Speech source localization using a multi-resolution technique

This paper presents a new speech source localization method in an adverse environment for microphone array systems. This method is applied to an octave-band decomposition of the signals obtained by the wavelet transform which uses a fast algorithm with short prototype filters. This method is applied in sub-bands separately and consists of two stages. First, a coarse region where the speech source is present is detected. Then the multi-beamforming operation is used to pinpoint the speaker´s location. Both stages are based on the examination of the energy level and its variation. The complete algorithm provides relatively small errors in the source localization estimate