An artificial intelligence-based english oral english intelligent correction method and system
By collecting and analyzing users' pronunciation signals through an artificial intelligence system, pronunciation errors can be corrected in a targeted manner, solving the problem that students cannot correct themselves and achieving efficient error correction and improvement of English speaking skills.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN POLYTECHNIC
- Filing Date
- 2026-02-04
- Publication Date
- 2026-06-05
AI Technical Summary
In current English speaking practice, students are unable to correct their own pronunciation errors, resulting in inaccurate pronunciation and affecting their English speaking ability.
By collecting users' pronunciation signals through an artificial intelligence system, comparing and analyzing them, predicting pronunciation accuracy, correcting pronunciation in a targeted manner, providing personalized pronunciation correction data, and correcting errors through differentiated emphasis on pronunciation.
It improves the efficiency and effectiveness of English speaking practice, enabling users to better correct pronunciation errors and improve their English speaking skills.
Smart Images
Figure CN122157694A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, specifically to an intelligent error correction method and system for spoken English based on artificial intelligence. Background Technology
[0002] Spoken English is a core communication skill in the era of globalization. It is not only a language tool, but also a key driver for personal development, career advancement, cross-cultural communication, and the expansion of thinking. Its importance spans multiple dimensions, including personal, professional, academic, and daily life.
[0003] Current English pronunciation practice generally involves face-to-face practice and shadowing between students or between students and teachers. This method has many drawbacks. Students are prone to inaccurate pronunciation during shadowing and are unable to correct their own pronunciation. Over time, this can lead to poor spoken English, more Chinglish, and negatively impact English listening and communication skills. Summary of the Invention
[0004] Based on the above description, this invention provides an intelligent English oral pronunciation correction method based on artificial intelligence, which can exaggerate the pronunciation of non-standard syllables of users in a targeted manner, so as to achieve a better error correction effect.
[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: An AI-based intelligent error correction method for spoken English involves guiding users to repeat the pronunciation of several sets of test syllables selected in the current course before practice begins, thereby collecting audio signals when users pronounce each set of test syllables; the collected audio signals are then identified and compared with standard pronunciation signals to measure the accuracy of the user's pronunciation of each test syllable. After training begins, a training example sentence is selected from the learning database and its standard pronunciation is played; after playback, the user is reminded to repeat it for the first time. Simultaneously, based on the pronunciation accuracy measured above, the proficiency of each word in the training example sentence is predicted; based on the predicted proficiency, each word in the training example sentence is sorted from low to high to obtain a word sequence; a certain proportion of words are selected from the word sequence, i.e., target words. Obtain the standard pronunciation data of the training example sentences, and correct the pronunciation data of the test syllables in the target words according to the proficiency of each target word, the stress, volume and pronunciation duration of the target words, to obtain personalized word correction pronunciation data; Play the corrected and emphasized pronunciation based on the word-corrected pronunciation data to the user, and remind the user to repeat it a second time.
[0006] As a preferred approach: the collected audio signals of each group of syllables are input into a pre-trained evaluation model, which then compares the collected audio signals with the corresponding standard audio signals of the syllables to determine the accuracy of the user's pronunciation of each group of syllables. Pronunciation accuracy calculation formula: D=a×J+b×K, where D represents the accuracy value, J represents the similarity of pronunciation, K represents the similarity of intonation, and a and b are preset calculation coefficients.
[0007] As a preferred approach, the word proficiency is predicted using the formula S=p×F×(D1+D2+……+Dn) / n, where S represents the predicted word proficiency, n represents the number of test syllables in the word, F represents the value corresponding to the numerical range of the number of test syllables, and p is a random factor with a value range of 0.7-1.
[0008] As a preferred option, the random factor p takes values in a segmented probability formula, with its value range divided into three segments: 0.7-0.8, 0.8-0.9, and 0.9-1. When a word contains only stressed syllables, the probability of p being in the range of 0.7-0.8 is m1%, the probability of p being in the range of 0.8-0.9 is m2%, and the probability of p being in the range of 0.9-1 is m3%. When a word contains only linked syllables or only weak syllables, the probability of p being in the range of 0.7-0.8 is m2%, the probability of p being in the range of 0.8-0.9 is m3%, and the probability of p being in the range of 0.9-1 is m1%. When a word contains both linked and weak syllables and no stressed syllable, the probability of p being in the range of 0.7-0.8 is m3%, the probability of p being in the range of 0.8-0.9 is m2%, and the probability of p being in the range of 0.9-1 is m2%. Where m1+m2+m3=100, and m3>m2>m1.
