摘要 |
PROBLEM TO BE SOLVED: To provide a processing method of an acoustic tube of a noisy speech, etc., rapidly processing a power component and a noise component of an observed acoustic signal with a small distortion. SOLUTION: Many acoustic signal values yk=xk+rk consisting of a speech component x and the noise component r are recorded at a sampling intervalτ(101), and a time delay vector is formed (104), and adjacent vectors are formed/extracted (105), and a correlation between the time delay vectors is decided, and the time delay vector is projected to a number Q of a peculiar vector (107). Then, an effective signal value forming a speech signal nearly corresponding to the speech component xk and/or a noise signal nearly corresponding to the noise component rk is decided, and a correlated signal component is allocated to the power component of the acoustic signal, and a not correlated signal component is allocated to the noise component of the acoustic signal (109).
|