Method and device for determining wake-up model

Abstract

The invention provides a method and device for determining a wake-up model, and the method comprises the steps of inputting any batch of training wake-up data in a training set into a parameter whichis based on an identification model, and determining the current state of the last layer of a hidden layer of a neural network; carrying out interpolation processing on the parameter of the last layerof the hidden layer of the recognition model in the previous state and the parameter of the current state, determining an interpolation, and updating the interpolation into the parameter of the current state; Inputting training wake-up data of other batches in the training set into an identification model, and updating parameters of the current state until training wake-up data of all batches inthe training set are input into the identification model; and determining the interpolation of the parameter of the current state and the parameter of the previous state, and updating the parameter ofthe current state of the last layer of the hidden layer of the neural network, thereby determining a wake-up model. The parameter of the last layer of the hidden layer of the neural network is updated, the updating amount is small, and a more accurate wake-up model can be obtained.

Description

technical field [0001] The invention relates to the field of neural networks, in particular to a method and device for determining an awakening model. Background technique [0002] At present, the establishment of the wake-up module is generally to specially record voice information about the wake-up word for training the neural network. And in the training process of the neural network, each parameter of each layer of the entire neural network is updated. This generally takes a lot of time and costs, and the training calculation of the neural network is too large, which is prone to errors. Therefore, the accuracy of the obtained wake-up model is not high. Contents of the invention [0003] (1) Technical problems to be solved [0004] The object of the present invention is to provide a method and device for determining an awakening model, so as to solve at least one of the above technical problems. [0005] (2) Technical solutions [0006] An embodiment of the present ...

AI Technical Summary

Problems solved by technology

This generally takes more time and costs more, and the training calculation of the neural network is too large, and errors are prone to occur. Therefore, the accuracy of the obtained wake-up model is not high.

Images