English listening and speaking training data processing system and method based on artificial intelligence

Abstract

The invention discloses an English listening and speaking training data processing system and an English listening and speaking training data processing method based on artificial intelligence, and relates to the field of data processing. Dialect category labels of each row of data in a preprocessed first oral English audio data matrix are generated, so that a classification standard is provided for subsequently constructing an LEAN neural network model; by constructing a final LEAN neural network model, an identification model is provided for subsequent dialect sound identification of collected listening and speaking audio data, and then targeted analysis is performed on the listening and speaking audio data with dialect sound, so that a trainer is helped to correct; characteristic extraction is carried out on listening and speaking audio data corresponding to audio category label data with the maximum occurrence probability, extracted characteristic data is compared with characteristic data of source audio, clustering operation is carried out on comparison results, and the comparison data of the category with the maximum data size serves as training reference data. And the reference data is more credible and accurate.

Description

Technical Field

[0001] The present invention belongs to the field of data processing, and in particular, relates to an English listening and speaking training data processing system and method based on artificial intelligence. Background Art

[0002] Chinese patent CN114547154B discloses an intelligent interactive English training method, system and computer-readable medium, which starts with vocabulary training and improves comprehensive English ability through reading, listening, writing and speaking exercises; during the learning process, learners can not only strengthen their mastery of new words, but also avoid mechanical memorization of words; the difficulty will be dynamically adjusted according to the user's training situation to ensure that they are always trained at a difficulty level suitable for them; at the same time, the user's training situation is recorded to more clearly understand the training effect.

[0003] At present, in the process of training English listening and speaking skills, due to the personal growth environment of students, they may have a certain degree of dialect accent, which has a great influence on the correct pronunciation of English; and if these English listening and speaking training data with dialect accents cannot be processed in a targeted manner, it will make it difficult for students to correct their accents in a targeted manner; ultimately resulting in students' English accents being non-standard. Summary of the invention

[0004] In view of the problems in the related art, the present invention proposes an English listening and speaking training data processing system and method based on artificial intelligence to overcome the above-mentioned technical problems existing in the existing related art.

[0005] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:

[0006] The present invention is an artificial intelligence-based English listening and speaking training data processing method, comprising the following steps:

[0007] S1, collecting English spoken audio data of multiple different dialect sound types and non-dialect sound types to obtain a first English spoken audio data matrix; preprocessing the first English spoken audio data matrix;

[0008] S2, generating a dialect category label for each row of data in the preprocessed first spoken English audio data matrix to obtain a first dialect category label data set; using the preprocessed first spoken English audio data matrix and the corresponding dialect category label data to construct a final LEAN neural network model;

[0009] S3, collect multiple copies of English listening and speaking source audio data, listening and speaking audio data corresponding to the source audio data, and text data corresponding to the source audio data, to obtain a source audio data set, a listening and speaking audio data matrix, and a source audio text data set; use the final LEAN neural network model to perform audio type recognition on the preprocessed listening and speaking audio data matrix to obtain a second audio category label data matrix;

[0010] S4, judging whether the preprocessed listening and speaking audio data matrix meets the training requirements according to the probability of occurrence of each audio category label data in the second dialect category label data matrix, and obtaining an audio feature average data matrix;

[0011] S5. Perform a clustering operation on the audio feature average data matrix to obtain an audio feature average data classification matrix; and generate training reference standard data according to the audio feature average data classification matrix;

[0012] By collecting English spoken audio data of various dialect and non-dialect types, data support is provided for the subsequent construction of a model that can identify the dialect type of English spoken audio data; by preprocessing the collected English spoken audio data, the quality of the English spoken audio data is made higher, so that the subsequent construction of the final LEAN neural network model can recognize the dialect more accurately; by generating a dialect category label for each row of data in the preprocessed first English spoken audio data matrix, a classification standard is provided for the subsequent construction of the LEAN neural network model; by constructing the final LEAN neural network model, a recognition model is provided for the subsequent dialect recognition of the collected listening and speaking audio data, and then the listening and speaking audio data with dialect is analyzed in a targeted manner to help trainers make corrections; by extracting features from the listening and speaking audio data corresponding to the audio category label data with the highest probability of occurrence, and comparing the extracted feature data with the feature data of the source audio, and then clustering the comparison results, the comparison data of the category with the largest amount of data is used as training reference data, so that the reference data is more credible and accurate.

[0013] Preferably, the S1 comprises the following steps:

[0014] S11. Set a variety of different dialect sound types to obtain a dialect sound type set a i Indicates the set i-th dialect type, Indicates the total number of the set dialect sound types; then sets the non-dialect sound types;

[0015] For the dialect sound type set For each dialect sound type and non-dialect sound type in the, a plurality of English spoken audio data with corresponding dialect sounds and without dialect sounds are collected to obtain a first English spoken audio data matrix;

[0016] S12: performing a preprocessing operation on each English spoken audio data in the first English spoken audio data matrix to obtain a preprocessed first English spoken audio data matrix. as follows,

[0017]

[0018] in, represents the jth preprocessed English spoken audio data collected for the i-th dialect type, represents the total number of English spoken audio data collected for the type; a′ represents the total number of English spoken audio data types in the first English spoken audio data matrix after preprocessing,

[0019] By collecting English spoken audio data of various different dialect types and English spoken audio data without dialect sounds, the collected data is made more comprehensive, so that the final LEAN neural network model constructed using the data can simultaneously identify the dialect types of English spoken audio data with dialect sounds and English spoken audio data without dialect sounds, making the model more universal.

