Artificial intelligence-based foreign language teaching effect evaluation method and system

Abstract

The application discloses a foreign language teaching effect evaluation method and system based on artificial intelligence, and the content comprises data collection, speech feature extraction, multi-modal fusion evaluation modeling and teaching effect evaluation. The application relates to the technical field of intelligent evaluation of teaching effect, and particularly discloses a foreign language teaching effect evaluation method and system based on artificial intelligence. The scheme constructs a pronunciation articulation composite index through formant difference analysis, calculates acoustic semantic coupling degree, and evaluates phoneme transfer stability, so that the articulation, accuracy and coherence of speech can be accurately quantified in a multi-noise environment. The scheme introduces modal attention, realizes bidirectional interaction of multi-modal information by using full-interconnection cross-modal cross-attention, and innovatively introduces a complementary gating mechanism to adaptively adjust the modal fusion ratio. Local scores are generated by combining a lightweight network, meta-decision network dynamically allocates weights, and finally, end-to-end optimization is realized by using a differentiable weighting strategy, so that the teaching effect evaluation of fineness and accuracy is realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent evaluation technology for teaching effectiveness, specifically to a method and system for evaluating the effectiveness of foreign language teaching based on artificial intelligence. Background Technology

[0002] With the development of artificial intelligence technology, especially breakthroughs in natural language processing, speech recognition, and machine learning, it has become possible to build more intelligent, automated, and accurate teaching effectiveness evaluation systems. Currently, in foreign language teaching effectiveness evaluation, existing methods are insufficient in assessing pronunciation clarity, accuracy, and fluency, and are easily affected by noise in noisy environments. Furthermore, traditional evaluation methods suffer from limited processing and fixed fusion mechanisms for the multimodal features of learning behavior, grades, and speech, making it difficult to meet the needs of personalized and high-precision evaluation. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an artificial intelligence-based method and system for evaluating the effectiveness of foreign language teaching. Addressing the problem that traditional speech feature extraction methods cannot accurately quantify pronunciation clarity, accuracy, and fluency, this solution constructs a composite index of pronunciation clarity through spectral subtraction noise reduction and short-time energy and formant difference analysis; calculates acoustic-semantic coupling degree based on the probabilistic coupling of acoustic and language models; and evaluates phoneme transfer stability by constructing phoneme feature vectors using frame-level Mel-frequency cepstral coefficients and their first and second-order differences. This enables precise quantification of speech clarity, accuracy, and fluency in noisy environments. Furthermore, it addresses the limitations of traditional foreign language teaching evaluation methods. Traditional methods for evaluating learning effectiveness suffer from several drawbacks. They treat features of different modalities in isolation, employ a single fusion method with fixed weights, making it difficult to uncover the deep dependencies between learning behavior, performance, and speech. They are also susceptible to noise interference and cannot meet personalized needs. This approach addresses these issues by uniformly mapping heterogeneous modal features to a latent space to eliminate scale differences, introducing intramodal self-attention to deeply capture internal dependencies, and utilizing fully interconnected cross-modal cross-attention to achieve bidirectional interaction of multimodal information. Furthermore, it adaptively adjusts the intramodal and cross-modal fusion ratio through a complementary gating mechanism, combines a lightweight quantum network to quickly generate local scores, and a meta-decision network to dynamically allocate personalized weights. Finally, it optimizes end-to-end using a differentiable weighted strategy to achieve a refined and accurate evaluation of teaching effectiveness.

[0004] The technical solution adopted in this invention is as follows: a foreign language teaching effectiveness evaluation method based on artificial intelligence, which includes the following steps:

[0005] Step S1: Data Acquisition;

[0006] Step S2: Speech feature extraction;

[0007] Step S3: Multimodal fusion evaluation modeling;

[0008] Step S4: Evaluation of teaching effectiveness.

[0009] Furthermore, in step S1, the data collection specifically involves collecting foreign language learners' foreign language reading audio data, foreign language learning behavior data, and foreign language test score data; the foreign language learning behavior data includes learning duration, learning frequency, and learning task completion rate; the foreign language test score data includes foreign language vocabulary test scores, foreign language reading comprehension test scores, and foreign language listening test scores.

[0010] Furthermore, in step S2, the speech feature extraction specifically includes the following:

[0011] Step S21: Standardization processing. The original speech signal is standardized, background noise is removed by spectral subtraction, effective speech segments are extracted based on speech activity detection based on short-time energy and zero-crossing rate, high-frequency attenuation is compensated, and then the preprocessed signal is divided into short-time frames and a Hamming window is applied to each frame.

