| 摘要 |
PROBLEM TO BE SOLVED: To separate a sound at each sound source by high S/N even in an environment that the number of sound sources exceeds the number of microphones, and slight background noise, an echo and reverberation are generated. SOLUTION: The power of noise components included in a separation sound is considered as a cost value other than an I1 norm used as a cost function when a ¾1 norm minimization method separates the sound. In addition, in the ¾1 norm minimization method, the cost function is defined by a hypothesis that a voice does not have relevance in the time direction, in the present invention, the cost function is defined by hypothesis that the voice has relevance in the time direction and a mechanism by which a solution having relevance in the time direction becomes easy to be selected is provided. COPYRIGHT: (C)2007,JPO&INPIT |