[0020] Preferably, the S12 comprises the following steps:

[0021] S121, obtaining the signal data of each English spoken audio data in the first English spoken audio data matrix to obtain a first signal data matrix; performing frame segmentation and windowing processing on each signal data in the first signal data matrix to obtain a processed first signal frame data set matrix b; as follows,

[0022]

[0023] Among them, b ij Express A signal frame data set obtained after frame segmentation and windowing processing;

[0024] b ijk Indicates b ij The kth signal frame data in b i ' j Indicates b ij The total number of signal frame data in the middle;

[0025] S122, perform fast Fourier transform on each signal frame data set in the processed first signal frame data set matrix b to obtain an amplitude spectrum data matrix and a phase spectrum data matrix b2′; then perform average processing on each amplitude spectrum data in the amplitude spectrum data matrix to obtain an average amplitude spectrum data matrix b3′; b2′ and b3′ are respectively as follows,

[0026]

[0027] Among them, b1′ ij , b2′ ij Respectively represent b ij Amplitude spectrum data and phase spectrum data obtained by fast Fourier transform; b3′ ij Indicates b1′ ij Average amplitude spectrum data obtained by averaging;

[0028] S123, performing multi-window spectrum estimation on each signal frame data set in the processed first signal frame data set matrix b to obtain a power spectrum density data matrix; then performing average processing on each power spectrum density data in the power spectrum density data matrix to obtain an average power spectrum density data matrix b5′;

[0029] Determine the frame length occupied by the noise segment in each English spoken audio data in the first English spoken audio data matrix to obtain a noise segment frame length data matrix; calculate the average power spectral density of the noise of each English spoken audio data in the first English spoken audio data matrix according to each noise segment frame length data in the noise segment frame length data matrix to obtain a noise average power spectral density data matrix b6′; b5′ and b6′ are respectively as follows,

[0030]

[0031] Among them, b4′ ij Indicates b ij The power spectral density data obtained by multi-window spectrum estimation; b5′ ij Indicates b4′ ij Average power spectrum density data obtained by averaging; b6′ ij represents the average power spectral density of the noise in the j-th English spoken audio data collected from the i-th dialect type;

[0032] S124, calculating the gain factor data according to the average power spectrum density data matrix b5′ and the noise average power spectrum density data matrix b6′, to obtain the gain factor data matrix b1′; as follows,

[0033]

[0034] Among them, b1′ij represents the calculated b ij The gain factor data is as follows:

[0035]

[0036] In the formula, α and β represent the over-reduction factor and gain compensation factor respectively;

[0037] S125, calculating the amplitude spectrum data after spectrum subtraction according to the gain factor data matrix b1′ and the average amplitude spectrum data matrix b3′, and obtaining the amplitude spectrum data matrix b4′ after spectrum subtraction; as follows,

[0038]

[0039] Among them, b4′ ij represents the calculated b ij The amplitude spectrum data after spectrum subtraction; the calculation formula is as follows,

[0040] b4′ ij =b1′ ij b3′ ij ;

[0041] S126, performing inverse fast Fourier transform on each post-spectral-subtraction amplitude spectrum data in the post-spectral-subtraction amplitude spectrum data matrix b4′ according to the phase spectrum data matrix b2′, to obtain a processed first signal frame data set matrix b;

[0042] By performing frame windowing, fast Fourier transform, averaging and inverse fast Fourier transform on the audio signal of each English spoken audio data in the first English spoken audio data matrix, the noise signal in the audio signal of the English spoken audio data is removed, thereby avoiding the influence of the noise data during the subsequent model construction.

[0043] Preferably, S2 comprises the following steps:

[0044] S21, setting the pre-processed first English spoken audio data matrix The dialect category label of each row of data in is obtained to obtain the first dialect category label dataset The matrix representing the first English spoken audio data after preprocessing is The dialect category label data of the i-th row of data;

[0045] Set a first training data ratio; and perform preprocessing on the first English spoken audio data matrix according to the first training data ratio. Performing data division to obtain a first audio training data matrix and a first audio test data matrix;

[0046] S22, constructing an initial LEAN neural network model and setting a first training error threshold; combining the first audio training data matrix and the first dialect category label data set Input into the initial LEAN neural network model for training; during the training process, when the training error is less than the first training error threshold, stop the training to obtain a trained LEAN neural network model;

[0047] S23, setting a first accuracy threshold; combining the first audio test data matrix and the first dialect category label data set Input into the trained LEAN neural network model for testing; after the test is completed, the first accuracy rate is obtained;

[0048] When the first accuracy rate is greater than or equal to the first accuracy rate threshold, the trained LEAN neural network model is used as the final LEAN neural network model; otherwise, returning to S22 to train the trained LEAN neural network model until the first accuracy rate is greater than or equal to the first accuracy rate threshold, to obtain the final LEAN neural network model;

[0049] By using the preprocessed first English spoken audio data matrix and the corresponding dialect category label data to iteratively train and test the initial LEAN neural network model, the LEAN neural network model's ability to recognize the dialect sounds contained in the preprocessed first English spoken audio data matrix is ​​greatly enhanced, thereby providing a recognition model for the subsequent dialect sound recognition of the collected listening and speaking audio data.

[0050] Preferably, S3 comprises the following steps:

[0051] S31. Collect multiple English listening and speaking source audio data and corresponding text data to obtain source audio data sets and source audio text data sets c′ represents the total number of English listening and speaking source audio data collected; Indicates the text data corresponding to the i-th English listening and speaking source audio data collected;

[0052] An English listening and speaking trainer is set; the English listening and speaking trainer performs several listening and speaking trainings on each English listening and speaking source audio data in the source audio data set; the audio of the listening and speaking training is recorded to obtain a listening and speaking audio data matrix;

[0053] S32, using S12 method to preprocess the source audio data set and each audio data in the audio and video data matrix, to obtain a preprocessed source audio data set c = {c1, c2, ..., c i ,...,c c′} and the preprocessed listening and speaking audio data matrix c i represents the English listening and speaking source audio data obtained after preprocessing the i-th English listening and speaking source audio data collected; as follows,

[0054]

[0055] in, represents the preprocessed listening and speaking audio data recorded from the jth listening and speaking training of the i-th English listening and speaking source audio data, d i represents the total number of times the listening and speaking training is performed on the i-th English listening and speaking source audio data;

[0056] S33, using the final LEAN neural network model to pre-process the listening and speaking audio data matrix Each audio data in the audio is identified by audio type, and the second audio category label data matrix is ​​obtained. as follows,

[0057]

[0058] in, represents the dialect category label data obtained by performing dialect sound type recognition on the preprocessed listening and speaking audio data recorded during the jth listening and speaking training of the i-th English listening and speaking source audio data;

[0059] By collecting multiple English listening and speaking source audio data, the corresponding text data, and the data of several listening and speaking trainings conducted by English listening and speaking trainees on each English listening and speaking source audio data set, a data basis is provided for analyzing the English listening and speaking pronunciation problems of English listening and speaking trainees.

