Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Method of setting up speech recognition model, speech recognition method and corresponding device

A speech recognition model and speech recognition technology, applied in the field of speech search, can solve the problems of slow update speed, limit the volume of language model, reduce the search for new things and information, etc., and achieve the effect of high search and fast real-time dynamic update

Active Publication Date: 2014-06-18
BEIJING BAIDU NETCOM SCI & TECH CO LTD
View PDF5 Cites 28 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, although the above-mentioned first method of the prior art has a faster speech recognition speed, it often takes a lot of time and computer memory, which limits the size of the language model that can be used.
And because the language layer and the acoustic layer are coupled together, each update of the language layer involves the update of the entire network, resulting in a very slow update speed, which greatly reduces the ability to search for new events and information
The recognition speed of the second method is slow, and the construction of two WFST networks leads to the update of the language layer involving the update of the two networks, and the update speed is also very slow, which also affects the ability to search for new things and information

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 of setting up speech recognition model, speech recognition method and corresponding device
  • Method of setting up speech recognition model, speech recognition method and corresponding device
  • Method of setting up speech recognition model, speech recognition method and corresponding device

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0060] figure 1 The flow chart of the method for establishing a speech recognition model provided by Embodiment 1 of the present invention, such as figure 1 As shown, the method mainly includes the following steps:

[0061] Step 101: The dictionary and the acoustic model information are fused to obtain an acoustic layer space network.

[0062] The purpose of this step is to establish an acoustic layer space network representing acoustic model information, which is used to organize all acoustic-related information content in speech recognition into a network connected by a large number of nodes that is easy for computer processing.

[0063] The resources required for the construction of the acoustic layer space network are dictionaries and acoustic model information, without any language model information.

[0064] Specifically, the method for constructing the acoustic layer space network specifically includes: after arranging the words in the dictionary, constructing a jump-...

Embodiment 2

[0078] image 3 The flow chart of the voice recognition method provided by Embodiment 2 of the present invention, such as image 3 As shown, the method may include the following steps:

[0079] Step 301: Acoustic feature extraction is performed on the input speech.

[0080] In this step, the acoustic feature extraction of the input speech can be performed in any way in the prior art, and there is no specific limitation here, such as the extraction of linear predictive cepstral coefficients (LPCC) and Mel frequency cepstral coefficients (MFCC) Wait.

[0081] Step 302: Based on the extracted acoustic features, search for nodes on the acoustic layer space network and the language layer network, and use the language model prediction network to cut the found nodes during the search process, and construct the optimal The decoding path serves as the recognition result of the input speech.

[0082] This step is the core content of speech recognition, in which the search for the ac...

Embodiment 3

[0095] Figure 5 The structure diagram of the device for establishing the speech recognition model provided by Embodiment 3 of the present invention, such as Figure 5 As shown, the device may include: an acoustic layer construction unit 500 , a language layer construction unit 510 and a prediction model construction unit 520 .

[0096] The acoustic layer construction unit 500 fuses the dictionary and the acoustic model information to obtain the acoustic layer space network. The resources required for the construction of the acoustic layer space network are dictionaries and acoustic model information, without any language model information.

[0097] Figure 6 An implementation of the acoustic layer construction unit 500 is shown in , as Figure 6 As shown, it may specifically include: a first construction subunit 501 , a second construction subunit 502 and an optimization subunit 503 .

[0098] After arranging the words in the dictionary, the first construction subunit 501...

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 method of setting up a speech recognition model, a speech recognition method and a corresponding device; a dictionary and acoustics model information are combined to obtain an acoustics layer space network; language model information is abstracted in a finite-state machine and optimized to obtain a language layer network; the acoustics layer space network and the language layer network can form a language model prediction network; the acoustics layer space network, the language layer network, and the language model prediction network form a speech recognition model. The speech recognition model can isolate a coupling relation between speech layer information and acoustics layer information, so independent network can be formed to realize fast and dynamical update of language layer information; speech research realized according to the speech recognition model has higher newly generated events and information searching capacity.

Description

【Technical field】 [0001] The invention relates to voice search technology in the field of computer applications, in particular to a method for establishing a voice recognition model, a voice recognition method and a corresponding device. 【Background technique】 [0002] Voice search is a novel search technology emerging recently, which brings a brand-new search experience to the vast number of Internet users. Users can use voice to search and query. Voice search uses voice recognition technology to recognize the user's voice content into text, and then uses text search technology to return the search results to the user. It can be seen that voice recognition is the key core link in voice search. [0003] The existing speech recognition technology mainly adopts the following technologies: [0004] First, the speech recognition system based on the weighted finite state machine (WFST), using WFST technology to integrate the acoustic layer information and language layer informat...

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): G10L15/06G10L15/08
Inventor 贾磊钱胜万广鲁
Owner BEIJING BAIDU NETCOM SCI & TECH CO LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com