CN109979257A - A method of partition operation is carried out based on reading English auto-scoring and is precisely corrected - Google Patents
A method of partition operation is carried out based on reading English auto-scoring and is precisely corrected Download PDFInfo
- Publication number
- CN109979257A CN109979257A CN201910346958.9A CN201910346958A CN109979257A CN 109979257 A CN109979257 A CN 109979257A CN 201910346958 A CN201910346958 A CN 201910346958A CN 109979257 A CN109979257 A CN 109979257A
- Authority
- CN
- China
- Prior art keywords
- word
- value
- voice
- record
- pronunciation
- 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
Links
Classifications
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
- G09B5/00—Electrically-operated educational appliances
- G09B5/04—Electrically-operated educational appliances with audible presentation of the material to be studied
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
- G09B7/00—Electrically-operated teaching apparatus or devices working with questions and answers
- G09B7/02—Electrically-operated teaching apparatus or devices working with questions and answers of the type wherein the student is expected to construct an answer to the question which is presented or wherein the machine gives an answer to the question presented by a student
Abstract
A method of partition operation is carried out based on reading English auto-scoring and is precisely corrected, the present invention is marking basis with HMM posterior probability algorithm, whole sentence is carried out to reading English voice, word, three stratum level operational analyses of syllable phoneme, from the whole sentence phonetic acquisition phonetic feature of reading English, with standard reference model alignment score, whole sentence voice underproof to score value carries out partition operational analysis, obtain the word of bright read error, and then phoneme is carried out to pronunciation of words, the partition of syllable level, the position of precise positioning pronunciation mistake, standard pronunciation and correlated knowledge point are retrieved simultaneously and is corrected for user learns, the main points for allowing the quick master English of user to read aloud.
Description
Technical field
The present invention relates to voice knowledges to read aloud marking technical field, more particularly to a kind of based on the progress of reading English auto-scoring
The method that study is precisely corrected in partition operation.
Background technique
The research of China's speech recognition technology develops rapidly in recent years, and the level of the Research of Speech Recognition is basic and the world is same
Step, is provided with the advantage in China in certain subdivision fields, reaches advanced world standards, but automates marking assessment in reading English
The research in this speech recognition subdivision field, which just seems, to relatively lag behind, and patent retrieval and the analysis official website of office is known by state, respectively
" voice and marking " and the various combinations of the words such as " reading English ", " spoken language ", " English " are retrieved, resulting result is very few,
Technical solution wherein representative, that massage voice reading marking aspect is representative has: CN200810226674 one kind is used for
A kind of relevant Oral English Practice of text of automatic evaluation and diagnosis method of text reading level, CN201711200048 of oral test
Pronunciation error detection and quality score method, a kind of postgraduates'english oral teaching voice of CN201811030689 are assessed flat automatically
Implementation method of platform etc., these technical solutions provide the technical solution of many effective automatic assessments and marking, but existing
In reality, people are in English exercise, not only it should be understood that existing reading level, with greater need for the mistake understood in reading English
Place, the relevant knowledge point of learning and mastering are efficiently learnt by the knowledge point of study high quality statement, correct English rapidly
Language reads aloud the habit of middle mistake, improves the reading level of the Oral English Practice of itself.And the prior art it is rarely seen in this respect efficiently, just
Victory, system, practical technical solution.
Summary of the invention
In view of problem described in background technique, the present invention is marking basis with HMM posterior probability algorithm, to reading English language
The whole sentence of sound progress, three word, syllable phoneme level operational analyses, the first whole sentence phonetic acquisition phonetic feature from reading English,
With standard reference model alignment score, whole sentence voice underproof to score value carries out partition operational analysis, obtains bright read error
Word, and then the partition of phoneme, syllable is carried out to pronunciation of words, the position of precise positioning pronunciation mistake, and phoneme is established respectively
Specific combination phonetic feature pattern library relevant to pronunciation law.The knowledge element of pronunciation of English profession itself is fused to English
It reads aloud in marking test practice, retrieves standard pronunciation and correlated knowledge point for user and correct study, quickly grasp and read aloud knowledge.
English is based on text and reads aloud sentence marking, student English text is read aloud one group of observation sequence y of voice=(,, ,…..), multiple groups status switch s in standard reference model=( , , ,…..), then model s is produced
The probability of raw observation sequence y is, Viterbi algorithm is used in decoding process, after phoneme is aligned, selection most may be used
Can status switch S corresponding with observation sequence y, be thus calculated and be based onHidden MarkovThe log posterior of statistical model is general
The algorithm of rate: phonemePosterior probability calculating method formula 1 under i-th section of each frame of voice:
Taking logarithm, then accumulative superposition can be obtained by phonemeIn the log posterior of i-th section of time point corresponding voice segments
Probability marking calculation formula 2:
WhereinIndicate phonemeThe initial time of i-th section of corresponding voice, Z represent factor total number in voice,To observe vector under given phoneme qProbability distribution phoneme sum, in this way include pair of all phoneme section voices
The score mean value of number posterior probability is formula 3:
WhereinFor the frame number of k-th of phoneme duration；Pass through a Passing Criteria score value of above-mentioned calculating score value and default
Compare size and determines whether read aloud speech score up to standard.
