Multiple sound sources tracking method based on Subspace Tracking

In this paper, we propose a novel method for tracking two talkers. The tracking is performed using a microphone array and is based on algorithms implemented successively, namely, the PAST (Projection Approximation Subspace Tracking) and IPLS (Interior Point Least Square) algorithms. When multiple sound sources exist, the order of eigenvectors estimated by the PAST might change at each point in the time-frequency domain. This prevents us from separating and localizing each source. In addition, the appropriate initial values for the IPLS algorithm are required even for the silent-speech sections or in the low-energy domain because of the sparseness of speech signals. To overcome such difficulties, we propose a new method for assigning eigenvectors in the appropriate order and a method for setting the suitable initial value in the IPLS. Several results of experiments performed in an actual room environment show the effectiveness of the proposed method.