Research Of CS-ACELP Coding Algorithm And Implementation On DSP

CS-ACELP speech codec algorithm of G.729 standard was developed by ITU-T(International Telecommunication Union)in 1996.It is a relatively new voice codec algorithm with low latency, good sound quality advantages.In the rate of 8kb/s,it obtains on the equivalent of 32kb/s of the ADPCM (differential pulse code modulation)voice quality, and it has broad application prospects.Although the CS-ACELP algorithm with a variety of advantages, but there is still a high computational complexity, proceedings of the algorithm are analyzed in detail in this paper, each module in the complexity of the algorithm is assessed, improvement methods are given to the parts of open loop pitch analysis, and searching of codebooks. Without affecting the pitch periods, this paper reduces the calculation of the calculation of autocorrelation function. In the search of codebooks, local optimum search ideas are used, a lower computation is achieved by a slight loss of quality.The main work of this paper is as follows:(1) Firstly, the status, background, significance of the research of speech codec algorithm are introduced. It analyzes the speech codec technology development trends and confronts problems. Speech coding theories, techniques commonly used in speech treatment and coding quality evaluation systems are described in the paper.(2) Focuses on the CS-ACELP coding algorithm's basic working principle and proceedings, this paper does description in detail.On the linear prediction analysis and quantization, pitch analysis, adaptive codebook search, fixed codebook search modules,this paper makes a detailed analysis showing that these areas with high computational complexity.(3) In the proceeding of analysis of the pitch, involved large number of autocorrelation calculations, this paper uses three-level clipping combined with the modified method of the autocorrelation function to improve it. Compared to the method of autocorrelation function and the method of the center clipping combined with modified autocorrelation function,It find that this method does not affect the value of pitch delay, but it reduces the amount of computation in the proceeding of analysis of the pitch.(4) Relative higher complexity of the adaptive codebook search, fixed codebook search part, this paper uses the idea of local optimal to improve them. By the characteristics of the adaptive code book, this paper changes the step length of search, it obtains a local optimal search results, compared to the method of one step recursion and the standard CS-ACELP algorithm,it found that computation has been reduced. In the part of the fixed codebook search, this paper reduces the size of the codebook by using the value of judgement formula, based on this, this paper introduces an grouping search method to contrast with the above method, choose the best one as the final search results. with the comparison to the standard search algorithm find that computation has been improved. Improved algorithm is simulated found that the improved algorithm by a slight loss of voice quality, to achieve the improvement of the complexity.(5) In the TMS320VC5416 DSP processors the CS-ACELP speech coding algorithm is put into practise.In DSP-specific integrated development environment-CCS,coding the program by the languages of assembler and C,debugging and running, the experimental results well,shows that the improved algorithm with a certain degree of feasibility.