| 摘要 |
<P>PROBLEM TO BE SOLVED: To provide a sound source separation device, a sound source separation program and a sound source separation method that can obtain high sound source separating performance irrelevantly to a state of localization of each sound source signal in a mixed audio signal generated by superposing a plurality of sound source signals and further lighten a processing load. <P>SOLUTION: An FDICA sound source separation processing section 11 which finds an output signal yi(t) performs sound source separation processing based upon an FDICA method of subjecting to the sound source separation processing signal x1(f) and x2(f) generated by previously processing signals x1(f) and x2(f) of channels L and R through discrete Fourier transformation to a frequency range. Further, the signals x1(t) and x2(t) of the plurality of channels in the frequency range after the discrete Fourier transformation are each branched into two, and one is used to perform sequential update calculation (learning calculation) of a separate matrix W(f). In parallel to that, signals x1'(t) and x2'(t) each obtained by delaying the other are subjected to separation processing based upon the separate matrix W(f) found through the sequential update calculation. <P>COPYRIGHT: (C)2007,JPO&INPIT |