[0060] Preferably, S4 comprises the following steps:

[0061] S41, calculating the second dialect category label data matrix The probability of each audio category label data appearing in the second category probability dataset f = {f1, f2, ..., f i ,...,f a′}, f i represents the probability value of the i-th audio category appearing in the second dialect category label data matrix e~;

[0062] S42, when the second category probability data set f={f1,f2,...,f i ,...,f a′ The audio category label corresponding to the maximum probability value in} is the dialect type set When the second category probability data set f = {f1, f2, ..., f i ,...,f a′}, the audio category label corresponding to the maximum probability value is used as the final audio category label and enters S43;

[0063] When the second category probability data set f={f1,f2,...,f i ,...,f a′ The audio category label corresponding to the maximum probability value in} is not a dialect type set When the second category probability data set f = {f1, f2, ..., f i ,...,f a′}, the audio category label corresponding to the probability value second only to the maximum probability value is taken as the final audio category label and enters S43;

[0064] When the second category probability data set f={f1,f2,...,f i ,...,f a′} and the category corresponding to the probability data is a non-dialect type, a pre-trained recurrent neural network model is used to pre-process the listening and speaking audio data matrix Each audio data in the audio and video is converted into text to obtain the audio and video conversion text data matrix Enter S44; as follows,

[0065]

[0066] in, Indicates that a pre-trained recurrent neural network model is used to The audio-visual conversion text data obtained after the text conversion;

[0067] S43: Cooperate with the second audio category label data matrix From the preprocessed audio data matrix Select the listening and speaking audio data corresponding to the final audio category label, and obtain the listening and speaking audio data matrix f′ with dialect sound; as follows,

[0068]

[0069] Among them, f ij ′ represents the jth listening and speaking audio data with dialect sounds obtained after the listening and speaking training of the i-th English listening and speaking source audio data; represents the total number of listening and speaking audio data with dialect sounds obtained after the listening and speaking training of the i-th English listening and speaking source audio data;

[0070] Extract and process the features of each audio data in the audio data matrix f′ with dialect sounds to obtain an audio feature average data matrix

[0071] S44, converting the audio and video into a text data matrix Each audio-to-text data in the audio-to-text dataset is the same as the source audio-to-text dataset Compare the corresponding source audio text data in to obtain the training accuracy data matrix g; as follows,

[0072]

[0073] Among them, g ij Indicates that and Compare the training accuracy data obtained;

[0074] Calculate the average value of each row of the training accuracy data matrix g to obtain the training average accuracy data set g′={g1′,g′2,...,g i ′,...,g c ″},g i ' represents the average value of the i-th row data in the training accuracy data matrix g,

[0075]

[0076] S45, respectively set a correct rate threshold for each source audio data in the source audio data set to obtain a correct rate threshold set

[0077] When the training average accuracy data set g′={g1′,g′2,...,g i ′,...,g c The i-th training average accuracy data in ″} is greater than or equal to the accuracy threshold set When the i-th accuracy rate threshold in , the listening and speaking training of the i-th source audio data in the source audio data set meets the requirement; otherwise, the requirement is not met;

[0078] By calculating the probability of occurrence of each audio category label data in the second dialect category label data matrix, the audio category label data with the largest probability must be the main problem of English trainees' English listening and speaking pronunciation. Therefore, the audio category label corresponding to the maximum probability value in the second category probability data set is used as the final audio category label, so that the main pronunciation problems of English trainees can be solved in a targeted manner in the future, thereby improving the efficiency of pronunciation correction; at the same time, the pronunciation of English trainees without dialect accents is also converted into text, and the converted text is compared with the standard text, and the pronunciation problems of English trainees without dialect accents are analyzed according to the comparison results; in summary, this scheme comprehensively considers various pronunciation problems that may arise for English trainees, making the scheme more adaptable and universal.

[0079] Preferably, in S43, extracting and processing the features of each audio and video data in the audio and video data matrix f′ with dialect to obtain the audio feature average data matrix comprises the following steps:

[0080] S431. Setting English audio feature type set d i ′ represents the set i-th English audio feature type, Indicates the total number of set English audio feature types;

[0081] According to the English audio feature type set For the preprocessed source audio dataset c={c1,c2,...,c i ,...,c c′} and the preprocessed listening and speaking audio data matrix Extract the features of each audio data in to obtain the source audio feature data matrix And the listening and speaking audio feature dataset matrix e; respectively as follows,

[0082]

[0083] in, Indicates c i The jth feature data obtained by feature extraction; e ij Express The feature data set obtained by feature extraction, e ijk Indicates e ij The kth feature data in;

[0084] S432: the listening and speaking audio feature data set matrix e and the source audio feature data matrix Perform a difference operation on each corresponding audio feature data in to obtain the audio feature difference data set matrix e′; as follows,

[0085]

[0086] Among them, e i ' j Indicates e ij A matrix with the source audio feature data The audio feature difference data set is obtained by subtracting the i-th row data in; e i ' jk Indicates e i ' j The difference data of the kth audio feature in is calculated as follows:

[0087]

[0088] S433, calculating the average value of the same type of audio feature data in each row of the audio feature difference data set matrix e′, to obtain an audio feature average data matrix

[0089]

[0090] in, represents the average value of the k-th type of audio feature data in the audio feature difference data set in the i-th row in the audio feature difference data set matrix e′; the calculation formula is as follows:

[0091]

[0092] English audio feature types include pitch, speaking speed, intonation and other feature types. These feature types are highly correlated with the correct pronunciation of English. Therefore, by extracting these features from audio with dialect sounds, the pronunciation problems of English trainees can be analyzed from a more subtle perspective, providing more specific and accurate reference data for subsequent correction of English trainees' pronunciation.

[0093] Preferably, the S5 comprises the following steps:

[0094] S51, averaging the audio feature data matrix Perform clustering operations to obtain the audio feature average data classification matrix set Represents the average data matrix for the audio features The i-th audio feature average data classification matrix obtained by clustering, h represents the average data matrix of the audio feature The total number of audio feature average data classification matrices obtained by clustering;

[0095] S52: Counting the audio feature average data classification matrix set The total number of rows of the average data classification matrix for each audio feature in is obtained, and the audio feature classification number set h′={h1′,h2′,...,h i ′,...,h h ′},h i ′ represents the audio feature average data classification matrix set The total number of rows of the i-th audio feature average data classification matrix in the audio feature average data classification matrix set corresponding to the maximum value in the audio feature classification number set is selected as the training reference standard data;

[0096] S53, the English listening and speaking trainee adjusts his / her English listening and speaking training process according to the training reference standard data;

[0097] By performing clustering operations on the audio feature average data matrix, similar data in the audio feature average data matrix are clustered into one category; therefore, the audio feature average data in the category with the most data can reflect the main problems of English trainees, and therefore the training reference standard data generated based on the audio feature average data in the category with the most data is more accurate.