The present invention by massage voice reading be divided into test pattern and practice Job evaluation mode, test pattern only to read aloud voice into
Row test marking, and when practicing Job evaluation mode, read aloud English sentence speech score it is up to standard when, it was demonstrated that student has learned to this article
This content is read aloud, and is handled there is no need to further, and reading aloud for next content of text is directly entered.
In score value situation not up to standard, it is necessary to which the accurate word of operational analysis pronunciation mistake or syllable allow student to know specifically
Where mistake, it can accurately learn the result corrected mistake and convince massage voice reading marking assessment, it is therefore desirable to whole sentence language
Sound carries out partition to syllable, word, it is therefore desirable to carry out partition identification holophrase segment, the operations such as syllable verbal audio section to voice.
Whole sentence voice is split into word, prior art Herman Kamper, Aren Jansen and Sharon in 2016
The unsupervised Bayesian model that Goldwater is proposed, can be split and then be clustered into virtual phrase for unlabelled voice.It should
The error rate of model about 20%, but for reading English marking, it is also unable to satisfy requirement, reading English marking is one
As be text based, the range of text is confined to the range of very little, therefore is split into single syllable to sentence voice and improves
More full and accurate basis, the invention circulation recursion marking partition method of identification, is first split into group of words for text, obtains
Received pronunciation and its acoustic feature of word etc. be used as standard reference model, first assume word read aloud when a length of standard when
It is long, the voice segments of the duration are successively decoupled on tested voice, the voice segments for obtaining highest score are compared, then carry out to
The amendment of plus-minus duration after forward direction obtains the ideal matched voice segments of word.The voice segments of perfect match.Specific implementation
Step:
Step 1, English text and Chinese-character text are different, and English text text, come separated words, therefore passes through split by space
Equal functions, are the character for identifying substring boundary using space, English text are become to the word number being made of each word
Group a, i.e. a=Split(text)；Containing dummy suffix notation ' contiguous alphabet combination regard a word as.
Step 2, the voice that specified English word is obtained by third-party speech interface, embodiment: by English text
Post submits to the network address of Baidu's voice development platform, obtains the voice document for returning to the formats such as mp3；(or language is turned by text
The voice of the acquisition particular text word such as engine of sound).
Step 3 simultaneously obtains phonetic feature by preanalysis, is converted into new standard reference model M, while recording holophrase
The duration S of sound, and pre-suppose that the tested duration S read aloud of word text.
Step 4, take it is tested read aloud in voice, initial time 1, end time are that this section S is new tested reads aloud
Operation is compared in M in voice and step 3, calculates score value J by formula 1, formula 2, formula 3,
Step 5 takes to be tested and read aloud in voice, successively adds for 1, end time 1 this section is added to be new be tested initial time
Voice group is read aloud, until operation is compared equal to the duration of original tested voice, respectively with the M in step 3 in the end time, is led to
It crosses formula 1, formula 2, formula 3 and calculates score value.
The score value that step 4 and step 5 calculate is compared step 6, obtains the numerical value A of maximum value, and with maximum value phase
The parameters such as corresponding initial time T1 and end time T2.
Above step reads aloud duration based on word and standard duration is equal, it is therefore desirable to correction is optimized to result, this
Invention is given a mark by comparison using the initial time of above-mentioned steps 6 and end time extension downwards upwards respectively and is obtained optimal value
To correct period and the duration of word pronunciation.Following back to back step is embodied:
Step 7, take it is tested read aloud in voice, initial time successively adds to T1 circulation subtracts 1, this section end time T2 is new
It is tested read aloud voice group, until successively decreasing, 1 initial time is equal to 1, the acoustic feature and step of gained voice segments in loop computation
M in rapid 3 is compared operation and obtains score value, and the A score value in score value and step 6 compares, and if score value is greater than A, then sets the value of A
It is set to current score values and T1 is set as the corresponding initial time of current score values, score value, which is less than A, is then jumped out initial time and successively decrease 1
Circulation.
Step 8, successively end time T2 circulation subtracted 1, this section initial time T1 be it is new it is tested read aloud voice group,
Until successively decreasing, 1 end time is equal to T1, and the M in loop computation in the acoustic feature and step 3 of gained voice segments is compared
Operation obtains score value, and the A score value in score value and step 7 compares, and if score value is greater than A, then the value of A is set as current score values and incited somebody to action
T2 is set as the current score values corresponding end time, score value be less than A then jump out initial time successively decrease 1 circulation.
Step 9 takes to be tested and read aloud in voice, and initial time is successively added T1 circulation plus 1, this section end time T2
For it is new it is tested read aloud voice group, until being incremented by plus 1 initial time is equal to T2, the acoustics of gained voice segments is special in loop computation
M in step 3 of seeking peace is compared operation and obtains score value, and the A score value in score value and step 8 compares, if score value is greater than A, then by A
Value be set as current score values and T1 be set as the corresponding initial time of current score values, score value is less than A and then jumps out initial time
Successively decrease 1 circulation.
