Method and system for recognizing and classifying speech data related to a contradiction dispute

Abstract

The application discloses a kind of identification classification method and system to the speech data involving contradictory dispute, it is related to data identification field.The steps of the method include: after the feature extraction of received speech data, the basic features capable of expressing the speech content of current speech data are obtained, the mel frequency cepstrum coefficient information in all basic features is determined;By the sentiment analysis model obtained by training according to mel frequency cepstrum coefficient information, the mel frequency cepstrum coefficient information currently determined is classified into contradictory dispute event;Sentiment analysis model includes simple sentiment model based on bag-of-words model and complex sentiment model based on deep learning model.The contradictory dispute event in the present application can be more accurately identified in speech, and intelligent classification is carried out;Through sentiment analysis model based on deep learning model, semantic information can be effectively extracted, so as to improve the classification accuracy.

Description

Technical Field

[0001] This invention relates to the field of data recognition, specifically to a method and system for recognizing and classifying voice data involving conflicts and disputes. Background Technology

[0002] Significant progress has been made in the field of speech recognition technology in recent years, with many algorithms based on deep learning and natural language processing being widely applied in speech recognition systems. These technologies have enabled major advancements in speech recognition for areas such as voice search, voice assistants, and voice translation.

[0003] Currently, existing speech recognition technology performs semantic recognition through grammatical and statistical models based on preset rules. Its shortcomings include:

[0004] (1) Limitations of semantic understanding: grammatical and statistical models based on preset rules are difficult to capture complex semantic information in speech; in classifying disputes using the above techniques, there are still limitations in understanding the context and semantic information in speech, resulting in low classification accuracy.

[0005] (2) Insufficient handling of diversity: In classifying voice data of conflicts and disputes, due to the diversity of people's voice expression and speaking habits, the above-mentioned grammatical and statistical models based on preset rules have limited adaptability to the voice features and language habits of different speakers, thus reducing the stability of classification.

[0006] (3) Real-time and efficiency issues: Some speech recognition systems experience delays when processing large-scale speech data, failing to meet the real-time requirements for automatic classification of disputes. Furthermore, the computational complexity of the aforementioned methods is high when processing large-scale speech data, resulting in low system efficiency and failing to meet the needs of large-scale real-time speech classification. Summary of the Invention

[0007] In view of the deficiencies in the existing technology, the technical problem solved by the present invention is: how to improve the accuracy of classification when recognizing and classifying voice data involving conflicts and disputes.

[0008] To achieve the above objectives, in a first aspect, embodiments of this application provide a method for recognizing and classifying voice data involving conflicts and disputes, comprising the following steps: after extracting features from the received voice data, obtaining basic features that can express the voice content of the current voice data, and determining the Mel frequency cepstral coefficient information among all basic features; classifying the conflict and dispute events based on the currently determined Mel frequency cepstral coefficient information using a sentiment analysis model trained according to the Mel frequency cepstral coefficient information; the sentiment analysis model includes a simple sentiment model based on the bag-of-words model and a complex sentiment model based on a deep learning model.

[0009] In conjunction with the first aspect, in one implementation, the training process of the sentiment analysis model includes: constructing Mel-frequency cepstral coefficient information samples of several speech data corresponding to different speech features and language habits; extracting training features from each Mel-frequency cepstral coefficient information sample as sample data; using 75%–85% of the sample data as a training set, 10%–20% of the sample data as a validation set, and 3%–10% of the sample data as a test set; adjusting the model parameters by selecting a cross-entropy loss function during training on the training set; and optimizing the model parameters based on performance evaluation results during validation on the validation set.

[0010] In conjunction with the first aspect, in one implementation, the process of extracting features from the received voice data includes: segmenting the voice data according to a specified data length, and then extracting features from each segment of voice data according to preset feature extraction conditions.

[0011] In conjunction with the first aspect, in one implementation, the process of segmenting the voice data according to a specified data length includes: segmenting the voice data after denoising.

[0012] In conjunction with the first aspect, in one implementation, the process of classifying conflict events based on the currently determined Mel frequency cepstral coefficient information using a sentiment analysis model trained on Mel frequency cepstral coefficient information includes: defining the task type of Mel frequency cepstral coefficient information with a data length below a specified threshold as a simple sentiment analysis task, and defining the task type of Mel frequency cepstral coefficient information with a data length greater than a specified threshold as a complex sentiment analysis task; processing the Mel frequency cepstral coefficient information of the task type simple sentiment analysis task using a sentiment analysis model based on a bag-of-words model; and processing the Mel frequency cepstral coefficient information of the task type complex sentiment analysis task using a sentiment analysis model based on a deep learning model.

