| 摘要 |
<p>PURPOSE:To improve a recognition rate and to allow the recognition of diversified voices by processing the characteristic data of the voices extracted by a characteristic extracting section with a neural network and executing voice recognition. CONSTITUTION:The zero order input data of the characteristic information extracted from the digital data patterns corresponding to input vocal sounds is supplied to a recognizing section 100 and is stored in a data memory 110. The data is multiplied by the weight of the 1st neural layer of the memory 110 in a multiplier circuit 121 in an output data forming section 120 having the association effect of the neural network, is accumulated in an adder circuit 122 and is compared with the threshold stored in the memory 110 in a threshold processing section 123, by which the data is recognized to form the 1st order neuron which is stored as the primary output data into the memory 110. The secondary output data of the 2nd order neuron is stored into the memory by the similar processing from this output data and the weight of the 2nd neural layer. The voice recognition is executed by the similar iteration. The erroneous recognition is decreased by the processing with such neural network, by which the recognition rate is enhanced. In addition, diversified kinds of the voice recognition are executed by the association effect.</p> |
