Eureka AIR delivers breakthrough ideas for toughest innovation challenges, trusted by R&D personnel around the world.

Speech recognition model word segmentation training method, system, mobile terminal and storage medium

A speech recognition model and training method technology, applied in speech recognition, speech analysis, instruments, etc., to achieve high recognition performance and reduce impact

Active Publication Date: 2022-05-17
XIAMEN KUAISHANGTONG TECH CORP LTD
View PDF4 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] The technical problem to be solved by the embodiments of the present invention is how to reduce manual participation as much as possible without using external resources, and improve word segmentation performance so as to further improve the performance of the speech recognition language model

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
  • Speech recognition model word segmentation training method, system, mobile terminal and storage medium
  • Speech recognition model word segmentation training method, system, mobile terminal and storage medium
  • Speech recognition model word segmentation training method, system, mobile terminal and storage medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0044] see figure 1 , is a flowchart of the speech recognition model word segmentation training method provided by the first embodiment of the present invention, including steps:

[0045] Step S10, collecting locally stored text corpus, setting the total number of word groups, and calculating the combination frequency between each current word and the next word in the text corpus;

[0046] Among them, preferably, due to the large number of words in the Chinese dictionary, even commonly used words also have more than 4000 words, and the number of words formed by these words increases exponentially. Therefore, it is necessary to collect a large amount of text corpus with a wide range of topics to The language model based on N-grams trained through the text corpus is credible, otherwise there will be great prejudice and poor generalization ability, resulting in the final speech recognition effect not meeting the requirements;

[0047] Specifically, in this step, the text corpus ...

Embodiment 2

[0057] see figure 2 , is a flow chart of the speech recognition model word segmentation training method provided by the second embodiment of the present invention, including steps:

[0058] Step S11, collecting locally stored text corpus, performing special character recognition on the text corpus, and deleting characters according to the recognition result;

[0059] Among them, there may be characteristic characters stored in the collected text corpus, but because special characters generally do not exist in Chinese speech, in order to ensure the accuracy of subsequent word-to-word combination calculations, special characters need to be removed, so , before performing the subsequent word formation step, by performing special character recognition, so as to improve the accuracy of the speech recognition model word segmentation training method;

[0060] Step S21, performing punctuation recognition on the text corpus, and converting the recognized punctuation marks into line b...

Embodiment 3

[0079] see image 3 , is a schematic structural diagram of the speech recognition model word segmentation training system 100 provided by the third embodiment of the present invention, including: a frequency calculation module 10, a word group control module 11, a dictionary merging module 12 and a word segmentation training module 13, wherein:

[0080] Frequency calculation module 10 is used to collect the text corpus of local storage, set the total number of group words, and calculate the combination frequency between each current word and the next word in the text corpus respectively;

[0081] Word grouping control module 11, for when judging that described combination frequency is greater than the first frequency threshold value, carry out word grouping with described current character and described next character, and the word mark after grouping word as a whole, until final If the long combination frequency is less than or equal to the second frequency threshold, word fo...

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 present invention is applicable to the technical field of speech recognition, and provides a speech recognition model word segmentation training method, system, mobile terminal and storage medium. The method includes: collecting locally stored text corpus, setting the total number of word groups, and calculating the sum of each current text and Combination frequency between the next text; when it is judged that the combination frequency is greater than the first frequency threshold, the current text and the next text are combined and marked as a whole, until the longest combination frequency is less than or equal to the second frequency threshold, stop forming words , to obtain a word dictionary; merge the word dictionary with the original dictionary, and perform word segmentation on the text corpus according to the merged dictionary to obtain word segmentation data; train the speech recognition model according to the word segmentation data. The present invention expands the original dictionary by constructing a word-group dictionary, so that there is no need to additionally collect training corpus for substring labeling learning, and it is not necessary to perform manual word segmentation in advance or use other word segmentation algorithms for word segmentation.

Description

technical field [0001] The invention belongs to the technical field of speech recognition, and in particular relates to a speech recognition model word segmentation training method, system, mobile terminal and storage medium. Background technique [0002] In Chinese, a character is the smallest unit, but in many cases, words combined with characters have stronger semantics. Therefore, in most applications, it is more reasonable to regard words as a whole, but in Chinese There is no space for semantic segmentation like in English, which makes the computer unable to distinguish the boundaries between words and words, and words and words, thus giving birth to the research field of Chinese word segmentation. Chinese word segmentation is currently an essential processing step in the fields of Chinese search engines, Chinese natural language processing, including Chinese speech recognition language model modeling. [0003] Chinese word segmentation methods can be divided into rul...

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 Patents(China)
IPC IPC(8): G10L15/06
CPCG10L15/063G10L2015/0633
Inventor 徐敏李稀敏肖龙源蔡振华刘晓葳王静
Owner XIAMEN KUAISHANGTONG TECH CORP LTD
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
Eureka Blog
Learn More
PatSnap group products