CN101739867B - Method for scoring interpretation quality by using computer - Google Patents
Method for scoring interpretation quality by using computer Download PDFInfo
- Publication number
- CN101739867B CN101739867B CN2008102266733A CN200810226673A CN101739867B CN 101739867 B CN101739867 B CN 101739867B CN 2008102266733 A CN2008102266733 A CN 2008102266733A CN 200810226673 A CN200810226673 A CN 200810226673A CN 101739867 B CN101739867 B CN 101739867B
- Authority
- CN
- China
- Prior art keywords
- model
- translation
- recognition result
- scoring
- tested
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000013519 translation Methods 0.000 claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000005516 engineering process Methods 0.000 claims abstract description 15
- 230000008569 process Effects 0.000 claims description 13
- 239000013598 vector Substances 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 6
- 244000188472 Ilex paraguariensis Species 0.000 claims description 3
- 238000013461 design Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 239000000463 material Substances 0.000 claims description 2
- 230000004927 fusion Effects 0.000 claims 1
- 238000012821 model calculation Methods 0.000 claims 1
- 238000012545 processing Methods 0.000 abstract description 8
- 238000011156 evaluation Methods 0.000 abstract description 7
- 230000000694 effects Effects 0.000 abstract description 3
- 230000007246 mechanism Effects 0.000 abstract description 3
- 238000012790 confirmation Methods 0.000 abstract 1
- 238000004422 calculation algorithm Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 239000000654 additive Substances 0.000 description 2
- 230000000996 additive effect Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 241000209202 Bromus secalinus Species 0.000 description 1
- 240000007711 Peperomia pellucida Species 0.000 description 1
- 235000012364 Peperomia pellucida Nutrition 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000001303 quality assessment method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 230000001755 vocal effect Effects 0.000 description 1
Images
Abstract
The invention discloses a method for scoring interpretation quality by using a computer, which comprehensively uses computer voice identification, voice pronunciation estimation, text translation quality confirmation technology to obtain the interpretation quality of a testee. The method comprises the following steps: establishing a database aiming at the characteristics of the tested pronunciation crowd; training the database by using a large vocabulary continuous speech acoustic model training platform to obtain an acoustic model; collecting corresponding expert knowledge and translated text linguistic data for each translation question to form a language model, a scoring model and a standard adjustment model required for identification; and finally, integrating the output result of a speech identifier and a linguistic processing mechanism to output the score of the interpretation quality of the testee, and providing feedback suggestion. The effect of machine evaluation already basically reaches the level close to expert evaluation; and meanwhile, the method also can provide some suggestions for testee pronunciation, vocabulary use and sentence pattern use during evaluating to guide the testee to correct.
Description
Technical field
The present invention proposes a kind of method of using computing machine that spoken translation quality is marked, belong to speech recognition, voice signal processing, computer-assisted language learning field.Be meant that specifically the utilization computing machine records to the personnel to be tested of Interpreter's topic type,, obtain the scoring and the feedback opinion of Interpreter's quality then through Computer Processing.
Background technology
The informationization of social life and economic globalization become increasingly conspicuous the importance of English.English has become most popular language in human lives's every field as one of most important information carrier.Many countries put in outstanding status all the important component part of English education as citizens'quality education, and with it in the basic education development strategy.
At present, the maximum verbal learning form of domestic employing is aspectant classroom instruction, because English teacher's shortage, the student is difficult to obtain man-to-man English study environment.In the face of this situation; The how tame unit in home and abroad has developed to use a computer and has carried out the system of Oral English Practice pronunciation diagnosis; These systems generally judge in short whether say whether pronounce fluent, and whether pronounce correct; But these technical ability are more elementary, are difficult to satisfy the increasingly high requirement of English learner.Because open topic type more can be understood personnel's to be tested thought process and language ability than the objective item of highly structural; In corresponding decision process, some more Useful Informations can be provided also; And the also less influence that receives exam-oriented education, cheats at one's exam of open topic type, have very positive meaning in English study with examining.In recent years, used a computer and carry out text composition and appraise and come into vogue, but also not have to occur system that spoken language is write a composition and marked.For this reason, the present invention is directed to a kind of topic type commonly used in the study of SET and living English, Interpreter's topic type; Carry out the computing machine scoring; This is compared with other software has had a very big raising, because except pronouncing fluently to the enunciator, pronounce accurately to judge; But also can the enunciator be passed judgment on, and can provide feedback opinion by vocabulary use, grammer use, whole translation situation.This is domestic or all is beyond example abroad.
