Method and system for improving accuracy of speech recognition

A speech recognition and accuracy technology, applied in speech recognition, speech analysis, instruments, etc., can solve problems such as no increase in probability, uncorrectable person name context recognition errors, uncorrectable context content recognition errors, etc., to improve the recognition accuracy. Effect

Active Publication Date: 2014-07-02
讯飞医疗科技股份有限公司
View PDF13 Cites 57 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

It can be seen that the hot word incentive method only improves the cumulative historical probability of the hot word node, improves the distinction between the hot word node and other active nodes at the same time, but does not increase the probability

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
  • Method and system for improving accuracy of speech recognition
  • Method and system for improving accuracy of speech recognition
  • Method and system for improving accuracy of speech recognition

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0024] figure 2 It is a flow chart of a method for improving speech recognition accuracy provided by an embodiment of the present invention. The method mainly includes the following steps:

[0025] Step 201: Match candidate words in the path set obtained through speech decoding according to user preset information to obtain a new path set.

[0026] After receiving the voice information, perform voice decoding and obtain a path set, which can include several paths, and each path can contain several nodes, and every two adjacent nodes form an arc, and each arc corresponds to a Candidate words.

[0027] Since there are a large number of homophones in Chinese, such as ("晨旭" and "晨旭"), it is necessary to match the candidate words in the path set obtained by voice decoding according to the user preset information; A new path and a candidate word corresponding to the new path are added to the starting and ending nodes corresponding to the candidate word, so as to obtain a new path...

Embodiment 2

[0033] In order to introduce the present invention more specifically, below in conjunction with appendix Figure 3-5 The present invention is further described. Such as image 3 shown, including the following steps:

[0034] Step 301, decoding for the first time. This process is for regular speech decoding, such as Figure 4 Shown is the result of decoding for the first time when voice input "This is the room where Chen Xu used to live".

[0035] Firstly, the voice signal is received, at this time, the continuous voice signal can be digitally sampled as a series of discrete energy values ​​and stored in the data buffer. Further, the front-end noise reduction preprocessing can be performed on the collected original voice signal to eliminate the noise in the voice signal, so as to improve the signal processing capability of the subsequent system.

[0036] Then, extract the acoustic feature sequence in the speech signal, and use a fast search algorithm to roughly match the e...

Embodiment 3

[0063] Figure 6 A schematic diagram of a system for improving speech recognition accuracy provided by Embodiment 3 of the present invention, the system mainly includes:

[0064] The matching module 61 is used to match the candidate words in the path set obtained by voice decoding according to user preset information to obtain a new path set;

[0065] A correction module 62, configured to correct the language model probability of the candidate words in the new path set by using the classification language model constructed with the user preset information as an element;

[0066] The decoding module 63 is configured to perform speech decoding processing according to the corrected language model probability of the candidate words.

[0067] Wherein, the matching module 61 may include:

[0068] A conversion sub-module 611, configured to convert all candidate word strings in the user preset information and path sets into pinyin;

[0069] The fuzzy matching sub-module 612 is 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
Login to view more

PUM

No PUM Login to view more

Abstract

A method and an apparatus for improving accuracy of voice recognition. The method comprises: (201) matching, according to user preset information, a candidate word in a path set obtained by using voice decoding, and obtaining a new path set; (202) correcting the linguistic model probability of a candidate word in the new path set by using a classification-based linguistic model established by using the user preset information as elements; and (203) performing voice decoding processing according the corrected the linguistic model probability of the candidate word. By using the method, the recognition accuracy of specific user information and content of a context of the information is improved.

Description

technical field [0001] The invention relates to the field of speech signal processing, in particular to a method and system for improving speech recognition accuracy. Background technique [0002] With the popularity of voice input functions and applications on smart terminals such as mobile phones, users have more and more demands for voice input on smart terminals such as mobile phones, and the accuracy of user personalized information, especially the identification of contacts in the address book, is also increasing. put forward higher requirements. However, due to the limitations of language model training methods and recognition methods, traditional continuous speech recognition systems may not be able to provide correct word results for Chinese speech signals with polyphonic characters, especially in the recognition of personal name information. There are further restrictions: firstly, there are a large number of common Chinese names, and the continuous speech recogni...

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
IPC IPC(8): G10L15/18G10L15/26
CPCG10L15/22G10L15/18G10L15/183
Inventor 潘青华何婷婷余健鹿晓亮王智国胡国平胡郁刘庆峰
Owner 讯飞医疗科技股份有限公司
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