech always comparatively complicated, also increase the difficulty of application. In the wide-band speech coding algorithm complexity influence factors, the fixed codebook searching computation occupies a large proportion. Therefore, we need to study and improve the fixed codebook searching method, to reduce the whole complexity of the wide-band speech coding.In this paper, we introduced some basic knowledge about wide-band speech codec, knew some common traditional encoder model, then take wide-band encoder G.722.2as the target of studying to launch the research work, After researching the fixed codebook structure and its searching method, we combined with the pulse replacement method and the codebook section search method and so on, then put forward the following improved fixed codebook search methods:First of all, we put forward the first tree pulse replacement method. It used the good quality characteristics of pulse replacement method and the fast searching speed characteristics of the depth-first tree search method. The experimental results show that, though the speech qualities have a little drops, but complexity and the whole encode running time all decreased considerably.The second, we improved the original codebook subsection search method. It used the reference signal to determine the initial codebook and the way vector cross section. The experimental results show that, after the improvement, the running time become more shorter, but the speech quality drops comparatively heavy and the experimental results of male and female have a great disparity, so, we should improve it further more.At last, we proposed sub-position sub-codevector replacement method, used the idea of basic sub-position into IFCDS, mainly applying Sub-Position Replacement Method into subcodevector innovation, in order to reach the target of reducing running time with appropriate speech quality. The experiment result shows that this method has relatively the best effects. The synthesized speech has similar quality with the depth first tree pulse replacement method synthesized speech quality, but the coding time is much shorter than the depth first tree pulse replacement method.