End time T2 circulation plus 1, this section initial time T1 are successively new tested to read aloud voice by step 10
Group, until progressively increasing, 1 end time is equal to the duration of original tested voice entirety, the acoustics spy of gained voice segments in loop computation
M in step 3 of seeking peace is compared operation and obtains score value, and the A score value in score value and step 9 compares, if score value is greater than A, then by A
Value be set as current score values and T2 be set as the current score values corresponding end time, score value be less than A then jump out initial time
Progressively increase 1 circulation.
Step 11, record word and resulting through the above steps read aloud corresponding starting, end time on voice
And the data such as score value, step 2-10 is repeated, obtains all words of step 1 partition when being read aloud corresponding starting on voice
BetweenAnd the end timeAnd corresponding score value, serial number of the word in text sentence is wherein designated as under i.
Step 12, word score are lower than the error threshold of system setting, i.e., qualitative unqualified to read aloud pronunciation, then recall
The text that current word maps in step is stated, specific user interface is shown to, reminds user's pronunciation of words mistake, and be arranged
Broadcast click function mark be linked to step 2 formation word pronunciation position, configure corresponding formula allow student click described in broadcast
The word pronunciation that can hear standard is remembered in issue of bidding documents.And the partition analysis of phoneme, syllable rank is carried out to word.
Phoneme is the smallest sound bite that feature of having any different can be recognized by people of syllabication, is divided from sound quality
The smallest linear phonetic unit out.According to the articulation analysis in syllable, an articulation constitutes a phoneme；
English intemational phonetic symbols share 48 phonemes, wherein vowel phoneme 20, consonant phoneme 28.English letter shares 26, wherein
There are vowel 5, semivowel letter 2, consonant 19；Therefore, English Phonetics is decoupled to syllable phoneme, word
Level can find the core standard of English Phonetics most original, and the user for helping user to be especially just English learning precisely quickly slaps
The element for holding reading English, solving the problems, such as to study English to read aloud exists.In order to further help user's precise positioning mistake, entangle
Pronunciation is just being read aloud, the sound bite of partition to phoneme word carries out operational analysis, user more effectively can accurately be helped to learn English
Language is read aloud, and the prior art carries out the resulting phonemes of technological means such as framing adding window, decoding, direct computation of DFT transformation according to voice
Only observation operation based on probability as a result, can not be applied as teaching material standard, it is each it is even more impossible to match pronunciation of English
Kind rule, the present invention are split to syllable, sound for the unqualified word word further progress of pronunciation scoring described in above-mentioned steps 12
Plain level, side group user analyze.It is embodied as follows:
Letter is decoupled out according to byte in S1, word word, embodiment: using MID function, MID character string function, effect be from
Intercept out the character of specified quantity in one character string, MID (text, start_num, num_chars) text be need by
The word start_num of fractionation intercepts num_chars from 1 character length of interception (is used to the right from left to right since the 1st from left to right
Numeral expression), by the circulation being incremented by from 1 to word character length, the character array X (len-1) of word is obtained, len is single
Word byte length.
S2, the knowledge base for creating english phoneme phonetic symbol, wherein comprising 48 phonemes of the International Phonetic Symbols, front vowel/i/, // ,/
e/,/æ/；Medial vowel ... ..;Back vowel ... ..；Open and close diphthong :/e/,/a/, // ,/a/, //；......
Voiceless consonant ...；Nasal sound :/m/ ,/n/, //；.........Corresponding classification is set for each phoneme record, reads aloud and knows
Know point, the matched received pronunciation storage database tables column such as path and its speech acoustics feature, records knowledge point for example :/e/
It is single vowel front vowel, symbol of this phonetic symbol in English phonetic symbol is /e/, and the American corresponding diacritic of phonetic symbol is
[], specific skill of pronouncing have: 1) lip is slightly separated to two sides, and a little finger point can be about accommodated between upper lower tooth
Distance；2) tongue front is lifted during the pronunciation process, and the tip of the tongue slightly contacts the lower back of tooth；3) chin moves gradually downward when pronouncing,
Rack band, sending/e/ sound.Note :/e/ short vowel, pay attention to //, // difference, such knowledge point, other
Phoneme equally record its relevant knowledge.
The control library of S3, creation about phonetic symbol, letter, monogram corresponding relationship, first will be 26 in pronunciation of English rule
A letter and common monogram and its corresponding phonetic symbol increase to the corresponding table of database corresponding to rule base, and increase
Such as two identical vowels are added to be arranged together, such as oo, ee special word of a read-only phonetic symbol in foot, meet word
Mother's combination, while the columns such as alphabetic sort are created, the information such as belonging to the classifications such as the alphabetical vowel of record, consonant, that is to say, that will
English Conventional wisdom English phonetic and the information input of letter and monogram control compare library.
