A teacher classroom instruction recognition method and system based on classifier integration

Abstract

The present invention discloses a method and system for identifying teacher's classroom instructions based on classifier integration, which obtains the teacher's speech to be identified during classroom teaching, pre-processes the acquired speech to be identified to obtain an audio segment set, then performs content recognition on the audio segment set to obtain a text set, extracts content attributes from the text set to obtain a feature vector set, and finally inputs the feature vector set into a preset integrated classifier to obtain the instruction classification result corresponding to the speech to be identified, wherein the preset integrated classifier is integrated by multiple base classifiers, and is trained with the feature vector set of speech training data as training samples and the instruction classification result corresponding to the speech training data as sample labels. The present application can realize accurate recognition of teacher's voice instructions in the classroom, which is helpful to better evaluate the teaching effect of teachers in the classroom, and measure the teacher's control over the classroom and the classroom atmosphere of the teacher.

Description

Technical Field

[0001] The present invention relates to the technical field of classroom teaching, and in particular to a method and system for identifying teacher classroom instructions based on classifier integration. Background Art

[0002] In smart classroom teaching, in order to understand, evaluate, adjust and promote teaching services, teaching evaluation is a very important link. Teaching evaluation in classroom teaching, as the basis for teaching activities, can directly reflect the level of teachers' teaching and directly affect the learning effect of students. Among them, the evaluation of the effect of teacher-student interaction is becoming an increasingly important part. The effect of teacher-student interaction refers to the response of students to teachers' instructions. For example, when teachers emphasize that a certain part of knowledge is important and needs to be noted, do students record the content in time? Or when teachers ask questions, do students think about and answer teachers' questions in time? An important part of the evaluation of the effect of teacher-student interaction comes from the issuance of teachers' classroom instructions. The behavioral analysis of teachers' classroom instructions during classroom teaching can analyze the teacher's control over the classroom and the classroom atmosphere from an overall perspective, which is of great significance for improving teaching methods, improving teaching quality, and creating a good classroom atmosphere.

[0003] At present, in the field of voice command signal processing, the text instructions corresponding to the voice commands are usually obtained first, and then the text of the instructions is classified. Traditional machine learning text classification methods mainly include latent Dirichlet distribution, K-nearest neighbor method, support vector machine, etc. These methods are relatively mature, but the classification effect is heavily dependent on the extracted features and model parameter tuning, and the whole process is time-consuming and labor-intensive. With the popularization of neural networks, many scholars have applied them to the field of natural language processing, such as applying convolutional neural networks to sentence classification, and combining recurrent neural networks with convolutional neural networks in text classification. In the classification of short texts with a single model, the above methods have high model complexity. Although good results have been achieved, there is limited room for improvement.

[0004] Therefore, how to achieve accurate recognition of the teacher's voice instructions in the classroom is a technical problem that technical personnel in this field need to solve. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for identifying teacher's classroom instructions based on classifier integration, which can realize accurate recognition of teacher's voice instructions in the classroom, help to better evaluate the teacher's classroom teaching effect, and measure the teacher's control over the classroom and the teacher's classroom atmosphere.

[0006] In order to achieve the above object, the present invention provides a method for identifying teacher classroom instructions based on classifier integration, comprising:

[0007] Obtain the teacher's voice to be recognized during classroom teaching;

[0008] Preprocessing the speech to be recognized to obtain a set of preprocessed audio clips;

[0009] Performing content recognition on the audio clip set to obtain a recognized text set;

[0010] Extracting content attributes from the text set to obtain an extracted feature vector set;

[0011] The feature vector set is input into a preset integrated classifier to obtain the instruction classification result corresponding to the speech to be recognized; wherein the integrated classifier is integrated by multiple base classifiers, and is trained with the feature vector set of speech training data as training samples and the instruction classification results corresponding to the speech training data as sample labels.

[0012] Preferably, the preprocessing of the speech to be recognized to obtain a set of preprocessed audio segments includes:

[0013] Performing noise reduction processing on the speech to be recognized by using wavelet transform;

[0014] The speech to be recognized after the noise reduction process is cut and segmented to obtain a set of audio segments in the form of several sentences.

