PURPOSE: To obtain a caller recognizing telephone device in which it is not necessary to limit an objective voice to a preliminarily learned voice content at the time of recognizing a caller.

CONSTITUTION: At the time of recognizing the caller by using a neural network, a caller recognizing part 13 of a caller recognizing telephone device 10 inputs the series of a vector indicating the shape of a short time spectrum, synthesizes the series of a network output by the sum, product, and majority or the like of a recognized result obtained from each output, and obtains one recognized result. Thus, a phoneme well-balanced text can be used as the data for learning of the network.