Amr to G.729a Speech Transcoding with Fast Codebook Search

Speech transcoding scheme is needed in the voice system over internet. Full decoding technique is an intuitive and traditional speech transcoding method, but it requires high computational complexity and long processing time. In this paper, we propose a partial decoding technique with fast codebook search, which utilizes the pulse replacement method, on ACELP coding architecture. There is no need to redo all the decoding and encoding processes. Partial decoding method can be directly applied to ACELP based speech coding, such as AMR[1] and G.729A[2] speech standards. The proposed method decodes the parameters from the input bit-stream, which includes line-spectral pair (LSP), pitch delay, fixed codevector and codebook gain. It achieves excellent voice quality as the full decoding method does while it only requires 7.2% computation loading calculated byclockticks per frame. In this paper, we propose a partial decoding technique with fast codebook search, which utilizes the pulse replacement method, and describe its working procedure in Section 2. In Section 3, we use C simulation to demonstrate the quality measurement of our proposed approach. Finally, Section 4 concludes the paper. Keyword : speech transcoding, AMR, G.729A, ACELP Figure 1. Integration of 3GPP and IP network 2. PROPOSED TRANSCODING SCHEME FROM AMR TO G.729A Figure 2. Two speech transcoding schemes (a) Full decoding (b) Partial decoding To translate one frame from AMR to G.729A, the direct solution is to cascade the decoder of AMR and the encoder of G.729A. However, this conventional method has several problems, including of computation complexity, coding delay, and so on: 1. Computation complexity: The conventional full decoding method needs to perform decode and encoder at least once. The computation load is tremendous in some compression procedure. 2. Coding delay:Some processing delay is generated by frame buffering and windowing look-ahead from LPC analysis. Therefore, this conventional transcodng method increases the processing delay. To solve these problems, we make use of the similarities between both codecs. Both AMR and G.729A are based on ACELP. ACELP transmits four kinds of parameters that correspond with the speech characteristics:LPC coefficients, pitch delay, index of fixed codebook, and gains of the adaptive codebook, and fixed codebook. AMR and G.729A are different from each other in frame size and method of quantization, as shown in Table I and Table II . Table I. Specification of AMR and G.729A coding standards AMR G.729A Algorithm ACELP CS-ACELP Bit-rate 4.75, 5.15, 5.9, 6.7, 7.4, 7.95, 10.2, 12.2 kbits/s 8 kbits/s Frame size 20ms 10ms Subfrmae size 5ms 5ms Table II. Difference of techniques between AMR and G.729A AMR G.729A LSP SMQ VQ Adaptive codebook lag 12.2k mode: 1/6 1/3 Other modes: 1/3 Fixed codebook pulses 2~10 pulses 4 pulses