[0015] Preferably, performing content recognition on the audio clip set to obtain a recognized text set includes:

[0016] The audio segment set is subjected to content recognition of fixed-form sentences using speech recognition technology, and the audio segment set is converted into text to obtain a recognized text set.

[0017] Preferably, the integration method of the multiple base classifiers is specifically as follows:

[0018] The plurality of base classifiers are subjected to an attention mechanism to generate a weight coefficient for each base classifier;

[0019] Based on the weight coefficients, a plurality of base classifiers are integrated by weighted summation.

[0020] Preferably, the multiple base classifiers include: a convolutional neural network base classifier, a bidirectional long short-term memory neural network base classifier, a convolutional long short-term memory neural network base classifier and a regional convolutional neural network base classifier.

[0021] Another aspect of the present invention provides a teacher classroom instruction recognition system based on classifier integration, comprising:

[0022] An acquisition module is used to acquire the teacher's speech to be recognized during classroom teaching;

[0023] A preprocessing module, used for preprocessing the speech to be recognized to obtain a set of preprocessed audio segments;

[0024] A recognition module, used to perform content recognition on the audio clip set to obtain a recognized text set;

[0025] An extraction module, used to extract content attributes from the text set to obtain an extracted feature vector set;

[0026] The output module is used to input the feature vector set into a preset integrated classifier to obtain the instruction classification result corresponding to the speech to be recognized; wherein the integrated classifier is integrated by multiple base classifiers, and is trained with the feature vector set of speech training data as training samples and the instruction classification result corresponding to the speech training data as sample labels.

[0027] Preferably, the preprocessing module comprises:

[0028] A noise reduction processing unit, used for performing noise reduction processing on the speech to be recognized by using wavelet transform;

[0029] The segmentation unit is used to segment the speech to be recognized after the noise reduction process to obtain a set of audio segments in the form of several sentences.

[0030] Preferably,

[0031] The recognition module is specifically used to use speech recognition technology to perform content recognition of fixed-form sentences on the audio segment set, convert the audio segment set into text, and obtain a recognized text set.

[0032] Preferably, the integration method of the multiple base classifiers is specifically as follows:

[0033] The plurality of base classifiers are subjected to an attention mechanism to generate a weight coefficient for each base classifier;

[0034] Based on the weight coefficients, a plurality of base classifiers are integrated by weighted summation.

[0035] Preferably, the multiple base classifiers include: a convolutional neural network base classifier, a bidirectional long short-term memory neural network base classifier, a convolutional long short-term memory neural network base classifier and a regional convolutional neural network base classifier.

[0036] The present invention has at least the following beneficial effects:

[0037] The present invention obtains the teacher's speech to be recognized during classroom teaching, and pre-processes the acquired speech to be recognized to obtain an audio segment set, then performs content recognition on the audio segment set to obtain a text set, and then performs content attribute extraction on the text set to obtain a feature vector set, and finally inputs the feature vector set into a preset integrated classifier to obtain the instruction classification result corresponding to the speech to be recognized, wherein the preset integrated classifier is integrated by multiple base classifiers, and is trained with the feature vector set of speech training data as training samples and the instruction classification result corresponding to the speech training data as sample labels. Due to the idea of ​​integrating multiple base classifiers, when classifying and predicting the feature vector set of the speech to be recognized, as much information as possible can be captured and the real hypothesis space can be learned, thereby improving the recognition accuracy of the instruction, thereby realizing accurate recognition of the teacher's voice instruction in the classroom, helping to better evaluate the classroom teaching effect of the teacher, and measuring the teacher's control over the classroom and the classroom atmosphere of the teacher. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0039] Figure 1 It is a flow chart of a method for identifying teacher's classroom instructions based on classifier integration in an embodiment of the present invention;

[0040] Figure 2 A schematic diagram of an integrated method of integrating classifiers in an embodiment of the present invention;

[0041] Figure 3 It is a structural diagram of a teacher classroom instruction recognition system based on classifier integration in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0043] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only and do not represent the only implementation method.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0045] See also Figure 1 On the one hand, an embodiment of the present invention provides a method for identifying teacher classroom instructions based on classifier integration, comprising:

[0046] S110, obtaining the teacher's speech to be recognized during classroom teaching.

