Target sound source extraction using spectral subtraction with a two-channel microphone array

Abstract

A novel method for extracting a target sound source from the measurements of a two-channel microphone array is presented. In this paper, it is assumed that there is a single target sound source located directly in front of the array and a single noise source located in the other direction. The proposed method exploits the power spectrum of the difference signal between the signals observed at two microphones. Since the arrival times of the target sound are equal and those of the noise source are different, the difference signal contains only the noise signal. In the proposed method, the power spectrum of the difference signal is linearly modified on the assumption that the two microphones are located sufficiently close to each other, and then, the target source is extracted by using a spectral subtraction in which the power spectrum of the difference signal is subtracted from that of the observed signal. The proposed method requires neither delay estimation nor direction-of-arrival estimation, which are generally used for modification of the spectrum of the difference signal. Therefore, the proposed method is performed by a simple algebraic calculation.