For the localization problem of far-field sound source, the TDoA positioning scheme is proposed by using short base-line sensor network in the paper. Usually there is a kind of localization algorithms which do not need an estimated initial point; on the other hand, there is another kind of algorithms which depend on an initial estimation. In the paper the two kinds of methods are combined, and several combinational localization algorithms are developed. The main idea is that the first kind of localization algorithms are applied to finish a coarse positioning process and output an initial estimation location, which is inputted as the initial value of the second kind of algorithms with high positioning precision to take an advantage and supply a gap from them. The circular error probability of combinational localization algorithms is proposed to demonstrate the feasibility of the performance evaluation criterion, which is based on Monte-Carlo method and a simple case of geometry deployment. The circular error probability of combinational algorithms is numerically analyzed, and it is concluded that the combination of spherical interpolation and least squared equation error has an optimal performance. The box-plot character is experimented about this combination algorithm. According to the simulation experiments of different distances and standard deviation of sound range difference, and some sound source data obtained from the actual field scene, the conclusion of the paper has been verified.