[0047] In an embodiment of the present invention, high-precision digital monitoring microphones are installed on both sides of the teacher's podium and connected to the computer host. When the teacher is teaching in the classroom, the digital monitoring microphones collect the teacher's voice data in real time and save it in .wav format audio files according to time segments as the voice to be recognized.

[0048] S120 , preprocessing the speech to be recognized to obtain a set of preprocessed audio segments.

[0049] In an embodiment of the present invention, in order to avoid interference from external noise, after obtaining the teacher's voice to be recognized, the obtained voice to be recognized is subjected to pre-processing methods such as noise reduction and enhancement to obtain relatively clean voice data, and an audio segmentation operation is performed to split the continuous voice into a set of segmented audio clips.

[0050] S130: Perform content recognition on the audio clip set to obtain a recognized text set.

[0051] In an embodiment of the present invention, in order to perform interactive instruction recognition and classroom speech analysis, the preprocessed audio clip set can be subjected to content recognition through speech recognition technology, and the audio data can be translated into text data to obtain a recognized text set, thereby effectively capturing the language interaction behavior between the teacher and the students in the classroom and conveying instructions through language.

[0052] S140: extract content attributes from the text set to obtain an extracted feature vector set.

[0053] In the embodiment of the present invention, a vector space model is used to represent the text content with vectors in a multidimensional space, and the semantic similarity is expressed by the spatial similarity, which is intuitive, easy to understand, concise and efficient. Specifically, the word2vec tool is used to generate a corpus for training word vectors, and word vectors are trained for short texts. The trained corpus is used to extract the content attribute feature vectors of the recognized text set to obtain a feature vector set corresponding to the speech to be recognized.

[0054] S150. Input the feature vector set into a preset integrated classifier to obtain the instruction classification result corresponding to the speech to be recognized; wherein the integrated classifier is integrated by multiple base classifiers, and is trained with the feature vector set of speech training data as training samples and the instruction classification result corresponding to the speech training data as sample labels.

[0055] In the embodiment of the present invention, a plurality of base classifiers are first combined in parallel to construct a classifier integration system. Then, the feature vector set of speech training data is used as a training sample, and the instruction classification result corresponding to the speech training data is used as a sample label to train each base classifier in the classifier integration system to obtain a plurality of trained base classifiers. Then, according to the importance of each base classifier, the plurality of base classifiers are integrated to obtain a final integrated classifier. Finally, the feature vector set corresponding to the speech to be recognized is input into the integrated classifier, and the feature vector is subjected to text classification to obtain the instruction classification result corresponding to the speech to be recognized.

[0056] From the above, it can be seen that the method for identifying classroom instructions of teachers based on classifier integration provided by the embodiment of the present invention obtains the teacher's voice to be recognized during classroom teaching, and pre-processes the acquired voice to be recognized to obtain an audio segment set, then performs content recognition on the audio segment set to obtain a text set, and then extracts content attributes from the text set to obtain a feature vector set, and finally inputs the feature vector set into a preset integrated classifier to obtain the instruction classification result corresponding to the voice to be recognized, wherein the preset integrated classifier is integrated by multiple base classifiers, and is trained with a feature vector set of voice training data as a training sample and an instruction classification result corresponding to the voice training data as a sample label. Due to the idea of ​​integrating multiple base classifiers, when classifying and predicting the feature vector set of the voice to be recognized, as much information as possible can be captured and the real hypothesis space can be learned, thereby improving the recognition accuracy of the instructions, thereby realizing accurate recognition of the teacher's voice instructions in the classroom, which is helpful to better evaluate the classroom teaching effect of the teacher, and measure the teacher's control over the classroom and the classroom atmosphere of the teacher.

[0057] In one embodiment of the present invention, the above step S120 is introduced as an implementation method for preprocessing the speech to be recognized. Optionally, the process may include:

[0058] S1201, performing noise reduction processing on the speech to be recognized by using wavelet transform.

[0059] In the embodiment of the present invention, wavelet transform is used to reduce noise of the acquired speech to be recognized, and the blank speech area when no one is speaking in the classroom and the relatively messy speech area during free discussion or break are removed to obtain relatively clean speech data. Among them, the coefficients of each layer after wavelet decomposition that are greater than and less than the set threshold are processed respectively, and the part of the wavelet coefficients below the threshold is set to zero, and then the inverse transform is performed, so as to achieve the purpose of removing noise.

