A speech recognition method, device, equipment and medium for campus anti-bullying

Abstract

The present invention relates to the field of big data processing technology, and specifically to a speech recognition method, device, equipment and medium for campus anti-bullying. The present invention performs subsequent anti-bullying detection only when the radar simultaneously recognizes a human body after determining that there is a preset command word in the voice data in a preset environment, so as to prevent false detection caused by noise; subsequently, the first and second natural language models obtained by training the voice data separately after removing the business scene and under the preset business scene are used to perform sentiment analysis on the voice data, so as to exclude false detection caused by mentioning the preset command word in normal conversation, and the first and second natural language models are respectively applicable to universal scenes and scene-specific sentiment distinction, and finally, in the process of identifying the bullying type by speech recognition, a technical effect of having both high scene applicability and judgment accuracy is achieved.

Description

Technical Field

[0001] The present invention relates to the field of big data processing technology, and in particular to a speech recognition method, device, equipment and medium for preventing bullying on campus. Background Art

[0002] With the development of society, school bullying incidents are increasingly taken seriously by schools and parents. The current ways to prevent school bullying can be roughly divided into two categories: machine vision detection and voice recognition.

[0003] Among them, the anti-bullying detection under voice recognition usually detects whether there are bullying-related words in the voice data to determine whether a bullying incident has occurred and issue a timely warning. However, the existing anti-bullying warning method of voice recognition will produce false alarms when bullying-related words are mentioned during normal conversations between people in the detection area. In addition, since voice recognition technology can only perceive sound, when the sound generated by wind, water flow or moving objects in the detection area is similar to the pronunciation of bullying-related words, false alarms will also be generated in the face of such phenomena as auditory hallucination.

[0004] Therefore, the current anti-bullying method based on voice recognition has the defect of low accuracy. Summary of the invention

[0005] In view of this, the embodiments of the present invention provide a speech recognition method, apparatus, device and medium for campus anti-bullying, so as to solve the problem of low accuracy of the anti-bullying method under the current speech recognition mode.

[0006] In a first aspect, an embodiment of the present invention provides a speech recognition method for preventing bullying in schools, the speech recognition method comprising:

[0007] Real-time collection of environmental sound data in a preset environment, speaker separation of the environmental sound data to obtain voice data, and when a preset command word is detected in the voice data, obtaining radar monitoring data of the preset environment, performing human body recognition on the radar monitoring data to obtain a human body recognition result;

[0008] If at least one human body is detected in the human body recognition result, speech recognition is performed on the speech data to obtain a speech recognition result, and the first natural language model obtained by business scenario training is removed, and sentiment analysis is performed on the speech recognition result to obtain a positive emotion class representing positive emotions and a class of confidence and a negative emotion class representing negative emotions and a class of confidence;

[0009] If the confidence levels of the positive emotion class and the negative emotion class are both within a preset confidence interval, a second natural language model trained with a preset business scenario is used to perform sentiment analysis on the speech recognition result to determine a second confidence level representing positive emotions and a second confidence level representing negative emotions;

[0010] Determining gain data according to the two-category confidence level representing the positive emotion and the two-category confidence level representing the negative emotion, and performing gain processing on the speech recognition result using the gain data to obtain a speech recognition result after gain;

[0011] Using the first natural language model, sentiment analysis is performed on the gained speech recognition result to obtain a sentiment analysis result, and based on the sentiment analysis result, a recognition result representing the type of bullying is determined.

[0012] In a second aspect, an embodiment of the present invention provides a speech recognition device for preventing bullying in schools, the speech recognition device comprising:

[0013] A voice data acquisition module is used to collect environmental sound data in a preset environment in real time, perform speaker separation on the environmental sound data to obtain voice data, and when a preset command word is detected in the voice data, obtain radar monitoring data of the preset environment, perform human body recognition on the radar monitoring data, and obtain a human body recognition result;

[0014] A sentiment analysis module, for performing speech recognition on the speech data to obtain a speech recognition result if at least one human body is detected in the human body recognition result, removing the first natural language model obtained by business scenario training, performing sentiment analysis on the speech recognition result to obtain a positive emotion class representing positive emotions and a class of confidence and a negative emotion class representing negative emotions and a class of confidence;

[0015] A secondary sentiment analysis module, for performing sentiment analysis on the speech recognition result using a second natural language model trained with a preset business scenario to determine a second confidence level representing positive emotions and a second confidence level representing negative emotions if both the first confidence levels of the positive emotion class and the first confidence level of the negative emotion class are within a preset confidence interval;

[0016] A speech recognition result gain module, used to determine gain data according to the two-category confidence level representing positive emotions and the two-category confidence level representing negative emotions, and use the gain data to perform gain processing on the speech recognition result to obtain a speech recognition result after gain;

[0017] The bullying type recognition module is used to use the first natural language model to perform sentiment analysis on the gained speech recognition result to obtain a sentiment analysis result, and determine a recognition result representing the bullying type based on the sentiment analysis result.

[0018] In a third aspect, an embodiment of the present invention provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the speech recognition method for preventing bullying in schools as described in the first aspect is implemented.