[0012] Step S22: Pronunciation clarity assessment, specifically including the following steps:

[0013] Step S221: Normalize the short-time energy of the speech to obtain the normalized short-time energy;

[0014] Step S222: Extract the frequencies of the first three formants using the linear predictive coding method, and then normalize the frequency difference between the second and third formants;

[0015] Step S223: Combine the energy spectrum with the resonance peak difference to assess the articulation;

[0016] Step S23: Calculate the acoustic semantic coupling degree. First, transcribe the speech signal to obtain a word sequence, and use a pronunciation dictionary to map each word to the corresponding phoneme sequence. Then, align the phonemes and words. First, use a forced alignment tool to obtain the correspondence between phonemes and words. Then, use an acoustic model to output the probability of phonemes and a pre-trained language model to output the probability of words. Finally, align the acoustic model and language model probabilities to generate the acoustic semantic coupling degree.

[0017] Step S24: Calculate the phoneme transfer stability index. First, construct the acoustic feature vector of the phoneme. Combine the average Mel frequency cepstral coefficient, first-order difference coefficient and second-order difference coefficient of each phoneme with the average Mel frequency of all frames contained in the phoneme to form the acoustic feature vector. Then evaluate the smooth transition between adjacent phoneme features to generate the phoneme transfer stability index.

[0018] Step S25: Generate the final speech feature vector.

[0019] Furthermore, in step S3, the multimodal fusion evaluation modeling specifically includes the following steps:

[0020] Step S31: Data preparation. Foreign language learning behavior data (learning duration, learning frequency, and learning task completion rate) are combined into a learning behavior feature vector; foreign language exam score data (vocabulary exam scores, reading comprehension exam scores, and listening comprehension exam scores) are combined into a learning performance feature vector; finally, the final speech vector is obtained. Generate source data for teaching effectiveness evaluation;

[0021] Step S32: Modal feature unification mapping, synchronously mapping heterogeneous features to a unified latent space;

[0022] Step S33: Same-modal self-attention calculation, using self-attention within the same modality to uncover dependencies between internal features;

[0023] Step S34: Cross-modal attention fusion, using a fully interconnected cross-attention structure, enabling each modality to exchange information with all other modalities simultaneously;

[0024] Step S35: Generate dynamic gating vectors, introduce complementary gating mechanism, learn adaptive weights based on the original and cross-modal representations, and control the information fusion ratio;

[0025] Step S36: Feature fusion representation, designing differentiable weighted fusion, fusing same-modal and cross-modal features according to gating coefficients;

[0026] Step S37: Local scoring, design a light quantum network for each mode and introduce a nonlinear mapping to convert the fused features into evaluation scores;

[0027] Step S38: Adaptive calculation of decision weights, construction of meta-decision network, inputting local scores and learning performance feature vectors together to generate fusion weights;

[0028] Step S39: Generate the final model, use the final scoring decision weights to weight the local scores to obtain the overall score, and generate a multimodal fusion evaluation model.

[0029] Furthermore, in step S4, the teaching effectiveness evaluation specifically involves acquiring foreign language reading speech data, foreign language learning behavior data, and foreign language test score data of the users being evaluated for foreign language teaching effectiveness. First, speech features are extracted from the data, and then the data is input into a multimodal fusion evaluation model. The multimodal fusion evaluation model outputs a comprehensive evaluation score of the user's foreign language ability, with a score range of 0 to 1. A higher score indicates a better foreign language teaching effect.

[0030] The foreign language teaching effectiveness evaluation system based on artificial intelligence provided by this invention includes a data acquisition module, a speech feature extraction module, a multimodal fusion evaluation modeling module, and a teaching effectiveness evaluation module;

[0031] The data acquisition module collects foreign language learners' foreign language reading speech data, foreign language learning behavior data, and foreign language test score data, and sends the data to the speech feature extraction module and the multimodal fusion evaluation and modeling module.

[0032] The speech feature extraction module receives data sent by the data acquisition module, performs standardization processing on foreign language reading speech data, evaluates pronunciation clarity, calculates acoustic semantic coupling degree and phoneme transfer stability index, generates final speech feature vector, and sends the data to the multimodal fusion evaluation and modeling module.