S4, creation pronunciation of English rule base, various pronunciation of English rules are sorted out and facilitate formula can be with logical operation
Expression-form, and classify in the way of it can carry out logical operation: 1, character property feature: contain these characteristic key words
Phonetic symbol of this i.e. record be that the default of corresponding letter or alphabetical monogram is pronounced, and this characteristic key words, by with spy
Sign symbol such as & symbol is separately combined in same record, example: open syllable & stress；2, it enumerates: by the word containing current record
The pronunciation of female or monogram is the word or sentence of the phonetic symbol of current record, is enumerated to this record, and different words or sentence are used
Characteristic symbol separates.
S5, X (len-1) character array that judgement S1 step obtain regular based on the aphonic pronunciation of English of word end e
Whether last position is e, then forces character array to subtract last member: len=len-1 in this way.
S6, creation two variables of A, Z, and assign initial value A=0.
(conventional alpha combination is preferably at most 4 alphabetical combination tion, pronunciation when the value that S7, such as len subtract A is greater than or equal to 4
Phonetic symbol be []) then Z=4, otherwise Z=len-A.By A again assignment: A=A+1, A is taken in X (len-1) character array member
A to the Z character combination combines into the control library of common pronunciation monogram and retrieves, located respectively according to search result
Reason.
1, the character of present combination is then submitted to the custom function of S9 step when retrieving multiple record
The character of guizefunction(present combination, word, A, Z) insertion operation；Work as A+Z > len and directly executes S10 step otherwise general
A assignment: A=A+Z-1 simultaneously restarts this step.
2, it only retrieves unique record then to record the phonetic symbol of this record and A, Z value together, works as A+Z > len and directly execute
Otherwise S10 step to A again assignment: A=A+Z-1 and restarts this step.
3, it does not retrieve record and then enters next step.
S8, as jumped to 7 step of S above if Z=1, Z is otherwise assigned a value of Z-1, takes wherein A to Z character, group
It closes and is retrieved in the rule base for coming common pronunciation monogram, it is handled respectively according to search result.
1, it only retrieves unique record then to record the phonetic symbol of this record and A, Z value together, by A again assignment: A=A+Z-
1 and directly jump execute S7 step.
2, multiple custom function for recording and the character of present combination then being submitted to S9 step are retrieved
The character of guizefunction(present combination, word, A, Z) insertion operation；A=A+Z-1 simultaneously directly jumps execution S7 step.
3, it then repeats to start this step without record, carries out cycle analysis, until Z value is 1.
S9, regular operation custom function guizefunction(str, str1, Index1, index2),
A, str character string is retrieved in pronunciation of English rule base, and is to the content for " enumerating " column record in the record in retrieval
No includes current word, if returned comprising if phonetic symbol in current record as this function as a result, by phonetic symbol and
Index1, index2 value record together, and terminate the operation of this function, do not carry out next record inspection then；
B, determine whether str its latter byte location letter in str1 is consonant, two character variables are first set
Tex, texx, tex=right(word if index1+index2+1 > len (word), 1) otherwise tex=MID (str1,
Index1+index2+1,1), texx is assigned a value of " open syllable " and otherwise examines in rule base if tex is " r " or " w " or " y "
Rope letter or monogram be tex record, by the record on " alphabetic sort column " in record if comprising " vowel " texx assignment
For " open syllable ", otherwise texx is assigned a value of " closed syllable ".
C, str character is retrieved in rule base, and the content on characteristic key words column whether there is in verifying record one by one
The content of texx returns to this function if there is this is then recorded corresponding phonetic symbol, by phonetic symbol and Index1, index2 value one
Record is played, qualified next record is otherwise verified.
S10, will decouple out with English text word, letter and monogram and its corresponding phonemic transcription, retrieve phase
The knowledge point of pass, is shown to user interface, allows user's learning and mastering.
In practical reading English, phoneme often changes the pronunciation of original standard because of upper and lower phoneme, more with
Syllable reads aloud the basic unit of pronunciation as studying English, thus can effectively help user to grasp just word fractionation syllabication
True reads aloud.
According to English word pronunciation law: word splits the principle abided by of syllabication: 1) one return after, refer to two syllables it
Between, if there is consonant, this letter incorporates the latter syllable into, 2) two points open, refer between two syllables, if there are two consonants
Letter then incorporates former and later two syllables into respectively；Fallibility point: vowel and vowel, which mix up syllabification, to be drawn by vowel,
If vowel is mute, if that cannot syllabication there are two vowels together, but only send out member
Sound still calculates a syllable.Syllabification boundary rule: vowel phoneme is the main body of syllabication, and consonant is the boundary of syllable
Line.Each vowel phoneme can constitute a syllable, therefore more decoupling to syllable more accurately to help user to learn
Reading English, the implementation of English word syllable partition:
Step 1, pass through more than word split into phoneme technical solution record composition and English word text letters and word
Mother combines corresponding one group of phonetic symbol: , , ,…..And the initial position of the corresponding letter of each phonetic symbol: , ,,….., end position: , , ,…..。
Step 2 retrieves phonetic symbol in knowledge base respectively, obtains the phoneme that one or a set of phonetic symbol is classified as vowel.
Step 3, according to the corresponding position numerical value of step 1 phoneme, successively using the functions such as mid obtain the corresponding letter of vowel or
Initial position value corresponding to latter one vowel is then subtracted 1 such as only one letter by the letter between alphabetical group, and such as two
It is alphabetical then the corresponding end position value of previous vowel is added 1.