[0019] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the speech recognition method for preventing bullying in schools as described in the first aspect is implemented.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] First, after determining that there are preset command words in the voice data in the preset environment, the present invention does not directly carry out bullying identification but additionally combines radar for human body identification. Only when the existence of a human body is identified by radar is the subsequent anti-bullying detection carried out to prevent the occurrence of bullying false detections due to noise similar to the pronunciation of the preset command words in the preset environment. Subsequently, the present invention removes the business scenario and trains the first and second natural language models respectively under the preset business scenario to perform sentiment analysis on the voice data, thereby eliminating false detections caused by mentioning preset command words in normal conversations, and the first and second natural language models are respectively applicable to universal scenarios and scenario-specific sentiment differentiation. Finally, in the process of identifying and segmenting bullying types by voice recognition, the technical effect of both high scenario applicability and judgment accuracy is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, 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 of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0023] Figure 1 This is a schematic diagram of an application environment of a speech recognition method for campus anti-bullying provided by Embodiment 1 of the present invention;

[0024] Figure 2 It is a flow chart of a speech recognition method for preventing bullying in schools provided by the first embodiment of the present invention;

[0025] Figure 3 is a flow chart of a method for obtaining gain data provided in Embodiment 2 of the present invention;

[0026] Figure 4 1 is a schematic diagram of the structure of a speech recognition device for preventing bullying in schools provided by Embodiment 3 of the present invention;

[0027] Figure 5 It is a structural diagram of a computer device provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0028] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0029] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0030] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0031] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0032] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0033] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0034] It should be understood that the size of the serial numbers of the steps in the following embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0035] In order to illustrate the technical solution of the present invention, specific embodiments are provided below for illustration.

[0036] A speech recognition method for campus anti-bullying provided by the first embodiment of the present invention can be applied in the following aspects: Figure 1 In the application environment, the sensing device is a device installed in a preset environment and used to collect the corresponding environmental sound data and point cloud data in the preset environment. For example, the sensing device can be a microphone and a radar, wherein the radar can specifically be a millimeter wave radar, a laser radar, etc. The sensing device is connected to the server through a network, etc., so that the server can obtain the corresponding data and perform data processing. The computer equipment corresponding to the server can be implemented with an independent server or a server cluster composed of multiple servers. The server, the sensing device, etc. form the entire speech recognition system, and the server is used to collect data and perform data analysis and processing, thereby realizing bullying identification. Of course, the speech recognition method for campus anti-bullying of the present application can also be applied to the sensing device. The prerequisite is that the sensing device has a processing chip capable of performing analytical operations. At this time, the sensing device only needs to send the processing results to the server to realize anti-bullying supervision.

[0037] See also Figure 2 , is a flow chart of a speech recognition method for campus anti-bullying provided in Example 1 of the present application. The speech recognition method for campus anti-bullying is applied to Figure 1 The server or sensing device in Figure 2 As shown, the speech recognition method for preventing bullying in schools may include the following steps:

[0038] Step S101, collect environmental sound data in a preset environment in real time, perform speaker separation on the environmental sound data to obtain voice data, and when it is detected that the voice data contains a preset command word, obtain radar monitoring data of the preset environment, perform human body recognition on the radar monitoring data, and obtain a human body recognition result.

[0039] In this embodiment, the preset environment may refer to the installation scenario of the sensing device. For example, the set area may be a student dormitory, a corridor or a staircase in a teaching building or a dormitory building, wherein the sensing device is a sound pickup device of various forms and appearances, which is used to obtain ambient sound data in the preset environment.

[0040] The ambient sound data may have different decibel values ​​depending on the number of people in the preset environment. When the ambient sound data collected from the preset environment is less than the preset decibel threshold, it can be considered that there are no people in the preset environment at this time, so subsequent processing is only performed when the decibel of the ambient sound data is greater than the preset decibel threshold. Of course, in other embodiments, in order to further ensure the anti-bullying effect, the decibel value of the ambient sound data can be judged without performing subsequent processing.

[0041] After obtaining the environmental sound data and its decibel value is greater than the preset decibel threshold, considering that the environmental sound data may contain sound data generated by multiple people speaking at the same time, in order to ensure the accuracy of subsequent anti-bullying recognition, the environmental sound data is first separated by speakers to obtain voice data that distinguishes the sound data of different speakers according to the identity of the speaker.

[0042] For the obtained voice data, it is detected whether there are preset command words in it. The preset command words are words related to bullying. The detection of preset command words is essentially the existing voice wake-up (Keyword Spotting) process. When the preset command words are detected in the voice data, it means that a bullying incident may have occurred at this time.

[0043] In order to avoid wasting recognition computing resources, and more importantly, to prevent misjudgment caused by the presence of sound data in the environmental sound data that is similar to the pronunciation of the above-mentioned preset command words related to bullying due to wind, water flow or object movement, after detecting the presence of preset command words in the voice data, this embodiment also additionally calls the radar deployed in the preset environment to perform human body recognition. The radar monitoring data is used to perform human body recognition to assist in determining whether there are actually people in the preset environment, so as to complete the judgment of whether the prerequisites for the occurrence of a bullying incident are met.

[0044] Step S102, if it is detected that there is at least one human body in the human recognition result, speech recognition is performed on the speech data to obtain a speech recognition result, and the first natural language model obtained by business scenario training is removed, and sentiment analysis is performed on the speech recognition result to obtain a positive emotion class representing positive emotions and a class of confidence and a negative emotion class representing negative emotions and a class of confidence.

[0045] When there is at least one human body in the human body recognition result, it means that there is a prerequisite for a bullying incident to occur. At this time, the voice data is subjected to voice recognition to obtain the voice recognition result. The process of voice recognition is to convert the vocabulary in the voice into computer-readable data, such as text.

