CN109101499A - Artificial intelligent voice learning method neural network based - Google Patents

Artificial intelligent voice learning method neural network based Download PDF

Info

Publication number
CN109101499A
CN109101499A CN201810874085.4A CN201810874085A CN109101499A CN 109101499 A CN109101499 A CN 109101499A CN 201810874085 A CN201810874085 A CN 201810874085A CN 109101499 A CN109101499 A CN 109101499A
Authority
CN
China
Prior art keywords
word
translation
chinese
foreign language
association
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201810874085.4A
Other languages
Chinese (zh)
Other versions
CN109101499B (en
Inventor
王大江
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Zhongke Huilian Technology Co ltd
Original Assignee
Individual
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Individual filed Critical Individual
Priority to CN201810874085.4A priority Critical patent/CN109101499B/en
Publication of CN109101499A publication Critical patent/CN109101499A/en
Application granted granted Critical
Publication of CN109101499B publication Critical patent/CN109101499B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/40—Processing or translation of natural language
    • G06F40/58—Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/40—Processing or translation of natural language
    • G06F40/42—Data-driven translation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/26—Speech to text systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Machine Translation (AREA)

Abstract

In order to further increase the efficiency and accuracy when translation on line, the present invention provides a kind of artificial intelligent voice learning methods neural network based, including (1) based on using the progress foreign language-Chinese text artificial intelligence study of self-adaptive growth type neural network；(2) voiced translation of foreign language-Chinese is carried out.The present invention can be by big data foreign language-Chinese bilingual dictionary that machine learning obtains, semantic and context matching is carried out based on 6 rank depth probability analysis methods, to which way compared with prior art reduces the operand more than 40% or more, translation efficiency is improved while ensuring translation accuracy.

