G.729协议是ITU-T制定的一个能在8kb/s速率上实现高质量语音编码的建议，因其高质量，低码率而得到广泛应用。在IP电话网关中，G.729协议被用来实现实时语音编码处理。 G.729算法实现的重点在于编解码算法的实现以及数据的采集、存储。国际电信联盟ITU在推出G.729编解码器算法建议的同时，给出了G.729短发的一种软件参考实现。该软件采用ANSI C语言编写的，可以在许多硬件平台上编译、运行，但是，该软件没有针对任何平台的任何优化。 本文简要介绍了G.729协议中用到的一些基本技术，在详述阐述G.729编解码器原理的基础上，通过已有的对G.729编解码器的性能的测试，基于这些测试统计结果，得出结论：G.729编解码器满足了在大多数使用情况下提供长途质量的基本要求。 针对G.729编解码算法的优化工作，本文首先对ITU—T提供的CS—ACELP编码器的16比特定点标准C语言软件仿真的解码端进行分析。得出结论：程序实现了G.729协议的主体功能，但是没有给出出错（坏帧）处理的部分。 为了解决优化问题，本文采用TMS320C54x指令改写C语言程序的方法，首先是对存储区进行优化，在此基础上通过代码优化器对代码进行优化，然后利用TMS320C54x指令把C语言程序改写为汇编程序。 在优化实验中，采用了675帧的语言信息作为测试数据，对其进行编、解码时间测试，进行10次，最后将得到的数据作一个统计平均，使得系统影响在时间上趋于一致。由测试的结果来看，编解码的运算时间小于15ms的时延，证明该优化方案已经可以实时应用。

G.729 agreement is made by ITU-T to realize speech code with good quality at the rate of 8kb/s. and used widely because of its good quality and low coode rate. In VoIP gateway, G.729 agreement is used to realize instant speech code process. The realized emphasis of G.729 arithmatic lies in realization of coder and decoder and collection and storage of data.ITU put forward the agreement as well as its software reference application written with ANSI.C,the software can be translated and run on many hardware platform.However,the software is not optimized to any platform. The article briefly introduces some basic techniques of G.729 and states the theory of coder and decoder in details.Based on the result of testing the capability of coder and decoder, it concludes that G.729 meets the basic need of long-distance quality in most cases. The article analyses emulated C language software of decoder that is CS-ACELP with 16bit fixed point, then come to conclusion that the software realizes the main function of G.729 agreement, but it fails to provide concealment of frame erasures. To solve the problem of optimizing, the article uses the instruction of TMS320C54x to rewrite C language process. Firstly, storage section is optimized, then code by optimizing equipment at last C language process id rewritten in compilation process with the instruction of TMS320C54x. In the optimizing test, 675 frames speech information is used as test data to test its coded and decoded time. The test is dome 10 times and the data. From the test is averaged in order to make the system affect speech information consistently. From the result of the test, it resches the conclusion that postponement time between coder and decoder is less than 15ms, which proves that the optimizing method can be used in reality.