[0046] After obtaining the speech recognition results, in order to prevent false alarms of bullying incidents due to the mention of bullying-related words in normal conversations between people in the preset scenarios, sentiment analysis is continued on the speech recognition results. The sentiment analysis results are used to distinguish which are normal conversation contents and which are conversation contents in real bullying scenarios.

[0047] Specifically, this embodiment uses the first natural language model trained by removing the business scenario to perform sentiment analysis on the speech recognition results, and obtains a positive emotion class representing positive emotions and a confidence level and a negative emotion class representing negative emotions and a confidence level. Removing the business scenario here means not targeting a specific scenario, that is, the first natural language model in this embodiment is a sentiment analysis model trained using speech data in a general scenario as a training sample, and its training samples include various scenarios, not specific occasions, so the first natural language model obtained by training has scenario generalization, which improves its scenario applicability.

[0048] Also, positive emotion class and negative emotion class refer to various emotions under the positive emotion classification and various emotions under the negative emotion classification. The positive emotion class can include specific emotions such as happiness, recognition, excitement, etc., while the negative emotion class can include specific emotions such as sadness, disgust, anger, fear and disappointment. These specific emotions can be summarized and summarized from a large amount of training data. They are only listed here rather than being exhaustive.

[0049] Correspondingly, the positive emotion class representing positive emotions and its confidence level refer to the confidence level of a certain positive emotion class in the positive emotion direction after sentiment analysis of the speech recognition results, such as the confidence level of happy emotions; the negative emotion class representing negative emotions and its confidence level refer to the confidence level of a certain negative emotion class in the negative emotion direction after sentiment analysis of the speech recognition results, such as the confidence level of fear emotions.

[0050] Step S103, if the first-class confidence of the positive emotion class and the first-class confidence of the negative emotion class are both in the preset confidence interval, a second natural language model trained by a preset business scenario is used to perform sentiment analysis on the speech recognition result to determine the second-class confidence representing positive emotions and the second-class confidence representing negative emotions.

[0051] After the first natural language model performs sentiment analysis on the speech recognition result, if the sentiment analysis result shows that the sentiment classification of the current speech recognition result belongs to a category of positive emotion and negative emotion, and the confidence is low and in the preset confidence interval, it indicates that it is very likely to belong to the neutral emotion classification result. Among them, the preset confidence interval can be adaptively set by the operator in combination with the accuracy requirements of the sentiment analysis judgment. In this embodiment, the preferred preset confidence interval refers to an interval with a confidence less than a preset confidence threshold, and the specific value of the preset confidence threshold is 0.5, that is, the preferred preset confidence interval of this embodiment is [0,0.5]. In other embodiments, the minimum value of the preset confidence interval can also be a value greater than 0, such as 0.1.

[0052] However, although the first natural language model analyzes the speech recognition results and believes that the confidence, that is, the possibility, of the emotion classification contained in the speech recognition results belonging to the positive emotion class and the negative emotion class is not high, because the first natural language model is a generalized universal model, although it has high scene applicability, it is precisely because it is a generalized model that the emotion analysis results obtained by its analysis, that is, the possibility of the above speech recognition results belonging to the positive emotion class and the negative emotion class is not high, does not necessarily mean that the true emotion classification of the speech recognition results is indeed neutral emotion. There is also a situation where the first natural language model sacrifices part of the analysis accuracy to improve scene applicability, and mistakenly judges the positive emotion class or the negative emotion class as neutral emotion.

[0053] To this end, this embodiment continues to use the second natural language model trained with a preset business scenario to re-analyze the speech recognition results when the possibility of the emotion classification of the speech recognition results belonging to the positive emotion class and the negative emotion class is not high.

[0054] The so-called preset business scenario here refers to a scenario that contains both voice data under the positive emotion category and voice data under the opposite negative emotion category, such as a scenario that includes both conversation content in a laughing situation and conversation content in a crying situation with opposite emotions, or a scenario that includes both conversation content in an approval situation and conversation content in a disgust situation with opposite emotions.

[0055] The second natural language model trained with the preset business scenario is more effective and accurate in distinguishing between ambiguous positive and negative emotions because its training samples are completely constructed from positive emotions and negative emotions. After re-analyzing the emotion of speech recognition results based on the second natural language model, the second-class confidence representing positive emotions and the second-class confidence representing negative emotions based on the second natural language model can be obtained.

[0056] It should also be noted that, in most cases, after the first natural language model performs sentiment analysis on the speech recognition results, the above-mentioned ambiguous situation in which the confidence levels of the positive emotion class and the negative emotion class are both within the preset confidence interval will not occur. Instead, clear and reliable positive or negative emotion judgment results will be directly generated.

[0057] Therefore, optionally, after performing sentiment analysis on the speech recognition result to obtain a positive emotion class representing positive emotions and a class of confidences thereof and a negative emotion class representing negative emotions and a class of confidences thereof, the method further includes:

[0058] If the confidence level of the positive emotion class is less than the minimum value of the preset confidence interval, and the confidence level of the negative emotion class is greater than the maximum value of the preset confidence interval, then detecting whether the confidence level of the negative emotion class exceeds the preset value;

[0059] If it is detected that a confidence level of the negative emotion class exceeds the preset value, the first mapping table is called to perform keyword recognition on the speech recognition result to obtain a recognition result representing the bullying type.

