LEARNING DEVICE, METHOD, AND PROGRAM: abstract simplified (18022901)

• This abstract for appeared for patent application number 18022901 Titled 'LEARNING DEVICE, METHOD, AND PROGRAM'

Simplified Explanation

The abstract describes a method for learning a speaker embedding neural network. This network is trained to minimize the difference between a speaker label and the output value it produces. The network consists of an input layer that receives voice signals, and an output layer that indicates the speaker. The network also includes a calculation network that determines the accuracy of the speaker identification based on the input features. This accuracy is then used to calculate an average and a posterior distribution accuracy, which is compared to a prior distribution accuracy.

Original Abstract Submitted

Learning means generates a speaker embedding extracting neural network by learning a weighting factor so as to minimize a loss function indicating an error between a speaker label indicating a speaker of a voice signal and an output value output from an output layer, with respect to a neural network including an input layer that receives an input of the voice signal and the output layer that outputs the output value indicating a speaker of the voice signal. The speaker embedding extracting neural network includes a network that calculates a first accuracy from a feature in units of frames and calculates an average and a second accuracy of a posterior distribution from an average and an accuracy of a prior distribution and the feature and the first accuracy.