Unlock instant, AI-driven research and patent intelligence for your innovation.

Wake-up word recognition model training method and device, wake-up word recognition method and device and medium

A technology of recognition model and training method, applied in speech recognition, speech analysis, instruments, etc., can solve the problem of high false awakening rate in wake word recognition

Pending Publication Date: 2022-04-19
青岛信芯微电子科技股份有限公司
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] The implementation of this application provides a wake-up word recognition model training, recognition method, device, and medium to solve the problems in the prior art that the wake-up word recognition has a high false wake-up rate, or excessively relies on the context information of the wake-up word

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Wake-up word recognition model training method and device, wake-up word recognition method and device and medium
  • Wake-up word recognition model training method and device, wake-up word recognition method and device and medium
  • Wake-up word recognition model training method and device, wake-up word recognition method and device and medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0041] figure 1 It is a schematic diagram of the process of the wake-up word recognition model training method provided in the embodiment of the present application. The process specifically includes the following steps:

[0042] S101: For each sample speech in the sample set, according to the length of the pre-saved receptive field and the length of the sample speech, determine the target label corresponding to each speech frame in the sample speech, and the target label is used to identify the corresponding speech frame. Whether the complete wake-up word in the sample speech is located in the receptive field of the model, and the current position of the wake-up word in the receptive field of the model; input each speech frame of the sample speech into the wake-up word recognition model, and obtain the wake-up word recognition An identification label for each frame of speech output by the model.

[0043] The wake-up word recognition model training method provided in the embo...

Embodiment 2

[0054] In order to ensure the recognition accuracy of the wake-up word recognition model, on the basis of the above-mentioned embodiments, in the embodiment of the present application, according to the length of the pre-saved receptive field and the length of the wake-up word in the sample speech, each The target labels corresponding to speech frames include:

[0055] Determine the sub-duration corresponding to each label according to the first duration of the wake-up word in the sample voice and the length of the pre-saved receptive field;

[0056] Determine the target time range corresponding to each label in the sample speech according to the sub-time length and the receptive field length;

[0057] According to the time range corresponding to each speech frame and the target time range corresponding to each tag, the target tag corresponding to each speech frame is determined.

[0058] In order to complete the training of the wake-up word recognition model and ensure the ac...

Embodiment 3

[0099] In order to reduce the false wake-up rate of wake-up word recognition and avoid over-reliance on the context information of the wake-up word in the wake-up word recognition process, on the basis of the above-mentioned embodiments, in the embodiment of the present application, a wake-up word recognition method is provided. Figure 9 It is a process schematic diagram of a wake-up word recognition method provided in the embodiment of the present application, and the process includes the following steps:

[0100] S901: Receive a target speech to be recognized.

[0101] The wake-up word recognition method provided in the embodiment of the present application is applied to an electronic device, and the electronic device may be a mobile terminal, a PC, or a smart home appliance supporting a voice function, such as an air conditioner, a TV, and the like.

[0102] After the wake-up word recognition model training is completed, the wake-up word recognition can be carried out. In ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The embodiment of the invention provides a wake-up word recognition model training method and device, a wake-up word recognition method and device and a medium. In the wake-up word recognition model training process, a target label corresponding to each voice frame in a sample voice is determined according to the pre-stored receptive field length and the length of a wake-up word in the sample voice; inputting each voice frame of the sample voice into the wake-up word recognition model to obtain a recognition tag of each voice frame output by the wake-up word recognition model, and determining a loss value corresponding to each voice frame according to the recognition tag of each voice frame of the sample voice and the target tag, therefore, the wake-up word recognition model is trained according to the loss value corresponding to each voice frame, a more accurate wake-up word recognition result is output, and dependence on context information of the wake-up word based on the wake-up word recognition model is avoided.

Description

technical field [0001] The present application relates to the technical field of speech recognition, and in particular to a wake-up word recognition model training, recognition method, device and medium. Background technique [0002] At present, wake-up word recognition mainly relies on deep learning technology, that is, the features in the voice information are input into the neural network model, and the neural network model is trained based on the wake-up word features in the input features. Because people speak at different speeds, the range that the neural network model "sees" is not only the wake-up word itself, but also the context information around it. The range that includes the wake-up word and its context information is called the range of the neural network model. The receptive field, these contextual information is an important factor affecting the performance of the neural network model in recognizing wake words. For example, for a neural network model that s...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G10L15/06G10L15/22
CPCG10L15/063G10L15/22G10L2015/223
Inventor 李程帅周全徐涛
Owner 青岛信芯微电子科技股份有限公司