[0098] Preferably, the S51 comprises the following steps:

[0099] S511, average data matrix from the audio feature Randomly select h rows of data from the matrix to obtain an initial classification center data matrix; calculate the Euclidean distance from each row of data in the audio feature average data matrix except the data in the initial classification center data matrix to each row of data in the initial classification center data matrix to obtain an initial Euclidean distance matrix;

[0100] S512, constructing a squirrel population; setting the maximum number of iterations of the squirrel population to l1′ and the current number of iterations to l2′;

[0101] The initial position of each squirrel in the squirrel population is set according to the initial classification center data matrix to obtain an initial position matrix set;

[0102] S513, setting the fitness function of the squirrel population according to the initial Euclidean distance matrix;

[0103] S514, start the iteration operation, and set the current iteration number l2′ to 1 before the iteration; calculate the fitness of each squirrel according to the fitness function of the squirrel population in each round of iteration to obtain a fitness set; take the maximum fitness value in the fitness set as the global optimal fitness; and take the position of the squirrel corresponding to the global optimal fitness as the global optimal position;

[0104] The position of each squirrel in the squirrel population is updated according to the global optimal fitness and the global optimal position; after the position update is completed, the current number of iterations l2′ is increased by one; and the next round of iteration is entered;

[0105] S515, when l2′≥l1′, stop iteration; output the final global optimal position matrix; use the final global optimal position matrix as the final classification center data matrix;

[0106] S516, calculating the Euclidean distance from each row of data in the audio feature average data matrix to each row of data in the final classification center data matrix, to obtain a final Euclidean distance matrix; selecting the row data of the audio feature average data matrix corresponding to the smallest Euclidean distance data in each row of the final Euclidean distance matrix and classifying them as the row data of the final classification center data matrix corresponding to the smallest Euclidean distance data; obtaining an audio feature average data classification matrix set;

[0107] By using the Squirrel algorithm to iteratively optimize multiple classification center data of the audio feature average data matrix for multiple times, after classifying the data in the audio feature average data matrix according to the iteratively optimized classification center data, the overall discreteness is minimized, thereby achieving the optimal classification effect, making the subsequently generated training correction reference data more accurate.

[0108] An English listening and speaking training data processing system based on artificial intelligence comprises a first audio data acquisition module, an audio preprocessing module, a dialect category label generation module, a LEAN neural network model construction module, a second audio data acquisition module, an audio type recognition module, a determination module, a clustering module and a dialect sound adjustment reference data generation module.

[0109] The present invention has the following beneficial effects:

[0110] 1. In the present invention, by generating a dialect category label for each row of data in the first English spoken audio data matrix after preprocessing, a classification standard is provided for the subsequent construction of a LEAN neural network model; by constructing the final LEAN neural network model, a recognition model is provided for the subsequent dialect recognition of the collected listening and speaking audio data, and then a targeted analysis is performed on the listening and speaking audio data with dialect sounds to help trainers make corrections; by extracting features from the listening and speaking audio data corresponding to the audio category label data with the highest probability of occurrence, and comparing the extracted feature data with the feature data of the source audio, and then clustering the comparison results, the comparison data of the category with the largest amount of data is used as training reference data, so that the reference data is more credible and accurate.

[0111] 2. The present invention extracts these features from the audio with dialect sounds, thereby analyzing the pronunciation problems of English trainees from a more subtle perspective, providing more specific and accurate reference data for the subsequent correction of the pronunciation of English trainees.

[0112] 3. In the present invention, the squirrel algorithm is used to perform multiple iterative optimizations on multiple classification center data of the audio feature average data matrix, so that after the data in the audio feature average data matrix is ​​classified according to the iteratively optimized classification center data, the overall discreteness is minimized, thereby achieving the optimal classification effect, making the subsequently generated training correction reference data more accurate.

[0113] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0114] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying creative work.

[0115] Figure 1 The present invention is a flowchart of an English listening and speaking training data processing system based on artificial intelligence for processing English listening and speaking training data. DETAILED DESCRIPTION

[0116] The following will be combined with the drawings in the embodiments of the invention to clearly and completely describe the technical solutions in the embodiments of the invention. Obviously, the described embodiments are only part of the embodiments of the invention, not all of the embodiments. Based on the embodiments in the invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the invention.

[0117] In the description of the present invention, it is necessary to understand that the terms "opening", "upper", "lower", "top", "middle", "inside" and the like indicating orientation or positional relationship are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the invention.

[0118] Embodiment 1

[0119] This embodiment is an English listening and speaking training data processing method based on artificial intelligence, comprising the following steps:

[0120] S1, collecting English spoken audio data of multiple different dialect sound types and non-dialect sound types to obtain a first English spoken audio data matrix; preprocessing the first English spoken audio data matrix;

[0121] The S1 comprises the following steps:

[0122] S11. Set a variety of different dialect sound types to obtain a dialect sound type set a i express

[0123] The i-th dialect type is set, Indicates the total number of the set dialect sound types; then sets the non-dialect sound types;

[0124] For the dialect sound type set For each dialect sound type and non-dialect sound type in the, a plurality of English spoken audio data with corresponding dialect sounds and without dialect sounds are collected to obtain a first English spoken audio data matrix;

[0125] S12: performing a preprocessing operation on each English spoken audio data in the first English spoken audio data matrix to obtain a preprocessed first English spoken audio data matrix. as follows,

[0126]

[0127] in, represents the jth preprocessed English spoken audio data collected for the i-th dialect type, represents the total number of English spoken audio data collected for the type; a′ represents the total number of English spoken audio data types in the first English spoken audio data matrix after preprocessing,

[0128] The S12 comprises the following steps:

[0129] S121, obtaining the signal data of each English spoken audio data in the first English spoken audio data matrix to obtain a first signal data matrix; performing frame segmentation and windowing processing on each signal data in the first signal data matrix to obtain a processed first signal frame data set matrix b; as follows,

[0130]

[0131] Among them, b ij Express A signal frame data set obtained after frame segmentation and windowing processing;