[0033] The multimodal fusion evaluation modeling module receives data sent by the data acquisition module and the speech feature extraction module, and sequentially performs data preparation, unified mapping of modal features, calculation of self-attention within the same modality, cross-modal attention fusion, generation of dynamic gating vectors, feature fusion representation, adaptive calculation of local scores and decision weights, generates the final model, completes the multimodal fusion evaluation modeling, and sends the data to the teaching effectiveness evaluation module.

[0034] The teaching effectiveness evaluation module receives data sent by the multimodal fusion evaluation modeling module, uses the multimodal fusion evaluation model to output a comprehensive evaluation score of the foreign language ability of the user, with the score ranging from 0 to 1. The higher the score, the better the foreign language teaching effect.

[0035] The beneficial effects achieved by adopting the above solution are as follows:

[0036] (1) In view of the problem that traditional speech feature extraction cannot accurately quantify speech clarity, pronunciation correctness and speech coherence, this scheme constructs a speech clarity composite index by spectral subtraction noise reduction and short-time energy and formant difference analysis, calculates acoustic semantic coupling degree based on the probability coupling of acoustic model and language model, and uses frame-level Mel frequency cepstral coefficients and their first and second order differences to construct phoneme feature vectors to evaluate phoneme transfer stability. Thus, it can accurately quantify speech clarity, accuracy and coherence in a noisy environment, and significantly improve the objectivity, reliability and applicability of foreign language teaching assessment.

[0037] (2) In view of the problems that traditional foreign language teaching assessment methods treat each modality feature in isolation, have a single fusion method and fixed weights, make it difficult to explore the deep dependence between learning behavior, performance and speech, and are also susceptible to noise interference and cannot meet personalized needs, this scheme eliminates scale differences by uniformly mapping heterogeneous modal features to the latent space, introduces the same modal self-attention to deeply capture internal dependencies, and then uses fully interconnected cross-modal cross attention to realize bidirectional interaction of multimodal information. It also adaptively adjusts the same and cross-modal fusion ratio through complementary gating mechanism, combines light quantum network to quickly generate local scores and meta-decision network to dynamically allocate personalized weights, and finally optimizes end-to-end with differentiable weighting strategy, thereby significantly improving the precision and accuracy of assessment, enhancing the adaptability, real-time performance and interpretability of the system, and providing more comprehensive and reliable technical support for personalized foreign language ability assessment. Attached Figure Description

[0038] Figure 1 A schematic diagram illustrating the artificial intelligence-based foreign language teaching effectiveness evaluation method provided by this invention;

[0039] Figure 2 A schematic diagram of the artificial intelligence-based foreign language teaching effectiveness evaluation system provided by the present invention;

[0040] Figure 3 This is a schematic diagram of step S2;

[0041] Figure 4 This is a schematic diagram of step S3.

[0042] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0043] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0044] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0045] Example 1, see Figure 1 The present invention provides an artificial intelligence-based method for evaluating the effectiveness of foreign language teaching, which includes the following steps:

[0046] Step S1: Data collection, collecting foreign language learners' foreign language reading voice data, foreign language learning behavior data, and foreign language test score data;

[0047] Step S2: Speech feature extraction. The final speech feature vector is generated by standardizing the foreign language reading speech data, evaluating pronunciation clarity, calculating acoustic semantic coupling degree, and calculating phoneme transfer stability index.

[0048] Step S3: Multimodal fusion evaluation modeling, which involves data preparation, unified mapping of modal features, calculation of self-attention within the same modality, cross-modal attention fusion, generation of dynamic gating vectors, feature fusion representation, adaptive calculation of local scores and decision weights, generating the final model and completing the multimodal fusion evaluation modeling;

[0049] Step S4: Teaching effectiveness evaluation. Using a multimodal fusion evaluation model, output the comprehensive evaluation score of the foreign language ability of the user. The score range is 0 to 1, and the higher the score, the better the foreign language teaching effect.

[0050] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the data collection specifically involves collecting foreign language learners' foreign language reading audio data, foreign language learning behavior data, and foreign language test score data. The foreign language learning behavior data includes learning duration, learning frequency, and learning task completion rate. The foreign language test score data includes foreign language vocabulary test scores, foreign language reading comprehension test scores, and foreign language listening test scores.

