CN109979257A - A method of partition operation is carried out based on reading English auto-scoring and is precisely corrected - Google Patents

Abstract

A method for accurate correction of split operation based on automatic scoring of English reading aloud. The present invention uses HMM posterior probability algorithm as the scoring basis, and carries out three-level operation analysis on English reading speech at three levels: whole sentence, word, and syllable and phoneme, and reads aloud from English. The whole sentence speech obtained by the system obtains the phonetic features, compares and scores with the standard reference model, conducts split operation analysis on the speech of the whole sentence with unqualified scores, obtains the words that are read aloud, and then separates the pronunciation of the word at the phoneme and syllable levels. Accurately locate the location of pronunciation errors, and at the same time retrieve the standard pronunciation and related knowledge points for users to correct and learn, allowing users to quickly master the essentials of English reading.

Description

A method for accurate correction of split operation based on automatic scoring of English reading

technical field

The invention relates to the technical field of speech recognition, reading, and scoring, in particular to a method for accurate correction and learning of split operation based on automatic scoring of English reading.

Background technique

The research on speech recognition technology in China has developed rapidly in recent years. The level of speech recognition research is basically synchronized with the international level. In some sub-fields, it has the advantages of China and has reached the international advanced level. The research in different fields seems to be relatively lagging behind. Through the official website of the State Intellectual Property Office for patent search and analysis, various combinations of words such as "voice and scoring" and "English reading", "spoken language" and "English" were searched, and the results obtained were obtained. There are very few, and the representative technical solutions in terms of speech reading and scoring include: CN200810226674 A text-reading level automatic evaluation and diagnosis method for oral language testing, CN201711200048 A text-related English oral pronunciation error detection and quality scoring Methods, CN201811030689 A realization method of an automatic evaluation platform for postgraduate oral English teaching, etc. These technical solutions provide many practical and effective technical solutions for automatic evaluation and scoring, but in reality, when people practice English, not only need to understand the current situation. For some reading levels, it is more necessary to understand the mistakes in English reading, learn to master the relevant knowledge points, learn efficiently by learning the knowledge points expressed in high quality, quickly correct the wrong habits in English reading, and improve their spoken English. Reading level. However, in the prior art, there are few efficient, convenient, systematic and practical technical solutions in this regard.

SUMMARY OF THE INVENTION

In view of the problems described in the background art, the present invention takes the HMM posterior probability algorithm as the scoring basis, and carries out three-level operation analysis on the whole sentence, word, and syllable phoneme of the English reading speech, and first obtains the phonetic features from the whole sentence pronunciation of the English reading, Compare and score with the standard reference model, split and analyze the speech of the whole sentence with unqualified scores, obtain words that are pronounced incorrectly, and then divide the pronunciation of words into phonemes and syllables to accurately locate the wrong pronunciation. Establish a speech feature benchmark library related to specific phoneme combinations and pronunciation laws. The knowledge elements of the English pronunciation major are integrated into the English reading and scoring test exercises, and the standard pronunciation and related knowledge points are retrieved for users to correct and learn, and quickly master the reading knowledge.

English is scored based on text-speaking sentences, and a set of observation sequences y=( , , ,… ) , in the standard reference model, multiple sets of state sequences s = ( , , ,… ), then the probability that the model s produces the observation sequence y is , the Viterbi algorithm is used in the decoding process, after aligning the phonemes, the state sequence S that is most likely to correspond to the observation sequence y is selected, and the algorithm of the logarithmic posterior probability based on the hidden Markov statistical model is calculated: phoneme The posterior probability calculation method formula 1 under each frame of the i-th speech:

Take the logarithm and accumulate it to get the phoneme The logarithmic posterior probability score calculation formula 2 of the speech segment corresponding to the i-th time point:

in Represents a phoneme The corresponding start time of the i-th speech, Z represents the total number of factors in the speech, Observe the vector for a given phoneme q The probability distribution of the total number of phonemes, such that the fractional mean of the log-posterior probability of speech including all phoneme segments is Equation 3:

in is the number of frames that the kth phoneme lasts; by comparing the above calculated score with a standard score set by the system to determine whether the reading voice score meets the standard.

The invention divides the speech reading and scoring into a test mode and a practice evaluation mode. The test mode only tests and scores the reading voice, and in the practice evaluation mode, when the pronunciation of reading an English sentence reaches the standard, it proves that the student has learned to read the text content. Therefore, no further processing is required, and the reading of the next text content is directly entered.