[0060] Among them, a confidence level of the positive emotion class is less than the minimum value of the preset confidence interval, and a confidence level of the negative emotion class is greater than the maximum value of the preset confidence interval. This means that after the first natural language model performs sentiment analysis on the speech recognition result, an emotion judgment result that is not ambiguous and requires further analysis is obtained. Specifically, a judgment result that can be determined to belong to a negative emotion is obtained. At this time, a judgment is made on whether the confidence level of the negative emotion class is greater than the preset value, so as to further clarify whether the current voice data belongs to a negative emotion scene under bullying circumstances. If it is greater than the preset value, the first mapping table constructed by the mapping relationship between preset keywords and bullying types is called, and keyword recognition is performed on the speech recognition result. The bullying type is determined according to the recognized keywords by comparing the first mapping table.

[0061] Step S104, determining gain data according to the two-category confidence level representing the positive emotion and the two-category confidence level representing the negative emotion, and performing gain processing on the speech recognition result using the gain data to obtain a gained speech recognition result.

[0062] Although the second natural language model can more accurately judge the ambiguous sentiment analysis results of the first natural language model, it is easy to understand that the real emotions corresponding to the language recognition results are still neutral emotions. Therefore, it is necessary to judge the size of the second category of confidence before deciding whether to perform gain processing, and the sentiment analysis results cannot be directly assumed to have only positive and negative emotions.

[0063] To this end, a gain threshold can be preset to determine whether the two-class confidence is greater than the preset gain threshold. Only when it is greater than, it indicates that the emotional type of the speech recognition result is positive or negative, and the credibility is high enough. When no two-class confidence is greater than the preset gain threshold, it indicates that the emotional type of the speech recognition result is likely to be neutral. In this implementation, the specific value of the preset gain threshold is preferably 0.5, and in other embodiments, it can be specifically set by the implementer according to the requirements for the accuracy of bullying recognition.

[0064] When one or more of the two-category confidence representing positive emotions or the two-category confidence representing negative emotions is greater than a preset gain threshold, gain processing is performed on the corresponding emotion type. Any feasible method of gain processing can be selected. For example, when the two-category confidence representing a certain emotion is greater than the preset gain threshold, the perception sensitivity of the first natural language model to the emotion can be improved. Specifically, a weight gain value is set and multiplied by a coefficient related to the perception of the emotion to improve the perception sensitivity of the first natural language model to the emotion. The weight gain value set at this time is the gain data; or, when the two-category confidence representing a certain emotion is greater than the preset gain threshold, since the language recognition result is a text content, the number of related texts under the emotion can be directly increased in the language recognition result to complete the gain of the speech recognition result. The additional related text under the emotion added at this time is the gain data, and so on.

[0065] That is, there are various methods to complete the determination of gain data and obtain the gained speech recognition result based on the gain data. The principle is to first determine which emotion's corresponding second-category confidence level needs gain, and then determine any feasible gain data construction method to complete the gain and obtain the gained speech recognition result.

[0066] Step S105: Use the first natural language model to perform sentiment analysis on the gained speech recognition result to obtain a sentiment analysis result, and determine a recognition result representing the type of bullying based on the sentiment analysis result.

[0067] Based on the second natural language model, the above steps complete the sentiment analysis of the speech recognition results by the first natural language model, and the positive emotion class representing positive emotions and its confidence level and the negative emotion class representing negative emotions and its confidence level are both within the preset confidence interval. This further enhances the distinction between the positive and negative emotion classes where the distinction is not clear, and correspondingly obtains the gained speech recognition results that can more effectively and fully represent the positive and negative emotion classes.

[0068] Therefore, by using the first natural language model again to perform sentiment analysis on the gained speech recognition results, a more accurate sentiment analysis result can be obtained compared to the initial sentiment analysis of the speech recognition results using the first natural language model. Based on the obtained sentiment analysis results, the recognition results representing the bullying type can be determined.

[0069] Optionally, determining, based on the sentiment analysis result, an identification result representing the type of bullying includes:

[0070] Detecting whether the confidence level of the sentiment analysis result for the negative sentiment class exceeds a preset value;

[0071] If it is detected that the confidence of the sentiment analysis result for the negative emotion class exceeds the preset value, the first mapping table corresponding to the negative emotion class is called to perform keyword recognition on the speech recognition result to obtain a recognition result representing the bullying type. The first mapping table is the first mapping table that records the mapping relationship between the preset keywords and the bullying type.

[0072] It should be emphasized that the object of keyword recognition here is the speech recognition result obtained after speech recognition of the speech data, that is, the speech recognition result before gain. The reason is that the speech recognition result after gain includes additional added data that does not belong to the environmental sound data. Therefore, in order to ensure the accuracy of bullying type recognition, keyword recognition is selected for the speech recognition result before gain.

[0073] After completing the keyword recognition, the aforementioned first mapping table is called, and the recognized keyword is compared with the first mapping table. The bullying type corresponding to the recognized keyword is determined by searching the first mapping table to obtain a recognition result representing the bullying type.

[0074] According to the obtained recognition results representing the bullying type, the anti-bullying warning process can be completed, and the specific processing method can be set by the implementer according to the needs and on-site conditions. A better processing method is to define different severity levels of bullying incidents for different bullying types, so optionally, after determining the recognition results representing the bullying type based on the sentiment analysis results, it also includes:

[0075] The alarm level corresponding to the recognition result is obtained, corresponding alarm voice data is determined according to the alarm level, and the alarm voice data is broadcast.

[0076] Preferably, the bullying severity levels corresponding to the bullying type identification results can be divided into three levels. The first level can correspond to a verbal warning, the second level can correspond to calling the manager corresponding to the preset environment and forwarding the manager's voice, and the third level can correspond to calling the police and forwarding the telephone voice.

