| 摘要 |
PROBLEM TO BE SOLVED: To provide a speech extracting device capable of extracting a speech component from a speech signal of a single microphone by properly removing noise components without using a plurality of microphones. SOLUTION: A signal decomposition part 45 uses a plurality of filters FL0, FL1, and FL2 passing signal components of mutually different frequency bands to extract mutually independent signal components y0, y1, and y2 from the speech signal inputted to the single microphone. A signal composition part 47 puts together signal components y0, y1, and y2 according to a 1st rule to generate a 1st composite signal Y1 and puts together the signal components y0, y1, and y2 according to a 2nd rule to generate a 2nd composite signal Y2. The 1st and 2nd rules are so determined that the difference between a probability density function p1(z) of the composite signal Y1 and a probability density function p2(z) of the composite signal Y2 becomes maximum. A selective output part 49 selectively outputs a composite signal having a larger difference from a Gaussian distribution between the composite signals Y1 and Y2. COPYRIGHT: (C)2005,JPO&NCIPI
|