[0060] S1202: Segment the speech to be recognized after the noise reduction process to obtain a set of audio segments in the form of several sentences.

[0061] In an embodiment of the present invention, the audioread function in Matlab is used to read the speech to be recognized after noise reduction processing, and it is cut and segmented in the time domain. The sampling rate multiplied by the starting number of seconds is set, and the sampling rate multiplied by the ending number of seconds is set, and the speech to be recognized is divided into short audio streams of one sentence at a time to obtain a set of audio clips in the form of several sentences.

[0062] In another embodiment of the present invention, the aforementioned step S130 is introduced, in which the content of the audio clip set is recognized to obtain the recognized text set:

[0063] The speech recognition technology is used to identify the content of fixed-form sentences in the audio clip set, and the audio clip set is converted into text to obtain a recognized text set.

[0064] In the embodiment of the present invention, the SAPI speech recognition technology of Speech SDK is used to perform content recognition of fixed-form sentences on the audio segment set, capture the text information of the audio segment set, and translate the audio data into text data to obtain a recognized text set. Among them, Speech SDK is a software development resource package for voice application development, developed based on the COM standard, and the underlying protocol is completely independent of the application layer in the form of a COM component.

[0065] The SAPI programming interface for speech recognition is used to obtain the recognized text results, which mainly includes the following parts: (1) IspRecognizer: speech recognition engine interface. It is mainly used for the recognition and creation of recognition engines. There are two types of recognition engines: shared engines and private engines. We use a private recognition engine, which can only be used by the application we created. (2) IspRecoContext: recognition context interface, which is mainly used to receive and send event messages related to speech recognition messages, and load and unload recognition grammar resources. (3) IspRecoGrammer: speech recognition grammar interface. The private speech recognition engine requires an XML grammar document. The program loads and activates XML grammar rules through this interface. The grammar rules define the words and sentences that need to be recognized. (4) IspRecoResult: speech recognition result interface. This interface can infer and recognize the recognized text of the acquired speech information, and then obtain the recognized text results. At the same time, it can also provide relevant information about incorrect recognition and prompt relevant results.

[0066] In the specific implementation, in the aforementioned step S140, when extracting content attributes from a text set to obtain a feature vector set after extraction, when training word vectors for short texts, each word can be represented by a trained word vector:

[0067] W i =(w 1 , w 2 , ..., w k );

[0068] In the formula, w i represents the weight of the i-th dimension in the word vector, and k represents the dimension of the word vector obtained after word2vec training.

[0069] Each sentence can be represented by a concatenation of words:

[0070]

[0071] In the formula, represents the concatenation operator, and N represents the length of the sentence, i.e. the number of words it contains.

[0072] In this way, a matrix representation of each sentence can be obtained as the input data of each model.

[0073] See also Figure 2 , which shows a schematic diagram of an integrated classifier integration method. Figure 2 In another embodiment of the present invention, an integration process of multiple base classifiers is introduced:

[0074] S1501. Multiple base classifiers are subjected to an attention mechanism to generate a weight coefficient for each base classifier.

[0075] In the embodiment of the present invention, considering that different base classifiers have different classification effects on samples of different categories and different regions, for the multiple base classifiers after training, we use the attention mechanism to obtain the weight coefficient of each base classifier, which represents the importance of each base classifier. Since the weights of different base classifiers are adaptively adjusted using the attention mechanism, it is more conducive to selecting the combined weights of different base classifiers for different data, thereby giving play to the classification advantages of different base classifiers on different data, improving the recognition accuracy of text instructions, and thus improving classification performance. Among them, the attention mechanism is a mechanism that makes weighted changes to the target data through encoding and decoding, so that the system knows more clearly where to focus.

[0076] S1502: Based on the weight coefficients, multiple base classifiers are integrated by using a weighted summation method.

[0077] In the embodiment of the present invention, multiple base classifiers are integrated by weighted summation according to the weight coefficient of each base classifier to obtain a final integrated classifier, which is expressed as follows:

[0078]

[0079] In the formula, H(x) represents the integrated classifier, h i (x) represents the i-th base classifier, ω i ≥0 represents the weight coefficient of the i-th base classifier, and T represents the number of base classifiers.