Step 4, the value that one group of phoneme and new initial position and end position are obtained by step 3, according to this group of phoneme
Corresponding initial position and end position, which calculate, obtains corresponding letter or alphabetical group, user interface is output to, as word
Decouple the result of syllable.
Especially statement: " embodiment " etc. described in the present specification refers to the specific spy for combining embodiment description
Sign, element or feature include in the embodiment of the application generality description.There is table of the same race in multiple places in the description
It states and non-limiting refer in particular to is the same embodiment.That is, in conjunction with any embodiment describe a specific features, element or
When person's feature, what is advocated is to realize that this feature, element or feature are contained in the present invention in conjunction with other embodiments
In the scope of the claims for applying for protection；Embodiment is multiple explanatory embodiments referring to logical architecture of the present invention and thinking
Invention has been described, but scope of protection of the present invention is not limited thereto, and those skilled in the art are in the technology of the present invention
Can be designed that a lot of other modification and implementations under solution framework, can to technical solution want point transformation combination/or
Layout carries out a variety of non-intrinsically safe variations and modifications, and to those skilled in the art, other purposes also will be apparent,
The unsubstantiality change or replacement of implementation can be readily occurred in, these modifications and implementations will fall in principle model disclosed in the present application
Within enclosing and being spiritual.
Detailed description of the invention
Fig. 1 carries out the method overall logic that study is precisely corrected in partition operation based on reading English auto-scoring to be a kind of
Frame diagram.
Claims (4)
1. a kind of carry out the method that study is precisely corrected in partition operation based on reading English auto-scoring, feature includes the steps that
With to be known as: massage voice reading is divided into test pattern and practice Job evaluation mode, test pattern is only surveyed to reading aloud voice
Examination marking, and when practicing Job evaluation mode, read aloud English sentence speech score it is up to standard when, be directly entered the bright of next content of text
It reads；In score value situation not up to standard, the word of operational analysis pronunciation mistake first will using circulation recursion marking partition method of identification
Text is split into group of words, obtains received pronunciation and its acoustic feature of word etc. as standard reference model, first assumes word
The when a length of standard duration read aloud successively decouples the voice segments of the duration on tested voice, acquisition best result is compared
The voice segments of value；
Step 1, English text and Chinese-character text are different, and English text text, come separated words, therefore passes through split by space
Equal functions, are the character for identifying substring boundary using space, English text are become to the word number being made of each word
Group a, i.e. a=Split(text)；Containing dummy suffix notation ' contiguous alphabet combination regard a word as；
Step 2, the voice that specified English word is obtained by third-party speech interface, embodiment: English text post is mentioned
The network address of Baidu's voice development platform is given, the voice document for returning to the formats such as mp3 is obtained；(or drawing by text-to-speech
The voice for obtaining particular text word such as hold up)；
Step 3 simultaneously obtains phonetic feature by preanalysis, is converted into new standard reference model M, while recording word pronunciation
Duration S, and pre-suppose that the tested duration S read aloud of word text；
Step 4, take it is tested read aloud in voice, initial time 1, end time be this section S be it is new it is tested read aloud voice,
Operation is compared with the M in step 3, score value J is calculated by formula 1, formula 2, formula 3；
Step 5 takes to be tested and read aloud in voice, successively adds for 1, end time 1 this section is added to be new be tested initial time
Voice group is read aloud, until operation is compared equal to the duration of original tested voice, respectively with the M in step 3 in the end time, is led to
It crosses formula 1, formula 2, formula 3 and calculates score value；
The score value that step 4 and step 5 calculate is compared step 6, obtains the numerical value A of maximum value, and corresponding with maximum value
Initial time T1 and the parameters such as end time T2.
