Keyword model construction method, decoding network construction method, detection method and related equipment

A construction method and decoding network technology, applied in speech analysis, speech recognition, instruments, etc., can solve problems such as poor performance, poor decoding network performance, poor keyword detection effect, etc., to achieve less voice samples and solve sample problems The effect of difficulty in collection and reduction of time and labor costs

Pending Publication Date: 2020-08-14
合肥讯飞数码科技有限公司
View PDF14 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] However, there is currently no scheme that can construct a keyword model with better performance. It is understandable that if the performance of the constructed keyword model is not good, it will lead to poor performance of the decoding network, which will lead to the detection of keywords. not effectively

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
  • Keyword model construction method, decoding network construction method, detection method and related equipment
  • Keyword model construction method, decoding network construction method, detection method and related equipment
  • Keyword model construction method, decoding network construction method, detection method and related equipment

Examples

Experimental program
Comparison scheme
Effect test

no. 1 example

[0074] See figure 1 , Shows a schematic flow chart of the keyword model construction method provided in this embodiment, which may include:

[0075] Step S101: Extract model information corresponding to each keyword in the keyword list to be detected from the acoustic model of the source domain.

[0076] Among them, the list of keywords to be detected is composed of keywords to be detected. The list of keywords to be detected may include one keyword or multiple keywords. The keywords contained in the list of keywords to be detected and specific The testing requirements of the application are related.

[0077] In this embodiment, the model information corresponding to each keyword in the keyword list to be detected includes the model information of each triphone in the triphone sequence string of the keyword.

[0078] In addition, it should be noted that the acoustic model of the source domain is an existing acoustic model, which is obtained by training with a large number of samples, ...

no. 2 example

[0092] This embodiment introduces step S101 and step S102 in the foregoing embodiment.

[0093] The implementation process of "step S101, extracting the model information corresponding to each keyword in the keyword list to be detected from the acoustic model of the source domain" may include:

[0094] Step a1: Obtain the triphone sequence string of each keyword in the keyword list.

[0095] Among them, the triphone sequence string of any keyword is a sequence string composed of all the triphones of the keyword in order.

[0096] Step a2: According to the triphone sequence string of each keyword in the keyword list, extract the model information corresponding to each keyword in the keyword list from the acoustic model of the source domain.

[0097] Specifically, for each keyword in the keyword list: extract the model information of each triphone in the triphone sequence string of the keyword from the acoustic model of the source domain; The model information of the triphone is used as ...

no. 3 example

[0105] On the basis of the keyword model construction method provided in the above embodiment, this embodiment provides a decoding network construction method, please refer to figure 2 , Shows a schematic flow chart of the decoding network construction method, which may include:

[0106] Step S201: Construct a keyword model of the target domain for the list of keywords to be detected.

[0107] For the specific implementation process of step S201, reference may be made to the above-mentioned first embodiment and the second embodiment, and this embodiment will not be repeated here.

[0108] Step S202: Construct a garbage model of the target domain.

[0109] Among them, the garbage model of the target domain (ie, the Filler model) is used to absorb non-keywords.

[0110] Specifically, using unlabeled speech samples randomly sampled from the target domain, an unsupervised maximum expectation algorithm (EM) is used to learn a garbage model for absorbing non-keywords, and the garbage model m...

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 invention provides a keyword model construction method, a decoding network construction method, detection method and related equipment, and the keyword model construction method can comprise the steps: extracting the model information corresponding to each keyword in a to-be-detected keyword list from an acoustic model of a source domain; generating the keyword model of the source domain according to the model information corresponding to each keyword in the keyword list; and performing training to obtain the keyword model of the target domain by using the voice sample of each keyword in the keyword list and taking the source domain keyword model as a basic model. According to the keyword model construction method provided by the invention, the keyword model with better performance canbe constructed by using a small number of keyword voice samples.

Description

Technical field [0001] This application relates to the technical field of keyword detection, in particular to a keyword model construction method, a decoding network construction method, a keyword detection method, device, equipment and storage medium. Background technique [0002] Keyword detection technology refers to the detection of designated keywords from the voice data, and if the designated keywords are detected, the location of the detected designated keywords in the voice data is determined. [0003] The general idea of ​​the current keyword detection scheme is to construct a keyword model for the keyword model to be detected, build a decoding network on this basis, and use the constructed decoding network for keyword detection. It should be noted that the key to the above keyword detection scheme is to construct a keyword model with better performance. [0004] However, there is currently no solution that can construct a keyword model with better performance. It is unders...

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): G10L15/26G10L15/02G10L15/06G10L15/08
CPCG10L15/26G10L15/02G10L15/063G10L15/083G10L2015/025G10L2015/0631
Inventor 方磊杨军杨帆方昕方四安
Owner 合肥讯飞数码科技有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap