Research On Transcoding From AMR To G.729A Speech Coder

With the rapid development of modern communication technologies, the merging between the wireless network and Internet becomes a very important issue. Because different networks always employ different speech codecs, bit stream generated by one coder should be translated seamlessly to that of the other coder. This kind of technology is called transcoding. Presently the traditional Tandem algorithm for transcoding is through a so-called Decode-Then-Encode approach. However, there are mainly three disadvantages in this method. Firstly, the computational complexity is quite high. Secondly, the synthesized speech quality descends. Lastly, additional delay will be needed because of the recoding.Combined with the latest research result of the world, a parametric domain transcoding method from AMR to G.729A speech codecs is introduced in this paper. AMR codec is adopted in the WCDMA mobile communication and the G.729A is the most favorable speech coder in VoIP application. Compared with the Tandem algorithm, the proposed method not only produces equivalent speech quality but also gives lower computational complexity and shorter system delay. The outline of the paper is organized as follows:(1) Speech codec used in the modern telecommunication field is introduced briefly.(2) The background of the transcoding is roughly described.(3) Models for the speech signal and the methods to evaluate the quality of the speech are discussed.(4) The normal methods of the speech codec are analyzed in detail, including LSP (Linear Spectrum Pair), VQ (Vector Quantization) and A-B-S (Analysis by Synthesis).(5) The differences of frame structure and coding method between AMR and G.729A speech codec are detailed discussed.(6) According to the character of the frame structure of the two speech codecs, a scheme is proposed to transcode from AMR to G.729A.(7) Simulations of the parametric domain transcoding system are carried out with the float-point C programming language.Compared with the traditional transcoding scheme, the proposed scheme has almost the same speech quality, while the computational complexity is reduced remarkably.