Research And Implementation Of Low Rate Speech Codec Based On EX-CELP

The research and application of speech coding technology is greatly promoted by market demand and communication technology development.In the communication system,the band resources are very valuable.After coding by the low-rate speech codec,the speech will occupy the less transmission bandwidth.Hereby,low-rate speech coding with less bandwidth resources is utilized to improve band resources utilization as the primary method.Low rates and high quality speech codec can be obtained by applying the Code Excited Linear Prediction technology,which has been widely used in mobile communications and multimedia communications.The Selectable Mode Vocoder(SMV),published by the 3rd Generation Partnership,uses the Extend Code Excited Linear Prediction technology and can achieve a good synthetic voice quality under the condition of low-rate coding.But the complexity of the SMV is very high,thus the application is limited.This thesis study on SMV's complexity reduction,and implement SMV on DSP.Firstly,this thesis analyzes the basic principles and implementation process of SMV in detail.Then,adjusting the algorithm structure and selecting the coding rate to realize the variable rate speech codec based on eX-CELP with the coding rate of 4kb/s.Secondly,this thesis improves both the voice activation detection module and fixed codebook search module of SMV.Considering the complexity of voice activation detection algorithm,this thesis compares the speech signal energy with the threshold level which is based on the energy of the speech signal energy and the background noise level.The result shows that the complexity is reduced and the accuracy of the voice activation detection algorithm is maintained to a certain extent.SMV's fixed codebook utilizes the multi-sub codebook structure.Fixed codebook search operation is huge.In this thesis,we improve a pitch enhancement algorithm of fixed codebook search,which limits the number of sub-frames of enhanced pitch.Then,we need to select the sub-codebook before searching the sub-codebook when the pulse sub-codebook search begins.Meanwhile,according to the frame classification information and other speech parameters to narrow the range of sub-codebook selection so that the complexity of fixed codebook search is reduced,and a improved variable rate speech codec with the coding rate of 4kb/s is achieved.On this basis,the performance of the codec is tested.Finally,in order to test the complexity of the algorithm,the algorithm is transplanted to the TMS320C6713 DSK hardware platform and optimized.After testing the algorithm complexity,the results showed that the calculation of improved voice activation detection module is 28.99% less than the original module,the computation of the improved fixed codebook search module is reduced 47.74% than the original module,and the improved codec's complexity is 18.7% lower than the original one.