) to the speaker by power and zero-crossing-numbers of input signals which are acquired by four directional microphones. Second, three inter-microphone level differences between M_

and other three microphones are computed to obtain a power spectrum of the target speaker. By adding the signal of microphone M_

to the obtained power spectrum, we can get the speech with less distortion. We conducted both subjective and objective evaluation experiments. The results show that the system can suppress undesired signals with little speech distortion.