Binaural sound source localization based on upward zero-crossing points detection in reverberant fields

This paper mainly deals with the problem of sound source localization based on upward zero-crossing points detection with two microphones. Gammatone filter bank are used to divide the signal into different frequencies, SNR can be obtained by the statistical properties of the time difference samples, and then reliable time difference samples can be selected according to the SNR limitation. Finally, the practical time difference is achieved by the analysis of the time difference samples. Consequently, the time differences are transformed into azimuths by use of geometrical relationship. Experiments are conducted with different sounds, and sounds with wider spectrum seem to perform more precisely; Evidence also shows that the method is robust to sound signals with different SNR, and experiments in different rooms reveal that reverberation apparently degrades the localization accuracy.