Summary of the invention
Weak point to Oral English Practice examination and teaching the purpose of this invention is to provide the method that a kind of interactivity utilization computing machine good, that do not receive the time site limitation is marked to spoken translation quality.
For reaching said purpose, the present invention provides the utilization method that computing machine carries out the scoring of Oral English Practice translation quality, and the present invention is made up of two parts:
The training part: training department divides and comprises the training acoustic model, language model, and scoring model and scoring characteristic are to the standard adjustment model of final scoring.
The scoring part: utilization digital signal processing theory and Computer Language Processing technology are marked to Interpreter's recording of people to be tested, and the final scoring that makes machine provide is marked near expert's manual work as much as possible.
The present invention realizes through following technical scheme:
1) collects and sets up database to personnel's group characteristic to be tested;
2) on Basis of Database, use big vocabulary continuous speech acoustic training model platform, obtain acoustic model;
3) each translation topic type is collected corresponding expertise and cypher text language material, in order to generate the language model that recognition system needs;
4) each translation topic type is collected corresponding corpus of text, form Interpreter's assessment data storehouse, generate the model that scoring needs then;
5) use the expert of the existing examination paper knowledge of giving a mark to carry out the adjustment of machine marking standard, obtain the standard adjustment model, be used to improve system performance;
6) voice to personnel to be tested carry out speech recognition, obtain recognition result;
7) from each model bank, obtain Interpreter's quality score characteristic with recognition result;
8) comprehensive all scoring characteristics, through using the standard adjustment model, the voice that obtain personnel to be tested are finally marked, and provide feedback opinion.
Method of the present invention, can be used for using SET or Oral Training software Interpreter, picture talk, see the video spoken language scoring with similar topic type of speaking; Can apply to the right spoken quality scores of various language such as Chinese-English, English-Chinese, Chinese-French.
Beneficial effect of the present invention: judge the performance quality of computer examination, evaluating system, according to general in the world convention, be through judge manual work that scoring that computing machine provides and expert provide divide between correlativity judge.Owing to also have certain error between the expert, so generally adopt average mark that 3-5 experts appraise as people's work point.
On test data, to process statistics on 7 experts' the artificial divided data storehouse, what correlativity was minimum between the different experts has 0.80, and maximum has 0.92.The scoring that the spoken translation quality points-scoring system of computing machine provides reaches 0.90 with the correlativity of expert's average mark, can find out that therefore effect that machine is given a mark has reached the approaching level of giving a mark with the expert basically.
The machine evaluation and test not only can substitute expert's scoring, can also in the evaluation and test process, provide some suggestions of personnel's pronunciation to be tested, vocabulary use, sentence pattern use, instructs personnel to be tested to correct.
Description of drawings
Fig. 1 is the system chart of the methods of marking of the embodiment of the invention.
Fig. 2 is the process flow diagram of the embodiment of the invention.
Embodiment
Below in conjunction with accompanying drawing and specific embodiment the present invention is done and to describe in further detail.
Terminological interpretation
Speech recognition: speech recognition technology is to let machine pass through some relevant technology of artificial intelligence; The technology that voice signal changes the corresponding text of journey or orders, the related field of speech recognition technology comprises: signal Processing, pattern-recognition, theory of probability and information theory, sound generating mechanism and hearing mechanism, artificial intelligence etc.
BLEU: the present the most widely used method for automatically evaluating in MT evaluation field is BLEU (Bilingual Evaluation Understudy), just based on the typical case's representative in these class methods of n unit coupling, is proposed in 2002 by IBM.Similarly method also comprises the NIST method; This method is proposed and name by Unite States Standard (USS) and technical institute; It is on the basis of BLEU method; Taken all factors into consideration the weight of each n unit speech, composed to higher weight to embody the quantity of information that it is comprised for those occurrence number speech still less in reference translation.
