Research On CS-ACELP Speech Codec Algorithm And Implementation On DSP

With the rapid development of modern mobile communication, the bandwidth utilization and speech quality have been the crucial problems. The valid approach to solve this problem is low rate, low cost and high quality speech codec. G.729 protocol was enacted by ITU(International Telecom United) to meet these requirements in 1996. G.729, which is based on Conjugate-Structure Algebraic-Code-Excited Linear-Prediction (CS-ACELP) at the rate of 8kbit/s, has the advantage of low delay, high reconstructed speech quality. Owing to its high quality, G.729 protocol has been widely used in many fields, such as satellite communication, multimedia communication, mobile communication and IP phone nowadays.In this thesis, firstly, the background knowledge of low-rate speech coding is introduced, and especially G.729 standard is presented systematically. Moreover, some important techniques of algorithm including Linear-Prediction, Perceptual Weighting, Vector Quantization, Gain Quantization, Adaptive-codebook Search, and Fixed-codebook Search have been analyzed theoretically. Furthermore, some methods are put forward to improve the algorithm, such as improving the quantization of LSP coefficients, improving open-loop searching methods, simplifying the codebook quantization of the gains, optimizing the codebook search.Secondly, the characteristics of low power loss and high-performance 16-bit fixed-point digital signal processing chip TMS320VC5402 are introduced. DSP (Digital Signal Processor) as a special element for digital signal processing has been widely used in many fields such as multimedia, telecommunication. Because DSP has many advantages, such as compactness, low power consumption, high operation speed and low price, it is very suitable for real-time implementation on speech and video compression. In this thesis, the G.729 algorithm is implemented on the TMS320VC5402 fixed-point DSP of Texas Instruments, and the modules of the source code with larger computational quantity have been optimized.Finally, the simulation results show that the computational complexity of the optimized G.729 algorithm has been reduced to 1/14 that of the original algorithm. At the same time, by means of subjective judgment, the output speech has high reconstructed quality with the optimized G.729 algorithm.