Neural network model establishing method and voice waking method, device thereof, medium and equipment

Abstract

The invention provides a neural network model establishing method and a voice waking method, a device thereof, a medium and equipment. The model establishing method comprises the following steps: extracting acoustic features of each frame in a speech signal; obtaining a feature spectrum based on the acoustic features; inputting the feature spectrum into the neural network model and performing treatment through a convolution layer, a pooling layer and a fully connected layer to obtain a wake-up word probability value; and correcting the model parameters according to the comparison result of thewake-up word probability value and a preset value. The voice waking method includes the steps of receiving the voice signal input by a user; extracting an acoustic feature of the voice signal; and obtaining the feature spectrum based on the acoustic spectrum; processing the feature spectrum by the neural network model to determine the wake-up word probability value; and determining whether the voice waking is performed or not based on the wake-up word probability value. The method obtains the feature spectrum by extracting the acoustic features of the speech signal, and processes the featurespectrum by using the neural network to obtain the wake-up word probability value, which improves the computational efficiency while reducing the computational complexity.

Description

technical field [0001] Embodiments of the present invention relate to the field of neural network models and voice wake-up, more specifically, embodiments of the present invention relate to neural network model building methods and voice wake-up methods, devices, media and equipment. Background technique [0002] This section is intended to provide a background or context for implementations of the invention that are recited in the claims. The descriptions herein are not admitted to be prior art by inclusion in this section. [0003] Traditional voice wake-up usually adopts a method based on automatic speech recognition (ASR) (Automatic Speech Recognition) technical route. The voice wake-up solution based on the ASR technology route can be practical in limited scenarios, but when the application environment is relatively harsh, such as far-field sound pickup, ultra-low signal-to-noise ratio, and very short wake-up words, the traditional voice wake-up solution is often effec...

AI Technical Summary

Problems solved by technology

[0009] The embodiment of the present invention provides a method for establishing a neural network model and a voice wake-up method, device, medium, and equipment to solve the problems of complex training process and low computing efficiency in existing technical solutions

Images