The system chart of the methods of marking of the embodiment of the invention as shown in Figure 1; Interpreter's points-scoring system of realizing present embodiment runs on Microsoft's Window operating system, comprises Interpreter's expertise 1 Interpreter examinee's text 2, cypher text database 3, language model 4, scoring characteristic model 5, standard adjusting module 6, pronunciation extracting module 7, sound identification module 8, scoring characteristic extracting module 9, finally marks 10, acoustic model 11, pronunciation dictionary 12, feedback module 13 as a result.After personnel Interpreter voice to be tested are recorded; At first get into pronunciation extracting module 7 and extract the characteristic of dividing frame; This phonetic feature comprises energy and MFCC characteristic (the Mei Er cepstrum coefficient is promoted and the acoustic feature of derivation by people's auditory system achievement in research), and every frame is totally 39 dimensional features; Get into sound identification module 8 then,, select acoustic model 11, pronunciation dictionary 12, the language model 4 of use, utilize speech recognition engine that characteristic sequence is discerned according to personnel's to be tested sex and current Interpreter's topic type; The characteristic extracting module 9 of entering then to mark, this scoring characteristic extracting module 9 are extracted each corresponding characteristic of scoring characteristic models 5: the model answer match condition of current recognition result, BLEU Model Matching situation, scoring characteristic model 5 match condition; Get into standard adjusting module 6 at last; Scoring characteristic according to last module; Use standard adjustment model 6 and parameter obtain the personnel's to be tested final scoring 10 of voice, and feedback module 13 comes out some suggestion feedbacks that personnel's to be tested pronunciation, vocabulary use, sentence pattern use as a result.Comprise also the acoustic model that obtains 11, language model 4 and scoring characteristic model 5 are saved in the system that each existing model of only need reloading that uses does not need training pattern again.
Complete flow process is made up of two parts:
The training part: training department divides and comprises training acoustic model 11, language model 4, scoring characteristic model 5 and the scoring characteristic standard adjustment model 6 to final scoring 10.
The scoring part: utilization digital signal processing theory and Computer Language Processing technology are marked to Interpreter's recording of people to be tested, and the final scoring 10 that the system that makes provides is marked near expert's manual work as much as possible.
The present invention realizes through following technical scheme:
The method of Interpreter's quality score comprises a model training part, and training process may further comprise the steps:
(a) foundation is following to the speech database usage policy of personnel's group characteristic to be tested:
A.1 divide sex, seek a collection of agematched crowd according to correspondence personnel crowd's to be tested age distribution;
A.2 the principle according to the phoneme balance designs voice sample;
A.3 the designated person records according to the recording text, and voice are related with the foundation of corresponding text, and the voice document name is got identical filename with text, and the different files suffix like this can be rapidly from its corresponding content of text of voice document acquisition;
(b) use continuous speech acoustic training model platform training to obtain acoustic model 11:
B.1 every training utterance is extracted 12 rank Mel cepstrums, normalized energy and constitute that totally 13 dimensions obtain 39 dimensional features through single order and second order difference then as essential characteristic;
B.2 through forcing alignment algorithm and front and back to be estimated, obtain the sub-acoustic model of single-tone to algorithm;
B.3 pass through design decision tree and front and back to algorithm, training obtains the three-tone acoustic model;
B.4 through discrimination model training algorithm, train the three-tone acoustic model that obtains having discrimination information;
(c) use irrelevant N gram language model 4 interpolation of topic relevant N gram language model 4 and topic to improve Interpreter's quality assessment performance.Each translation topic type is collected corresponding corpus of text, in order to generate the language model 4 that identification needs:
C.1 each translation topic type is collected corresponding corpus of text training and is obtained corresponding ternary language model 4, and corpus of text is from cypher text database 3;
C.2 compatible for the identification that increases dialogue outlying content, adopt the training of common english text to obtain one three gram language model 4;
C.3 each translation topic type corresponding ternary language model 4 is through obtaining final three gram language model 4 of its appropriate translation topic type with general three gram language model, 4 interpolation;
(d) each translation topic type is collected corresponding corpus of text, forms Interpreter's assessment data storehouse, generate the characteristic model 5 that scoring needs then:
