Voice call group call capture processing method and system

Abstract

The invention discloses a voice call group call capture processing method and system, which can ensure the security and privacy protection of call content and avoid sensitive information leakage by encrypting and storing recording data and text records, and can effectively improve the call quality by using a voice recognition technology and a natural language processing technology. The key words and key information in the call can be automatically extracted, and support is provided for subsequent manual docking or automatic processing.

Description

Technical Field

[0001] The present invention relates to the field of electromechanical equipment, and in particular to a method and system for capturing and processing a voice call group call. Background Art

[0002] Voice calls are a form of communication that uses voice and transmission media. Common ones include landline calls, mobile phone calls, intercom calls, and voice chats on the Internet. It can be divided into two forms: consuming traffic and consuming phone charges. In the last century, voice calls were limited to phones, mobile phones, and intercoms. Now, it has gradually developed into a common means of communication. Not only through phones, but also chat software and even mobile phone software can realize voice calls. At present, for group calls, due to the limited number of human workers, it takes a long time to connect and the processing time is slow, which is not conducive to solving the corresponding problems. With the development of Internet technology, the demand for multi-person online voice calls is increasing. However, the existing voice call system has problems such as low recording quality and difficulty in extracting key information.

[0003] In summary, a method and system for capturing and processing a voice call group call is needed to address the deficiencies in the prior art. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method and system for capturing and processing a voice call group call, aiming to solve the above problems.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for capturing and processing a voice call group call, comprising the following steps: Step S1: establishing a multi-party voice call connection through a voice call system according to the multi-user voice call requirements; Step S2: Record the call voice and upload the recorded data to cloud storage; Step S3: using a speech recognition model to convert the recorded data in real time to generate a text record; Step S4: Analyze the text records in real time to extract keywords of key information; Step S5: Analyze the number and keyword information and determine whether it is the target group. If so, switch to manual connection according to the conversion words. Otherwise, extract the keywords of key information from the voice call system recording and make a real-time judgment. Step S6: The key words of the key information are associated with the original recording data and stored to form a complete record.

[0006] Optionally, the speech recognition model is established in step S3 in the following manner: Step S31: data preparation, collecting a large amount of voice data from multiple sources; Step S32: data preprocessing, converting the audio file to a uniform sampling rate and format, and removing background noise; Step S33: feature extraction, performing feature extraction on the pre-processed audio file and extracting spectrum information; Step S34: extracting text features, converting the audio file into text, and converting the text into a sequence of characters or words; Step S35: training the speech recognition model according to the extracted audio features and text sequence, and enhancing the training data by adding noise or changing the speed; Step S36: performing sequence-to-sequence modeling on the speech recognition model through a loss function, and optimizing the speech recognition model through an optimization algorithm; Step S37: Test the speech recognition model through an independent test set until the speech recognition model is verified.

[0007] Optionally, step S4 is implemented in the following manner: Step S41: Through the microservice architecture, the processing logic is divided into multiple microservices, and the message queue technology is used to transmit the text records to the processing system in real time; Step S42: batch processing multiple text records and using an asynchronous processing model to increase processing speed; Step S43: Encrypt and store the text records and keyword results.

[0008] Optionally, in step S5, determining whether it is the target group is specifically performed in the following manner: Step S51: First, identify the number and determine whether it is in the restricted list; Step S52: judging based on the tone, emotion and patience of the caller, and judging whether it is the target group based on the number of times the keywords and context of the text appear; Step S53: Conduct a multi-party voice call through the voice call system, and estimate the time required to end the call based on the needs of the callers.

[0009] Optionally, the S52 target group triggering condition is: Keyword matching: the extracted keywords are compared with the keyword standards of the preset target group. If they meet the keyword standards, manual docking is triggered; Model prediction: prediction is made through the target group prediction model of the voice call system. If the prediction is the target group, manual docking is triggered.

[0010] Optionally, the multi-party voice call in step S53 is performed in the following manner: Step A1: Monitor the problems of multiple parties in real time through the voice call system, and display the problems in the form of text; Step A2: Determine whether the question is a routine question. If so, the voice call system extracts the question response script and simulates a human voice to respond. Otherwise, the voice system selects the caller who asked an unconventional question and responds with an artificial voice.

[0011] A voice call group call capture and processing system adopts a voice call group call capture and processing method, including a multi-party call module, a recording module, a voice-to-text module, a key information extraction module and a data storage module; Multi-party call module, used for voice calls among multiple users and establishing multi-party voice call connections; The recording module is used to record voice calls in real time and upload the recorded data to the cloud for storage; The speech-to-text module is used to convert the recorded data in real time using speech recognition to generate text records; Keyword extraction module, which is used to analyze text records using natural language processing to extract key information; The data storage module is used to associate key information with the original recording data to form a complete record.

