摘要 |
PURPOSE: To reproduce a wide-band voice signal with a small computing amount and without increasing the band width by applying a nonlinear processing to the predictive residual signal of a narrow-band voice and utilizing a higher harmonic component to be generated at this time. CONSTITUTION: A nonlinear processing part 45 applies the nonlinear processing to the predictive residual signal of the narrow-band voice every sample to generate a wide-band predictive residual signal. A wide-band voice signal is generated by passing the predictive residual signal made to be the wide-band through a predictive synthetic filter 46 whose coefficient is made to be the wide-band predictive coefficient from a neural network part 44. The low frequency component and the high frequency component of a voice signal are respectively extracted by passing the wide-band voice signal through band-pass filters 48, 49. Then, the wide-band voice signal is generated by adding these two kinds of extracted components and the mid-band frequency component taken out by a band-pass filter 47 from the narrow-band input signal in a synthesis part 50. |