D.1 the foundation in Interpreter's model answer storehouse: each sense-group in the translation topic through a comprehensive N expert's cypher text, obtains the weight of model answer network He this network of current sense-group;
D.2 the foundation of Interpreter's model answer BLEU model bank: through a comprehensive N expert's cypher text, set up a BLEU model bank, cypher text is from cypher text database 3;
D.3 the foundation in Interpreter's text vector hierarchy model storehouse: Interpreter's expertise 1 of the cypher text through a comprehensive N expert and Interpreter's text 2 of the original examinee of each grade; Set up the lexicon grammar vector storehouse of each grade, Interpreter's expertise 1 of expert's cypher text and examinee Interpreter text at different levels 2 are from cypher text database 3;
(e) foundation in standards of grading adjustment model 6 standards adjustment storehouse: use the expert of the existing examination paper knowledge of giving a mark to carry out the adjustment of machine marking standard; Be used to improve system performance, every type of Interpreter's topic type is collected the M road by examinee's examination question, provide scoring through L expert; And with the manual work scoring of expert's average mark as the per pass topic; Introduce then based on forecast method, set up the mapping relations of each parameter, and the parameter that obtains is preserved to artificial mark; As the mapping model of the last machine scoring of each parameter, this process is the process of a standard adjustment.The effect that these parameters and model will play adjustment and proofread and correct.
Oral English Practice translation points-scoring system, this system comprises a scoring part, scoring process may further comprise the steps:
(f) speech recognition:
F.1 according to people's to be tested sex and topic type, select corresponding pronunciation dictionary, acoustic model, language model;
F.2 obtain recognition result through big vocabulary continuous speech voice recognition device;
F.3 recognizer is exported the confidence level of each word among the result simultaneously, starting and ending time, and the confidence level of this each phoneme of word, starting and ending time.Recognition result for recognizer output refuses to know function, because the language model of identification framework relatively is partial to the correct translation content, so for translation error, random situation about translating, occur sometimes being identified as correct translation vocabulary, causes the virtual height of scoring.Therefore, system has used and has refused to know function, uses the degree of confidence technology of discerning to suppress the generation of this situation.
(g) Interpreter's obtaining of characteristic of marking:
G.1 carry out weighting through the recognition result and the matching degree in indication model answer storehouse d.1 and obtain matching degree, weighting is obtained the characteristic of matching degree as system; The matching degree of said recognition result and model answer is: to each sense-group of translation topic type; At first the multidigit expert is translated content and generate a model answer network; In test process, use the technology of Dynamic matching; Search the content that Interpreter's recognition result and this sense-group mate most, with the characteristic of matching degree as system.
G.2 mate through recognition result and indication BLEU model d.2, calculate the matching degree under the BLEU pattern, the degree of the coupling under the BLEU pattern a characteristic as system; The BLEU model that said recognition result and model answer obtain; The matching degree of calculating under the BLEU pattern is: to each translation topic type, according to the model answer that expertise is put in order out, train the BLEU model; According to Interpreter's recognition result, calculate BLEU Model Matching degree then.
G.3 the lexicon grammar vector storehouse model that obtains through recognition result and each grade cypher text of indication d.3 carries out classification and differentiates; Go out matching degree according to classification situation and classification distance calculation, with the matching degree of classification situation and classification distance a characteristic as system; The lexicon grammar vector model that said recognition result and each grade cypher text obtain is: the text of translating out according to Interpreter's expertise 1 and each grade examinee 2; Generate the lexicon grammar vector of each grade; Then according to Interpreter's recognition result; Select a vector that matees most, provide matching degree simultaneously.
G.4 obtain current tested personnel's the fluent situation of partial through word speed, the voice quality that calculates recognition result, with the characteristic of the fluent situation of partial as system; Personnel's voice quality to be tested, pronunciation fluency are on model answer network matching content, to measure; With the correct translation irrelevant contents on do not pass judgment on, so voice quality and fluency are just to the personnel to be tested that possess certain Interpreter's ability.
