In this thesis, a novel transcoding algorithm for G.729A and Seletable Mode Vocoder (SMV) speech codecs is proposed.
As in tandem transcoding which transcodes by synthesizing speech via decoding of the bit stream using the decoder of the source codec and then re-encoding using the encoder of the target codec, the proposed algorithm directly transforms the parameters of source codec to those of the target codec without performing the encoding and decoding processes twice and completely synthesizing the speech in process.
Since G.729A and SMV are both based on Code Excited Linear Prediction (CELP), the parameters of both are similar. Therefore the parameters of one can be transformed to those of the other directly.
The transformation of these parameters is the main theme of this thesis. Line Spectral Pairs (LSP) are converted via linear interpolation. The search space for pitch delay of target codec is confined to a small interval around the pitch delay of source codec.
Whereas a fixed frame rate is used in G.729A, SMV varies the transmission rate of each frame depending on both the characteristic of the frame and the input from the network. When transcoding from G.729A to SMV, a novel rate selection algorithm is used to determine the characteristic of the frame and the frame rate.
The transcoder is implemented using C programming language. The program works on the Win32 platform but uses fixed-point format for the digital signal processor (DSP) implementation. The performance of proposed algorithm is evaluated using various measures. The evaluation results show that the speech quality of proposed algorithm is equivalent to that of tandem transcoding algorithm while requiring less computational load and shorter delay.