| 摘要 |
PURPOSE:To realize a noise suppressing device which suppresses the noise contained in the speech signals in a condition where the relative position between a noise source and a speech source frequently varies and furthermore, no bothersome noise is left in the speech after the noise suppression process and no deterioration exists in the suppression effect even though the time-wise pattern of the input signals varies. CONSTITUTION:Input signals, which are band divided by a band dividing means 120, are inputted to a neural network 130 and the signals, on which noise is superimposed, are mapped onto the signals without noise. So far as the adjustment of this device is concerned, a back propagation method is used to determine the weighting coefficients of the neural network 130 so as to minimize the errors of the output signals and the no noise signals while inputting the noise superimposed speech signals to this device. |