[0132] bijk Indicates b ij The kth signal frame data in b i ' j Indicates b ij The total number of signal frame data in the middle;

[0133] S122, perform fast Fourier transform on each signal frame data set in the processed first signal frame data set matrix b to obtain an amplitude spectrum data matrix and a phase spectrum data matrix b2′; then perform average processing on each amplitude spectrum data in the amplitude spectrum data matrix to obtain an average amplitude spectrum data matrix b3′; b2′ and b3′ are respectively as follows,

[0134]

[0135] Among them, b1′ ij , b2′ ij Respectively represent b ij Amplitude spectrum data and phase spectrum data obtained by fast Fourier transform; b3′ ij Indicates b1′ ij Average amplitude spectrum data obtained by averaging;

[0136] S123, performing multi-window spectrum estimation on each signal frame data set in the processed first signal frame data set matrix b to obtain a power spectrum density data matrix; then performing average processing on each power spectrum density data in the power spectrum density data matrix to obtain an average power spectrum density data matrix b5′;

[0137] Determine the frame length occupied by the noise segment in each English spoken audio data in the first English spoken audio data matrix to obtain a noise segment frame length data matrix; calculate the average power spectral density of the noise of each English spoken audio data in the first English spoken audio data matrix according to each noise segment frame length data in the noise segment frame length data matrix to obtain a noise average power spectral density data matrix b6′; b5′ and b6′ are respectively as follows,

[0138]

[0139] Among them, b4′ ij Indicates b ij The power spectral density data obtained by multi-window spectrum estimation; b5′ ij Indicates b4′ ij Average power spectrum density data obtained by averaging; b6′ ij represents the average power spectral density of the noise in the j-th English spoken audio data collected from the i-th dialect type;

[0140] S124, calculating the gain factor data according to the average power spectrum density data matrix b5′ and the noise average power spectrum density data matrix b6′, to obtain the gain factor data matrix b1′; as follows,

[0141]

[0142] Among them, b1′ ij represents the calculated b ij The gain factor data is as follows:

[0143]

[0144] In the formula, α and β represent the over-reduction factor and gain compensation factor respectively;

[0145] S125, calculating the amplitude spectrum data after spectrum subtraction according to the gain factor data matrix b1′ and the average amplitude spectrum data matrix b3′, and obtaining the amplitude spectrum data matrix b4′ after spectrum subtraction; as follows,

[0146]

[0147] Among them, b4′ ij represents the calculated b ij The amplitude spectrum data after spectrum subtraction; the calculation formula is as follows,

[0148] b4′ ij =b1′ ij b3′ ij ;

[0149] S126, performing inverse fast Fourier transform on each post-spectral-subtraction amplitude spectrum data in the post-spectral-subtraction amplitude spectrum data matrix b4′ according to the phase spectrum data matrix b2′, to obtain a processed first signal frame data set matrix b;

[0150] S2, generating a dialect category label for each row of data in the preprocessed first spoken English audio data matrix to obtain a first dialect category label data set; using the preprocessed first spoken English audio data matrix and the corresponding dialect category label data to construct a final LEAN neural network model;

[0151] The S2 comprises the following steps:

[0152] S21, setting the pre-processed first English spoken audio data matrix The dialect category label of each row of data in is obtained to obtain the first dialect category label dataset The matrix representing the first English spoken audio data after preprocessing is The dialect category label data of the i-th row of data;

[0153] Set a first training data ratio; and perform preprocessing on the first English spoken audio data matrix according to the first training data ratio. Performing data division to obtain a first audio training data matrix and a first audio test data matrix;

[0154] S22, constructing an initial LEAN neural network model and setting a first training error threshold; combining the first audio training data matrix and the first dialect category label data set Input into the initial LEAN neural network model for training; during the training process, when the training error is less than the first training error threshold, stop the training to obtain a trained LEAN neural network model;

[0155] S23, setting a first accuracy threshold; combining the first audio test data matrix and the first dialect category label data set Input into the trained LEAN neural network model for testing; after the test is completed, the first accuracy rate is obtained;

[0156] When the first accuracy rate is greater than or equal to the first accuracy rate threshold, the trained LEAN neural network model is used as the final LEAN neural network model; otherwise, returning to S22 to train the trained LEAN neural network model until the first accuracy rate is greater than or equal to the first accuracy rate threshold, to obtain the final LEAN neural network model;

[0157] S3, collect multiple copies of English listening and speaking source audio data, listening and speaking audio data corresponding to the source audio data, and text data corresponding to the source audio data, to obtain a source audio data set, a listening and speaking audio data matrix, and a source audio text data set; use the final LEAN neural network model to perform audio type recognition on the preprocessed listening and speaking audio data matrix to obtain a second audio category label data matrix;

[0158] The S3 comprises the following steps:

[0159] S31. Collect multiple English listening and speaking source audio data and corresponding text data to obtain source audio data sets and source audio text data sets c′ represents the total number of English listening and speaking source audio data collected; Indicates the text data corresponding to the i-th English listening and speaking source audio data collected;

[0160] An English listening and speaking trainer is set; the English listening and speaking trainer performs several listening and speaking trainings on each English listening and speaking source audio data in the source audio data set; the audio of the listening and speaking training is recorded to obtain a listening and speaking audio data matrix;

[0161] S32, using S12 method to preprocess the source audio data set and each audio data in the audio and video data matrix, to obtain a preprocessed source audio data set c = {c1, c2, ..., c i ,...,c c′} and the preprocessed listening and speaking audio data matrix c i represents the English listening and speaking source audio data obtained after preprocessing the i-th English listening and speaking source audio data collected; as follows,

[0162]

[0163] in, represents the preprocessed listening and speaking audio data recorded from the jth listening and speaking training of the i-th English listening and speaking source audio data, d i represents the total number of times the listening and speaking training is performed on the i-th English listening and speaking source audio data;

[0164] S33, using the final LEAN neural network model to pre-process the listening and speaking audio data matrix Each audio data in the audio is identified by audio type, and the second audio category label data matrix is ​​obtained. as follows,

[0165]

[0166] in, represents the dialect category label data obtained by performing dialect sound type recognition on the preprocessed listening and speaking audio data recorded during the jth listening and speaking training of the i-th English listening and speaking source audio data;

