Blind source separation without permutation and scaling indeterminacy

Independent component analysis (ICA) can separate unknown sources using only their mixtures if they are statistically independent. However, the scaling and permutation problem occur at each frequency when the frequency domain ICA is adopted; it is essential to solve this problem peculiar to ICA. In the present paper, a permutation correction rule is proposed by using locational information on the sources under a multi-source and multi-microphone configuration. A method for extracting human speech exclusively is also proposed.