For multiple sound source localization and counting under noisy and reverberant environment, two modules, local signal-to-noise ratio tracking and coherence test, were introduced to extract the time-frequency bins less affected by noise and mutual interference of the sound sources in classical generalized cross-correlation angular spectrum; then the traditional peak search method was replaced by the dual-width matching pursuit to improve the following localization and counting accuracy. Through the simulation results, the proposed algorithm of multiple sound source localization and counting using both the filtered angular spectrum and dualwidth matching pursuit method is proved to be more accurate and robust than the traditional one, especially under the environment with lower signaltonoise ratio, stronger reverberation and more sound sources.