[0009] As a preferred approach, the process of obtaining corrected pronunciation data is as follows: After obtaining the target word, determine which test syllables exist in the target word, extract the pronunciation data of these test syllables from the standard pronunciation data of the target word to obtain each set of data packets, restore the audio data in the data packets to analog audio signals, and perform pitch shifting, amplitude amplification, and waveform splicing processing on the analog audio signals. The processed analog audio signals have a heavier emphasis, higher volume, and longer pronunciation duration. Then, convert the processed analog audio signals into digital signals and splice and package the processed digital signals with the pronunciation digital signals corresponding to other syllables in the word to obtain the corrected pronunciation data of each set of target words.
[0010] Another objective of this application is to provide an AI-based English oral error correction system that performs the above-described methods, characterized in that the system comprises: The sound acquisition module is used to acquire the user's pronunciation and output audio signals; The microprocessor module is used to preprocess the audio signal; compare the acquired audio signal of the test syllable with the standard pronunciation signal of the test syllable, and output the accuracy of the user's pronunciation of the test syllable; predict the word proficiency based on the pronunciation accuracy of the test syllable, and correct the stress, volume and pronunciation duration of the test syllable in words with low proficiency, so as to obtain personalized corrected pronunciation data. The storage module, connected to the microprocessor module, is used to store pronunciation data and calculation parameters; The display screen, connected to the microprocessor module, is used to display test syllables and training example sentences; The sound module, connected to the microprocessor module, is used to play sound; The power module is used to supply power to the system.
[0011] As a preferred embodiment: the microprocessor module includes a signal processing unit, a sound recognition unit, a computational processing unit, a data correction unit, and a signal restoration unit; the signal processing unit is used to perform filtering and noise reduction preprocessing on the audio signal; the sound recognition unit is used to recognize the audio signal of the acquired test syllable and compare it with the standard pronunciation signal of the test syllable, and output the user's pronunciation accuracy of the test syllable; the computational processing unit is used to predict the proficiency of words based on the pronunciation accuracy of the test syllables contained in the training example sentences, and sort the words in each group of test example sentences according to the proficiency to obtain a word sequence; the data correction unit is used to select a certain proportion of words from the word sequence, i.e., target words, and correct the stress, volume, and pronunciation duration of the test syllables in the target words according to the proficiency of each target word to obtain personalized corrected pronunciation data.
[0012] As a preferred embodiment, the microprocessor module further includes a probability value-taking unit, which is used to randomly select values for the calculation parameters in the calculation processing unit based on the stress, linking, and weak pronunciation of the target word.
[0013] Compared with existing technologies, the technical solution of this application has the following beneficial effects: This application predicts the proficiency of each word in the training example sentences. For words with low proficiency, the pronunciation of the test syllables they contain is specifically corrected. When the corrected words are restored and pronounced, the test syllables are exaggerated, giving different words a differentiated auditory experience. This can effectively remind and correct users, deepening their impression of correcting pronunciation errors for unfamiliar words. Compared with traditional oral training methods that rely on repeated listening and reading for training and correction, which is monotonous and tedious, this application allows words to have different auditory experiences, providing users with a better effect on correcting oral pronunciation errors. Attached Figure Description
[0014] Figure 1 This is a flowchart of the method in Example 1; Figure 2 This is a system block diagram from Example 2. The attached diagram lists the components represented by each number as follows: 1. Sound acquisition module; 2. Microprocessor module; 3. Sound output module; 4. Display screen; 5. Storage module; 6. Power supply module. Detailed Implementation
[0015] Example 1: Refer to Figure 1 An AI-based intelligent error correction method for spoken English, specifically: Before formal practice begins, a basic pronunciation test is required to assess the user's pronunciation level. The test process involves pre-selecting several sets of test syllables for each course based on its difficulty level. During the test, the most basic course is automatically selected, and the user is guided to repeat the pronunciation of several sets of test syllables selected for the current course to collect audio signals from the user's pronunciation of each set of test syllables.
[0016] The pre-trained speech recognition model identifies the collected audio signals and compares them with standard pronunciation signals to measure the accuracy of the user's pronunciation of each test syllable. This step allows for a test of the user's basic pronunciation level.