[0077] Among them, if no one answers the manager's phone call, the manager's voice can be generated and broadcast based on the recorded manager's voice.

[0078] The above processing is all processing means for the scenario where the sentiment analysis result of the speech recognition result after gain is a negative emotion after bullying occurs. It is easy to understand that there may be a scenario where the sentiment analysis result of the speech recognition result after gain does not belong to a negative emotion after bullying occurs. Therefore, optionally, after detecting whether the confidence of the sentiment analysis result being the negative emotion class exceeds a preset value, it also includes:

[0079] If it is detected that the sentiment analysis result is the positive emotion class, or the confidence level of the sentiment analysis result being the negative emotion class does not exceed the preset value, the ambient sound data and the radar monitoring data are sent to the background, so that the background compares the ambient sound data and the radar monitoring data with the historical data received last time to obtain a comparison result, and determines whether to generate an alarm based on the comparison result.

[0080] This situation shows that even after being recognized by the second natural language model, which is a targeted and accurate model specifically designed to distinguish between positive and negative emotions, it is still impossible to determine that the emotion category corresponding to the current speech recognition result is negative, and it is very likely that no bullying incident has occurred.

[0081] However, considering the possibility of failure of the relevant hardware equipment used to implement anti-bullying identification, it is preferred to send the environmental sound data and radar monitoring data in this case to the background, so that the background can compare the current data received with the historical data received last time. Since anti-bullying detection is real-time, the interval between the acquisition of two adjacent data will not be very large. If the data obtained this time is significantly different from the previous historical data, that is, the environmental sound data or radar monitoring data corresponding to the above comparison result has a significant change between the two adjacent acquisition times, it indicates that the hardware equipment is likely to have failed. Therefore, it is set to decide whether to issue an alarm based on the comparison result to remind maintenance personnel to inspect the hardware equipment.

[0082] The implementer may adopt any implementation method in the prior art for the specific form of the comparison result and how to determine whether to generate an alarm based on the comparison result, which will not be described in detail here.

[0083] In summary, after acquiring the voice data in the preset environment and detecting the presence of the preset command words, the present embodiment performs human body recognition through radar to prevent the false detection of the preset command words due to the noise interference in the preset environment, and then performs sentiment analysis on the voice recognition result by using the first natural language model obtained by removing the business scenario training, and continues to call the second natural language model obtained by training with the preset business scenario to perform sentiment analysis on the voice recognition result for the ambiguous sentiment analysis result obtained by the first natural language model analysis, and gains the voice recognition result with the sentiment analysis result obtained by the second natural language model, and again uses the first natural language model to perform sentiment analysis on the gained voice recognition result, and determines the specific type of bullying based on the analysis result. The present embodiment specifically constructs the first natural language model obtained by removing the business scenario training and the second natural language model obtained by training with the preset business scenario, achieving the technical effect of having both high scenario applicability and judgment accuracy in the process of bullying recognition by voice recognition, and eliminates the misjudgment caused by environmental noise, thereby improving the judgment accuracy.

[0084] See also Figure 3 , is a specific method for determining gain data in step S104 given in the second embodiment of the present application, and the method includes:

[0085] Step S201 : determining an emotion with a greater confidence as a target emotion from the two types of confidences representing positive emotions and the two types of confidences representing negative emotions.

[0086] As described in step S103, when the sentiment analysis results obtained after the first natural language model performs sentiment analysis on the speech recognition results belong to a category of positive emotion and a category of negative emotion, and the confidence levels are both within a preset confidence interval, it is possible that the true emotion of the speech recognition result is neutral, or it may belong to one of the positive emotion category or the negative emotion category.

[0087] Therefore, after using the second natural language model to perform sentiment analysis on the speech recognition results to obtain the second category confidence representing positive emotions and the second category confidence representing negative emotions, it is also necessary to judge the size of their confidences. Only when the second category confidence representing positive emotions is greater than the preset gain threshold, the confidence that the emotion classification of the currently judged speech recognition result should belong to positive emotions is high enough, or when the second category confidence representing negative emotions is greater than the preset gain threshold, the confidence that the emotion classification of the currently judged speech recognition result should belong to negative emotions is high enough. The size of the preset gain threshold can be set by the implementer, and in this embodiment, it is preferably 0.5.

[0088] At the same time, considering that the emotion corresponding to the speech recognition result should be one of the positive emotions or the negative emotions, in order to adapt to the actual situation and reduce unnecessary gain processing calculation process, it is further preferred to determine the emotion with greater confidence as the target emotion from the two types of confidence representing positive emotions and the two types of confidence representing negative emotions, and only use the text under the target emotion as gain data.

[0089] Step S202: If the target emotion is a positive emotion, a positive text representing the positive emotion is determined according to the two-category confidence level representing the positive emotion, and the positive text is used as gain data; if the target emotion is a negative emotion, a negative text representing the negative emotion is determined according to the two-category confidence level representing the negative emotion, and the negative text is used as gain data.

[0090] Since the speech recognition result obtained after performing speech recognition on the speech data is a kind of text data, the data type of the gain data can correspond to the data type of the speech recognition result, which is also text data. In this case, when the target emotion is a positive emotion, the positive text representing the positive emotion can be determined according to the size of the second-category confidence of the positive emotion and used as the gain data; or when the target emotion is a negative emotion, the negative text representing the positive emotion can be determined according to the size of the second-category confidence of the negative emotion and used as the gain data.