[0080] In a specific implementation, the multiple base classifiers include: a convolutional neural network base classifier, a bidirectional long short-term memory neural network base classifier, a convolutional long short-term memory neural network base classifier and a regional convolutional neural network base classifier.

[0081] In the embodiment of the present invention, the final integrated classifier is composed of a CNN (English full name: Convolutional Neural Networks, Chinese full name: Convolutional Neural Networks) base classifier, a B-LSTM (English full name: Bi-directional Long Short-Term Memory, Chinese full name: Bidirectional Long Short-Term Memory Neural Network) base classifier, a C-LSTM (English full name: Convolutional-Long Short Term Memory, Chinese full name: Convolutional Long Short-Term Memory Neural Network) base classifier, and an R-CNN (English full name: Region-Convolutional Neural Networks, Chinese full name: Regional Convolutional Neural Network) base classifier. Since these four base classifiers have their own areas of expertise, the prediction effects produced when processing the same text set have relatively independent characteristics, which is more conducive to classifying and predicting the teacher's instructions.

[0082] Another aspect of the embodiment of the present invention provides a teacher classroom instruction recognition system based on classifier integration. The system described below can be referenced in correspondence with the method described above.

[0083] See also Figure 3 The system comprises:

[0084] An acquisition module 310 is used to acquire the teacher's speech to be recognized during classroom teaching;

[0085] A preprocessing module 320, used to preprocess the speech to be recognized to obtain a set of preprocessed audio segments;

[0086] The recognition module 330 is used to perform content recognition on the audio segment set to obtain a recognized text set;

[0087] An extraction module 340 is used to extract content attributes from a text set to obtain an extracted feature vector set;

[0088] The output module 350 is used to input the feature vector set into a preset integrated classifier to obtain the instruction classification result corresponding to the speech to be recognized; wherein the integrated classifier is integrated by multiple base classifiers, and is trained with the feature vector set of speech training data as training samples and the instruction classification result corresponding to the speech training data as sample labels.

[0089] Optionally, in one embodiment of the present invention, the preprocessing module 320 includes:

[0090] A noise reduction processing unit 3201 is used to perform noise reduction processing on the speech to be recognized by using wavelet transform;

[0091] The segmentation unit 3202 is used to segment the speech to be recognized after the noise reduction process to obtain a set of audio segments in the form of several sentences.

[0092] Optionally, in one embodiment of the present invention, the recognition module 330 is specifically configured to use speech recognition technology to perform content recognition of fixed-form sentences on the audio segment set, convert the audio segment set into text, and obtain a recognized text set.

[0093] Optionally, in one embodiment of the present invention, the integration method of multiple base classifiers is specifically as follows:

[0094] Multiple base classifiers are passed through the attention mechanism to generate the weight coefficient of each base classifier;

[0095] Based on the weight coefficients, multiple base classifiers are integrated using weighted summation.

[0096] Optionally, in one embodiment of the present invention, the multiple base classifiers include: a convolutional neural network base classifier, a bidirectional long short-term memory neural network base classifier, a convolutional long short-term memory neural network base classifier, and a regional convolutional neural network base classifier.

[0097] From the above, it can be seen that the teacher classroom instruction recognition system based on classifier integration provided by the embodiment of the present invention obtains the teacher's voice to be recognized during classroom teaching, and pre-processes the acquired voice to be recognized to obtain an audio segment set, then performs content recognition on the audio segment set to obtain a text set, and then extracts content attributes from the text set to obtain a feature vector set, and finally inputs the feature vector set into a preset integrated classifier to obtain the instruction classification result corresponding to the voice to be recognized, wherein the preset integrated classifier is integrated by multiple base classifiers, and is trained with a feature vector set of voice training data as a training sample and an instruction classification result corresponding to the voice training data as a sample label. Due to the idea of ​​integrating multiple base classifiers, when classifying and predicting the feature vector set of the voice to be recognized, as much information as possible can be captured and the real hypothesis space can be learned, thereby improving the recognition accuracy of the instruction, thereby realizing accurate recognition of the teacher's voice instruction in the classroom, which is helpful to better evaluate the teacher's classroom teaching effect, and measure the teacher's control over the classroom and the teacher's classroom atmosphere.