[0017] The speech recognition model in this embodiment has a preset formula for calculating pronunciation accuracy: D = a × J + b × K, where D represents the accuracy value, J represents the pronunciation similarity, K represents the intonation similarity, and a and b are preset calculation coefficients. The speech recognition model calculates the pronunciation similarity J by comparing the waveform of the user's test syllable pronunciation with the waveform of the standard pronunciation of that syllable; it then averages the frequency of the user's test syllable pronunciation with the frequency of the standard pronunciation, and calculates the difference between the two average frequencies. Several pre-defined ranges of average frequency differences are set, and each range is assigned a value, defined as the intonation similarity value K. Larger values at the beginning and end of the range result in smaller assigned values. After obtaining the values of J and K, the pronunciation accuracy of the current test syllable can be calculated using the above formula.
[0018] After completing the basic pronunciation level test, you can begin formal pronunciation training.
[0019] It should be noted that a learning database needs to be established in advance, with corresponding training example sentences preset for different courses, and these training example sentences saved in the learning database. Most of the words in the training example sentences for each course contain the test syllables selected for that course.
[0020] Once training begins, a training example sentence is selected from the learning database corresponding to the current course, and its standard pronunciation is played. After the playback ends, the user is reminded to repeat it for the first time.
[0021] While reminding users to repeat after the speaker, the system predicts the proficiency level of each word in the training example sentences based on the previously measured pronunciation accuracy.
[0022] In this embodiment, the word proficiency is predicted using the formula S=p×F×(D1+D2+……+Dn) / n, where S represents the predicted word proficiency, n represents the number of test syllables in the word, F represents the value assigned to the numerical range of the number of test syllables, and p is a random factor with a value range of 0.7-1.
[0023] For example, if the current word contains three sets of test syllables, and the pronunciation accuracy of these three sets of test syllables is retrieved to obtain the values of D1, D2, and D3, then n=3 in the above formula.
[0024] In this embodiment, several numerical intervals are set for the number of test syllables contained in a word, and a value is assigned to each numerical interval. The larger the first and last values of the numerical interval, the smaller the assigned value F. During calculation, the numerical interval in which n belongs is determined based on the value of n, and then the assigned value F of the corresponding numerical interval is retrieved. Then, a random factor p is randomly generated, and the proficiency S of the current word is predicted using the above calculation formula.
[0025] As a preferred embodiment, the random factor p in this embodiment is a segmented probability value, and its value range is divided into three segments: 0.7-0.8, 0.8-0.9, and 0.9-1.
[0026] When a word contains only stressed syllables, the probability of p being in the range of 0.7-0.8 is m1%, the probability of p being in the range of 0.8-0.9 is m2%, and the probability of p being in the range of 0.9-1 is m3%. When a word contains only linked syllables or only weak syllables, the probability of p being in the range of 0.7-0.8 is m2%, the probability of p being in the range of 0.8-0.9 is m3%, and the probability of p being in the range of 0.9-1 is m1%. When a word contains both linked and weak syllables and no stressed syllable, the probability of p being in the range of 0.7-0.8 is m3%, the probability of p being in the range of 0.8-0.9 is m2%, and the probability of p being in the range of 0.9-1 is m2%. Where m1 + m2 + m3 = 100, and m3 > m2 > m1. This strategy can improve the reliability of predictions.
[0027] The above steps can be used to predict the proficiency level of each word in the training example sentences.
[0028] Based on the predicted proficiency, the words in the training example sentences are sorted from low to high to obtain a word sequence.
[0029] Select a certain percentage of words from the word sequence (e.g., 30% of the words; if the percentage is not an integer, round it off). These are the target words.
[0030] Then, the standard pronunciation data of the current training example sentence is obtained, and the stress, volume and pronunciation duration of the test syllables in the target words are corrected according to the proficiency of each target word to obtain personalized corrected pronunciation data.
[0031] The specific processing steps are as follows: After obtaining the target word, determine which test syllables exist in the target word, extract the pronunciation data of these test syllables from the standard pronunciation data of the target word to obtain each set of data packets, restore the audio data in the data packets to analog audio signals, and perform pitch boosting, amplitude amplification, and waveform splicing processing on the analog audio signals. The processed analog audio signals have a heavier emphasis, higher volume, and longer pronunciation duration. Then, convert the processed analog audio signals into digital signals and splice and package the processed digital signals with the pronunciation digital signals corresponding to other syllables in the word to obtain the corrected pronunciation data of each set of target words.
