Blind Source Separation Based on Rotation of Joint Distribution Without Inversion of Positive and Negative Sign

Abstract

This paper proposes a blind source separation method based on rotation of a joint distribution of observed mixture signals with microphones. Our previous method can separate the sound sources using only the information of observed signals. The separation method has sign inversion problem in some cases. The problem means that the positive and negative signs of the estimated signal are inverted. In the case of using the signal from the start to the end of the human speech, the sign of the estimated signal is not a big problem. However, in the case of short-time frame processing, inversion of a sign at the junction point of the separated signal becomes a big problem. Therefore, we propose a new separation method without indeterminacy of positive and negative sign of the estimated signal.