The characteristic that comprehensive g mentions through the standard adjustment model of using e to obtain, obtains finally scoring, and provides feedback opinion.
Fig. 2 is the process flow diagram of present embodiment, and is as shown in the figure, may further comprise the steps:
Step 101 is carried out speech signal collection, and the analog-signal transitions that personnel to be tested are pronounced is a digital signal, and is kept in the computing machine.
Step 102 divides frame to handle to the digital signal of voice, and every frame extraction energy, and the MFCC parameter is totally 39 dimensional features.In the present embodiment, this characteristic adopts the prior art means to extract, and has instrument HCopy to extract 39 dimensional features in Hidden Makov Model Toolkit 3.4 versions like univ cambridge uk's issue.The additive method that person skilled was known under the extraction of certain above information was also available obtains.
Step 103 according to personnel's to be tested sex and current Interpreter's topic type, is selected the language model 4 and acoustic model 11 that use; Utilize speech recognition engine that characteristic sequence is discerned; Identification obtains a string word, generates the degree of confidence of each word simultaneously, the initial termination time; And the degree of confidence of each phoneme in the word, the initial termination time.Big vocabulary continuous speech recognition can obtain through the prior art means; As in Hidden Makov ModelToolkit 3.4 versions through univ cambridge uk issue instrument HVite or HDecode being arranged, " integrated prediction searching method that Chinese continuous speech is discerned " of the patent No. 00124971.1 that perhaps proposes through one of inventor carries out.
Step 105 according to the characteristic that step 104 obtains, uses corresponding forecast model to calculate final scoring.Forecast model can be that the method that linear regression forecast model, SVM forecast model, neural network prediction model or other affiliated person skilled are known realizes.
Step 106 adjusts this test according to above result, passes judgment on, and gives personnel to be tested some suggestion feedbacks that pronunciation, vocabulary use, sentence pattern use.
The above; Be merely the embodiment among the present invention, but protection scope of the present invention is not limited thereto, anyly is familiar with this technological people in the technical scope that the present invention disclosed; Can understand conversion or the replacement expected; All should be encompassed in of the present invention comprising within the scope, therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.
Claims (10)
1. method of using computing machine that spoken translation quality is marked; It is characterized in that: the speech recognition technology that comprehensively uses a computer, sound pronunciation assessment technology, text translation quality determination technology obtain Interpreter's quality of personnel to be tested; Provide feedback opinion simultaneously, it is following to comprise step:
Step 1: the speech database of collecting and establish object crowd age distribution characteristics to be tested;
Step 2: on the speech database basis, use big vocabulary continuous speech acoustic training model platform, obtain acoustic model;
Step 3: each translation topic type is collected corresponding expertise and cypher text language material, in order to generate N gram language model, scoring characteristic model and the standards of grading adjustment model that recognition system needs;
Step 4: use above-mentioned model, the output recognition result of integrated voice recognizer obtains Interpreter's quality score characteristic with recognition result from each model bank; Comprehensive all scoring characteristics through using the standard adjustment model, are exported the score of personnel Interpreter quality to be tested, and are provided feedback opinion.
2. method according to claim 1 is characterized in that: comprise also the acoustic model that obtains, N gram language model and scoring characteristic model are saved in the system that each existing model of only need reloading that uses does not need training pattern again.
3. method according to claim 1; It is characterized in that: set up to the speech database usage policy of object crowd age distribution characteristics to be tested following: the branch sex, seek a collection of suitable crowd according to object age distribution to be tested and also carry out voice recording according to the recording script of the phoneme balance that designs.
4. method according to claim 2 is characterized in that: the step that obtains language model is following:
Step 311: each translation topic type is collected the relevant corpus of text of topic, and training obtains corresponding N gram language model;
Step 312: adopt the training of common english text to obtain general N gram language model;
Step 313: the corresponding N gram language model of each translation topic type is through obtaining the final language model of its appropriate translation topic type with general N gram language model interpolation.
