Acoustic model training method, device and storage medium

Abstract

The present invention provides a training method, device and storage medium for an acoustic model, wherein the method includes: using Chinese training data to train an initial model to obtain a first neural network; obtaining a second neural network according to the first neural network; wherein, the second neural network The hidden layer is initialized with the first hidden layer of the first neural network, and the weights of N+S nodes are initialized randomly; the second neural network is trained with Chinese training data to obtain the third neural network; Chinese training data and English training data are used The acoustic model is obtained by training the third neural network. In the acoustic model training method, device and storage medium provided by the present invention, when the trained acoustic model is applied to the recognition of Chinese and English mixed speech data, the recognition rate of the acoustic model for Chinese speech data is improved.

Description

Technical field [0001] The present invention relates to electronic technology, and more particularly to a training method, apparatus, and storage medium of an acoustic model. Background technique [0002] With the development of electronic technology, speech recognition technology has gradually entered various fields such as industrial, home appliances, communications, automotive electronics, medical, home services, and consumer electronics products in people's lives. As a living scenario with people's lives, common electronic devices generally have voice recognition functions to implement the interaction of voice content between the user and the electronic device, so that the user can pass the address book in the voice control of the electronic device. Input method, map, car navigation and other software implement related functions. Most of the electronic devices achieve the speech content of the user through the machine through the machine learning by the speech recognition mod...

AI Technical Summary

Problems solved by technology

[0004] However, in the process of using the prior art to train the acoustic model, due to the characteristics of backpropagation in the neural network training process, the weight of each node in the output layer will participate in the calculation of the hidden layer nodes of the neural network, and for the input layer of the neural network at the same time The Chinese and English mixed speech data input to the neural network by the nodes will affect each other due to different pronunciation methods

The acoustic model obtained by actual training has a low recognition rate for Chinese speech data in the subsequent process of recognizing speech data

Images