2. one kind carries out partition operation based on reading English auto-scoring and precisely corrects learning method according to claim 1,
The step of its feature also includes and element: then the voice segments based on the highest score that claim 1 obtains carry out forward backward
Plus-minus duration amendment, obtain the ideal matched voice segments of word, immediately the step of claim one:
Step 7, take it is tested read aloud in voice, initial time successively adds to T1 circulation subtracts 1, this section end time T2 is new
It is tested read aloud voice group, until successively decreasing, 1 initial time is equal to 1, the acoustic feature and step of gained voice segments in loop computation
M in rapid 3 is compared operation and obtains score value, and the A score value in score value and step 6 compares, and if score value is greater than A, then sets the value of A
It is set to current score values and T1 is set as the corresponding initial time of current score values, score value, which is less than A, is then jumped out initial time and successively decrease 1
Circulation；
Step 8, successively end time T2 circulation subtracted 1, this section initial time T1 be it is new it is tested read aloud voice group, until
Successively decrease 1 end time is equal to T1, and operation is compared in the M in loop computation in the acoustic feature and step 3 of gained voice segments
Score value is obtained, the A score value in score value and step 7 compares, and if score value is greater than A, then the value of A is set as current score values and sets T2
Be set to the current score values corresponding end time, score value be less than A then jump out initial time successively decrease 1 circulation；
Step 9 takes to be tested and read aloud in voice, and initial time is successively added T1 circulation plus 1, this section end time T2 is new
It is tested read aloud voice group, until being incremented by plus 1 initial time is equal to T2, in loop computation the acoustic feature of gained voice segments and
M in step 3 is compared operation and obtains score value, and the A score value in score value and step 8 compares, if score value is greater than A, then by the value of A
It is set as current score values and T1 is set as the corresponding initial time of current score values, score value, which is less than A, is then jumped out initial time and successively decrease 1
Circulation；
Step 10, successively by end time T2 circulation plus 1, this section initial time T1 be it is new it is tested read aloud voice group, directly
To progressively increasing, 1 end time is equal to the duration of original tested voice entirety, in loop computation the acoustic feature of gained voice segments with
M in step 3 is compared operation and obtains score value, and the A score value in score value and step 9 compares, if score value is greater than A, then by the value of A
It is set as current score values and T2 is set as the current score values corresponding end time, score value, which is less than A, is then jumped out initial time and progressively increase 1
Circulation；
Step 11, record word and through the above steps it is resulting read aloud on voice it is corresponding starting, the end time and point
The data such as value repeat step 2-10, obtain all words that step 1 decouples and are being read aloud corresponding initial time on voice
And the end timeAnd corresponding score value, serial number of the word in text sentence is wherein designated as under i；
Step 12, word score are lower than the error threshold of system setting, i.e., qualitative unqualified to read aloud pronunciation, then recall above-mentioned step
The text of current word mapping, is shown to specific user interface in rapid, reminds user's pronunciation of words mistake, and be arranged and broadcast
Click function mark broadcast is linked to step 2 formation word pronunciation position, configure corresponding formula allow student click described in broadcast
The word pronunciation that can hear standard is remembered in issue of bidding documents, and the partition analysis of phoneme, syllable rank is carried out to word.
3. a kind of method that the unqualified word of pronunciation scoring is carried out partition analytic operation, feature includes the steps that and element
Have:
S1, word word decouple out letter according to byte；
S2, the knowledge base for creating english phoneme phonetic symbol, wherein including 48 phonemes of the International Phonetic Symbols, for each phoneme record setting phase
The classification answered reads aloud knowledge point, the matched received pronunciation storage database tables column such as path and its speech acoustics feature, record
Knowledge point；
The control library of S3, creation about phonetic symbol, letter, monogram corresponding relationship, first by 26 words in pronunciation of English rule
Female and common monogram and its corresponding phonetic symbol increase to the corresponding table of database corresponding to rule base, and increase all
Such as two identical vowels are arranged together special letter combination, while creating the columns such as alphabetic sort, the alphabetical such as vowel of record,
The information such as belonging to the classifications such as consonant；
S4, creation pronunciation of English rule base, various pronunciation of English rules are sorted out and facilitate formula can be with the expression of logical operation
Form, and classifying in the way of it can carry out logical operation: 1, character property feature: containing these characteristic key words i.e.
The phonetic symbol of this record is the default pronunciation of corresponding letter or alphabetical monogram, and this characteristic key words, by with character symbol
Number such as & symbol is separately combined in same record, example: open syllable & stress；2, enumerate: by the letter containing current record or
Monogram pronunciation is the word or sentence of the phonetic symbol of current record, is enumerated to this record, different words or sentence feature
Symbol separates；
S5, it is based on the aphonic pronunciation of English rule of word end e, X (len-1) character array for determining that S1 step obtains is last
Whether one be e, then forces character array to subtract last member: len=len-1 in this way；
S6, creation two variables of A, Z, and assign initial value A=0；
S7, Z=4 if when the len value for subtracting A is greater than or equal to 4, otherwise Z=len-A；By A again assignment: A=A+1, in X (len-
1) A are taken in character array member to the Z character combination, are combined into the control library of common pronunciation monogram and are examined
Rope is handled respectively according to search result；
The character of present combination is then submitted to the custom function guizefunction of S9 step when retrieving multiple record
(character of present combination, word, A, Z) is embedded in operation；Work as A+Z > len and directly execute S10 step otherwise for A assignment: A=A+Z-1
And restart this step；It only retrieves unique record then to record the phonetic symbol of this record and A, Z value together, it is straight to work as A+Z > len
Connect and execute S10 step otherwise by A assignment: A=A+Z-1 simultaneously restarts this step；Record is not retrieved then to enter in next step
Suddenly；
S8, as jumped to 7 step of S above if Z=1, Z is otherwise assigned a value of Z-1, takes wherein A to Z character, group closes
It comes in the rule base of common pronunciation monogram and retrieves, handled respectively according to search result: only retrieving unique record and then will
The phonetic symbol and A, Z value of this record record together, by A assignment: A=A+Z-1 simultaneously directly jumps execution S7 step；
Multiple records are retrieved then to work as the custom function guizefunction(that the character of present combination submits to S9 step
Preceding combined character, word, A, Z) insertion operation；A=A+Z-1 simultaneously directly jumps execution S7 step, does not record, repeats to start
This step carries out cycle analysis, until Z value is 1；
S9, regular operation custom function guizefunction(str, str1, Index1, index2),
A, str character string is retrieved in pronunciation of English rule base, and is to the content for " enumerating " column record in the record in retrieval
No includes current word, if returned comprising if phonetic symbol in current record as this function as a result, by phonetic symbol and
Index1, index2 value record together, and terminate the operation of this function, do not carry out next record inspection then；
B, determine whether str its latter byte location letter in str1 is consonant, two character variables are first set
Tex, texx, tex=right(word if index1+index2+1 > len (word), 1) otherwise tex=MID (str1,
Index1+index2+1,1), texx is assigned a value of " open syllable " and otherwise examines in rule base if tex is " r " or " w " or " y "
Rope letter or monogram be tex record, by the record on " alphabetic sort column " in record if comprising " vowel " texx assignment
For " open syllable ", otherwise texx is assigned a value of " closed syllable "；
C, str character is retrieved in rule base, and the content on characteristic key words column whether there is texx's in verifying record one by one
Content returns to this function if there is this is then recorded corresponding phonetic symbol, and by phonetic symbol and Index1, index2 value records together,
Otherwise qualified next record is verified；
S10, will decouple out to English text word, letter and monogram and its corresponding phonemic transcription, retrieve it is relevant
Knowledge point is shown to user interface, allows user's learning and mastering.