5. method according to claim 2 is characterized in that: obtain the scoring model and be to use a plurality of scoring Feature Fusion, it is following to obtain a plurality of scoring characteristic concrete steps:
Step 321: the matching degree weighting of recognition result and model answer network is obtained matching degree;
Step 322: the BLEU model with recognition result and model answer obtain, calculate the matching degree under the BLEU pattern;
Step 323: the lexicon grammar vector storehouse model that recognition result and each grade cypher text are obtained carries out classification to be differentiated, and goes out matching degree according to classification situation and classification distance calculation;
Step 324: calculate pronunciation accuracy, the pronunciation fluency of recognition result, obtain the current tested personnel situation of pronouncing.
6. method according to claim 5; It is characterized in that: the matching degree of said recognition result and model answer is: to each sense-group of translation topic type; At first the multidigit expert is translated content and generate a model answer network; In test process, use the technology of Dynamic matching, search the content that this sense-group of Interpreter's recognition result and translation topic type matees most, the characteristic of matching degree as system.
7. method according to claim 5; It is characterized in that: the matching degree under the BLEU Model Calculation BLEU pattern that said recognition result and model answer obtain is: to each translation topic type; The model answer of putting in order out according to expertise; Train the BLEU model,, calculate BLEU Model Matching degree then according to Interpreter's recognition result.
8. method according to claim 5; It is characterized in that: the lexicon grammar vector model that said recognition result and each grade cypher text obtain is: the text of translating out according to expertise and each grade examinee; Generate the lexicon grammar vector of each grade; According to Interpreter's recognition result, select a vector that matees most then, provide matching degree simultaneously.
9. method according to claim 1 is characterized in that: have for the recognition result of recognizer output and refuse to know function, use the degree of confidence technology of identification to suppress translation error, disorderly translation is identified as correct translation vocabulary.
10. method according to claim 1 is characterized in that: personnel's voice quality to be tested, pronunciation fluency are on model answer network matching content, to measure, and are used for to the personnel to be tested that possess certain Interpreter's ability.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008102266733A CN101739867B (en) | 2008-11-19 | 2008-11-19 | Method for scoring interpretation quality by using computer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008102266733A CN101739867B (en) | 2008-11-19 | 2008-11-19 | Method for scoring interpretation quality by using computer |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101739867A CN101739867A (en) | 2010-06-16 |
| CN101739867B true CN101739867B (en) | 2012-03-28 |
Family
ID=42463292
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2008102266733A Active CN101739867B (en) | 2008-11-19 | 2008-11-19 | Method for scoring interpretation quality by using computer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101739867B (en) |
Families Citing this family (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102034475B (en) * | 2010-12-08 | 2012-08-15 | 安徽科大讯飞信息科技股份有限公司 | Method for interactively scoring open short conversation by using computer |
| CN102930866B (en) * | 2012-11-05 | 2014-05-21 | 广州市神骥营销策划有限公司 | Evaluation method for student reading assignment for oral practice |
| CN103810158A (en) * | 2012-11-07 | 2014-05-21 | 中国移动通信集团公司 | Speech-to-speech translation method and device |
| CN104516870B (en) * | 2013-09-29 | 2018-08-07 | 北大方正集团有限公司 | A kind of translation inspection method and its system |
| CN103729349A (en) * | 2013-12-23 | 2014-04-16 | 武汉传神信息技术有限公司 | Analyzing method for affecting factors on translation quality |
