METHOD AND DEVICE FOR SEPARATING BLIND SIGNAL

• Main Authors: OISHI KUNIO, SAITO SHINYA, KUBOTA HAJIME
• Format: Patent
• Language: English
• Published: 21.04.2011
• Subjects: ACOUSTICS; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

PROBLEM TO BE SOLVED: To provide a blind signal separation method and device, for estimating an unknown signal source signal from an observation signal by using a solution of a least-squares type simultaneous diagonalization problem and a solution of a permutation problem utilizing correlation with an adjacent frequency bin. SOLUTION: After short-time discrete Fourier transform of an observation signal multiplied by a window function, an inverse matrix of a spatial correlation matrix of the observation signal weighted by the way of an exponential function is determined in each block, and a plurality of blocks are used as objects for the least-squares type simultaneous diagonalization problem. An approximate solution of the least-squares type simultaneous diagonalization problem is determined adaptively by combining a least-squares method, an exponentiation and a repetition method in each frequency bin, and a separation matrix having high signal separation performance is generated. In a frequency bin having high signal separation accuracy, the permutation problem is solved adaptively by utilizing the fact that an adjacent frequency bin of a signal generated from the same signal source has correlation; and in a frequency bin having remarkably-low signal separation accuracy, permutation of a proximity frequency bin is copied to improve the signal separation performance. COPYRIGHT: (C)2011,JPO&INPIT