オオイシ　クニオ   大石　邦夫    所属   コンピュータサイエンス学部 コンピュータサイエンス学科    職種   教授
言語種別	英語
発行・発表の年月	2017/03
形態種別	学術講演予稿集（学会、研究会を含む）
標題	An efficient approximate joint diagonalization algorithm and its application to blind source separation
執筆形態	共著
掲載誌名	IEICE Technical Report
掲載区分	国内
出版社・発行元	The Institute of Electronics, Information and Communication Engineers
巻・号・頁	EA2016(141),pp.327-332
著者・共著者	Shinya Saito, Kunio Oishi, and Toshihiro Furukawa
概要	Fast implementation of diagonalization implemented by minimizing the direct and indirect least-squares criteria (DMDI) algorithm (FDMDI) is presented. The DMDI algorithm is one of approximate joint diagonalization (AJD) approaches for blind separations of convolutive speech mixtures. These approaches are based on the alternative minimization of indirect and direct least-squares criteria for diagonal matrices on the first phase and for mixing matrix on the second phase, respectively. In the FDMDI algorithm, the reduction in computation on the second phase arises from using a non-iterative method for estimating the mixing matrix. Simulation result shows that the conventional DMDI algorithm is at least 13.3 times as complex as the FDMDI algorithm without separation performance degradation in overdetermined convolutive audio mixtures.