[0051] Example 3, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S2, the speech feature extraction specifically includes the following:

[0052] Step S21: Standardization processing. The original speech signal is standardized, and background noise is removed by spectral subtraction. Speech activity is detected and effective speech segments are extracted based on short-time energy and zero-crossing rate. High-frequency attenuation is compensated. Then, the preprocessed signal is divided into short frames of 25ms with a step size of 10ms, and a Hamming window is applied to each frame.

[0053] Step S22: Pronunciation clarity assessment, specifically including the following steps:

[0054] Step S221: Normalize the short-time energy of the speech to obtain the normalized short-time energy, as shown below:

[0055] ;

[0056] in, Let represent the normalized short-time energy, t represent the frame index of the speech signal, r represent the sampling point index of the frame, and LF represent the frame length. This represents the speech signal value at the r-th sampling point of the t-th frame. This represents the raw short-time energy of frame t. and These represent the mean and standard deviation of the raw short-time energy for all frames, respectively.

[0057] Step S222: Extract the frequencies of the first three formants using a linear predictive coding method. , and Then, the frequency difference between the second and third resonance peaks is normalized, as shown below:

[0058] ;

[0059] in, This represents the normalized frequency difference between the second and third formants. and These represent the frequencies of the 2nd and 3rd formants extracted in frame t, respectively. and These represent the mean and standard deviation of the frequency difference between the second and third formants in all frames, respectively.

[0060] Step S223: Combine the energy spectrum with the resonance peak difference to evaluate the articulation, as shown below:

[0061] ;

[0062] in, The complex index represents the articulation index, and TL represents the total number of frames in the speech signal. This represents the weight of the energy and resonance peak, with a value range of [0.4, 0.8].

[0063] Step S23: Calculate the acoustic semantic coupling degree. First, align phonemes and words. Use the forced alignment tool Kaldi to obtain phonemes. With words The correspondence, and then using an acoustic model Output phonemes The probability of word pairs is obtained by using the pre-trained language model BERT-MLM to output the word pair. The probabilities of the acoustic model and the language model are then aligned, and the tolerance for different speech rates is adjusted using a sensitivity parameter, as shown below:

[0064] ;

[0065] in, This represents the acoustic semantic coupling degree, where OL represents the phoneme sequence length and o represents the phoneme sequence index. This represents an exponential function with the natural constant as its base. This indicates taking the absolute value. This indicates the acoustic model for the o-th phoneme. The probability, This indicates that the language model represents the o-th word. The probability, Indicates coupling sensitivity;

[0066] Step S24: Calculate the phoneme transition stability index. First, construct the acoustic feature vector of each phoneme. Combine the average Mel-frequency cepstral coefficients, first-order difference coefficients, and second-order difference coefficients of all frames contained in the phoneme to form the acoustic feature vector. Then, evaluate the smooth transition between adjacent phoneme features to generate the phoneme transition stability index, which reflects the coherence of speech flow, as shown below:

[0067] ;

[0068] in, Indicates the phoneme transfer stability index. This represents the acoustic feature vector of the (o+1)th phoneme. This represents the acoustic feature vector of the o-th phoneme. This represents the stability adjustment coefficient, with a value range of [0.1, 1]. Indicates modulo;

[0069] Step S25: Generate the final speech feature vector: .

[0070] By performing the above operations, this solution addresses the problem that traditional speech feature extraction processes cannot accurately quantify speech clarity, pronunciation accuracy, and speech coherence. It constructs a composite speech clarity index through spectral subtraction noise reduction and short-time energy and formant difference analysis; calculates acoustic-semantic coupling degree based on the probabilistic coupling of acoustic and language models; and evaluates phoneme transfer stability by constructing phoneme feature vectors using frame-level Mel-frequency cepstral coefficients and their first and second-order differences. This enables precise quantification of speech clarity, accuracy, and coherence in noisy environments, significantly improving the objectivity, reliability, and applicability of foreign language teaching assessment.

