CMG Research And Implementation Of Evrc Codec In The System

With the booming of the third generation mobile telecommunication, the study has been focused on the variable rate speech coding technology recent years. Enhanced Variable Rate Codec (EVRC) is one of the most applied variable rate speech coding technologies. In this paper, the principle of EVRC algorithm is described in detail and the implementation process is given. And we ameliorate fixed codebook search method. The measure results indicate that the new codebook search algorithm can reduce quite a lot of the operation load, and the quality of synthesis speech is improved.EVRC is based upon the RCELP algorithm, composed with encoder and decoder. At the encoder side, the input speech vector is presented to the signal pre-processing module at first. Then the pre-processed speech is presented to the LPC analysis module and long-term prediction module to abstract the line spectrum frequencies (LSF) and the pitch. The LSFs are quantized with the split Vector quantization (SVQ) method. Based on the rate which is determined by the rate determining algorithm, the need speech parameters are calculated and packed with corresponding formats to send out. At the decoder side, At first the frame error detection module detect and correct the frame error, then the speech parameters are unpacked from the date packets and presented to a synthesis filter to rebuild speech signal. At last, the rebuilt speech signal is filtered by the adaptive post-filter to improve audio quality.In this paper an efficient EVRC full rate fixed codebook search method is given. Compared with the "depth-first-jointly search" search algorithm described in the IS-127 standard, the new algorithm is a "sequential search & pulse replacement" search algorithm based on two stage. At the first stage, a coarse codevector is selected using a fast pure sequential search method. At the second stage, the coarse codevector is justified by the pulse replacement process and at last we can find the optimal codevector. The search load of the new algorithm is only 21% of the standard algorithm. The new algorithm also can solve the problem about the codevector optimality, which can not insure in the old algorithm. So with the new search algorithm, the synthesis speech can keep more characteristic of the speaker.We implemented the enhanced variable rate codec in the VC++6.0 and measured its performance. The result indicate that the codec using the "sequential search & pulse replacement" full rate fixed codebook search algorithm meet the IS-718 requirement on average data rate, and compared with the old EVRC system,20.5% operation load is reduced At the same time, the quality of synthesis speech is improved. If this codec is utilized on the vocoder board in the CMG system, with the same number of EVRC sustained,we can reduce quite a few DSPs.