If the score does not meet the standard, it is necessary to calculate and analyze the pronunciation error and precise words or syllables, so that students can know the specific error, so that they can accurately learn to correct the error and be convinced of the results of the pronunciation reading score. Therefore, it is necessary to split the whole sentence. To syllables and words, it is necessary to split the speech to identify word speech segments, syllable speech segments and other operations.

The whole sentence is split into words. The unsupervised Bayesian model proposed by Herman Kamper, Aren Jansen and Sharon Goldwater in the prior art in 2016 can segment unlabeled speech and then cluster it into virtual phrases. The error rate of this model is about 20%, but for English reading aloud scoring, it cannot meet the requirements. English reading aloud scoring is generally based on text, and the range of text is limited to a small range, so the sentence is divided into two parts. The word syllable improves a more detailed basis. The invention creates a cyclic recursive scoring and splitting recognition method. First, the text is split into word groups, and the standard pronunciation of the word and its acoustic characteristics are obtained as a standard reference model. The duration of the reading is the standard duration, and the speech segments of this duration are divided in turn on the tested speech, and the speech segment with the highest score is compared, and then the forward and backward addition and subtraction duration correction is performed to obtain words. matching speech segment. Ideally matched speech segments. Specific implementation steps:

Step 1. English text and Chinese text are different. English text text is separated by spaces. Therefore, through functions such as split, spaces are used as characters that identify the boundaries of substrings, and English text is turned into a word array a composed of individual words. That is, a=Split(text); the continuous letter combination containing the abbreviation ' is regarded as a word.

Step 2, obtain the voice of the specified English word through the voice interface of the third party, embodiment: submit the English text post to the website of the Baidu voice development platform, obtain the voice file of the format such as returning mp3; (or through the engine of text-to-speech etc. to get the speech of a specific text word).

In step 3, the speech features are obtained through pre-analysis, converted into a new standard reference model M, and the duration S of the word speech is recorded at the same time, and the duration S of the word text is assumed to be read aloud by test in advance.

Step 4. Take the tested reading voice, the starting time 1 and the ending time are S. This interval is the new tested reading voice, and compare the operation with M in step 3, through formula 1, formula 2, formula 3 Calculate the score J,

Step 5. Take the tested reading voice, add 1 to the starting time and add 1 to the ending time as a new tested reading voice group, until the ending time is equal to the duration of the original tested voice, respectively and step 3. The M of the comparison operation is performed, and the score is calculated by formula 1, formula 2, and formula 3.

Step 6: Compare the scores calculated in Step 4 and Step 5 to obtain the maximum value A, and parameters such as the start time T1 and the end time T2 corresponding to the maximum value.

The above steps are based on the fact that the reading time of the word is equal to the standard time, so the results need to be optimized and corrected. The present invention uses the start time and end time of the above step 6 to expand upward and downward respectively, and obtain the optimal value by comparing and scoring to correct the pronunciation of the word. time period and length. The specific implementation is as follows:

Step 7. Take the tested reading voice, add T1 to the starting time and cyclically decrease 1, and the end time T2 is the new tested reading voice group, until the starting time decremented by 1 is equal to 1, and the cycle operation is obtained. The acoustic features of the speech segment are compared with the M in step 3 to obtain a score, and the score is compared with the A score in step 6. If the score is greater than A, the value of A is set as the current score and Set T1 as the starting time corresponding to the current score. If the score is less than A, the cycle of decreasing the starting time by 1 will jump out.

Step 8. Decrease the end time T2 cyclically by 1 and start time T1 as the new tested speech group until the end time decremented by 1 is equal to T1. The acoustic characteristics of the obtained speech segment in the cyclic operation are the same as those in step 3. The M is compared to obtain the score, and the score is compared with the A score in step 7. If the score is greater than A, the value of A is set as the current score and T2 is set as the corresponding value of the current score. If the score is less than A, it will jump out of the cycle where the start time is decremented by 1.

Step 9. Take the tested reading voice, add 1 to the starting time and T1 cyclically, and add 1 to the ending time T2 as a new tested reading voice group, until the starting time that increases by 1 is equal to T2. The acoustic features of the obtained speech segment are compared with the M in step 3 to obtain a score, and the score is compared with the A score in step 8. If the score is greater than A, the value of A is set as the current score. And set T1 as the starting time corresponding to the current score. If the score is less than A, it will jump out of the cycle of decreasing the starting time by 1.

