摘要 |
PROBLEM TO BE SOLVED: To separate a target sound from sound generated by a plurality of sound sources while reducing cost. SOLUTION: The signal processing device 1 comprises: a control section 10; a separation matrix arithmetic section 11; a direction specifying section 12; an interpolation section 13; and a signal separation section 14. A frequency band is divided into a plurality of bands by the control section 10, and each band is classified into a learning band group or an interpolation band group. Learning processing by an independent component analysis method is applied to the band which is classified into the learning band group by the separation matrix arithmetic section 11 and a learning separation matrix WG(f<SB>g</SB>) is calculated. On the other hand, DOA (Direction of Arrival) calculation is applied to the band which is classified into the interpolation band group so that a sound source direction is specified by the direction specifying section 12, and based on the specified sound direction, an interpolation separation matrix WH(f<SB>h</SB>) is obtained by the interpolation section 13. Moreover, a separation matrix W(f) for separating signals X<SB>1</SB>(f, t), X<SB>2</SB>(f, t) into signals Y<SB>1</SB>(f, t), Y<SB>2</SB>(f, t) is calculated by interpolating the learning separation matrix WG(f<SB>g</SB>) with the interpolation separation matrix WH(f<SB>h</SB>). COPYRIGHT: (C)2008,JPO&INPIT
|