4. a kind of method that the unqualified word of pronunciation scoring is carried out partition analytic operation according to claim 1, feature
The step of also including and element: the technical solution of English word syllable partition；
Step 1, the composition one group of phonetic symbol corresponding with English word text letters and monogram recorded by claim 3 and
The initial position of the corresponding letter of each phonetic symbol, end position；
Step 2 retrieves phonetic symbol in knowledge base respectively, obtains the phoneme that one or a set of phonetic symbol is classified as vowel；
Step 3, according to the corresponding position numerical value of step 1 phoneme, successively obtain the corresponding letter of vowel or letter using the functions such as mid
Initial position value corresponding to latter one vowel is then subtracted 1 such as only one letter by the letter between group, such as two letters
The corresponding end position value of previous vowel is then added 1；
Step 4, the value that one group of phoneme and new initial position and end position are obtained by step 3, it is corresponding according to this group of phoneme
Initial position and end position calculate and obtain corresponding letter or alphabetical group, be output to user interface, decoupled as word
The result of syllable.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910346958.9A CN109979257B (en) | 2019-04-27 | 2019-04-27 | Method for performing accurate splitting operation correction based on English reading automatic scoring |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910346958.9A CN109979257B (en) | 2019-04-27 | 2019-04-27 | Method for performing accurate splitting operation correction based on English reading automatic scoring |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109979257A true CN109979257A (en) | 2019-07-05 |
| CN109979257B CN109979257B (en) | 2021-01-08 |
Family
ID=67086631
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910346958.9A Active CN109979257B (en) | 2019-04-27 | 2019-04-27 | Method for performing accurate splitting operation correction based on English reading automatic scoring |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109979257B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110889987A (en) * | 2019-12-16 | 2020-03-17 | 安徽必果科技有限公司 | Intelligent comment method for correcting spoken English |
| CN113422825A (en) * | 2021-06-22 | 2021-09-21 | 读书郎教育科技有限公司 | System and method for assisting in culturing reading interests |
| CN113707178A (en) * | 2020-05-22 | 2021-11-26 | 苏州声通信息科技有限公司 | Audio evaluation method and device and non-transient storage medium |
| CN114245194A (en) * | 2021-12-23 | 2022-03-25 | 深圳市优必选科技股份有限公司 | Video teaching interaction method and device and electronic equipment |
| WO2022148176A1 (en) * | 2021-01-08 | 2022-07-14 | Ping An Technology (Shenzhen) Co., Ltd. | Method, device, and computer program product for english pronunciation assessment |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101739867A (en) * | 2008-11-19 | 2010-06-16 | 中国科学院自动化研究所 | Method for scoring interpretation quality by using computer |
| CN101739869A (en) * | 2008-11-19 | 2010-06-16 | 中国科学院自动化研究所 | Priori knowledge-based pronunciation evaluation and diagnosis system |
| CN101739868A (en) * | 2008-11-19 | 2010-06-16 | 中国科学院自动化研究所 | Automatic evaluation and diagnosis method of text reading level for oral test |
| CN107103915A (en) * | 2016-02-18 | 2017-08-29 | 广州酷狗计算机科技有限公司 | A kind of audio data processing method and device |
| CN107958673A (en) * | 2017-11-28 | 2018-04-24 | 北京先声教育科技有限公司 | A kind of spoken language methods of marking and device |
| CN108428382A (en) * | 2018-02-14 | 2018-08-21 | 广东外语外贸大学 | It is a kind of spoken to repeat methods of marking and system |
| US20180277098A1 (en) * | 2017-03-23 | 2018-09-27 | Samsung Electronics Co., Ltd. | Training method and apparatus for speech recognition |
| CN109300339A (en) * | 2018-11-19 | 2019-02-01 | 王泓懿 | A kind of exercising method and system of Oral English Practice |
-
2019
- 2019-04-27 CN CN201910346958.9A patent/CN109979257B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101739867A (en) * | 2008-11-19 | 2010-06-16 | 中国科学院自动化研究所 | Method for scoring interpretation quality by using computer |
| CN101739869A (en) * | 2008-11-19 | 2010-06-16 | 中国科学院自动化研究所 | Priori knowledge-based pronunciation evaluation and diagnosis system |
