A joint source-channel decoding algorithm using source parameters which were estimated by the soft output of Turbo code with Baum-Welch reestimated algorithm was proposed. The source parameters were obtained by the soft output of decoding the received noisy information sequence. The joint source-channel iterative decoding was implemented by combining the channel decoding and the source decoding with accurate probability structure of source model estimated by iteration. Discrimination information was suggested to measure the precision of the estimated source parameters and was used to determine the stop of the estimation iteration.