Description

Artificial intelligent voice learning method neural network based
Technical field
The present invention relates to voice control technology fields, more particularly, to a kind of artificial intelligence language neural network based Sound learning method.
Background technique
With the development of science and technology and economic globalization, either in daily life or sphere of learning is linked up, Translation on line exchange has existed more and more demands.Although having existed simultaneous interpretation, portable machine interpreting equipment etc. It comes into being, but in the usage scenario of the meeting or classroom for being related to professional domain etc., the accuracy of conventional machines interpreting equipment And the efficiency of simultaneous interpretation personnel makes people worried.Especially when certain side's word speed is very fast, machine translation will be difficult to be competent at, And simultaneous interpretation personnel then need to repeat the language being not kept pace with using the mode reaffirmed, to give some usage scenario bands Come the experience having some setbacks.
In order to meet the efficiency of translation on line and the raising demand of accuracy simultaneously, application No. is CN201710203439.8 Chinese invention patent application disclose a kind of multilingual intelligence pretreatment real-time statistics machine translation system, comprising: receive mould Block, preprocessing module, machine translation module and post-processing module.The receiving module includes text language receiving module and voice Recognition result receiving module；The preprocessing module includes Text Pretreatment module and speech recognition result preprocessing module；Machine Device translation module, the machine translation module are used to learn the translation of phrase-for-phrase, and to by preprocessing module processing Phrase finds out corresponding translation phrase, and phrase is connected into complete sentence；Post-processing module, the post-processing module are used It is handled in doing word lattice gauge, capital and small letter standardization and format specificationization to translation result, it is made to be more nearly target language The communicative habits of speech, and exported as final result.However, this system, which solves dynamics for the above-mentioned drawback of the prior art, to be had Limit.
Summary of the invention
In order to further increase the efficiency and accuracy when translation on line, the present invention provides a kind of neural network based Artificial intelligent voice learning method, comprising:
(1) based on using the progress foreign language-Chinese text artificial intelligence study of self-adaptive growth type neural network, the step Include:
(10) word library is established；
(20) voice prediction model is established；
(2) voiced translation of foreign language-Chinese is carried out.
Further, the step (2) includes: that (10) establish word library；
(20) voice prediction model is established；
(30) voice of input is converted into text；
(40) according to the word library and voice prediction model, translation text is determined.
Further, the step (10) include: according to dictionary establish foreign language word and it is corresponding with the foreign language word in The first association between the word of literary meaning, wherein with the first ordinal position mark in dictionary when the translation of Chinese word is multiple The Chinese translation word of knowledge is main Chinese translation word and the Chinese translation word of ordinal position is translated as secondary Chinese later Cliction language.
Further, the step (20) includes:
(201) word cutting is carried out according to foreign language article and obtains foreign language word and according to the Chinese translation word of the foreign language article, Establish the second level word connected after foreign language word and Chinese translation word and the Chinese translation word second is associated with；
(202) the first association is associated with second and is indexed；
Further, the step (201) includes: that machine learning is carried out in a manner of unsupervised learning according to foreign language article.
Further, the step (201) includes: to carry out machine to foreign language article and its translation using stochastic gradient descent method Device study.
Further, the step (202) includes:
It is associated as major key with first, the information relevant to the first association occurred from the second association is indexed.
Further, the information relevant to the first association that is described to be associated as major key with first, occurring from the second association It is indexed and includes:
(2021) major key information determines: assuming that English word Ei corresponds to main Chinese translation word Cj in the first association；And According to the second association, the second level word connected after word Cj constitutes set { Sm, Pm }, then using word Cj as major key, wherein Pm It is that word Sm appears in probability of the Cj later as the second level word connected, i, j and m are the natural number since 1；
(2022) probability of word Cj appearance is defined:
p(Sm|Cj)=χgh(pj),
Wherein
M=1,2,3,4,5,6；AndFor with forMean value, ξmFor The m rank diagonal matrix of variance,
(2023) according to Probability p (Sm|Cj) determine when word Cj takes current meaning and the matching degree of context:
It calculatesWherein p ' indicates to carry out difference to p；
It calculatesWhether less than the first preset threshold: when small Yu Shi, the position for determining that j is indicated in Cj meet the corresponding context of Ei, otherwise enable j=j+1, step (2022) are jumped to, if j Its maximum value is reached by traversal, then enable j=1 and is continued step (2024), u and v are natural number；
(2024) when correction second level word of the Sm as the connecting of Cj and the matching degree of context:
It calculatesWhether less than second Preset threshold: it when being less than, determines that the second level word of connecting of the Sm as Cj meets context, otherwise enables m=m+1, jump to step Suddenly (2022) enable m=1 if m reaches its maximum value by traversal.
Further, the step (30) includes:
(301) linear analysis is made to primary speech signal, obtains weighting cepstral coefficients as speech characteristic parameter；
(302) speech model is obtained according to speech characteristic parameter；
