Research On G.729 Speech Coding Algorithm And DSP Implemention

In recent years, with the rapid development of communication technology, the compression of speech coding technology has made breakthrough progress,also there emerges a lot of low bit-rate speech coding algorithm with good performance. The G.729 recommendation is based on conjunction structural Algebra code exciting linear prediction(CS-ACELP), G.729 has a low encoding rate and has a good synthesized speech effect, the high algorithm complexity results in the difficulty of real-time implement. However, because of the high complexity of the algorithm, the large data storage capacity and the excessive demands on hardware equipment, real time application has become a problem. With the rapidly development of(Digital Signal Processor, DSP) technology in recent decades, it has become possible to realize the more complex codec algorithm in a single chip.In the base of the research on the algorithm of G.729,this thesis propose realtime implementation of G.729 speech codec system on TMS320VC5402.According to the content of the research, this paper can be divided into two parts.The first part is the algorithm research. This paper briefly introduces the international standards and the performance metrics of the codec, then choose G.729 algorithm to implement the coding and the decoding, and next explains the mathematical model of speech production,on the basic of the speech processing theory, then give the detailed analysis on the principle of G.729 algorithm.The second part concentrates on the algorithm implementation on the DSP and the algorithm optimization.First introduce the hardware platform of the system which is based on the Texas Instruments of TMS320VC5402.Analog signal is transformed into digital signal by an audio A/D converter. The data streams are encoded in the TMS320VC5402.Then though the interface device, the data streams are transmitted to the receiver. Finally the bit streams will be decoded and converted into synthesized speech signal.Then introduce the software system, focus on the encoding and decoding, Mc BSP configuration and the serial interrupt.During the optimization algorithm,this paper propose the algorithm optimization scheme, the algorithm is optimized on the compiler level, C language level, assembler level and algorithmic level,and give the exanmlpes.After the algorithm migration on the DSP, we have the test.The test results show that the optimized algorithm c can be implemented on DSP real-time. The MOS score of the synthetic speech can achieve basic requirements of normal voice communication which has a high quality and gets good performance.