Design And Implention Of Low Bit Rate Speech Coding Algorithm Based On CELP

The most natural human communication information is speech which is the most effective and convenient, so most of the communication systems are included the function of the speech transmission. In recent decades, with the rapid development of information society and communication technology, the amount of transferred data that required is increasing exponentially at a dramatic rate, as a result, frequency resource has become increasingly important. But how to encode and transmit the speech information with lower coding rate and better voice quality has been a big problem for a long time.Code excited linear prediction(CELP) coding algorithm has high quality synthetic speech and excellent resistance to noise and multiple transfer performance. It will become very valuable in practical application if it can further reduce the CELP coding rate on the premise of guarantee the quality of synthesized speech, so the work about the study of CELP algorithm is very meaningful.This thesis focuses on how to further reduce the coding rate on the basis of guarantee the quality of voice, and conducts a series of research and experiment with the federal standards FED- STD-1016 algorithm. In the end, this thesis proposes a low bit rate speech coding plan and realizes it. The main research results are obtained:1. Research about the influence of frame length and numbers of the subframe on the synthetic speech quality. First of all, this thesis demonstrates the frame length and the change of subframe can adjust the rate of the speech coding through the analysis of theory. Then, this thesis fully verify the validity of the scheme through a large number of experiments, at the same time, the voice quality can be remained unchanged. Finally, this thesis realized the algorithm based on the Federal Standard 1016. Test and evaluation results have shown that the reconstructed speech of the designed algorithm is clear and natural, verify the feasibility of the algorithm.2. Improved the quantization method of the LSF parameters. It used independent and non-uniform quantization method to manage 10 LSF parameters of linear predictor in FED- STD 1016 algorithm. It requires a total of 34 bits to encode and transfer, and it is about a quarter of the total number of coding bits, in other words, many bits have been allocated for it. This thesis adopted vector quantization to quantify the LSF parameters and this method can use a dozen bits to describe the LSF parameters, also it can reduce coding rate and improve performance of the algorithm.3. The design and implementation of low rate speech coding scheme. The low rate speech coding scheme combines the change of the frame length and the subframe and the methods of improving the LSF parameters of quantitative. The scheme can achieve a variety of coding rate through thesis adjustment about frames and frames length and reasonable allocation of quantization bit number. This thesis implements 4kbps, 3.6kbps and 3.2kbps. The synthetic speech has good understandability and naturalness under those three rates through objective testing and subjective audition, at the same, the voice quality achieves the goals of expected.4. The fixed-point of the algorithm. It uses floating-point program to achieve the algorithm simulation, but the algorithm eventually need to apply to real applications, so it need to prepare for subsequent DSP platform transplantation by using of fixed-point algorithm. This thesis realizes the algorithm of the fixed point. This thesis made some optimization to improve on the original algorithm and further enhance the operational efficiency of the algorithm in the process of fixed point. Finally, compared the results of the fixed-point with the result of the floating-point program, on one hand, it verified the correctness of the fixed-point, on the other hand, it also verify the designed about the low rate of coding algorithm in this thesis.