(303) voice to be identified is matched with speech model, is searched for using frame synchronous network, to each frame voice An output probability value is determined for different models, is retained mulitpath in the matching process, is finally recalled matching result out；
(304) matched result is distributed with state duration and optimal path probability distribution carries out differentiating that rejection is fallen to know Voice except other range, obtains correct recognition result.
Further, the step (40) includes:
Voice is generated based on STT technology, using Chinese translation word.
The beneficial effect comprise that big data foreign language-Chinese bilingual dictionary that machine learning obtains, base can be passed through Semantic and context matching is carried out in 6 rank depth probability analysis methods, so that way compared with prior art reduces and is more than 40% or more operand, improves translation efficiency while ensuring translation accuracy.
Detailed description of the invention
Fig. 1 shows the flow chart of the method for the present invention.
Specific embodiment
As shown in Figure 1, preferred embodiment in accordance with the present invention, the present invention provides a kind of artificial intelligence neural network based Energy phonetic study method, comprising:
(1) based on using the progress foreign language-Chinese text artificial intelligence study of self-adaptive growth type neural network, the step Include:
(10) word library is established；
(20) voice prediction model is established；
(2) voiced translation of foreign language-Chinese is carried out.
Preferably, the step (2) includes:
(30) voice of input is converted into text；
(40) according to the word library and voice prediction model, translation text is determined.
Preferably, the step (10) includes: to establish foreign language word and Chinese corresponding with the foreign language word according to dictionary The first association between the word of meaning, wherein with the first ordinal position mark in dictionary when the translation of Chinese word is multiple Chinese translation word be main Chinese translation word and the Chinese translation word of ordinal position is as secondary Chinese translation later Word.
Preferably, the step (20) includes:
(201) word cutting is carried out according to foreign language article and obtains foreign language word and according to the Chinese translation word of the foreign language article, Establish the second level word connected after foreign language word and Chinese translation word and the Chinese translation word second is associated with；
(202) the first association is associated with second and is indexed；
Preferably, the step (201) includes: that machine learning is carried out in a manner of unsupervised learning according to foreign language article.
Preferably, the step (201) includes: to carry out machine to foreign language article and its translation using stochastic gradient descent method Study.
Preferably, the step (202) includes:
It is associated as major key with first, the information relevant to the first association occurred from the second association is indexed.The master Key is the major key for indicating the database of text corresponding relationship of foreign language and Chinese.
Preferably, described to be associated as major key with first, from the second association the information relevant to the first association that occurs into Line index includes:
(2021) major key information determines: assuming that English word Ei corresponds to main Chinese translation word Cj in the first association；And According to the second association, the second level word connected after word Cj constitutes set { Sm, Pm }, then using word Cj as major key, wherein Pm It is that word Sm appears in probability of the Cj later as the second level word connected, i, j and m are the natural number since 1；
(2022) probability of word Cj appearance is defined:
p(Sm|Cj)=χgh(pj),
Wherein
M=1,2,3,4,5,6；AndFor with forMean value, ξmFor The m rank diagonal matrix of variance,
(2023) according to Probability p (Sm|Cj) determine when word Cj takes current meaning and the matching degree of context:
It calculatesWherein p ' indicates to carry out difference to p；
It calculatesWhether less than the first preset threshold: when small Yu Shi, the position for determining that j is indicated in Cj meet the corresponding context of Ei, otherwise enable j=j+1, step (2022) are jumped to, if j Its maximum value is reached by traversal, then enable j=1 and is continued step (2024), u and v are natural number；
(2024) when correction second level word of the Sm as the connecting of Cj and the matching degree of context:
It calculatesWhether less than second Preset threshold: it when being less than, determines that the second level word of connecting of the Sm as Cj meets context, otherwise enables m=m+1, jump to step Suddenly (2022) enable m=1 if m reaches its maximum value by traversal.
Preferably, the step (30) includes:
(301) linear analysis is made to primary speech signal, obtains weighting cepstral coefficients as speech characteristic parameter；
(302) speech model is obtained according to speech characteristic parameter；
(303) voice to be identified is matched with speech model, is searched for using frame synchronous network, to each frame voice An output probability value is determined for different models, is retained mulitpath in the matching process, is finally recalled matching result out；
(304) matched result is distributed with state duration and optimal path probability distribution carries out differentiating that rejection is fallen to know Voice except other range, obtains correct recognition result.
Preferably, the step (40) includes:
Based on STT technology, i.e., Speech to Text technology, utilize Chinese translation word to generate voice.
The above-described embodiments merely illustrate the principles and effects of the present invention, and is not intended to limit the present invention.It is any ripe The personage for knowing this technology all without departing from the spirit and scope of the present invention, carries out modifications and changes to above-described embodiment.Cause This, institute is complete without departing from the spirit and technical ideas disclosed in the present invention by those of ordinary skill in the art such as At all equivalent modifications or change, should be covered by the claims of the present invention.