[0032] The lower the proficiency of the target word, the more pitches are raised, the greater the amplitude is amplified, and the longer the waveform splicing is when processing the test syllables it contains.
[0033] The technical effect achieved through the above steps is as follows: For words with low proficiency, the pronunciation of the test syllables is specifically corrected. When the corrected words are restored and pronounced, the test syllables are exaggerated, giving different words a differentiated auditory experience. This can effectively remind and correct users, deepening their impression of correcting pronunciation errors for unfamiliar words. Compared to traditional oral training methods that rely on repeated listening and reading for training and correction, which can be monotonous and tedious, this application allows words to have different auditory experiences, providing users with a better effect on correcting oral pronunciation errors.
[0034] Play the corrected and emphasized pronunciation based on the corrected pronunciation data to the user, and remind the user to repeat it a second time.
[0035] As users practice more, more challenging courses can be introduced to test and correct their mistakes, gradually helping to improve their spoken English.
[0036] Example 2: Refer to Figure 2 An AI-based intelligent error correction system for spoken English includes: Sound acquisition module 1, which is used to acquire the user's pronunciation and output audio signals; Microprocessor module 2 is used to preprocess the audio signal; compare the acquired audio signal of the test syllable with the standard pronunciation signal of the test syllable, and output the accuracy of the user's pronunciation of the test syllable; predict the proficiency of the word based on the accuracy of the pronunciation of the test syllable, and correct the stress, volume and pronunciation duration of the test syllable in words with low proficiency, so as to obtain personalized corrected pronunciation data.
[0037] Storage module 5, connected to the microprocessor module, is used to store pronunciation data and calculation parameters; Display screen 4, connected to the microprocessor module, is used to display test syllables and training example sentences; The sound module 3 is connected to the microprocessor module and is used to play sound; Power module 6 is used to supply power to the system.
[0038] The microprocessor module 2 in this embodiment includes a signal processing unit 201, a sound recognition unit 202, a calculation processing unit 203, a data correction unit 204, and a signal restoration unit 205.
[0039] The signal processing unit 201 is used to perform filtering and noise reduction preprocessing on the audio signal; the sound recognition unit 202 is used to recognize the audio signal of the acquired test syllable and compare it with the standard pronunciation signal of the test syllable, and output the user's pronunciation accuracy of the test syllable; the calculation processing unit 203 is used to predict the proficiency of words based on the pronunciation accuracy of the test syllables contained in the training example sentences, and sort the words in each group of test example sentences according to the proficiency to obtain a word sequence; the data correction unit 204 is used to select a certain proportion of words from the word sequence, i.e., target words, and correct the stress, volume, and pronunciation duration of the test syllables in the target words according to the proficiency of each target word to obtain personalized corrected pronunciation data; the signal restoration unit 205 is used to restore the digital audio signal to an analog audio signal.
[0040] The microprocessor module in this embodiment also includes a probability value-taking unit, which is used to randomly select values for the calculation parameters in the calculation processing unit based on the stress, linking, and weak pronunciation of the target word.
[0041] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. An intelligent error correction method for spoken English based on artificial intelligence, characterized by: Before the practice begins, the user is guided to read aloud several sets of test syllables selected in the current course to collect audio signals when the user reads each set of test syllables; the collected audio signals are identified and compared with standard pronunciation signals to measure the accuracy of the user's pronunciation of each test syllable. After training begins, a training example sentence is selected from the learning database and its standard pronunciation is played; after playback, the user is reminded to repeat it for the first time. Simultaneously, based on the pronunciation accuracy measured above, the proficiency of each word in the training example sentence is predicted; based on the predicted proficiency, each word in the training example sentence is sorted from low to high to obtain a word sequence; a certain proportion of words are selected from the word sequence, i.e., target words. Obtain the standard pronunciation data of the training example sentences, and adjust the stress, volume and pronunciation duration of the test syllables in the target words according to the proficiency of each target word to obtain personalized word pronunciation data; Play the corrected and emphasized pronunciation based on the word-corrected pronunciation data to the user, and remind the user to repeat it a second time.