[0013] In conjunction with the first aspect, in one implementation, the deep learning model employs an RNN model.

[0014] In conjunction with the first aspect, in one implementation, after classifying the conflict and dispute events, the method further includes the following steps: associating the classification results of the conflict and dispute events with the corresponding voice data and then storing them.

[0015] In conjunction with the first aspect, in one implementation, the classification result of the conflict / dispute event includes the dispute type, a timestamp of occurrence determined based on the voice data reception time, and a user identifier determined based on the user's identity.

[0016] In conjunction with the first aspect, in one implementation method, after classifying conflict events, the method further includes the following steps: generating display information for users to view based on the conflict event classification results.

[0017] Secondly, embodiments of this application provide a recognition and classification system for voice data involving conflicts and disputes, which is used to implement the method provided in the first aspect.

[0018] Compared with the prior art, the advantages of the present invention are as follows:

[0019] (1) This invention classifies conflict and dispute events by using a sentiment analysis model trained based on Mel frequency cepstral coefficient information. Mel frequency is proposed based on the characteristics of human hearing and has a non-linear correspondence with Hz frequency. Mel frequency cepstral coefficients (MFCC) are the Hz spectral features calculated using this relationship. Therefore, compared with the semantic recognition of existing technologies using grammatical and statistical models based on preset rules, this invention can more accurately identify conflict and dispute events in speech and perform intelligent classification. The sentiment analysis model obtained based on a deep learning model can effectively extract semantic information, thereby improving classification accuracy.

[0020] (2) The emotion analysis model of the present invention fully considers the diversity of speech data during training. By introducing diverse datasets for training and model optimization, it adopts flexible feature extraction and model training methods, which can adapt to speech input from different speakers and in different environments. It has strong generalization ability and improves the stability of the system.

[0021] Meanwhile, the sentiment analysis model trained by this invention can directly classify conflict events based on Mel frequency cepstral coefficient information, simplifying computational complexity and achieving low latency while ensuring classification accuracy. This meets the real-time requirements of conflict applications, which is of great significance for application scenarios that require rapid response. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating the method for recognizing and classifying voice data involving conflicts and disputes in an embodiment of the present invention.

[0024] Figure 2This is a modular schematic diagram of a voice data recognition and classification system involving conflicts and disputes in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0027] The method for recognizing and classifying voice data involving conflicts and disputes in this embodiment of the invention includes the following steps: after extracting features from the received voice data, basic features that can express the voice content of the current voice data are obtained, and the Mel frequency cepstral coefficient information among all basic features is determined; the conflict and dispute event is classified (e.g., family events, loan events, etc.) based on the currently determined Mel frequency cepstral coefficient information by using a sentiment analysis model trained based on the Mel frequency cepstral coefficient information; the sentiment analysis model includes a simple sentiment model based on the bag-of-words model and a complex sentiment model based on a deep learning model.

[0028] Therefore, this invention classifies conflicting events by using a sentiment analysis model trained on Mel-frequency cepstral coefficient information. Mel-frequency, proposed based on human auditory characteristics, has a non-linear relationship with Hz frequencies. Mel-frequency cepstral coefficients (MFCCs) are Hz spectral features calculated using this relationship. Thus, compared to existing technologies that use grammatical and statistical models based on preset rules for semantic recognition, this invention can more accurately identify conflicting events in speech and perform intelligent classification. Furthermore, the sentiment analysis model based on deep learning effectively extracts semantic information, thereby improving classification accuracy.

[0029] Preferably, the training process of the sentiment analysis model in this method includes:

[0030] We construct Mel-frequency cepstral coefficient (CFC) information samples for several speech data points corresponding to different speech features and language habits. From each CFC sample, we extract training features as sample data (if using a bag-of-words model, the extraction method is text-to-numerical feature conversion; if using a deep learning model, the extraction method is word embedding). We use 75%–85% of the sample data as the training set, 10%–20% as the validation set, and 3%–10% as the test set. During training on the training set, we adjust the model parameters by selecting a cross-entropy loss function; during validation on the validation set, we optimize the model parameters based on performance evaluation results.

[0031] It should be noted that, based on the disclosure of the above training process, those skilled in the art can derive specific training methods from their knowledge of the bag-of-words model and deep learning models.

[0032] Therefore, it can be seen that the sentiment analysis model of the present invention fully considers the diversity of speech data during training. By introducing diverse datasets for training and model optimization, it adopts flexible feature extraction and model training methods, which can adapt to speech input from different speakers and in different environments, has strong generalization ability, and improves the stability of the system.

[0033] Meanwhile, the sentiment analysis model trained by this invention can directly classify conflict events based on Mel frequency cepstral coefficient information, simplifying computational complexity and achieving low latency while ensuring classification accuracy. This meets the real-time requirements of conflict applications, which is of great significance for application scenarios that require rapid response.

