Research And Implementation On Transcoding From AMR To EFR And AMR To HR Speech Coders

Speech coding for communication system has new technologies and standards with the rapid development of wireless communication technology. Many countries, including china, use the GSM communication mobile network, its speech coding technologies are starting to use auto-adaptive multi-rate speech coding standard based on FR, EFR and HR speech coding algorithms. Thus, various speech coders coexist amoung different GSM mobile phones, meanwhile the use of different speech coders in transmitter and receiver of GSM network caused compatibility problem.The way to solve this problem is to develop a transcoding system between AMR and EFR/FR/HR. There are two methods to develop a transcoding system. One is Tandem algorithm based on decode-then-encode method; the other is parametric domain transcoding method. This paper studies the feasibility of parametric domain transcoding from AMR to EFR and HR, and presents a proposal to implementation of that transcoder.The outline of this thesis is organized as follows:(1) Introduced code excited linear prediction speech coding algorithm which is used by AMR, EFR, HR, and analysis its parameters, such as linear prediction coefficients, pitch delay, fixed-codebook parameters and codebook gain parameters.(2) According to similarity of parameters of each coder, proposed transcoding methods from AMR to EFR and from AMR to HR. On discontinuous speech transmission, the proposed AMR to EFR transcoding has two steps: LPC conversion and energy conversion, while AMR is directly mapped to EFR on the bit stream layer on continuous speech transmission; The proposed AMR to HR transcoding is composed of five steps: LPC conversion, MODE determination, pitch conversion, fixed codebook search and codebook gain conversion.(3) Simulations of the parametric domain transcoding system are carried out with the fixed-point C programming language on personal computer. The subjective and objectie listening test shows the proposed algorithm could produce equivalent speech quality to the Tandem algorithm but with at least 5ms shorter delay and 82% and 37.5% computational complexity in theory.