摘要 |
PURPOSE:To execute the decision as to whether an input acoustic signal is a voice signal or noise signal by a simple processing in a short period of time by constituting the filter in such a manner that the delayed acoustic signal existing within a voice section is outputted and the delayed acoustic signal existing in a noise section is not outputted in accordance with the identification signal from a neural network for voice section identification. CONSTITUTION:The cepstrum coefft. C(t) of 16-th order outputted from an acoustic analysis section 11 is inputted to a neural network 12 for vowel identification. Whether the acoustic signal of a frame (t) exists in the vowel section or not is identified in this network and the identification signal V(t) is outputted. This identification signal V(t) is inputted to a neural network 13 for voice section identification. Whether the acoustic signal of the frame (t-T) exists in the voice section or in the noise section is identified in this network. As a result, the identification signal NN(t)=1 is outputted if the acoustic signal of the frame (t-T) is identified to exist in the voice section. On the other hand, the identification signal NN(t)=0 is outputted when the frame (t-T) is identified to exist in the noise section. |