| CN104050160B (en) * | 2014-03-12 | 2017-04-05 | 北京紫冬锐意语音科技有限公司 | Interpreter's method and apparatus that a kind of machine is blended with human translation |
| CN104952444B (en) * | 2015-04-27 | 2018-07-17 | 桂林电子科技大学 | A kind of Chinese's Oral English Practice method for evaluating quality that text is unrelated |
| CN106356054A (en) * | 2016-11-23 | 2017-01-25 | 广西大学 | Method and system for collecting information of agricultural products based on voice recognition |
| CN108154735A (en) * | 2016-12-06 | 2018-06-12 | 爱天教育科技（北京）有限公司 | Oral English Practice assessment method and device |
| WO2019001418A1 (en) | 2017-06-26 | 2019-01-03 | 上海寒武纪信息科技有限公司 | Data sharing system and data sharing method therefor |
| CN109214616B (en) * | 2017-06-29 | 2023-04-07 | 上海寒武纪信息科技有限公司 | Information processing device, system and method |
| CN109426553A (en) | 2017-08-21 | 2019-03-05 | 上海寒武纪信息科技有限公司 | Task cutting device and method, Task Processing Unit and method, multi-core processor |
| CN110413551B (en) | 2018-04-28 | 2021-12-10 | 上海寒武纪信息科技有限公司 | Information processing apparatus, method and device |
| CN107885730A (en) * | 2017-09-25 | 2018-04-06 | 沈阳航空航天大学 | Translation knowledge method for distinguishing validity under more interpreter's patterns |
| CN108537246A (en) * | 2018-02-28 | 2018-09-14 | 成都优译信息技术股份有限公司 | A kind of method and system that parallel corpora is classified by translation quality |
| CN109241519B (en) * | 2018-06-28 | 2022-08-12 | 平安科技（深圳）有限公司 | Quality evaluation model acquisition method and device, computer equipment and storage medium |
| CN110718101A (en) * | 2018-07-12 | 2020-01-21 | 华创云科(深圳)教育科技有限公司 | Education system and method based on virtual reality technology and artificial intelligence technology |
| CN109118113B (en) * | 2018-08-31 | 2021-08-10 | 传神语联网网络科技股份有限公司 | ETM architecture and word-shifting distance |
| CN109658921A (en) * | 2019-01-04 | 2019-04-19 | 平安科技（深圳）有限公司 | A kind of audio signal processing method, equipment and computer readable storage medium |
| CN110085256B (en) * | 2019-03-21 | 2021-11-19 | 视联动力信息技术股份有限公司 | Information processing method and device |
| CN111031232B (en) * | 2019-04-24 | 2022-01-28 | 广东小天才科技有限公司 | Dictation real-time detection method and electronic equipment |
| CN109979257B (en) * | 2019-04-27 | 2021-01-08 | 深圳市数字星河科技有限公司 | Method for performing accurate splitting operation correction based on English reading automatic scoring |
| CN110245860B (en) * | 2019-06-13 | 2022-08-23 | 桂林电子科技大学 | Automatic scoring method based on virtual experiment platform |
| CN110674871B (en) * | 2019-09-24 | 2023-04-07 | 北京中科凡语科技有限公司 | Translation-oriented automatic scoring method and automatic scoring system |
| CN111639217A (en) * | 2020-05-12 | 2020-09-08 | 广东小天才科技有限公司 | Spoken language rating method, terminal device and storage medium |
| CN112116839A (en) * | 2020-08-14 | 2020-12-22 | 爱语科技(广州)有限公司 | Language learning interaction method, system and storage medium based on semantic feature symbols |
| CN113674764A (en) * | 2021-08-20 | 2021-11-19 | 广东外语外贸大学 | Interpretation evaluation method, system and equipment based on bidirectional cyclic neural network |
| CN113539246B (en) * | 2021-08-20 | 2022-10-18 | 贝壳找房(北京)科技有限公司 | Voice recognition method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1602483A (en) * | 2001-12-17 | 2005-03-30 | 内维尼·加雅拉特尼 | Real time translator and method of performing real time translation of a plurality of spoken word languages |
| CN1892724A (en) * | 2005-07-04 | 2007-01-10 | 李杰波 | Foreigh-language oral-speech computer teaching system |
| WO2007098055A2 (en) * | 2006-02-17 | 2007-08-30 | Google Inc. | Encoding and adaptive, scalable accessing of distributed models |
-
2008
- 2008-11-19 CN CN2008102266733A patent/CN101739867B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1602483A (en) * | 2001-12-17 | 2005-03-30 | 内维尼·加雅拉特尼 | Real time translator and method of performing real time translation of a plurality of spoken word languages |
| CN1892724A (en) * | 2005-07-04 | 2007-01-10 | 李杰波 | Foreigh-language oral-speech computer teaching system |
