摘要 |
PROBLEM TO BE SOLVED: To provide a signal component extracting method and device capable of surely extracting a signal component without being significantly affected by noise components. SOLUTION: Sound data M1 (n) and M2 (n) generated by sampling sound signals from a pair of microphones arranged at different collection positions are sequentially stored in a pair of ring buffers. In a silent section until a voiced section is started, the self-correlation value of voice signals collected by both the microphones, a cross-correlation value between both voice signals and a cross-correlation value between the both voice signals and a reference signal with a small amplitude are sequentially updated on the basis of stored contents each time the ring buffers are updated (S110 to S130). When the voiced section is started, a filter coefficient of a filter for converting a noise component into a signal with a small amplitude is calculated from the correlation value calculated in the silent section, and subsequently, the filter is applied to the voice data each time the ring buffers are updated to suppress a noise component (S140 to S180). COPYRIGHT: (C)2004,JPO
|