[0034] Preferably, the process of feature extraction of the received speech data in this method includes: segmenting the speech data according to the specified data length (set according to different processing capabilities), and then extracting features from each segment of speech data according to preset feature extraction conditions (such as preset keywords to be extracted).

[0035] Therefore, it can be seen that the present invention narrows the feature extraction range by extracting features in segments, thereby improving the extraction accuracy and subsequent classification accuracy.

[0036] Furthermore, the above process of segmenting speech data according to the specified data length includes: segmenting the speech data after denoising.

[0037] Preferably, the process of classifying conflict events based on the currently determined Mel-frequency cepstral coefficient information using the sentiment analysis model trained on the Mel-frequency cepstral coefficient information in the above method includes: defining the task type of Mel-frequency cepstral coefficient information with a data length below a specified threshold (the specific value depends on the computational power of the bag-of-words model and the deep learning model used) as a simple sentiment analysis task, and defining the task type of Mel-frequency cepstral coefficient information with a data length greater than the specified threshold as a complex sentiment analysis task. The sentiment analysis model based on the bag-of-words model processes Mel-frequency cepstral coefficient information for simple sentiment analysis tasks; the sentiment analysis model based on the deep learning model processes Mel-frequency cepstral coefficient information for complex sentiment analysis tasks.

[0038] Furthermore, the aforementioned deep learning model employs an RNN (Recurrent Neural Network) model.

[0039] Preferably, after classifying the conflict and dispute events, the method further includes the following steps: associating the classification results of the conflict and dispute events with the corresponding voice data and storing them, so as to ensure that they can be quickly retrieved when needed.

[0040] Furthermore, the above-mentioned classification results of disputes include the type of dispute, the timestamp of occurrence determined based on the time of voice data reception, and the user identifier determined based on the user's identity.

[0041] Furthermore, after classifying the conflict events, the method also includes the following steps: generating display information for users to view based on the classification results of the conflict events.

[0042] The method of the present invention will be specifically illustrated below through an embodiment.

[0043] See Figure 1 As shown, the steps in this embodiment include:

[0044] S1: Perform feature extraction on the received voice data. The specific process is as follows: After receiving the voice data, first denoise the voice data, then segment the denoised voice data according to the specified data length, and then extract features from each segment of voice data according to the preset feature extraction conditions to obtain the basic features that can express the voice content of the current voice data.

[0045] S2: Determine the Mel frequency cepstral coefficient information in all basic features.

[0046] S3: By using the sentiment analysis model trained based on the Mel frequency cepstral coefficient information, the Mel frequency cepstral coefficient information determined in S2 is used to classify conflict events, and the classification results of conflict events are obtained (including dispute type, occurrence timestamp and user identifier).

[0047] The selection process for sentiment analysis models in S3 includes: processing Mel-frequency cepstral coefficient information for simple sentiment analysis tasks using a sentiment analysis model based on a bag-of-words model; and processing Mel-frequency cepstral coefficient information for complex sentiment analysis tasks using a sentiment analysis model based on an RNN model.

[0048] The training process of the sentiment analysis model in S3 includes:

[0049] S301: Construct several speech data corresponding to different speech features and language habits by using historical data or by targeted self-generation.

[0050] S302: After standardizing the speech data (including removing punctuation marks, special symbols, and converting text to a specified format), determine the Mel frequency cepstral coefficient information samples in the speech data.

[0051] S303: Extract training features from each Mel frequency cepstral coefficient information sample as sample data (the bag-of-words model extracts text by converting it into numerical features, while the RNN model extracts them by embedding words).

[0052] S304: 80% of the sample data is used as the training set, 15% as the validation set, and 5% as the test set. During training on the training set, model parameters are adjusted by selecting a cross-entropy loss function to improve model performance and generalization ability. After training the model on the training set, it is validated on the validation set. During validation, the model parameters are optimized based on the performance evaluation results. The main performance metrics include accuracy, precision, recall, and F1 score.

[0053] S4: Store the classification results of conflict events and their corresponding voice data.

[0054] S5: Generate display information for users to view based on the classification results of conflict and dispute events.

[0055] The process of executing S1 to S5 also includes the following steps: real-time monitoring of the sentiment analysis model to ensure the performance and stability of the model in the production environment, and retraining or adjusting it as needed.

[0056] This invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the above-described method. It should be noted that the storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, ROM (Read-Only Memory), RAM (Random Access Memory), a magnetic disk, or an optical disk.

[0057] This invention also provides a system for recognizing and classifying voice data involving conflicts and disputes, which is used to implement the above method.