[0091] In the above process of determining positive text and negative text, a preferred setting manner may be: the number of positive text or negative text is proportional to the degree to which the corresponding second-class confidence exceeds the preset gain threshold.

[0092] Specifically, the interval between the preset gain threshold and 1 is segmented. For example, based on the preferred preset gain threshold of 0.5 in this embodiment, the interval between the preset gain threshold and 1 can be divided into 5 segments with a step size of 0.1, and as the segment gradually approaches 1, the corresponding number of set texts gradually increases. For example, for the case where the second-category confidence falls within the interval of (0.5, 0.6], a text vocabulary of the positive emotion class or the negative emotion class is selected as the gain data; for the case where the second-category confidence falls within the interval of (0.6, 0.7], two text vocabulary of the positive emotion class or the negative emotion class are selected as the gain data, and so on. The specifically selected text vocabulary corresponds to the positive emotion or negative emotion corresponding to the second-category confidence that exceeds the preset gain threshold. For example, when the second-category confidence representing the positive emotion exceeds the preset gain threshold, the text of the positive emotion class is selected as the gain data; when the second-category confidence representing the negative emotion exceeds the preset gain threshold, the text of the negative emotion class is selected as the gain data. Finally, a second mapping table of the mapping relationship between the second-category confidence representing the positive emotion and the corresponding number of positive texts, and a third mapping table of the mapping relationship between the second-category confidence representing the negative emotion and the corresponding number of negative texts can be constructed respectively. In the subsequent process of obtaining the gain data, it is only necessary to determine the specific size of the second-category confidence to search the second mapping table or the third mapping table to determine the matching positive text or negative text.

[0093] Alternatively, in the above process of determining positive text and negative text, another preferred setting method can be: the emotional intensity of the positive text or negative text is proportional to the degree to which the corresponding second-category confidence exceeds the preset gain threshold. That is, when the second-category confidence representing positive emotions exceeds the preset gain threshold to a greater extent, the positive text with greater emotional intensity is used as gain data, and when the second-category confidence representing negative emotions exceeds the preset gain threshold to a greater extent, the negative text with greater emotional intensity is used as gain data. A second mapping table of the mapping relationship between the second-category confidence representing positive emotions and the corresponding positive text emotional intensity, and a third mapping table of the mapping relationship between the second-category confidence representing negative emotions and the corresponding negative text emotional intensity can be constructed respectively.

[0094] That is, there are many methods for determining gain data according to the second type of confidence, and it is only necessary to keep the gain intensity corresponding to the gain data proportional to the size of the second type of confidence.

[0095] Since the speech recognition result and the gain data obtained after the speech recognition of the speech data are both text data, the gain data can be directly added to the speech recognition result to obtain the gained speech recognition result.

[0096] This embodiment provides a gain data acquisition method that is easier to operate, and in the specific process of determining the gain data, matches the amount of gain data with the size of the second category confidence, thereby improving the accuracy of the speech recognition results after obtaining the gain data, and ultimately improving the accuracy of anti-bullying recognition.

[0097] Corresponding to the speech recognition method for campus anti-bullying in the above embodiment, Figure 4 The structure block diagram of the speech recognition device for campus anti-bullying provided in the third embodiment of the present application is shown. The speech recognition device is applied to Figure 1 For ease of explanation, only the parts related to the embodiments of the present application are shown.

[0098] See also Figure 4 , the speech recognition device comprises:

[0099] The voice data acquisition module 31 is used to collect environmental sound data in a preset environment in real time, perform speaker separation on the environmental sound data to obtain voice data, and when a preset command word is detected in the voice data, obtain radar monitoring data of the preset environment, perform human body recognition on the radar monitoring data, and obtain a human body recognition result;

[0100] A primary sentiment analysis module 32, configured to perform speech recognition on the speech data to obtain a speech recognition result if at least one human body is detected in the human body recognition result, and to perform sentiment analysis on the speech recognition result to obtain a positive sentiment class representing positive emotions and a class of confidences thereof and a negative sentiment class representing negative emotions and a class of confidences thereof;

[0101] A secondary sentiment analysis module 33 is used to perform sentiment analysis on the speech recognition result using a second natural language model trained with a preset business scenario to determine a second confidence level representing positive emotions and a second confidence level representing negative emotions if both the first confidence levels of the positive emotion class and the first confidence level of the negative emotion class are within a preset confidence interval;

[0102] A speech recognition result gain module 34, used to determine gain data according to the two-class confidence level representing positive emotions and the two-class confidence level representing negative emotions, and use the gain data to perform gain processing on the speech recognition result to obtain a speech recognition result after gain;

[0103] The bullying type identification module 35 is used to use the first natural language model to perform sentiment analysis on the gained speech recognition result to obtain a sentiment analysis result, and determine a recognition result representing the bullying type based on the sentiment analysis result.

[0104] Optionally, the bullying type identification module 35 includes:

[0105] A negative emotion class confirmation unit, used to detect whether the confidence level of the sentiment analysis result being the negative emotion class exceeds a preset value;

[0106] A bullying type identification unit is used to call the first mapping table corresponding to the negative emotion class if it is detected that the confidence of the sentiment analysis result for the negative emotion class exceeds the preset value, and perform keyword recognition on the speech recognition result to obtain an identification result characterizing the bullying type, wherein the first mapping table records the mapping relationship between the preset keywords and the bullying type.

