Two-microphone voice activity detection in the presence of coherent interference

In this paper, we propose a two-microphone Voice Activity Detection (VAD) method in the presence of coherent interference. The proposed method is based on the Cross Power Spectrum Phase (CPSP) which is an implementation of the Phase Transform (PHAT) weighted cross correlation between two microphones. The PHAT weighting whitens the spectrum of input signals and makes the cross correlation dependent entirely on the phase of the cross spectrum. If we assume that the direction of desired speech signal is known and the time delay between microphones is compensated, the Averaged CPSP (A-CPSP) can be utilized as a VAD measure. In order to enhance the VAD performance in the presence of strong coherent interference from other direction, we propose a Maximum Partially Averaged Real CPSP (MPA-RCPSP) method which detects the cophased frequency region with high Signal-toInterference Ratio (SIR). Simulation results demonstrate that the proposed MPA-RCPSP is a more reliable measure to the conventional A-CPSP in the presence of strong coherent interference.