| CN101739868A (en) * | 2008-11-19 | 2010-06-16 | 中国科学院自动化研究所 | Automatic evaluation and diagnosis method of text reading level for oral test |
| CN107103915A (en) * | 2016-02-18 | 2017-08-29 | 广州酷狗计算机科技有限公司 | A kind of audio data processing method and device |
| US20180277098A1 (en) * | 2017-03-23 | 2018-09-27 | Samsung Electronics Co., Ltd. | Training method and apparatus for speech recognition |
| CN107958673A (en) * | 2017-11-28 | 2018-04-24 | 北京先声教育科技有限公司 | A kind of spoken language methods of marking and device |
| CN108428382A (en) * | 2018-02-14 | 2018-08-21 | 广东外语外贸大学 | It is a kind of spoken to repeat methods of marking and system |
| CN109300339A (en) * | 2018-11-19 | 2019-02-01 | 王泓懿 | A kind of exercising method and system of Oral English Practice |
Non-Patent Citations (2)
| Title |
|---|
| 吕军: "基于语音识别的汉语发音自动评分系统的设计与实现", 《数字化汉语教学的研究与应用》 * |
| 王娜: "汉语语音评分系统的设计与实现", 《绵阳师范学院学报》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110889987A (en) * | 2019-12-16 | 2020-03-17 | 安徽必果科技有限公司 | Intelligent comment method for correcting spoken English |
| CN113707178A (en) * | 2020-05-22 | 2021-11-26 | 苏州声通信息科技有限公司 | Audio evaluation method and device and non-transient storage medium |
| CN113707178B (en) * | 2020-05-22 | 2024-02-06 | 苏州声通信息科技有限公司 | Audio evaluation method and device and non-transient storage medium |
| WO2022148176A1 (en) * | 2021-01-08 | 2022-07-14 | Ping An Technology (Shenzhen) Co., Ltd. | Method, device, and computer program product for english pronunciation assessment |
| CN113422825A (en) * | 2021-06-22 | 2021-09-21 | 读书郎教育科技有限公司 | System and method for assisting in culturing reading interests |
| CN113422825B (en) * | 2021-06-22 | 2022-11-08 | 读书郎教育科技有限公司 | System and method for assisting in culturing reading interests |
| CN114245194A (en) * | 2021-12-23 | 2022-03-25 | 深圳市优必选科技股份有限公司 | Video teaching interaction method and device and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109979257B (en) | 2021-01-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109979257A (en) | A method of partition operation is carried out based on reading English auto-scoring and is precisely corrected | |
| CN102376182B (en) | Language learning system, language learning method and program product thereof | |
| CN106782603B (en) | Intelligent voice evaluation method and system | |
| JPS61177493A (en) | Voice recognition | |
| CN101551947A (en) | Computer system for assisting spoken language learning | |
| Gao et al. | A study on robust detection of pronunciation erroneous tendency based on deep neural network. | |
| Arora et al. | Phonological feature-based speech recognition system for pronunciation training in non-native language learning | |
| Peabody | Methods for pronunciation assessment in computer aided language learning | |
| Davel et al. | Pronunciation dictionary development in resource-scarce environments | |
| CN110415725B (en) | Method and system for evaluating pronunciation quality of second language using first language data | |
| Hu et al. | A new neural network based logistic regression classifier for improving mispronunciation detection of L2 language learners | |
| Scholten et al. | Learning to recognise words using visually grounded speech | |
| Duan et al. | Effective articulatory modeling for pronunciation error detection of L2 learner without non-native training data | |
| JP2004094257A (en) | Method and apparatus for generating question of decision tree for speech processing | |
| Jang | Speech rhythm metrics for automatic scoring of English speech by Korean EFL learners | |
| Hong et al. | A corpus-based analysis of English segments produced by Korean learners | |
| Sabu et al. | Automatic Assessment of Children's L2 Reading for Accuracy and Fluency. | |
| WO2019075827A1 (en) | Voice evaluation method and device | |
| CN111508522A (en) | Statement analysis processing method and system | |
| CN115440193A (en) | Pronunciation evaluation scoring method based on deep learning | |
| Díez et al. | Non-native speech corpora for the development of computer assisted pronunciation training systems | |
| Chen et al. | An Alignment Method Leveraging Articulatory Features for Mispronunciation Detection and Diagnosis in L2 English. | |
| JP2008083434A (en) | Voice learning support apparatus and voice learning support program | |
| Rouas et al. | Language and variety verification on broadcast news for Portuguese | |
| CN111899581A (en) | Word spelling and reading exercise device and method for English teaching |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |