摘要 |
PROBLEM TO BE SOLVED: To suppress unnecessary deterioration of accuracy of a processing result, while securely preventing overflow in the processing result of a digital signal. SOLUTION: Impulse response h(p×τ)(p=0 to N) of a processing section 170 for performing signal processing by digital processing is divided into a first part h<SB>1</SB>(q×τ)(q=0 to M(<N)), and a second part h<SB>2</SB>(r×τ)(r=M+1 to N) by an impulse response dividing section 183. After a speech data signal X(j×τ) which is read beforehand is convoluted with the first part h<SB>1</SB>(q×τ) by a convolution section 184, an absolute value Z<SB>2</SB>(j×τ) as a result of convolution is calculated by an absolute value calculation section 185. Meanwhile, an L<SB>1</SB>norm value NRM of the second part h<SB>2</SB>(r×τ) is calculated by an L<SB>1</SB>norm calculation section 186. Then, an inverse of a sum of the absolute value Z<SB>2</SB>(j×τ) and the norm value NRM are calculated by a gain control section 187, and based on the calculation result, a gain value G(T) is generated and forwarded to a multiplication section 150. COPYRIGHT: (C)2010,JPO&INPIT |