[0058] See Figure 2 As shown, the system specifically includes: a voice input module, a voice signal processing module, an automatic conflict classification module, a database module, and a user interface module.

[0059] The voice input module is used to receive user voice data input and transmit the voice data to the voice signal processing module. The voice input module employs a high-sensitivity microphone sensor, capable of capturing user voice input in different environments.

[0060] The speech signal processing module is used to: receive the speech signal from the speech input module, extract features from the received speech data according to the above method, obtain the basic features that can express the speech content of the current speech data, and determine the Mel frequency cepstral coefficient information among all basic features.

[0061] The automatic conflict and dispute classification module is used to classify conflict and dispute events based on the Mel frequency cepstral coefficient information determined by the speech signal processing module using the sentiment analysis model trained according to the above method, and then obtain the conflict and dispute event classification results.

[0062] The database module is used to store the classification results of conflict events and the corresponding voice data according to the above method.

[0063] The user interface module is used to: provide a user-friendly interface that can display the results of conflict classification and manage user personalization settings; the user interface module supports multiple platforms, including PC and mobile, to meet the user's needs on different devices.

[0064] An electronic device includes a memory and a processor, wherein the memory stores a computer program that runs on the processor, and the processor executes the computer program to implement the method described above.

[0065] Those skilled in the art will understand that all or some of the steps, systems, and apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software can be distributed on a computer-readable storage medium, which may include computer-readable storage media (or non-transitory media) and communication media (or transient media).

[0066] As is known to those skilled in the art, the term computer-readable storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer-readable storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0067] For example, the computer-readable storage medium may be an internal storage unit of the electronic device described in the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., provided on the electronic device.

[0068] The above are merely specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the scope of the claims.

Claims

1. A method for recognizing and classifying voice data involving conflicts and disputes, characterized in that, The method includes the following steps: after extracting features from the received speech data, the basic features that can express the speech content of the current speech data are obtained, and the Mel frequency cepstral coefficient information in all basic features is determined; The sentiment analysis models are trained based on Mel frequency cepstral coefficient information. These models include a simple sentiment model based on the bag-of-words model and a complex sentiment model based on deep learning. The task type of Mel frequency cepstral coefficient information with a data length below a specified threshold is defined as a simple sentiment analysis task, and the task type of Mel frequency cepstral coefficient information with a data length greater than a specified threshold is defined as a complex sentiment analysis task. The Mel frequency cepstral coefficient information of the task type of simple sentiment analysis task is processed by a sentiment analysis model based on the bag-of-words model, and the Mel frequency cepstral coefficient information of the task type of complex sentiment analysis task is processed by a sentiment analysis model based on a deep learning model.

2. The method for recognizing and classifying voice data involving conflicts and disputes as described in claim 1, characterized in that, The training process of the sentiment analysis model includes: constructing Mel-frequency cepstral coefficient information samples of several speech data corresponding to different speech features and language habits; extracting training features from each Mel-frequency cepstral coefficient information sample as sample data; using 75%~85% of the sample data as the training set, 10%~20% as the validation set, and 3%~10% as the test set; adjusting the model parameters by selecting a cross-entropy loss function during training on the training set; and optimizing the model parameters based on the performance evaluation results during validation on the validation set.

3. The method for recognizing and classifying voice data involving conflicts and disputes as described in claim 1, characterized in that, The process of extracting features from the received voice data includes: segmenting the voice data according to a specified data length, and then extracting features from each segment of voice data according to preset feature extraction conditions.

4. The method for recognizing and classifying voice data involving conflicts and disputes as described in claim 3, characterized in that, The process of segmenting speech data according to a specified data length includes: segmenting the speech data after denoising.

5. The method for recognizing and classifying voice data involving conflicts and disputes as described in claim 1, characterized in that: The deep learning model uses an RNN model.

6. The method for recognizing and classifying voice data involving conflicts and disputes as described in any one of claims 1 to 5, characterized in that, After classifying conflict events, the method also includes the following steps: associating the classification results of conflict events with the corresponding voice data and storing them.

7. The method for recognizing and classifying voice data involving conflicts and disputes as described in claim 6, characterized in that: The classification results of the conflict and dispute events include the type of dispute, the timestamp of occurrence determined based on the voice data reception time, and the user identifier determined based on the user's identity.

8. The method for recognizing and classifying voice data involving conflicts and disputes as described in claim 7, characterized in that, After classifying the conflict events, the method also includes the following steps: generating display information for users to view based on the classification results of the conflict events.

9. A system for recognizing and classifying voice data involving conflicts and disputes, characterized in that: The system is used to implement the method described in any one of claims 1 to 8.

Images