Smart Transcoding For Celp Based Speech Coders

With the development of wireless network and Internet, for a successful integration of the speech networks, the interoperability becomes more and more important. It is necessary for integrated speech networks, which support multiple speech coders, to translate bit streams between different speech coders seamlessly.Connecting two coders in tandem is the traditional way to realize bit streams translation. Tandem is to reconstruct speech signals by decoding bit streams of one codec and then to encode the speech re-constructed by another coder, however, tandem coding is associated with several problems such as high computational load and additional transmission delay.In this paper, after thorough study of the three popular CELP based speech coders including G.723.1, G.729 and AMR, a smart transcoding algorithm between G.723.1 and G.729 speech coders is proposed. This algorithm utilizes the commonness between these two speech coders to make a direct translation of LSP (Linear Spectral Pair) parameters, adaptive codebook parameters, fixed codebook parameters and codebook gain parameters. Therefore, the bit streams conversion is accomplished without reconstructing the speech signals. Informal subjective speech quality evaluations and analysis of the complexity of the transcoding algorithm are also given.Finally, a review of the already existed smart transcoding algorithm between G.729 and AMR speech coders is presented. Some further improvements on this algorithm are made, including fast fixed codebook search and codebook gain pre-search.