Step 10. Cycle the end time T2 by 1 and the start time T1 as the new tested speech group, until the end time incremented by 1 is equal to the overall duration of the original tested speech, and the speech segment obtained in the cyclic operation The acoustic features of the A are compared with the M in step 3 to obtain the score, and the score is compared with the A score in step 9. If the score is greater than A, the value of A is set as the current score and T2 Set as the end time corresponding to the current score. If the score is less than A, it will jump out of the cycle of incrementing the start time by 1.

Step 11, record the word and the data such as the corresponding start, end time and score on the read voice obtained by the above-mentioned steps, repeat steps 2-10, obtain all the words split in step 1 corresponding to the read voice. start time and end time , and the corresponding score , where the i subscript is the sequence number of the word in the text sentence.

Step 12, the word score is lower than the error threshold set by the system, that is, it is characterized as unqualified reading and pronunciation, then call out the text of the current word mapping in the above steps, display it to a specific user interface, remind the user that the word is pronounced incorrectly, and Set the advertisement click function mark to link to the word pronunciation position formed in step 2, and configure the corresponding program so that students can click the advertisement mark to hear the standard word pronunciation. And split the words at the phoneme and syllable levels.

A phoneme is the smallest speech segment that can be recognized by people and has distinctive features that constitute a syllable, and it is the smallest linear unit of speech divided from the sound quality. According to the analysis of pronunciation actions in syllables, one pronunciation action constitutes a phoneme; there are 48 phonemes in the English International Phonetic Alphabet, including 20 vowel phonemes and 28 consonant phonemes. There are 26 English letters in total, including 5 vowels, 2 semi-vowels, and 19 consonants; therefore, the most primitive core standard of English pronunciation can be found by dividing English pronunciation into syllables, phonemes and words. , to help users, especially those who are new to English, accurately and quickly master the elements of English reading aloud, and solve the problems existing in learning English reading aloud. In order to further help users to accurately locate errors and correct the pronunciation of reading aloud, the speech fragments of phoneme words are divided for operation analysis, which can help users learn English reading aloud more effectively and accurately. The phoneme obtained by technical means such as discrete Fourier transform is only the result of a probability-based observation operation, which cannot be used as a textbook standard, and can not match various rules of English pronunciation. It is further divided into syllables and phonemes to help group users analyze. The specific implementation is as follows:

S1. The word word is split into letters according to the bytes. Example: use the MID function, the MID string function, the function is to cut out a specified number of characters from a string, and MID(text, start_num, num_chars) text is required The split word start_num is intercepted from the 1st position from the left, num_chars is intercepted by 1 character length from the left to the right (expressed in numbers), and the character array X(len -1), len is the word byte length.

S2. Create a knowledge base of English phoneme phonetic symbols, which contains 48 phonemes of the International Phonetic Alphabet, front vowels /iː/, /ɪ/, /e/, /æ/; middle vowels...; back vowels. ....; opening and closing diphthongs: /eɪ/, /aɪ/, /ɔɪ/, /aʊ/, /əʊ/;...voiceless consonants....; nasal: / m/, /n/, /ŋ/;......... For each phoneme record, set the corresponding category, reading knowledge point, matching standard voice storage path and its voice acoustic characteristics and other database table columns, record knowledge points such as: /e/ is the vowel before the unit sound, this phonetic symbol is in the British style The symbol in the phonetic symbol is /e/, and the corresponding pronunciation symbol of the American phonetic symbol is [ɛ]. The specific techniques for its pronunciation are: 1) The lips are slightly separated to the sides, and the upper and lower teeth can accommodate about a little finger tip. distance; 2) the front part of the tongue is raised during pronunciation, and the tip of the tongue touches the back of the lower teeth slightly; 3) the jaw gradually moves down during pronunciation, vibrates the vocal cords, and makes the /e/ sound. Note: /e/ is a short vowel, pay attention to the difference from /ɜː/, /ə/, and other knowledge points, other phonemes also record their related knowledge.

S3. Create a comparison library about the correspondence between phonetic symbols, letters, and letter combinations. First, add the 26 letters and commonly used letter combinations and their corresponding phonetic symbols in the English pronunciation rules to the corresponding table of the database corresponding to the rule library, and add such as Two identical vowels are arranged together. For example, oo and ee can only read a special letter combination such as a phonetic symbol in the words foot and meet. At the same time, columns such as letter classification are created to record information such as the categories of letters such as vowels and consonants. That is to say, the information comparing English phonetic symbols and letters and letter combinations of English conventional knowledge is input into the comparison database.