Claims (10)

1. a kind of artificial intelligent voice learning method neural network based, comprising:
(1) based on using the progress foreign language-Chinese text artificial intelligence study of self-adaptive growth type neural network, the step packet It includes:
(10) word library is established；
(20) voice prediction model is established；
(2) voiced translation of foreign language-Chinese is carried out.
2. the method according to claim 1, wherein the step (2) includes:
(30) voice of input is converted into text；
(40) according to the word library and voice prediction model, translation text is determined.
3. the method according to claim 1, wherein the step (10) includes: to establish outer cliction according to dictionary The first association between language and the word of Chinese meaning corresponding with the foreign language word, wherein when the translation of Chinese word is multiple With the Chinese translation word of the first ordinal position mark in dictionary for main Chinese translation word and later in ordinal position Literary translation word is as secondary Chinese translation word.
4. according to the method described in claim 3, it is characterized in that, the step (20) includes:
(201) word cutting is carried out according to foreign language article and obtains foreign language word and according to the Chinese translation word of the foreign language article, foundation The second of the second level word connected after foreign language word and Chinese translation word and the Chinese translation word is associated with；
(202) the first association is associated with second and is indexed.
5. according to the method described in claim 4, it is characterized in that, the step (201) includes: according to foreign language article with no prison It superintends and directs mode of learning and carries out machine learning.
6. according to the method described in claim 4, it is characterized in that, the step (201) includes: using stochastic gradient descent method Machine learning is carried out to foreign language article and its translation.
7. according to the method described in claim 4, it is characterized in that, the step (202) includes:
It is associated as major key with first, the information relevant to the first association occurred from the second association is indexed.
8. going out from the second association the method according to the description of claim 7 is characterized in that described be associated as major key with first Existing information relevant to the first association, which is indexed, includes:
(2021) major key information determines: assuming that English word Ei corresponds to main Chinese translation word Cj in the first association；And according to Second association, the second level word that word Cj is connected later constitute set { Sm, Pm }, then using word Cj as major key, wherein Pm is word Language Sm appears in probability of the Cj later as the second level word connected, and i, j and m are the natural number since 1；
(2022) probability of word Cj appearance is defined:
p(Sm|Cj)=χgh(pj),
Wherein
For withFor mean value, ξmFor the m rank diagonal matrix of variance,
(2023) according to Probability p (Sm|Cj) determine when word Cj takes current meaning and the matching degree of context:
It calculatesWherein p ' indicates to carry out difference to p；
It calculatesWhether less than the first preset threshold: when being less than, The position for determining that j in Cj is indicated meets the corresponding context of Ei, otherwise enables j=j+1, jumps to step (2022), if j by time It goes through and reaches its maximum value, then enable j=1 and continue step (2024), u and v are natural number；
(2024) when correction second level word of the Sm as the connecting of Cj and the matching degree of context:
It calculatesIt is whether default less than second Threshold value: it when being less than, determines that the second level word of connecting of the Sm as Cj meets context, otherwise enables m=m+1, jump to step (2022), if m reaches its maximum value by traversal, m=1 is enabled.
9. the method according to claim 1, wherein the step (30) includes:
(301) linear analysis is made to primary speech signal, obtains weighting cepstral coefficients as speech characteristic parameter；
(302) speech model is obtained according to speech characteristic parameter；
(303) voice to be identified is matched with speech model, is searched for using frame synchronous network, each frame voice is directed to Different models determines an output probability value, retains mulitpath in the matching process, finally recalls matching result out；
(304) matched result is distributed with state duration and optimal path probability distribution carries out differentiating that rejection is fallen to identify model Voice except enclosing obtains correct recognition result.
10. the method according to claim 1, wherein the step (40) includes:
Voice is generated based on STT technology, using Chinese translation word.
CN201810874085.4A 2018-08-02 2018-08-02 Artificial intelligence voice learning method based on neural network Active CN109101499B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810874085.4A CN109101499B (en) 2018-08-02 2018-08-02 Artificial intelligence voice learning method based on neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810874085.4A CN109101499B (en) 2018-08-02 2018-08-02 Artificial intelligence voice learning method based on neural network

Publications (2)

Publication Number Publication Date
CN109101499A true CN109101499A (en) 2018-12-28
CN109101499B CN109101499B (en) 2022-12-16

