Today, there exist various wire and wireless communication networks where different speech coding standards are adopted. The development of an efficient transcoding algorithm for different speech coders is important since it is essential for the integration and interoperability of networks. Generally, the simplest way to solve the compatibility problem between two different speech coders is by the tandem transcoding algorithm : generate speech signal using a decoder of one speech coder and then re-encode the signal by the other speech coder. But this approach results in quality loss, increase of computational complexity and extra delay since the speech signal undergoes encoding and decoding processes twice. These problems can be solved by the tandemless transcoding algorithm in which the parameters that are commonly used in CELP vocoders are directly converted.
In this thesis, a novel tandemless transcoding algorithm for Adaptive Multi Rate(AMR) and Enhanced Variable Rate Codec(EVRC) speech coders is proposed. Before the parameters are transcoded, an input frame is determined whether it contains speech or not. For this kind of frame classification, the characteristics of the codebook gains are exploited. The input frame of non-speech region is transcoded efficiently using the DTX mode of AMR or Rate 1/8 mode of EVRC. In the frame of the speech region, the parameters should be transcoded exactly. The proposed transcoding algorithm consists of three parts --- conversion of LSP, conversion of pitch delay / pitch gain and conversion of fixed codebook vector / fixed codebook gain. The proposed algorithm is implemented in the fixed-point format using the C language in the Windows platform. The performance of the proposed algorithm is evaluated using the various measures. The evaluation results shows that the proposed algorithm achieves speech quality equivalent to that of the tandem algorithm while requiring less delay and computational load.
