Channel-codec optimized soft input source decoding and its application for low-bit-rate speech transmission

A novel joint source-channel soft decoding algorithm over a coded system is presented. Unlike the traditional channel decoding which needs to recover and delivers the hard outputs to the source decoder, in this proposed algorithm, the channel decoder produces and delivers its soft outputs directly to the source decoder for soft input source decoding without making any hard decision on the soft outputs. Vector quantization (VQ) is used by the source codec. The algorithm design consists of three parts, VQ partition optimization based on the error-correction ability of the channel codec over different channel SNR levels, derivation of soft information from the channel decoder, and optimal soft input source decoding. Objective and subjective evaluation results show that the proposed algorithm can improve the system performance considerably relative to the traditional hard-output/hard-input channel-source decoding.