| WO2007098055A2 (en) * | 2006-02-17 | 2007-08-30 | Google Inc. | Encoding and adaptive, scalable accessing of distributed models |
Non-Patent Citations (2)
| Title |
|---|
| 程葳 等.一种面向汉英口语翻译的双语语块处理方法.《中文信息学报》.2003,第17卷(第2期), * |
| 程葳 等.面向口语翻译的双语语块自动识别.《计算机学报》.2004,第27卷(第8期), * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101739867A (en) | 2010-06-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101739867B (en) | Method for scoring interpretation quality by using computer | |
| CN102354495B (en) | Testing method and system of semi-opened spoken language examination questions | |
| CN101740024B (en) | Method for automatic evaluation of spoken language fluency based on generalized fluency | |
| CN101751919B (en) | Spoken Chinese stress automatic detection method | |
| CN102034475B (en) | Method for interactively scoring open short conversation by using computer | |
| Saz et al. | Tools and technologies for computer-aided speech and language therapy | |
| CN101105939B (en) | Sonification guiding method | |
| CN103761975B (en) | Method and device for oral evaluation | |
| CN107103900A (en) | A kind of across language emotional speech synthesizing method and system | |
| WO2008033095A1 (en) | Apparatus and method for speech utterance verification | |
| CN106782603A (en) | Intelligent sound evaluating method and system | |
| CN102376182A (en) | Language learning system, language learning method and program product thereof | |
| Athanaselis et al. | Making assistive reading tools user friendly: A new platform for Greek dyslexic students empowered by automatic speech recognition | |
| Meng et al. | Development of automatic speech recognition and synthesis technologies to support Chinese learners of English: The CUHK experience | |
| Tong et al. | Multi-Task Learning for Mispronunciation Detection on Singapore Children's Mandarin Speech. | |
| Koudounas et al. | Italic: An italian intent classification dataset | |
| Liu et al. | AI recognition method of pronunciation errors in oral English speech with the help of big data for personalized learning | |
| Wang et al. | Predicting gradation of L2 English mispronunciations using crowdsourced ratings and phonological rules | |
| Hönig | Automatic assessment of prosody in second language learning | |
| Zechner et al. | Automatic scoring of children’s read-aloud text passages and word lists | |
| CN115240710A (en) | Neural network-based multi-scale fusion pronunciation evaluation model optimization method | |
| Li et al. | English sentence pronunciation evaluation using rhythm and intonation | |
| Lu et al. | Implementation of embedded unspecific continuous English speech recognition based on HMM | |
| KR20130067854A (en) | Apparatus and method for language model discrimination training based on corpus | |
| Lin et al. | Multi-lingual pronunciation assessment with unified phoneme set and language-specific embeddings |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20190724 Address after: 100190 52513, 5th floor, 6 Nansanjie, Zhongguancun, Haidian District, Beijing Patentee after: BEIJING ZIDONG COGNITIVE TECHNOLOGY Co.,Ltd. Address before: 100080 Zhongguancun East Road, Beijing, No. 95, No. Patentee before: Institute of Automation, Chinese Academy of Sciences |
|
| TR01 | Transfer of patent right | ||
| CP01 | Change in the name or title of a patent holder |
Address after: 100190 No. 52513, 5th floor, No. 6, South 3rd Street, Zhongguancun, Haidian District, Beijing Patentee after: Beijing Hongtu Cognitive Technology Co.,Ltd. Address before: 100190 No. 52513, 5th floor, No. 6, South 3rd Street, Zhongguancun, Haidian District, Beijing Patentee before: BEIJING ZIDONG COGNITIVE TECHNOLOGY Co.,Ltd. |
|
| CP01 | Change in the name or title of a patent holder | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20230421 Address after: 100190 automation building, No. 95, Zhongguancun East Road, Haidian District, Beijing Patentee after: BEIJING ZHONGZI SCIENCE AND TECHNOLOGY BUSINESS INCUBATOR CO.,LTD. Address before: 100190 No. 52513, 5th floor, No. 6, South 3rd Street, Zhongguancun, Haidian District, Beijing Patentee before: Beijing Hongtu Cognitive Technology Co.,Ltd. |
|
| TR01 | Transfer of patent right |