S4. Create a library of English pronunciation rules, sort out various English pronunciation rules into expressions that are convenient for programs that can be logically operated, and classify them according to the way that logical operations can be performed: 1. Character characteristics: those that contain these characteristic keywords are here The recorded phonetic symbol is the default pronunciation of the corresponding letter or letter-letter combination, and this characteristic keyword is combined in the same record with characteristic symbols such as & symbols at intervals, for example: open syllable &accent; 2. Enumeration: will contain the current record The letter or combination of letters is pronounced as the word or sentence of the phonetic symbol of the current record, listed in this record, and different words or sentences are separated by characteristic symbols.

S5. Based on the English pronunciation rule that the e at the end of the word is not pronounced, determine whether the last digit of the X(len-1) character array obtained in step S1 is e, and if so, force the character array to subtract the last member: len=len- 1.

S6. Create two variables, A and Z, and assign the initial value A=0.

S7. If the value of len minus A is greater than or equal to 4 (the conventional letter combination is a combination of 4 letters at most, and the phonetic symbol for pronunciation is [ʃ]), then Z=4, otherwise Z=len-A. Re-assign A: A=A+1, take the A-th to Z-th character combinations in the X(len-1) character array members, and combine them to search in the comparison library of commonly used pronunciation letter combinations. According to the search results, respectively deal with.

1. When multiple records are retrieved, submit the currently combined characters to the custom function guizefunction (currently combined characters, word, A, Z) of step S9 for embedded operation; when A+Z>len, execute step S10 directly, otherwise A assignment: A=A+Z-1 and restart this step.

2. If only the unique record is retrieved, record the phonetic symbol of this record together with the A and Z values. When A+Z>len, execute step S10 directly, otherwise reassign A: A=A+Z-1 and start this step again .

3. If no record is retrieved, go to the next step.

S8, if Z=1, jump to the above S7 step, otherwise assign Z to Z-1, take the A to Z characters, and combine them to search in the rule base of commonly used pronunciation letter combinations, according to the retrieval The results are processed separately.

1. If only the unique record is retrieved, record the phonetic symbol of this record together with the A and Z values, and reassign A: A=A+Z-1 and directly jump to step S7.

2. If multiple records are retrieved, submit the current combination of characters to the custom function guizefunction (currently combined characters, word, A, Z) in step S9 for embedded operation; A=A+Z-1 and directly jump to execute S7 step.

3. If there is no record, start this step repeatedly and perform cyclic analysis until the Z value is 1.

S9, the custom function of rule operation guizefunction (str, str1, Index1, index2),

a. Retrieve the str string in the English pronunciation rule base, and check whether the content recorded in the "List" column of the retrieved record contains the current word. If it does, return the phonetic symbol in the current record as the result of this function. Record with the values of Index1 and index2, and terminate the operation of this function, if not, perform the next record check;

b. Determine whether the letter in the next byte position of str in str1 is a consonant letter, first set two character variables tex, texx, such as index1+index2+1>len(word) then tex=right(word, 1) otherwise tex=MID(str1,index1+index2+1, 1), if tex is "r" or "w" or "y", then texx is assigned as "open syllable", otherwise it searches the rule base for letters or letter combinations that are tex Record, if the record in the "letter classification column" in the record contains "vowel", then texx is assigned as "open syllable", otherwise texx is assigned as "closed syllable".

C. Retrieve the str character in the rule base, and verify one by one whether the content of the characteristic keyword column in the record has the content of texx, if so, return the phonetic symbol corresponding to this record to this function, and combine the phonetic symbol with the values of Index1 and index2 record, otherwise verify the next record that meets the condition.

S10, splitting the English text words, letters and letter combinations and their corresponding phonemes and phonetic symbols, retrieving relevant knowledge points, and displaying them on the user interface, so that users can learn and master.

In actual English reading, phonemes often change the original standard pronunciation due to the upper and lower phonemes, and more syllables are used as the basic unit for learning English reading pronunciation. Therefore, splitting words into syllables can effectively help users master the correct reading. .

