Voice verification method and device, electronic equipment and computer readable storage medium

Abstract

The invention discloses a voice verification method and device, electronic equipment and a computer readable storage medium. The method comprises the steps of collecting a sound signal in real time; extracting audio features of the sound signals; inputting the audio features into a multi-classification verification model, and outputting a confidence coefficient set through the multi-classificationverification model, the confidence coefficient set comprising confidence coefficients of non-command words and confidence coefficients of a plurality of command words; selecting the maximum confidence in the confidence set; based on the condition that the maximum confidence coefficient is not the confidence coefficient of the non-command word, loading a binary classification verification model corresponding to the target command word to which the maximum confidence coefficient belongs; inputting the audio features into a binary classification verification model, and outputting a target confidence coefficient of the target command word through the binary classification verification model; and executing a function matched with the target command word based on the condition that the target confidence coefficient is greater than a specified threshold. Through the technical scheme of the invention, the equipment can ensure the accuracy of command word detection while saving power consumption and system resource consumption.

Description

technical field [0001] The invention relates to the technical field of neural networks, in particular to a voice verification method and device, electronic equipment and a computer-readable storage medium. Background technique [0002] At present, a variety of command words matching various commands are often set in voice control for the convenience of users. However, the neural network model used to verify various command words often has a complex structure, many parameters, a large amount of training data, and a long training time. It consumes a lot of power to operate. [0003] However, for low-power devices such as earphones, the high complexity and high computational load of the model will consume a lot of power, and if the complexity of the model is reduced to save power, it will affect the accuracy of command word recognition. Moreover, due to the computing limitations of low-power devices, the use of high-precision complex models will also cause high delays in voice...

AI Technical Summary

Problems solved by technology

[0002] At present, a variety of command words matching various commands are often set in voice control for the convenience of users. However, the neural network model used to verify various command words often has a complex structure, many parameters, a large amount of training data, and a long training time. It consumes a lot of power to operate

[0003] However, for low-power devices such as headphones, the high complexity and high computational load of the model will consume a lot of power, and if the complexity of the model is reduced to save power, it will affect the accuracy of command word recognition

Moreover, due to the computing limitations of low-power devices, the use of high-precision complex models will also cause high delays in voice wake-up, which will reduce the user experience of the device

Images