| With the development of multimedia technologies, the request for the speech code is more and more high, not only the low code rate, the low time delay, but also the very high quality of the synthesis pronunciation. ITU-T G.729 is based on the Conjugate-Structure Algebraic Code Excited Linear Prediction (CS-ACELP), which has 8kb/s rate, the low the delay and high quality of the synthesis pronunciation, So research and implement the CS-ACELP is of great significance.This paper introduces the signal model and the basic principle of linear prediction firstly, then the analysis model of the linear prediction code and the Levinson-Durbin algorithm which is one of the most commonly used highly effective recursion algorithm of the speech signal are established, based on the vector quantification, consciousness weighting filter, the Analysis By Synthesis, the Code Excited Linear Prediction coding (CELP), as well as the search process of auto-adapted code and fixed code, the detailed principle of Conjugate-Structure Algebraic Code Excited Linear Prediction (CS-ACELP) code are analysised, Finally, a implement scheme founded on the hardware platform composed of advanced TMS320VC5510 DSP is presented, the high performance characteristics of TMS320VC5510, the periphery connection, the initialization method as well as the development of driver of AIC23 are analysised, the test result indicates that the design project can satisfy the request of 8kb/s CS-ACELP. |