[0071] Example 4, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S3, the multimodal fusion evaluation modeling specifically includes the following steps:

[0072] Step S31: Data preparation. Foreign language learning behavior data (learning duration, learning frequency, and learning task completion rate) are combined into a learning behavior feature vector; foreign language exam score data (vocabulary exam scores, reading comprehension exam scores, and listening comprehension exam scores) are combined into a learning performance feature vector; finally, the final speech vector is obtained. Generate source data for teaching effectiveness evaluation;

[0073] Step S32: Modal feature unification mapping, synchronously mapping heterogeneous features to a unified latent space, as shown below:

[0074] ;

[0075] Where m represents the modality index. This represents the original eigenvector of mode m. This represents the vector after the mode m has undergone a unified mapping. Represents a unified mapping function. Represents the unified mapping matrix. Represents the uniform mapping bias vector;

[0076] Step S33: Same-modal self-attention calculation. Self-attention is used within the same modality to uncover dependencies between internal features, as shown below:

[0077] ;

[0078] Where i and j represent the indices of the sub-features, Let represent the attention weight of the i-th sub-feature in modality m on the j-th sub-feature. This represents the i-th sub-feature in mode m. This represents the query transformation matrix in self-attention. This represents the j-th sub-feature in mode m. This represents the key transformation matrix in self-attention, where T denotes the transpose sign. This represents the internal dimensions of the attention head. The index represents the sub-feature, and L represents the total number of sub-features. Represents the first mode in mode m. Individual characteristics;

[0079] Step S34: Cross-modal attention fusion, using a fully interconnected cross-attention structure, enables each modality to exchange information with all other modalities simultaneously, as shown below:

[0080] ;

[0081] in, This represents the attention weights after cross-modal attention fusion of modality m. This represents the normalized exponential function, where n represents the modality index. This represents the vector after the mode n has undergone a unified mapping. The query transformation matrix represents cross-modal attention. Show the key transformation matrix of cross-modal attention. The value transformation matrix represents cross-modal attention;

[0082] Step S35: Generate a dynamic gating vector, introduce a complementary gating mechanism, and learn adaptive weights based on the original and cross-modal representations to control the information fusion ratio, as shown below:

[0083] ;

[0084] in, The gating coefficients of mode m are represented. This represents the sigmoid function. and These represent the gating weight and gating bias, respectively. This indicates that the average pooling is performed row by row. This represents the vector concatenation operation;

[0085] Step S36: Feature fusion representation. Design a differentiable weighted fusion method to fuse same-modal and cross-modal features according to gating coefficients, as shown below:

[0086] ;

[0087] in, This represents the feature representation of mode m after gating coefficient fusion;

[0088] Step S37: Local scoring. A light quantum network is designed for each modality, and a nonlinear mapping is introduced to convert the fused features into evaluation scores, as shown below:

[0089] ;

[0090] in, This represents the local score of mode m. Represents the linear rectified function. and These represent the subnetwork weights and biases of mode m, respectively;

[0091] Step S38: Adaptive calculation of decision weights, construction of a meta-decision network, inputting local scores and learning performance feature vectors together to generate fusion weights, as shown below:

[0092] ;

[0093] in, Indicates the weight of the final scoring decision. This represents the weight of the meta-decision network for mode m. Represents the feature vector of academic performance. This represents the bias of the meta-decision network toward mode m. This represents the local score of mode n. This represents the bias of the meta-decision network toward mode n;

[0094] Step S39: Generate the final model. Utilize the final scoring decision weights to weight the local scores to obtain the overall score, generating a multimodal fusion evaluation model, as shown below:

[0095] ;

[0096] in, The output value of the multimodal fusion evaluation model is the comprehensive evaluation score of foreign language ability, which ranges from 0 to 1.

[0097] By performing the above operations, this solution addresses the problems of traditional foreign language teaching assessment methods, which isolate and process features of different modalities, employ a single fusion method with fixed weights, struggle to uncover the deep dependencies between learning behaviors, grades, and speech, are susceptible to noise interference, and fail to meet personalized needs. Instead, it eliminates scale differences by uniformly mapping heterogeneous modal features to a latent space, introduces intramodal self-attention to deeply capture internal dependencies, utilizes fully interconnected cross-modal cross-attention to achieve bidirectional interaction of multimodal information, adaptively adjusts the intramodal and cross-modal fusion ratio through a complementary gating mechanism, rapidly generates local scores using a lightweight quantum network, dynamically allocates personalized weights using a meta-decision network, and finally optimizes end-to-end using a differentiable weighting strategy. This significantly improves the granularity and accuracy of the assessment, enhances the system's adaptability, real-time performance, and interpretability, and provides more comprehensive and reliable technical support for personalized foreign language proficiency assessment.