[0012] Optionally, the system further includes a user interaction module and an identity authentication module. User interaction module, which provides a tool interface for users to view, edit and download records; The authentication module is used for user authentication so that only authorized users can access the records corresponding to the user.

[0013] Optionally, the system also includes a data encryption module for encrypting the recording data uploaded to the cloud and generating text records.

[0014] Optionally, applications of the keyword extraction module include but are not limited to sentiment analysis, entity recognition and keyword extraction.

[0015] Beneficial effects of the present invention: 1. In the present invention, by encrypting and storing the recording data and text records, the security and privacy of the call content are guaranteed, and the leakage of sensitive information is avoided. By using speech recognition technology and natural language processing technology, the keywords and key information in the call can be automatically extracted, which provides support for subsequent manual docking or automated processing; 2. In the present invention, by means of number recognition, keyword comparison and model prediction, the target group can be more accurately determined, thereby improving the pertinence and effectiveness of marketing activities or customer service: for non-target groups or routine questions, the system can directly process or reply, reducing the workload of manual customer service; for complex questions or target groups, they can be transferred to manual customer service in a timely manner, and human resources can be reasonably allocated; 3. In the present invention, key information is associated with the original recording data and stored, which not only forms a complete call record, but also provides a rich and accurate data basis for subsequent data mining and business analysis. The user interaction module allows users to easily view, edit and download call records, improving the convenience and satisfaction of user operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention is a flow chart of a method.

[0017] Figure 2 A system structure diagram of the present invention. DETAILED DESCRIPTION

[0018] In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] like Figure 1 As shown, a method for capturing and processing a voice call group call includes the following steps: Step S1: establishing a multi-party voice call connection through a voice call system according to the multi-user voice call requirements; Step S2: Record the call voice and upload the recorded data to cloud storage; Step S3: using a speech recognition model to convert the recorded data in real time to generate a text record; Step S4: Analyze the text records in real time to extract keywords of key information; Step S5: Analyze the number and keyword information and determine whether it is the target group. If so, switch to manual connection according to the conversion words. Otherwise, extract the keywords of key information from the voice call system recording and make a real-time judgment. Step S6: The key words of the key information are associated with the original recording data and stored to form a complete record.

[0020] The speech recognition model is established in step S3 in the following manner: Step S31: data preparation, collecting a large amount of voice data from multiple sources; Step S32: data preprocessing, converting the audio file to a uniform sampling rate and format, and removing background noise; Step S33: feature extraction, performing feature extraction on the pre-processed audio file and extracting spectrum information; Step S34: extracting text features, converting the audio file into text, and converting the text into a sequence of characters or words; Step S35: training the speech recognition model according to the extracted audio features and text sequence, and enhancing the training data by adding noise or changing the speed; Step S36: performing sequence-to-sequence modeling on the speech recognition model through a loss function, and optimizing the speech recognition model through an optimization algorithm; Step S37: Test the speech recognition model through an independent test set until the speech recognition model is verified.

[0021] Step S4 is implemented in the following manner: Step S41: Through the microservice architecture, the processing logic is divided into multiple microservices, and the message queue technology is used to transmit the text records to the processing system in real time; Step S42: batch processing multiple text records and using an asynchronous processing model to increase processing speed; Step S43: Encrypt and store the text records and keyword results.

[0022] In step S5, it is determined whether the target group is selected, specifically in the following manner: Step S51: First, identify the number and determine whether it is in the restricted list; Step S52: judging based on the tone, emotion and patience of the caller, and judging whether it is the target group based on the number of times the keywords and context of the text appear; Step S53: Conduct a multi-party voice call through the voice call system, and estimate the time required to end the call based on the needs of the callers.

[0023] S52 target group trigger conditions are: Keyword matching: the extracted keywords are compared with the keyword standards of the preset target group. If they meet the keyword standards, manual docking is triggered; Model prediction: prediction is made through the target group prediction model of the voice call system. If the prediction is the target group, manual docking is triggered.

[0024] Step S53: Multi-party voice call, in the following way: Step A1: Monitor the problems of multiple parties in real time through the voice call system, and display the problems in the form of text; Step A2: Determine whether the question is a routine question. If so, the voice call system extracts the question response script and simulates a human voice to respond. Otherwise, the voice system selects the caller who asked an unconventional question and responds with an artificial voice.

[0025] The voice call system allows multiple users to join voice calls at the same time, supports the recording and playback of voice calls, provides real-time voice-to-text services, and can identify and record key information of calls.

[0026] Front-end applications use modern front-end frameworks such as React or Vue to build user interfaces, provide friendly user experiences, implement functions such as user login and voice call control panel, and support WebRTC technology to achieve real-time communication between browsers.

[0027] The backend service uses the SIP protocol or WebRTC API to manage call establishment, maintenance, and termination, uses cloud storage services to automatically record and save call records, integrates speech recognition APIs to achieve real-time speech-to-text conversion, and is developed based on NLP technology to extract key information from the text of speech transcription.