2. The intelligent English speaking correction method based on artificial intelligence according to claim 1, characterized in that: The collected audio signals of each group of syllables are input into a pre-trained evaluation model. The evaluation model compares the collected audio signals with the corresponding standard audio signals of the syllables to determine the accuracy of the user's pronunciation of each group of syllables. Pronunciation accuracy calculation formula: D=a×J+b×K, where D represents the accuracy value, J represents the similarity of pronunciation, K represents the similarity of intonation, and a and b are preset calculation coefficients.
3. The English oral error correction method based on artificial intelligence according to claim 1, characterized in that: The word proficiency is predicted using the formula S=p×F×(D1+D2+……+Dn) / n, where S represents the predicted word proficiency, n represents the number of test syllables in the word, F represents the value corresponding to the numerical range of the number of test syllables, and p is a random factor with a value range of 0.7-1.
4. The English oral error correction method based on artificial intelligence according to claim 3, characterized in that: The random factor p takes values in a segmented probability formula, and its value range is divided into three segments: 0.7-0.8, 0.8-0.9, and 0.9-1. When a word contains only stressed syllables, the probability of p being in the range of 0.7-0.8 is m1%, the probability of p being in the range of 0.8-0.9 is m2%, and the probability of p being in the range of 0.9-1 is m3%. When a word contains only linked syllables or only weak syllables, the probability of p being in the range of 0.7-0.8 is m2%, the probability of p being in the range of 0.8-0.9 is m3%, and the probability of p being in the range of 0.9-1 is m1%. When a word contains both linked and weak syllables and no stressed syllable, the probability of p being in the range of 0.7-0.8 is m3%, the probability of p being in the range of 0.8-0.9 is m2%, and the probability of p being in the range of 0.9-1 is m2%. Where m1+m2+m3=100, and m3>m2>m1.
5. The English oral error correction method based on artificial intelligence according to claim 3, characterized in that, The process of obtaining corrected pronunciation data is as follows: After obtaining the target word, determine which test syllables exist in the target word, extract the pronunciation data of these test syllables from the standard pronunciation data of the target word to obtain each set of data packets, restore the audio data in the data packets to analog audio signals, and perform pitch shifting, amplitude amplification, and waveform splicing processing on the analog audio signals. The processed analog audio signals have a heavier emphasis, higher volume, and longer pronunciation duration. Then, convert the processed analog audio signals into digital signals and splice and package the processed digital signals with the pronunciation digital signals corresponding to other syllables in the word to obtain the corrected pronunciation data of each set of target words.
6. An artificial intelligence-based English oral error correction system, characterized in that, include: The sound acquisition module is used to acquire the user's pronunciation and output audio signals; The microprocessor module is used for preprocessing audio signals; The audio signal of the collected test syllable is compared with the standard pronunciation signal of the test syllable to output the accuracy of the user's pronunciation of the test syllable; the word proficiency is predicted based on the pronunciation accuracy of the test syllable, and the stress, volume and pronunciation duration of the test syllable in words with low proficiency are corrected to obtain personalized corrected pronunciation data. The storage module, connected to the microprocessor module, is used to store pronunciation data and calculation parameters; The display screen, connected to the microprocessor module, is used to display test syllables and training example sentences; The sound module, connected to the microprocessor module, is used to play sound; The power module is used to supply power to the system.
7. The English oral error correction system based on artificial intelligence according to claim 6, characterized in that: The microprocessor module includes a signal processing unit, a sound recognition unit, a computational processing unit, a data correction unit, and a signal restoration unit. The signal processing unit performs filtering and noise reduction preprocessing on the audio signal. The sound recognition unit identifies the audio signal of the acquired test syllable and compares it with the standard pronunciation signal of the test syllable, outputting the user's pronunciation accuracy of the test syllable. The computational processing unit predicts the word proficiency based on the pronunciation accuracy of the test syllables in the training example sentences, and sorts the words in each group of test example sentences according to the proficiency to obtain a word sequence. The data correction unit selects a certain proportion of words from the word sequence, i.e., target words, and corrects the stress, volume, and pronunciation duration of the test syllables in the target words according to the proficiency of each target word to obtain personalized corrected pronunciation data.
8. The English oral error correction system based on artificial intelligence according to claim 7, characterized in that: The microprocessor module also includes a probability value-taking unit, which is used to randomly select values for the calculation parameters in the calculation processing unit based on the stress, linking, and weak pronunciation of the target word.