According to the pronunciation rules of English words: the principles to be followed by dividing words into syllables: 1) After the first return, it refers to between two syllables. If there is a consonant letter, the letter is assigned to the next syllable; Between two syllables, if there are two consonants, they will be assigned to the two syllables before and after respectively; error prone: vowels and vowels are confused. Syllables are divided according to vowels, if vowels are not pronounced , it cannot form a syllable. If there are two vowels together, but only one vowel is pronounced, it still counts as a syllable. The rules of syllable division: vowel phonemes are the main body of syllables, and consonants are the boundaries of syllables. Each vowel phoneme can form a syllable, so it can be divided into syllables to help users learn English reading aloud more accurately. The implementation of syllable separation of English words:

Step 1, a group of phonetic symbols corresponding to English word text letters and letter combinations that are recorded by the technical scheme that the above words are split into phonemes: , , ,… , and the starting position of each phonetic symbol corresponding to the letter: , , ,… , ending at: , , ,… .

Step 2. Search phonemes in the knowledge base respectively to obtain one or a group of phonemes classified as vowels by phonemes.

Step 3. According to the position value corresponding to the phoneme in step 1, use functions such as mid in turn to obtain the letters between the letters corresponding to the vowels or the letter groups. If there is only one letter, then subtract 1 from the starting position value corresponding to the next vowel. , such as two letters, add 1 to the end position value corresponding to the previous vowel.

Step 4, obtain a group of phonemes and the values of the new starting position and ending position through step 3, calculate and obtain the corresponding letter or letter group according to the starting position and ending position corresponding to this group of phonemes, and output to the user interface as The result of word splitting syllables.

It is specially stated that "embodiments" and the like described in this specification refer to the specific features, elements or characteristics described in conjunction with the embodiments being included in the embodiments generally described in this application. The appearance of the same expression in various places in the specification is not intended to be limiting in particular to the same embodiment. That is to say, when a specific feature, element or characteristic is described in conjunction with any embodiment, what is claimed is to realize such feature, element or characteristic in conjunction with other embodiments, and it is included in the scope of the claims protected by the present application; The present invention is described with reference to multiple explanatory embodiments of the present invention's logical architecture and ideas, but the protection scope of the present invention is not limited to this, and those skilled in the art can design many other Modifications and implementations, various non-essential modifications and improvements can be made to the key point transformation combination/or layout of the technical solution, and other uses will also be obvious to those skilled in the art, and it is easy to think of the implementation of non-essential modifications and implementations. Substantial changes or substitutions, such modifications and embodiments will fall within the scope and spirit of the principles disclosed herein.

Description of drawings

Figure 1 is the overall logical frame diagram of a method for accurate correction and learning based on automatic scoring of English reading aloud.

Claims (4)