[0167] S4, judging whether the preprocessed listening and speaking audio data matrix meets the training requirements according to the probability of occurrence of each audio category label data in the second dialect category label data matrix, and obtaining an audio feature average data matrix;

[0168] The S4 comprises the following steps:

[0169] S41, calculating the second dialect category label data matrix The probability of each audio category label data appearing in the second category probability dataset f = {f1, f2, ..., f i ,...,f a′}, f i Represents the second dialect category label data matrix The probability value of the i-th audio category appearing in ;

[0170] S42, when the second category probability data set f={f1,f2,...,f i ,...,f a′ The audio category label corresponding to the maximum probability value in} is the dialect type set When the second category probability data set f = {f1, f2, ..., f i ,...,f a′}, the audio category label corresponding to the maximum probability value is used as the final audio category label and enters S43;

[0171] When the second category probability data set f={f1,f2,...,f i ,...,f a′ The audio category label corresponding to the maximum probability value in} is not a dialect type set When the second category probability data set f = {f1, f2, ..., f i ,...,f a′}, the audio category label corresponding to the probability value second only to the maximum probability value is taken as the final audio category label and enters S43;

[0172] When the second category probability data set f={f1,f2,...,f i ,...,f a′} and the category corresponding to the probability data is a non-dialect type, a pre-trained recurrent neural network model is used to pre-process the listening and speaking audio data matrix Each audio data in the audio and video is converted into text to obtain the audio and video conversion text data matrix

[0173] Enter S44; as follows,

[0174]

[0175] in, Indicates that a pre-trained recurrent neural network model is used to The audio-visual conversion text data obtained after the text conversion;

[0176] S43: Cooperate with the second audio category label data matrix From the preprocessed audio data matrix Select the listening and speaking audio data corresponding to the final audio category label, and obtain the listening and speaking audio data matrix f′ with dialect sound; as follows,

[0177]

[0178] Among them, f ij ′ represents the jth listening and speaking audio data with dialect sounds obtained after the listening and speaking training of the i-th English listening and speaking source audio data; represents the total number of listening and speaking audio data with dialect sounds obtained after the listening and speaking training of the i-th English listening and speaking source audio data;

[0179] Extract and process the features of each audio data in the audio data matrix f′ with dialect sounds to obtain an audio feature average data matrix

[0180] In S43, feature extraction and processing are performed on each audio and video data in the audio and video data matrix f′ with dialect to obtain an audio feature average data matrix, which includes the following steps:

[0181] S431. Setting English audio feature type set d i ′ represents the set i-th English audio feature type, Indicates the total number of set English audio feature types;

[0182] According to the English audio feature type set For the preprocessed source audio dataset c={c1,c2,...,c i ,...,c c′} and the preprocessed listening and speaking audio data matrix Extract the features of each audio data in to obtain the source audio feature data matrix And the listening and speaking audio feature dataset matrix e; respectively as follows,

[0183]

[0184] in, Indicates c i The jth feature data obtained by feature extraction; e ij Express The feature data set obtained by feature extraction, e ijk Indicates e ij The kth feature data in;

[0185] S432: the listening and speaking audio feature data set matrix e and the source audio feature data matrix Perform a difference operation on each corresponding audio feature data in to obtain the audio feature difference data set matrix e′; as follows,

[0186]

[0187] Among them, ei ' j Indicates e ij A matrix with the source audio feature data The audio feature difference data set is obtained by subtracting the i-th row data in; e i ' jk Indicates e i ' j The difference data of the kth audio feature in is calculated as follows:

[0188]

[0189] S433, calculating the average value of the same type of audio feature data in each row of the audio feature difference data set matrix e′, to obtain an audio feature average data matrix

[0190]

[0191] in, represents the average value of the k-th type of audio feature data in the audio feature difference data set in the i-th row in the audio feature difference data set matrix e′; the calculation formula is as follows:

[0192]

[0193] S44, converting the audio and video into a text data matrix Each audio-to-text data in the audio-to-text dataset is the same as the source audio-to-text dataset Compare the corresponding source audio text data in to obtain the training accuracy data matrix g; as follows,

[0194]

[0195] Among them, g ij Indicates that and Compare the training accuracy data obtained;

[0196] Calculate the average value of each row of the training accuracy data matrix g to obtain the training average accuracy data set g′={g1′,g′2,...,g i ′,...,g c ″},g i ' represents the average value of the i-th row data in the training accuracy data matrix g,

[0197]

[0198] S45, respectively set a correct rate threshold for each source audio data in the source audio data set to obtain a correct rate threshold set

[0199] When the training average accuracy data set g′={g1′,g′2,...,g i ′,...,g c The i-th training average accuracy data in ″} is greater than or equal to the accuracy threshold set When the i-th accuracy rate threshold in , the listening and speaking training of the i-th source audio data in the source audio data set meets the requirement; otherwise, the requirement is not met;

[0200] S5. Perform a clustering operation on the audio feature average data matrix to obtain an audio feature average data classification matrix; and generate training reference standard data according to the audio feature average data classification matrix;

[0201] The S5 comprises the following steps:

[0202] S51, averaging the audio feature data matrix Perform clustering operations to obtain the audio feature average data classification matrix set Represents the average data matrix for the audio features The i-th audio feature average data classification matrix obtained by clustering, h represents the average data matrix of the audio feature The total number of audio feature average data classification matrices obtained by clustering;

[0203] The S51 comprises the following steps:

[0204] S511, average data matrix from the audio feature Randomly select h rows of data to obtain the initial classification center data matrix as follows,

[0205]

[0206] in, represents the average data matrix from the audio features The average value of the k-th type of audio feature data in the i-th row of data randomly selected from;

[0207] Calculate the audio feature average data matrix In addition to the initial classification center data matrix Each row of data outside the data in is separately added to the initial classification center data matrix The Euclidean distance of each row of data in the initial Euclidean distance matrix is ​​obtained as follows,

[0208]

[0209] in, Represents the average data matrix of the audio features In addition to the initial classification center data matrix The j-th row of data outside the data in and the initial classification center data matrix The Euclidean distance between the i-th row of data in ;

[0210] S512. Constructing squirrel populations represents the jth squirrel in the squirrel population, l represents the total size of the squirrel population; the maximum number of iterations of the squirrel population is set to l1′, the current number of iterations is set to l2′, and the search space dimension is set to

