For the multi-source localization based on TDOA， because the measured TDOA values from the microphone array cannot associate with the target sources， the process of source localization will produce association ambiguity， thus affecting the accuracy of multi-source localization results. The GCC algorithm is used to estimate TDOA values of microphone array. Then all possible TDOAs sequences of microphone array are obtained by permutation algorithm， and all possible sound sources can be estimated based on Chan algorithm. A group of check sub-arrays is constructed by switching the reference microphone in the array， and the phantom sound sources are filtered out by the relative position relationship between the true sound sources and the array microphones. For all sound source locations screened by different check sub-arrays， redundancy check is carried out again according to the principle of maximum occurrence frequency， so as to improve the accuracy of the final real sound sources. The simulation and experiment results show that the proposed method can effectively eliminate the phantom sound sources in multi-sources localization by using the least number of conventional microphones. The localization accuracy and localization robustness of the proposed meth?od are higher than that of the comparison method with the same number of array microphones.