[0098] For the description of the relevant parts of the teacher classroom instruction recognition system based on classifier integration provided by an embodiment of the present invention, please refer to the detailed description of the corresponding parts in the teacher classroom instruction recognition method based on classifier integration provided by an embodiment of the present invention, and both have the corresponding effects of the teacher classroom instruction recognition method based on classifier integration provided by an embodiment of the present invention, which will not be repeated here.

[0099] Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for identifying teacher classroom instructions based on classifier integration, characterized in that: include: Obtain the teacher's voice to be recognized during classroom teaching; Preprocessing the speech to be recognized to obtain a set of preprocessed audio clips; Performing content recognition on the audio clip set to obtain a recognized text set; Extracting content attributes from the text set to obtain an extracted feature vector set; Input the feature vector set into a preset integrated classifier to obtain the instruction classification result corresponding to the speech to be recognized; wherein the integrated classifier is integrated by multiple base classifiers, and is trained with the feature vector set of speech training data as training samples and the instruction classification result corresponding to the speech training data as sample labels; The integration method of multiple base classifiers is specifically as follows: The plurality of base classifiers are subjected to an attention mechanism to generate a weight coefficient for each base classifier; Based on the weight coefficients, integrating the plurality of base classifiers by weighted summation; The multiple base classifiers include: a convolutional neural network base classifier, a bidirectional long short-term memory neural network base classifier, a convolutional long short-term memory neural network base classifier and a regional convolutional neural network base classifier; The expression of the integrated classifier is: In the formula, H(x) represents the integrated classifier, h i (x) represents the i-th base classifier, ω i ≥0 represents the weight coefficient of the i-th base classifier, and T represents the number of base classifiers.

2. The teacher classroom instruction recognition method based on classifier integration according to claim 1 is characterized in that: The preprocessing of the speech to be recognized to obtain a set of preprocessed audio segments includes: Performing noise reduction processing on the speech to be recognized by using wavelet transform; The speech to be recognized after the noise reduction process is cut and segmented to obtain a set of audio segments in the form of several sentences.

3. The teacher classroom instruction recognition method based on classifier integration according to claim 2 is characterized in that: The step of performing content recognition on the audio clip set to obtain a recognized text set includes: The audio segment set is subjected to content recognition of fixed-form sentences using speech recognition technology, and the audio segment set is converted into text to obtain a recognized text set.

4. A teacher classroom instruction recognition system based on classifier integration, characterized in that: include: An acquisition module is used to acquire the teacher's speech to be recognized during classroom teaching; A preprocessing module, used for preprocessing the speech to be recognized to obtain a set of preprocessed audio segments; A recognition module, used to perform content recognition on the audio clip set to obtain a recognized text set; An extraction module, used to extract content attributes from the text set to obtain an extracted feature vector set; An output module is used to input the feature vector set into a preset integrated classifier to obtain the instruction classification result corresponding to the speech to be recognized; wherein the integrated classifier is integrated by multiple base classifiers, and is trained with the feature vector set of speech training data as training samples and the instruction classification result corresponding to the speech training data as sample labels; The integration method of multiple base classifiers is specifically as follows: The plurality of base classifiers are subjected to an attention mechanism to generate a weight coefficient for each base classifier; Based on the weight coefficients, integrating the plurality of base classifiers by weighted summation; The multiple base classifiers include: a convolutional neural network base classifier, a bidirectional long short-term memory neural network base classifier, a convolutional long short-term memory neural network base classifier and a regional convolutional neural network base classifier; The expression of the integrated classifier is: In the formula, H(x) represents the integrated classifier, h i (x) represents the i-th base classifier, ω i ≥0 represents the weight coefficient of the i-th base classifier, and T represents the number of base classifiers.

5. The teacher classroom instruction recognition system based on classifier integration according to claim 4 is characterized in that: The preprocessing module comprises: A noise reduction processing unit, used for performing noise reduction processing on the speech to be recognized by using wavelet transform; The segmentation unit is used to segment the speech to be recognized after the noise reduction process to obtain a set of audio segments in the form of several sentences.

6. The teacher classroom instruction recognition system based on classifier integration according to claim 5 is characterized in that: The recognition module is specifically used to use speech recognition technology to perform content recognition of fixed-form sentences on the audio segment set, convert the audio segment set into text, and obtain a recognized text set.

Images