[0107] Optionally, the bullying type identification module 35 further includes:

[0108] The data retention and fault confirmation unit is used to send the ambient sound data and the radar monitoring data to the background if it is detected that the sentiment analysis result is the positive emotion class, or the confidence level of the sentiment analysis result being the negative emotion class does not exceed the preset value, so that the background compares the ambient sound data and the radar monitoring data with the historical data received last time to obtain a comparison result, and determines whether to generate an alarm based on the comparison result.

[0109] Optionally, the primary sentiment analysis module 32 further includes:

[0110] A primary emotion result confirmation unit, configured to detect whether the confidence level of the positive emotion class exceeds the preset value if the confidence level of the positive emotion class is less than the minimum value of the preset confidence interval and the confidence level of the negative emotion class is greater than the maximum value of the preset confidence interval;

[0111] The bullying determination unit after a sentiment analysis is used to call the first mapping table if it is detected that the confidence of a class of the negative emotion class exceeds the preset value, perform keyword recognition on the speech recognition result, and obtain a recognition result representing the bullying type.

[0112] Optionally, the speech recognition result gain module 34 includes:

[0113] A target emotion determination unit, configured to determine an emotion with a greater confidence as a target emotion from among the two types of confidences representing positive emotions and the two types of confidences representing negative emotions;

[0114] A gain data determination unit is used to determine a positive text representing the positive emotion according to the two-category confidence level representing the positive emotion if the target emotion is a positive emotion, and use the positive text as the gain data; if the target emotion is a negative emotion, determine a negative text representing the negative emotion according to the two-category confidence level representing the negative emotion, and use the negative text as the gain data.

[0115] Optionally, the gain data determining unit includes:

[0116] The positive text gain data acquisition subunit is used to query and obtain matching positive text from a preset second mapping table according to the two types of confidences representing positive emotions, wherein the second mapping table records the mapping relationship between different confidence values ​​and corresponding positive texts.

[0117] Optionally, the gain data determining unit includes:

[0118] The negative text gain data acquisition subunit is used to query and obtain matching negative text from a preset third mapping table according to the second type of confidence representing negative emotions, wherein the third mapping table records the mapping relationship between different confidence values ​​and corresponding negative texts.

[0119] Figure 5 This is a schematic diagram of the structure of a computer device provided in Example 4 of the present application. Figure 5 As shown, the computer device of this embodiment includes: at least one processor ( Figure 5 Only one is shown), a memory, and a computer program stored in the memory and executable on at least one processor, wherein when the processor executes the computer program, the steps in any of the above-mentioned embodiments of the speech recognition method for preventing bullying in schools are implemented.

[0120] The computer device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that Figure 5 These are merely examples of computer devices and do not constitute limitations on the computer devices. The computer devices may include more or fewer components than those shown in the figure, or a combination of certain components, or different components. For example, they may also include a network interface, a display screen, and an input device.

[0121] The processor may be a CPU, or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0122] The memory includes a readable storage medium, an internal memory, etc., wherein the internal memory may be the memory of a computer device, and the internal memory provides an environment for the operation of the operating system and computer-readable instructions in the readable storage medium. The readable storage medium may be a hard disk of a computer device, and in other embodiments, it may also be an external storage device of the computer device, for example, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device. Further, the memory may also include both an internal storage unit of the computer device and an external storage device. The memory is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of a computer program, etc. The memory may also be used to temporarily store data that has been output or is to be output.

[0123] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned device can refer to the corresponding process in the above-mentioned method embodiment, which will not be repeated here. If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned method embodiment can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device capable of carrying computer program code, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0124] The present application implements all or part of the processes in the above-mentioned embodiment method, and may also be completed through a computer program product. When the computer program product runs on a computer device, the steps in the above-mentioned method embodiment can be implemented when the computer device executes it.

[0125] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0126] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0127] In the embodiments provided in the present application, it should be understood that the disclosed devices / computer equipment and methods can be implemented in other ways. For example, the device / computer equipment embodiments described above are only schematic, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0128] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0129] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A speech recognition method for preventing bullying in schools, characterized in that: The speech recognition method comprises: Real-time collection of environmental sound data in a preset environment, speaker separation of the environmental sound data to obtain voice data, and when a preset command word is detected in the voice data, obtaining radar monitoring data of the preset environment, performing human body recognition on the radar monitoring data to obtain a human body recognition result; If at least one human body is detected in the human body recognition result, speech recognition is performed on the speech data to obtain a speech recognition result, and the first natural language model obtained by business scenario training is removed, and sentiment analysis is performed on the speech recognition result to obtain a positive emotion class representing positive emotions and a class of confidence and a negative emotion class representing negative emotions and a class of confidence, wherein the business scenario refers to a scenario that includes both speech data under the positive emotion class and speech data under the negative emotion class with opposing emotions; If the confidence levels of the positive emotion class and the negative emotion class are both within a preset confidence interval, and the preset confidence interval is [0, 0.5], it is considered that this is a neutral emotion classification result, and the second natural language model trained by the business scenario is used to perform sentiment analysis on the speech recognition result to determine the second confidence level representing the positive emotion and the second confidence level representing the negative emotion; Determining gain data according to the two-category confidence level representing the positive emotion and the two-category confidence level representing the negative emotion, and performing gain processing on the speech recognition result using the gain data to obtain a speech recognition result after gain; Using the first natural language model, performing sentiment analysis on the gained speech recognition result to obtain a sentiment analysis result, and determining a recognition result representing the type of bullying based on the sentiment analysis result; The step of determining gain data according to the two-class confidence level representing the positive emotion and the two-class confidence level representing the negative emotion comprises: From the two types of confidences representing positive emotions and the two types of confidences representing negative emotions, determine the emotion with a greater confidence and exceeding a gain threshold as the target emotion; If the target emotion is a positive emotion, a positive text representing the positive emotion is determined according to the two-category confidence representing the positive emotion, and the positive text is used as gain data; if the target emotion is a negative emotion, a negative text representing the negative emotion is determined according to the two-category confidence representing the negative emotion, and the negative text is used as gain data.