[0211] According to the initial classification center data matrix Setting the squirrel population The initial position of each squirrel in the matrix is ​​obtained. represents the initial position matrix of the j-th squirrel in the squirrel population; as follows,

[0212]

[0213] in, for represents the dimension component of the average value of the k-th type of audio feature data in the i-th classification center data; the calculation formula is as follows,

[0214]

[0215] In the formula, rand jik Indicates for Generated random numbers; and They respectively represent the lower limit and upper limit of the value of the k-th audio feature data;

[0216] S513, according to the initial Euclidean distance matrix Setting the squirrel population The fitness function m′ is as follows:

[0217]

[0218] S514, start the iteration operation, before the iteration, set the current iteration number l2′ to 1; in each iteration process, according to the squirrel population The fitness function m′ is used to calculate the fitness of each squirrel to obtain a fitness set; the maximum fitness value in the fitness set is used as the global optimal fitness; the position of the squirrel corresponding to the global optimal fitness is used as the global optimal position

[0219] The squirrel population is classified according to the global optimal fitness and the global optimal position. The position of each squirrel is updated; the position update formula m~ is as follows,

[0220]

[0221] in, They represent the position of the i-th squirrel in the squirrel population in the next iteration and the current iteration respectively; α, β, ε represent the random gliding distance, the sliding constant, and the probability of encountering a predator respectively; δ is a random number in [0,1]; Represents the random number generated for the i-th squirrel in the current iteration; Respectively represent the maximum and minimum values ​​of the positions of squirrels in the squirrel population;

[0222] After the position update is completed, the current iteration number l2′ is increased by one; and the next round of iteration is entered;

[0223] S515, when l2′≥l1′, stop iteration; output the final global optimal position matrix; use the final global optimal position matrix as the final classification center data matrix n; as follows,

[0224]

[0225] Among them, n ik Represents the average value of the k-th type of audio feature data in the i-th final classification center data;

[0226] S516: Calculate the audio feature average data matrix The Euclidean distance from each row of data in the final classification center data matrix n is obtained as follows:

[0227]

[0228] Among them, n i ' j Represents the average data matrix of the audio features The Euclidean distance between the j-th row of data in and the i-th row of data in the final classification center data matrix n;

[0229] Select the audio feature average data matrix corresponding to the minimum Euclidean distance data in each row of the final Euclidean distance matrix n′ The row data of is classified as the row data of the final classification center data matrix n corresponding to the minimum Euclidean distance data; the audio feature average data classification matrix set is obtained

[0230] S52: Counting the audio feature average data classification matrix set The total number of rows of the average data classification matrix for each audio feature in is obtained, and the audio feature classification number set h′={h1′,h2′,...,h i ′,...,h h ′},h i ′ represents the audio feature average data classification matrix set The total number of rows of the average data classification matrix of the i-th audio feature in;

[0231] Select the audio feature classification set h′={h1′,h2′,...,h i ′,...,h h The audio feature average data classification matrix set corresponding to the maximum value in ′} The classification center data of the audio feature average data classification matrix in is used as the training reference standard data;

[0232] S53: The English listening and speaking trainee adjusts his / her English listening and speaking training process according to the training reference standard data.

[0233] Embodiment 2

[0234] This embodiment discloses an English listening and speaking training data processing system based on artificial intelligence, which can implement the method of the above embodiment, including a first audio data acquisition module, an audio preprocessing module, a dialect category label generation module, a LEAN neural network model construction module, a second audio data acquisition module, an audio type recognition module, a determination module, a clustering module and a dialect sound adjustment reference data generation module;

[0235] The first audio data acquisition module is used to collect English spoken audio data of multiple different dialect sound types and non-dialect sound types to obtain a first English spoken audio data matrix;

[0236] The audio preprocessing module is used to preprocess the first English spoken audio data matrix;

[0237] The dialect category label generation module is used to generate a dialect category label for each row of data in the preprocessed first English spoken audio data matrix to obtain a first dialect category label data set;

[0238] The LEAN neural network model construction module is used to construct a final LEAN neural network model using the preprocessed first English spoken audio data matrix and the corresponding dialect category label data;

[0239] The second audio data collection module is used to collect multiple copies of English listening and speaking source audio data, listening and speaking audio data corresponding to the source audio data, and text data corresponding to the source audio data to obtain a source audio data set, a listening and speaking audio data matrix, and a source audio text data set;

[0240] The audio type recognition module is used to use the final LEAN neural network model to perform audio type recognition on the preprocessed listening and speaking audio data matrix to obtain a second audio category label data matrix;

[0241] The determination module is used to determine whether the preprocessed listening and speaking audio data matrix meets the training requirements according to the probability of occurrence of each audio category label data in the second dialect category label data matrix, and obtain an audio feature average data matrix;

[0242] The clustering module is used to perform a clustering operation on the audio feature average data matrix to obtain an audio feature average data classification matrix;

[0243] The dialect adjustment reference data generation module generates training reference standard data according to the audio feature average data classification matrix.

[0244] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0245] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the invention, so that those skilled in the art can understand and use the invention well.

Claims

1. A method for processing English listening and speaking training data based on artificial intelligence, characterized in that: The following steps are involved: S1, collecting English spoken audio data of multiple different dialect sound types and non-dialect sound types to obtain a first English spoken audio data matrix; preprocessing the first English spoken audio data matrix; S2, generating a dialect category label for each row of data in the preprocessed first spoken English audio data matrix to obtain a first dialect category label data set; using the preprocessed first spoken English audio data matrix and the corresponding dialect category label data to construct a final LEAN neural network model; S3, collect multiple copies of English listening and speaking source audio data, listening and speaking audio data corresponding to the source audio data, and text data corresponding to the source audio data to obtain a source audio data set, a listening and speaking audio data matrix, and a source audio text data set; The final LEAN neural network model is used to identify the audio type of the preprocessed listening and speaking audio data matrix to obtain the second audio category label data matrix; S4, judging whether the preprocessed listening and speaking audio data matrix meets the training requirements according to the probability of occurrence of each audio category label data in the second dialect category label data matrix, and obtaining an audio feature average data matrix; S5. Perform a clustering operation on the audio feature average data matrix to obtain an audio feature average data classification matrix; Training reference standard data is generated according to the audio feature average data classification matrix.

