音源の位置情報を利用した実環境下での雑音除去と目的音声の抽出

概要

論文の詳細を見る
It is essential to solve the scaling and permutation problem peculiar to ICA, when the frequency domain ICA Is applied to noise reduction in a real environment. This paper presents a solution to this problem in the case of two sources and two microphones under the assumption that a target source is on the side of one microphone and a noise source is on the side of the other microphone. First, it is shown that the split spectrum is uniquely expressed in terms of the product of a source and its transfer function to the corresponding microphone; this fact settles the scaling indeterminacy. Next, two permutation correction rules are derived, based on the absolute value and the argument of the split spectrum ratios. Finally, a new permutation correction method is proposed which uses each of the rules in its effective frequency band. In the same framework, a speech extraction is also proposed.
近畿大学の論文