1. a method based on automatic scoring of English reading aloud to carry out the method of splitting operation accurate correction learning, the steps and elements that it is characterized in include: the reading aloud scoring is divided into test mode and practice evaluation mode, and the test mode only carries out test scoring to the reading voice , and in the practice evaluation mode, when the pronunciation of the English sentence is up to the standard, it will directly enter the reading of the next text. The text is divided into word groups, and the standard speech and acoustic characteristics of the words are obtained as the standard reference model. First, assuming that the length of time the word is read aloud is the standard time length, and then divide the speech segments of this length in turn on the tested speech for comparison. The speech segment with the highest score; Step 1. English text and Chinese text are different. English text text is separated by spaces. Therefore, through functions such as split, spaces are used as characters that identify the boundaries of substrings, and English text is turned into a word array a composed of individual words. That is, a=Split(text); the continuous letter combination containing the abbreviation ' is regarded as a word; Step 2, obtain the voice of the specified English word through the voice interface of the third party, embodiment: submit the English text post to the website of the Baidu voice development platform, obtain the voice file of the format such as returning mp3; (or through the engine of text-to-speech etc. get the speech of a specific text word); Step 3, and obtain the voice feature by pre-analysis, convert into a new standard reference model M, record the duration S of the word voice simultaneously, and presuppose the duration S that the word text is read aloud by test; Step 4. Take the tested reading voice, the starting time 1 and the ending time are S. This interval is the new tested reading voice, and compare the operation with M in step 3, through formula 1, formula 2, formula 3 Calculate the score J; Step 5. Take the tested reading voice, add 1 to the starting time and add 1 to the ending time as a new tested reading voice group, until the ending time is equal to the duration of the original tested voice, respectively and step 3. The M of the comparison operation is performed, and the score is calculated by formula 1, formula 2, and formula 3; Step 6: Compare the scores calculated in Step 4 and Step 5 to obtain the maximum value A, and parameters such as the start time T1 and the end time T2 corresponding to the maximum value. 2. a kind of according to claim 1 based on English reading aloud automatic scoring to carry out the accurate correction learning method of split operation, the step and the element that it is characterized in that also comprise: the speech segment of the highest score value obtained based on claim 1, then carry out forward The correction of the backward addition and subtraction duration is to obtain the more ideally matched speech segment of the word, followed by the steps of claim 1: Step 7. Take the tested reading voice, add T1 to the starting time and cyclically decrease 1, and the end time T2 is the new tested reading voice group, until the starting time decremented by 1 is equal to 1, and the cycle operation is obtained. The acoustic features of the speech segment are compared with the M in step 3 to obtain a score, and the score is compared with the A score in step 6. If the score is greater than A, the value of A is set as the current score and Set T1 as the starting time corresponding to the current score. If the score is less than A, it will jump out of the cycle of decreasing the starting time by 1; Step 8. Decrease the end time T2 cyclically by 1 and start time T1 as the new tested speech group until the end time decremented by 1 is equal to T1. The acoustic characteristics of the obtained speech segment in the cyclic operation are the same as those in step 3. The M is compared to obtain the score, and the score is compared with the A score in step 7. If the score is greater than A, the value of A is set as the current score and T2 is set as the corresponding value of the current score. The end time of , if the score is less than A, jump out of the cycle where the start time is decremented by 1; Step 9. Take the tested reading voice, add 1 to the starting time and T1 cyclically, and add 1 to the ending time T2 as a new tested reading voice group, until the starting time that increases by 1 is equal to T2. The acoustic features of the obtained speech segment are compared with the M in step 3 to obtain a score, and the score is compared with the A score in step 8. If the score is greater than A, the value of A is set as the current score. And set T1 as the starting time corresponding to the current score, if the score is less than A, it will jump out of the cycle of decreasing the starting time by 1; Step 10. Cycle the end time T2 by 1 and the start time T1 as the new tested speech group, until the end time incremented by 1 is equal to the overall duration of the original tested speech, and the speech segment obtained in the cyclic operation The acoustic features of the A are compared with the M in step 3 to obtain the score, and the score is compared with the A score in step 9. If the score is greater than A, the value of A is set as the current score and T2 Set as the end time corresponding to the current score. If the score is less than A, it will jump out of the cycle of incrementing the start time by 1; Step 11, record the word and the data such as the corresponding start, end time and score on the read voice obtained by the above-mentioned steps, repeat steps 2-10, obtain all the words split in step 1 corresponding to the read voice. start time and end time , and the corresponding score , where the i subscript is the sequence number of the word in the text sentence; Step 12, the word score is lower than the error threshold set by the system, that is, it is characterized as unqualified reading and pronunciation, then call out the text of the current word mapping in the above steps, display it to a specific user interface, remind the user that the word is pronounced incorrectly, and Set the advertisement click function mark to broadcast the link to the word pronunciation position formed in step 2, configure the corresponding program so that students can click the advertisement mark to hear the standard word pronunciation, and conduct phoneme and syllable level split analysis of the word. 3. a method for splitting and analyzing the unqualified word of pronunciation score is carried out, and the step and the element that its feature comprises are: S1. The word word is split into letters according to the bytes; S2. Create a knowledge base of English phoneme