Family

ID=64848278

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810874085.4A Active CN109101499B (en) 2018-08-02 2018-08-02 Artificial intelligence voice learning method based on neural network

Country Status (1)

Country Link
CN (1) CN109101499B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6789057B1 (en) * 1997-01-07 2004-09-07 Hitachi, Ltd. Dictionary management method and apparatus
CN105183720A (en) * 2015-08-05 2015-12-23 百度在线网络技术（北京）有限公司 Machine translation method and apparatus based on RNN model
CN107102990A (en) * 2016-02-19 2017-08-29 株式会社东芝 The method and apparatus translated to voice
CN107315741A (en) * 2017-05-24 2017-11-03 清华大学 Bilingual dictionary construction method and equipment

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6789057B1 (en) * 1997-01-07 2004-09-07 Hitachi, Ltd. Dictionary management method and apparatus
CN105183720A (en) * 2015-08-05 2015-12-23 百度在线网络技术（北京）有限公司 Machine translation method and apparatus based on RNN model
CN107102990A (en) * 2016-02-19 2017-08-29 株式会社东芝 The method and apparatus translated to voice
CN107315741A (en) * 2017-05-24 2017-11-03 清华大学 Bilingual dictionary construction method and equipment

Also Published As

Publication number Publication date
CN109101499B (en) 2022-12-16

Similar Documents

Publication Publication Date Title
CN109359293B (en) Mongolian name entity recognition method neural network based and its identifying system
CN110717018A (en) Industrial equipment fault maintenance question-answering system based on knowledge graph
CN111709242B (en) Chinese punctuation mark adding method based on named entity recognition
CN109255113A (en) Intelligent critique system
CN112101010B (en) Telecom industry OA office automation manuscript auditing method based on BERT
CN109949799B (en) Semantic parsing method and system
CN111933127A (en) Intention recognition method and intention recognition system with self-learning capability
CN111489746A (en) Power grid dispatching voice recognition language model construction method based on BERT
CN113569553A (en) Sentence similarity judgment method based on improved Adaboost algorithm
CN110853669B (en) Audio identification method, device and equipment
CN109376355A (en) English word and sentence screening technique, device, storage medium and electronic equipment
Liu et al. AI recognition method of pronunciation errors in oral English speech with the help of big data for personalized learning
CN111090726A (en) NLP-based electric power industry character customer service interaction method
CN112528003B (en) Multi-item selection question-answering method based on semantic sorting and knowledge correction
CN114003700A (en) Method and system for processing session information, electronic device and storage medium
Chatziagapi et al. Audio and ASR-based filled pause detection
CN111046663A (en) Intelligent correction method for Chinese form
CN109101499A (en) Artificial intelligent voice learning method neural network based
CN115376547A (en) Pronunciation evaluation method and device, computer equipment and storage medium
CN109062911A (en) A kind of artificial intelligent voice modeling method
CN109241539A (en) The update method of machine learning artificial intelligence translation database
Wang et al. Phonological modeling of mispronunciation gradations in L2 English speech of L1 Chinese learners
CN105895075B (en) Improve the method and system of synthesis phonetic-rhythm naturalness
Zhang [Retracted] English Speech Recognition System Model Based on Computer‐Aided Function and Neural Network Algorithm
Zheng [Retracted] An Analysis and Research on Chinese College Students’ Psychological Barriers in Oral English Output from a Cross‐Cultural Perspective

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
TA01 Transfer of patent application right
TA01 Transfer of patent application right

Effective date of registration: 20221129

Address after: 100089 305, Zone 2, Building 9, No. 8, Dongbei Wangxi Road, Haidian District, Beijing

Applicant after: BEIJING ZHONGKE HUILIAN TECHNOLOGY Co.,Ltd.

Address before: No. 16, Elbow Group, Fruit Village, Liuxi Miao Township, Yiliang County, Zhaotong City, Yunnan Province 657600

Applicant before: Wang Dajiang

GR01 Patent grant
GR01 Patent grant