2. The method for processing English listening and speaking training data based on artificial intelligence according to claim 1, characterized in that: The S1 comprises the following steps: S11, setting a plurality of different dialect sound types to obtain a dialect sound type set; collecting a plurality of English spoken audio data with and without corresponding dialect sound for each dialect sound type and non-dialect sound type in the dialect sound type set to obtain a first English spoken audio data matrix; S12. Preprocessing the first spoken English audio data matrix to obtain a preprocessed first spoken English audio data matrix.

3. The method for processing English listening and speaking training data based on artificial intelligence according to claim 2, characterized in that: The S2 comprises the following steps: S21, setting the dialect category label of the preprocessed first spoken English audio data matrix to obtain a first dialect category label data set; performing data partitioning on the preprocessed first spoken English audio data matrix to obtain a first audio training data matrix and a first audio test data matrix; S22, constructing an initial LEAN neural network model and setting a first training error threshold; inputting the first audio training data matrix and the first dialect category label data set into the initial LEAN neural network model for training; during the training process, when the training error is less than the first training error threshold, stopping the training to obtain a trained LEAN neural network model; S23, setting a first accuracy threshold; inputting the first audio test data matrix and the first dialect category label data set into the trained LEAN neural network model for testing; after the test is completed, obtaining the final LEAN neural network model.

4. The method for processing English listening and speaking training data based on artificial intelligence according to claim 3, characterized in that: The S3 comprises the following steps: S31, collecting multiple copies of English listening and speaking source audio data and corresponding text data and listening and speaking training data to obtain a source audio data set, a source audio text data set and a listening and speaking audio data matrix; S32. Use the final LEAN neural network model to perform audio type recognition on the preprocessed listening and speaking audio data matrix to obtain a second audio category label data matrix.

5. The method for processing English listening and speaking training data based on artificial intelligence according to claim 4, characterized in that: The S4 comprises the following steps: S41, calculating the probability of occurrence of each audio category label data in the second dialect category label data matrix to obtain a second category probability data set; S42. When the audio category label corresponding to the maximum probability value in the second category probability data set is one of the dialect sound types, the audio category label corresponding to the maximum probability value in the second category probability data set is used as the final audio category label and enters S43; when the audio category label corresponding to the maximum probability value in the second category probability data set is not one of the dialect sound types, the audio category label corresponding to the probability value second only to the maximum probability value in the second category probability data set is used as the final audio category label and enters S43; when there is only one category probability data in the second category probability data set and the category corresponding to the probability data is a non-dialect sound type, a pre-trained recurrent neural network model is used to perform text conversion on the preprocessed listening and speaking audio data matrix to obtain a listening and speaking audio conversion text data matrix, and enter S44; S43, selecting the listening and speaking audio data corresponding to the final audio category label from the preprocessed listening and speaking audio data matrix in conjunction with the second audio category label data matrix to obtain a listening and speaking audio data matrix with dialect sounds; extracting and processing features of each listening and speaking audio data in the listening and speaking audio data matrix with dialect sounds to obtain an audio feature average data matrix; S44, comparing the audio-visual audio conversion text data matrix with the source audio text data set to obtain a training accuracy data matrix; calculating the average value of each row of data in the training accuracy data matrix to obtain a training average accuracy data set; S45. Set an accuracy threshold for each source audio data in the source audio data set to obtain an accuracy threshold set; when the i-th training average accuracy data in the training average accuracy data set is greater than or equal to the i-th accuracy threshold in the accuracy threshold set, the listening and speaking training of the i-th source audio data in the source audio data set meets the requirements; otherwise, the requirements are not met.

6. The method for processing English listening and speaking training data based on artificial intelligence according to claim 5, characterized in that: In S43, feature extraction and processing are performed on each audio and video data in the audio and video data matrix with dialect sounds to obtain an audio feature average data matrix, which includes the following steps: S431, setting an English audio feature type set; extracting features of each audio data according to the English audio feature type set to obtain a source audio feature data matrix and a listening and speaking audio feature data set matrix; S432, performing a difference operation on each corresponding audio feature data in the listening and speaking audio feature data set matrix and the source audio feature data matrix to obtain an audio feature difference data set matrix; S433. Calculate the average value of the audio feature data of the same type in the audio feature difference data set matrix to obtain an audio feature average data matrix.

7. The method for processing English listening and speaking training data based on artificial intelligence according to claim 6, characterized in that: The S5 comprises the following steps: S51, performing a clustering operation on the audio feature average data matrix to obtain an audio feature average data classification matrix set; S52, counting the total number of rows of each audio feature average data classification matrix to obtain an audio feature classification number set; selecting the classification center data of the audio feature average data classification matrix in the audio feature average data classification matrix set corresponding to the maximum value in the audio feature classification number set as the training reference standard data; S53. Adjust your own English listening and speaking training process based on the training reference standard data.

8. The method for processing English listening and speaking training data based on artificial intelligence according to claim 7, characterized in that: The S51 comprises the following steps: S511, randomly selecting a number of rows of data from the audio feature average data matrix to obtain an initial classification center data matrix; calculating the Euclidean distance from each row of data in the audio feature average data matrix except the data in the initial classification center data matrix to each row of data in the initial classification center data matrix to obtain an initial Euclidean distance matrix; S512, constructing a squirrel population; setting the maximum number of iterations of the squirrel population to l1′ and the current number of iterations to l2′; setting the initial position of each squirrel according to the initial classification center data matrix to obtain an initial position matrix array; S513, setting the fitness function of the squirrel population according to the initial Euclidean distance matrix; S514, start iterative operation, in each round of iteration, calculate the fitness of each squirrel according to the fitness function to obtain a fitness set; take the maximum fitness value in the fitness set as the global optimal fitness; take the position of the squirrel corresponding to the global optimal fitness as the global optimal position; update the position of each squirrel according to the global optimal fitness and the global optimal position; S515, when l2′≥l1′, stop iteration; output the final global optimal position matrix; use the final global optimal position matrix as the final classification center data matrix; S516. Calculate the Euclidean distance from each row of data in the audio feature average data matrix to each row of data in the final classification center data matrix to obtain a final Euclidean distance matrix; select the row data of the audio feature average data matrix corresponding to the smallest Euclidean distance data in each row of the final Euclidean distance matrix and classify them as the row data of the final classification center data matrix corresponding to the smallest Euclidean distance data; obtain an audio feature average data classification matrix set.

9. A system for implementing the artificial intelligence-based English listening and speaking training data processing method as described in any one of claims 1 to 8.

Images