[0098] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the teaching effect evaluation specifically involves acquiring the foreign language reading speech data, foreign language learning behavior data, and foreign language test score data of the user being evaluated for foreign language teaching effectiveness. First, speech features are extracted from the data, and then the data is input into a multimodal fusion evaluation model. The multimodal fusion evaluation model outputs a comprehensive evaluation score of the user's foreign language ability. The score range is from 0 to 1, and the higher the score, the better the foreign language teaching effect.

[0099] Example 6, see Figure 1 and Figure 2 Based on the above embodiments, the foreign language teaching effectiveness evaluation system based on artificial intelligence provided by the present invention includes a data acquisition module, a speech feature extraction module, a multimodal fusion evaluation modeling module, and a teaching effectiveness evaluation module.

[0100] The data acquisition module collects foreign language learners' foreign language reading speech data, foreign language learning behavior data, and foreign language test score data, and sends the data to the speech feature extraction module and the multimodal fusion evaluation and modeling module.

[0101] The speech feature extraction module receives data sent by the data acquisition module, performs standardization processing on foreign language reading speech data, evaluates pronunciation clarity, calculates acoustic semantic coupling degree and phoneme transfer stability index, generates final speech feature vector, and sends the data to the multimodal fusion evaluation and modeling module.

[0102] The multimodal fusion evaluation modeling module receives data sent by the data acquisition module and the speech feature extraction module, and sequentially performs data preparation, unified mapping of modal features, calculation of self-attention within the same modality, cross-modal attention fusion, generation of dynamic gating vectors, feature fusion representation, adaptive calculation of local scores and decision weights, generates the final model, completes the multimodal fusion evaluation modeling, and sends the data to the teaching effectiveness evaluation module.

[0103] The teaching effectiveness evaluation module receives data sent by the multimodal fusion evaluation modeling module, uses the multimodal fusion evaluation model to output a comprehensive evaluation score of the foreign language ability of the user, with the score ranging from 0 to 1. The higher the score, the better the foreign language teaching effect.

[0104] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0105] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0106] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for evaluating the effectiveness of foreign language teaching based on artificial intelligence, characterized in that, The method includes the following steps: Step S1: Data collection, collecting foreign language learners' foreign language reading voice data, foreign language learning behavior data, and foreign language test score data; Step S2: Speech feature extraction. The final speech feature vector is generated by standardizing the foreign language reading speech data, evaluating pronunciation clarity, calculating acoustic semantic coupling degree, and calculating phoneme transfer stability index. Step S3: Multimodal fusion evaluation modeling, which involves data preparation, unified mapping of modal features, calculation of self-attention within the same modality, cross-modal attention fusion, generation of dynamic gating vectors, feature fusion representation, adaptive calculation of local scores and decision weights, generating the final model and completing the multimodal fusion evaluation modeling; Step S4: Teaching effectiveness evaluation. Using a multimodal fusion evaluation model, output the comprehensive evaluation score of the foreign language ability of the user. The score range is 0 to 1, and the higher the score, the better the foreign language teaching effect.

2. The method for evaluating the effectiveness of foreign language teaching based on artificial intelligence according to claim 1, characterized in that: In step S2, the speech feature extraction specifically includes the following: Step S21: Standardization processing. The original speech signal is standardized by using spectral subtraction to remove background noise, and effective speech segments are extracted based on speech activity detection with short-time energy and zero-crossing rate. High-frequency attenuation is compensated. Then, the preprocessed signal is divided into short-time frames and a Hamming window is applied to each frame. Step S22: Pronunciation clarity assessment, specifically including the following steps: Step S221: Normalize the short-time energy of the speech to obtain the normalized short-time energy; Step S222: Extract the frequencies of the first three formants using the linear predictive coding method, and then normalize the frequency difference between the second and third formants; Step S223: Combine the energy spectrum with the resonance peak difference to assess the articulation; Step S23: Calculate the acoustic semantic coupling degree. First, transcribe the speech signal to obtain a word sequence, and use a pronunciation dictionary to map each word to the corresponding phoneme sequence. Then, align the phonemes and words. First, use a forced alignment tool to obtain the correspondence between phonemes and words. Then, use an acoustic model to output the probability of phonemes and a pre-trained language model to output the probability of words. Finally, align the acoustic model and language model probabilities to generate the acoustic semantic coupling degree. Step S24: Calculate the phoneme transfer stability index. First, construct the acoustic feature vector of the phoneme. Combine the average Mel frequency cepstral coefficient, first-order difference coefficient and second-order difference coefficient of each phoneme with the average Mel frequency of all frames contained in the phoneme to form the acoustic feature vector. Then evaluate the smooth transition between adjacent phoneme features to generate the phoneme transfer stability index. Step S25: Generate the final speech feature vector.