2. The speech recognition method for campus anti-bullying according to claim 1, characterized in that: Determining, based on the sentiment analysis result, an identification result representing the type of bullying, includes: Detecting whether the confidence level of the sentiment analysis result for the negative sentiment class exceeds a preset value; If it is detected that the confidence level of the sentiment analysis result for the negative emotion class exceeds the preset value, the first mapping table corresponding to the negative emotion class is called to perform keyword recognition on the speech recognition result to obtain an identification result characterizing the bullying type, wherein the first mapping table records the mapping relationship between the preset keywords and the bullying type.

3. The speech recognition method for preventing bullying in schools according to claim 2, characterized in that: After detecting whether the confidence level of the sentiment analysis result as the negative sentiment class exceeds a preset value, the method further includes: If it is detected that the sentiment analysis result is the positive emotion class, or the confidence level of the sentiment analysis result being the negative emotion class does not exceed the preset value, the ambient sound data and the radar monitoring data are sent to the background, so that the background compares the ambient sound data and the radar monitoring data with the historical data received last time to obtain a comparison result, and determines whether to generate an alarm based on the comparison result.

4. The speech recognition method for preventing bullying in schools according to claim 2, characterized in that: After performing sentiment analysis on the speech recognition result to obtain a positive emotion class representing positive emotions and a class of confidences thereof and a negative emotion class representing negative emotions and a class of confidences thereof, the method further includes: If the confidence level of the positive emotion class is less than the minimum value of the preset confidence interval, and the confidence level of the negative emotion class is greater than the maximum value of the preset confidence interval, then detecting whether the confidence level of the negative emotion class exceeds the preset value; If it is detected that a confidence level of the negative emotion class exceeds the preset value, the first mapping table is called to perform keyword recognition on the speech recognition result to obtain a recognition result representing the type of bullying.

5. The speech recognition method for preventing bullying in schools according to claim 1, characterized in that: The determining of the positive text representing the positive emotion according to the second-class confidence representing the positive emotion, and using the positive text as gain data, comprises: According to the second type of confidence representing the positive emotion, the matching positive text is queried from a preset second mapping table, and the second mapping table records the mapping relationship between different confidence values ​​and corresponding positive texts.

6. The speech recognition method for campus anti-bullying according to claim 1, characterized in that: The step of determining negative text representing negative emotions according to the second-class confidence level representing negative emotions, and using the negative text as gain data, includes: According to the second type of confidence that represents negative emotions, matching negative texts are queried from a preset third mapping table, in which a mapping relationship between different confidence values ​​and corresponding negative texts is recorded.

7. A speech recognition device for preventing bullying in schools, characterized in that: The speech recognition device comprises: A voice data acquisition module is used to collect environmental sound data in a preset environment in real time, perform speaker separation on the environmental sound data to obtain voice data, and when a preset command word is detected in the voice data, obtain radar monitoring data of the preset environment, perform human body recognition on the radar monitoring data, and obtain a human body recognition result; a primary sentiment analysis module, for performing speech recognition on the speech data to obtain a speech recognition result if at least one human body is detected in the human body recognition result, and performing sentiment analysis on the speech recognition result to obtain a positive emotion class representing positive emotions and a class of confidence and a negative emotion class representing negative emotions and a class of confidence, wherein the business scenario refers to a scenario containing both speech data under the positive emotion class and speech data under the negative emotion class with opposing emotions; A secondary sentiment analysis module is used for, if the confidence of the first category of the positive emotion class and the first category of the negative emotion class are both in a preset confidence interval, and the preset confidence interval is [0, 0.5], it is considered that this is a neutral emotion classification result, and then the second natural language model trained by the business scenario is used to perform sentiment analysis on the speech recognition result to determine the second category of confidence representing positive emotions and the second category of confidence representing negative emotions; A speech recognition result gain module, used to determine gain data according to the two-category confidence level representing positive emotions and the two-category confidence level representing negative emotions, and use the gain data to perform gain processing on the speech recognition result to obtain a speech recognition result after gain; a bullying type recognition module, configured to use the first natural language model to perform sentiment analysis on the speech recognition result after gaining, obtain a sentiment analysis result, and determine a recognition result representing the bullying type according to the sentiment analysis result; The step of determining gain data according to the two-class confidence level representing the positive emotion and the two-class confidence level representing the negative emotion comprises: From the two types of confidences representing positive emotions and the two types of confidences representing negative emotions, determine the emotion with a greater confidence and exceeding a gain threshold as the target emotion; If the target emotion is a positive emotion, a positive text representing the positive emotion is determined according to the two-category confidence representing the positive emotion, and the positive text is used as gain data; if the target emotion is a negative emotion, a negative text representing the negative emotion is determined according to the two-category confidence representing the negative emotion, and the negative text is used as gain data.

8. A computer device, characterized in that: The computer device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the speech recognition method for preventing bullying in schools as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the speech recognition method for preventing bullying in schools as described in any one of claims 1 to 6 is implemented.

Images