摘要 |
<p>PURPOSE:To improve the collating rate by discriminating whether or not a speaker is registered or not according to which of two output units of a discrimination neural network outputs a larger value. CONSTITUTION:This system has a collation neural network 30 which is equipped with an input unit 10 for inputting a speaker's voice and respective output units corresponding to respective registered speakers and the discrimination neural network 40 constituted while equipped with an output unit which outputs a discrimination making output indicating which of a registered speaker group and an unregistered speaker group a current speaker is in. Then the discrimination neural network decides whether the current speaker is registered or not according to which of output values of the registered speaker group and unregistered speaker group is larger. Consequently, the collating rate is improved.</p> |