Research And Optimization Of Vector Quantization In AMR-WB Coding

With the rapid development of the digital networks and audio-visual technique application, on the one hand, the requirement for speech quality is getting higher and higher, traditional narrowband speech, whose bandwidth is between300Hz and3400Hz, can no longer meet the people’s needs, people are increasingly in pursuit of the comfort of speech quality and the natural face-to-face communication, therefore, it is necessary to develop wideband speech coding whose bandwidth is between50Hz and7kHz. On the other hand, the speed of the traditional speech code is generally determined, if the encoding rate is higher, then it will occupy more network resources,whereas if the encoding rate is lower,then it will occupy less network resources,but the speech quality is worse. In order to solve the problems mentioned above, the variable rate speech coding is proposed.On the basis of the reasons above, it has become a hot topic to combine the wideband speech encoding and the variable rate speech coding. The adoption of AMR-WB is of significant importance for it is the speech code system adopted both by wireless and wireline service at first time, and ITU-T and3GPP both select it as wideband speech coding. AMR-WB is used widely in mobile communications, ISDN of broadband telephony, ISDN in areas such as video telephony and videoconferencing. The main research work of AMR-WB code in the paper is as follows:(1) First analyze the encoding principle of AMR-WB, and deeply study the pre-processing module, LPC module, open-loop pitch module, codebook search module and gain quantization module. Then describe the decoding principle of AMR-WB. Finally simulate the C++coding of AMR-WB by Visual Studio2008and the test sequence tst.inp provided by ITU-T, and lay a certain foundation for the further study of AMR-WB.(2) AMR-WB has high computational complexity, which limits its application in some areas with limited computing resources. To solve the problem, on the basis of the systemically study of the principle of AMR-WB codec algorithm, the paper focuses on the research of vector quantization. Based on the results of previous studies, the paper researches the codeword design、the codeword search and codeword index assigned. Among them, codebook design and the codeword search are deeply studied.(3) In AMR-WB vector quantization of LPC module, the codebook design is generated by LBG method, which leads to the slow convergence and high computational complexity.To solve the problem mentioned above, mean-based ranking is introduced to generate the initial codebook, the mean is a condition of aggregation in the method, which can gain similar amount aggregation in higher probability. The algorithm is applied to the AMR-WB. Compared to the random method and separation mean method, the results show that the improved clustering method can be performed at a faster speed, the quality of the finally obtained codebook is also better.(4) In AMR-WB vector quantization, the codeword search method is the full search algorithm, whose search accuracy is good. But the search algorithm needs to calculate the distortion of distance between the input vector and all codewords, thus leading to the high search complexity and the time-consuming codeword search. To deal with the problem, an improved algorithm is proposed which is based on partial distortion codeword search algorithm, in the method,it is needed to train the initial match codeword to make sure that the codeword search begins in a place close to the target vector. The simulation results show that the proposed algorithm can significantly reduce the complexity of the codebook search, and the calculation time of speech coding is significantly reduced. The improved fast codeword search algorithm is effective to achieve fast encoding and has a certain application value.