3. The method for evaluating the effectiveness of foreign language teaching based on artificial intelligence according to claim 1, characterized in that: In step S3, the multimodal fusion evaluation modeling specifically includes the following steps: Step S31: Data preparation. Combine foreign language learning behavior data (learning duration, learning frequency, and learning task completion rate) into a learning behavior feature vector; combine foreign language test score data (foreign language vocabulary test score, foreign language reading comprehension test score, and foreign language listening test score) into a learning performance feature vector; then obtain the final speech vector; generate source data for teaching effectiveness evaluation. Step S32: Modal feature unification mapping, synchronously mapping heterogeneous features to a unified latent space; Step S33: Same-modal self-attention calculation, using self-attention within the same modality to uncover dependencies between internal features; Step S34: Cross-modal attention fusion, using a fully interconnected cross-attention structure, enabling each modality to exchange information with all other modalities simultaneously; Step S35: Generate dynamic gating vectors, introduce complementary gating mechanism, learn adaptive weights based on the original and cross-modal representations, and control the information fusion ratio; Step S36: Feature fusion representation, designing differentiable weighted fusion, fusing same-modal and cross-modal features according to gating coefficients; Step S37: Local scoring, design a light quantum network for each mode and introduce a nonlinear mapping to convert the fused features into evaluation scores; Step S38: Adaptive calculation of decision weights, construction of meta-decision network, inputting local scores and learning performance feature vectors together to generate fusion weights; Step S39: Generate the final model, use the final scoring decision weights to weight the local scores to obtain the overall score, and generate a multimodal fusion evaluation model.

4. The method for evaluating the effectiveness of foreign language teaching based on artificial intelligence according to claim 1, characterized in that: In step S1, the data collection specifically involves collecting foreign language learners' foreign language reading audio data, foreign language learning behavior data, and foreign language test score data. The foreign language learning behavior data includes learning duration, learning frequency, and learning task completion rate. The foreign language test score data includes foreign language vocabulary test scores, foreign language reading comprehension test scores, and foreign language listening test scores.

5. The method for evaluating the effectiveness of foreign language teaching based on artificial intelligence according to claim 1, characterized in that: In step S4, the teaching effectiveness evaluation specifically involves acquiring foreign language reading speech data, foreign language learning behavior data, and foreign language test score data of the users being evaluated for foreign language teaching effectiveness. First, speech features are extracted from the data, and then the data is input into a multimodal fusion evaluation model. The multimodal fusion evaluation model outputs a comprehensive evaluation score of the user's foreign language ability, with a score range of 0 to 1. A higher score indicates a better foreign language teaching effect.

6. An AI-based foreign language teaching effectiveness evaluation system, used to implement the AI-based foreign language teaching effectiveness evaluation method as described in any one of claims 1-5, characterized in that: It includes a data acquisition module, a speech feature extraction module, a multimodal fusion evaluation and modeling module, and a teaching effectiveness evaluation module.

7. The foreign language teaching effectiveness evaluation system based on artificial intelligence according to claim 6, characterized in that: The data acquisition module collects foreign language learners' foreign language reading speech data, foreign language learning behavior data, and foreign language test score data, and sends the data to the speech feature extraction module and the multimodal fusion evaluation and modeling module. The speech feature extraction module receives data sent by the data acquisition module, performs standardization processing on foreign language reading speech data, evaluates pronunciation clarity, calculates acoustic semantic coupling degree and phoneme transfer stability index, generates final speech feature vector, and sends the data to the multimodal fusion evaluation and modeling module. The multimodal fusion evaluation modeling module receives data sent by the data acquisition module and the speech feature extraction module, and sequentially performs data preparation, unified mapping of modal features, calculation of self-attention within the same modality, cross-modal attention fusion, generation of dynamic gating vectors, feature fusion representation, adaptive calculation of local scores and decision weights, generates the final model, completes the multimodal fusion evaluation modeling, and sends the data to the teaching effectiveness evaluation module. The teaching effectiveness evaluation module receives data sent by the multimodal fusion evaluation modeling module, uses the multimodal fusion evaluation model to output a comprehensive evaluation score of the foreign language ability of the user, with the score ranging from 0 to 1. The higher the score, the better the foreign language teaching effect.

Images