phonetic symbols, which contains 48 phonemes of the International Phonetic Alphabet, and set the corresponding category, reading knowledge point, matching standard voice storage path and its voice acoustic characteristics and other database table columns for each phoneme record, and record the knowledge point; S3. Create a comparison library about the correspondence between phonetic symbols, letters, and letter combinations. First, add the 26 letters and commonly used letter combinations and their corresponding phonetic symbols in the English pronunciation rules to the corresponding table of the database corresponding to the rule library, and add such as Special letter combinations such as two identical vowels are arranged together, and columns such as letter classification are created at the same time to record information such as the categories of letters such as vowels and consonants; S4. Create a library of English pronunciation rules, sort out various English pronunciation rules into expressions that are convenient for programs that can be logically operated, and classify them according to the way that logical operations can be performed: 1. Character characteristics: those that contain these characteristic keywords are here The recorded phonetic symbol is the default pronunciation of the corresponding letter or letter-letter combination, and this characteristic keyword is combined in the same record with characteristic symbols such as & symbols at intervals, for example: open syllable &accent; 2. Enumeration: will contain the current record The letter or combination of letters is pronounced as the word or sentence of the phonetic symbol of the current record, listed in this record, and different words or sentences are separated by characteristic symbols; S5. Based on the English pronunciation rule that the e at the end of the word is not pronounced, determine whether the last digit of the X(len-1) character array obtained in step S1 is e, and if so, force the character array to subtract the last member: len=len- 1; S6. Create two variables, A and Z, and assign the initial value A=0; S7. If the value of len minus A is greater than or equal to 4, then Z=4, otherwise Z=len-A; reassign A: A=A+1, take the first A in the X(len-1) character array member Combinations of characters from the first to the Zth character are combined to search in the comparison library of commonly used pronunciation letter combinations, and are processed separately according to the search results; When multiple records are retrieved, the currently combined characters are submitted to the custom function guizefunction (currently combined characters, word, A, Z) of step S9 for embedded operation; when A+Z>len, step S10 is directly executed, otherwise A is assigned a value : A=A+Z-1 and restart this step; if only the unique record is retrieved, record the phonetic symbol of this record together with the A and Z values, when A+Z>len directly execute step S10, otherwise assign A value: A =A+Z-1 and restart this step; if no record is retrieved, go to the next step; S8, if Z=1, jump to the above S7 step, otherwise assign Z to Z-1, take the A to Z characters, combine them to search in the rule base of commonly used pronunciation letter combinations, and according to the search results Separate processing: if only the unique record is retrieved, record the phonetic symbol of this record together with the A and Z values, assign A value: A=A+Z-1 and directly jump to step S7; If multiple records are retrieved, submit the currently combined characters to the custom function guizefunction (currently combined characters, word, A, Z) of step S9 for embedded operation; A=A+Z-1 and directly jump to execute step S7, If there is no record, start this step repeatedly, and perform cyclic analysis until the Z value is 1; S9, the custom function of rule operation guizefunction (str, str1, Index1, index2), a. Retrieve the str string in the English pronunciation rule base, and check whether the content recorded in the "List" column of the retrieved record contains the current word. If it does, return the phonetic symbol in the current record as the result of this function. Record with the values of Index1 and index2, and terminate the operation of this function, if not, perform the next record check; b. Determine whether the letter in the next byte position of str in str1 is a consonant letter, first set two character variables tex, texx, such as index1+index2+1>len(word) then tex=right(word, 1) otherwise tex=MID(str1,index1+index2+1, 1), if tex is "r" or "w" or "y", then texx is assigned as "open syllable", otherwise it searches the rule base for letters or letter combinations that are tex Record, if the record in the "letter classification column" in the record contains "vowel", then texx is assigned as "open syllable", otherwise texx is assigned as "closed syllable"; C. Retrieve the str character in the rule base, and verify one by one whether the content of the characteristic keyword column in the record has the content of texx, if so, return the phonetic symbol corresponding to this record to this function, and combine the phonetic symbol with the values of Index1 and index2 record, otherwise verify the next record that meets the conditions; S10, splitting the English text words, letters and letter combinations and their corresponding phonemes and phonetic symbols, retrieving relevant knowledge points, and displaying them on the user interface, so that users can learn and master. 4. according to claim 1 a kind of method that pronunciation score unqualified word is carried out to split analysis operation, the step and element that it is characterized in also comprising: the technical scheme of English word syllable split; Step 1, the starting position, the ending position of one group of phonetic symbols corresponding to English word text letters and letter combinations and the corresponding letters of each phonetic symbol recorded by claim 3; Step 2, retrieve phonemes in the knowledge base respectively, obtain one or a group of phonemes classified as vowels; Step 3. According to the position value corresponding to the phoneme in step 1, use functions such as mid in turn to obtain the letters between the letters corresponding to the vowels or the letter groups. If there is only one letter, then subtract 1 from the starting position value corresponding to the next vowel. , such as two letters, add 1 to the end position value corresponding to the previous vowel; Step 4, obtain a group of phonemes and the values of the new starting position and ending position through step 3, calculate and obtain the corresponding letter or letter group according to the starting position and ending position corresponding to this group of phonemes, and output to the user interface as The result of word splitting syllables.