[0028] Data Collection Multi-source data: Collect a large amount of speech data from multiple sources, including public datasets (such as LibriSpeech, Common Voice), internal recording data, etc.

[0029] Diversity and representativeness: Ensure that the data covers different accents, speaking speeds, background noise, etc. to improve the generalization ability of the model.

[0030] Data preprocessing Audio Normalization: Convert all audio files to a uniform sampling rate (such as 16kHz) and format (such as WAV).

[0031] Noise Removal: Uses signal processing technology to remove background noise and improve audio quality.

[0032] Annotation: Manually annotate audio data and generate corresponding text transcriptions.

[0033] Audio Features Mel-frequency cepstral coefficient (MFCC): extracts the spectral features of audio and is often used in ASR tasks.

[0034] Filterbank: Uses multiple bandpass filters to extract spectral information from audio.

[0035] Acoustic features: including energy, zero-crossing rate, etc., which help capture the physical characteristics of audio.

[0036] Text Features Character-level or word-level representation: Convert text into a sequence of characters or words for training models.

[0037] Model Architecture Convolutional Neural Network (CNN): used to extract local features of audio.

[0038] Recurrent Neural Networks (RNNs): specifically Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU), used to capture dependencies in time series.

[0039] Transformer: Based on the self-attention mechanism, it is suitable for modeling long sequences.

[0040] Training process Data enhancement: Enhance the training data by adding noise, changing the speed, etc. to improve the robustness of the model.

[0041] Loss function: Use the CTC (Connectionist Temporal Classification) loss function, which is suitable for sequence-to-sequence modeling.

[0042] Optimization algorithm: Use optimization algorithms such as Adam and SGD, and adjust hyperparameters such as learning rate and batch size.

[0043] Regularization: Use techniques such as Dropout and L2 regularization to prevent overfitting.

[0044] decoding Greedy search: select the characters or words with the highest probability one by one.

[0045] Beam Search: retain multiple possible solutions and ultimately select the most likely sequence.

[0046] Language model fusion N-gram model: predicts the probability of the next word based on the previous n words.

[0047] Neural Language Models: Use RNN or Transformer to generate more complex language models.

[0048] Joint decoding: Combine the acoustic model and language model to generate the final transcription result.

[0049] Evaluation Metrics Word Error Rate (WER): measures the difference between the transcription result and the real text.

[0050] Character Error Rate (CER): used for character-level evaluation.

[0051] Test Set Diversity: Make sure your test set contains different accents, speaking speeds, background noise, etc.

[0052] Independence: The test set should have no overlap with the training set to ensure fair evaluation.

[0053] Real-time processing Stream Processing Message queue: Use message queue technologies such as Kafka and RabbitMQ to deliver text records to the processing system in real time.

[0054] Microservice architecture: Split the processing logic into multiple microservices to improve the scalability and reliability of the system.

[0055] Performance Optimization Batch processing: Process multiple text records in batches to reduce I / O overhead.

[0056] Asynchronous processing: Use asynchronous programming model to improve processing speed.

[0057] like Figure 2 As shown, a voice call group call capture and processing system adopts a voice call group call capture and processing method, including a multi-party call module, a recording module, a voice-to-text module, a key information extraction module and a data storage module; Multi-party call module, used for voice calls among multiple users and establishing multi-party voice call connections; The recording module is used to record voice calls in real time and upload the recorded data to the cloud for storage; The speech-to-text module is used to convert the recorded data in real time using speech recognition to generate text records; Keyword extraction module, which is used to analyze text records using natural language processing to extract key information; The data storage module is used to associate key information with the original recording data to form a complete record.

[0058] The system also includes a user interaction module and an identity authentication module. User interaction module, which provides a tool interface for users to view, edit and download records; The identity authentication module is used for user identity authentication so that only authorized users can access the records corresponding to the user. The system also includes a data encryption module for encrypting the recording data uploaded to the cloud and generating text records. The application of the keyword extraction module includes but is not limited to sentiment analysis, entity recognition and keyword extraction.

[0059] The present invention ensures the security and privacy of call content by encrypting and storing recording data and text records, avoiding leakage of sensitive information, and automatically extracts keywords and key information in calls by using speech recognition technology and natural language processing technology, providing support for subsequent manual docking or automated processing; Through number recognition, keyword comparison and model prediction, the target group can be determined more accurately, thereby improving the pertinence and effectiveness of marketing activities or customer service: For non-target groups or routine questions, the system can directly handle or reply, reducing the workload of manual customer service; for complex questions or target groups, they can be transferred to manual customer service in a timely manner, rationally allocating human resources; Storing key information in association with the original recording data not only forms a complete call record, but also provides a rich and accurate data foundation for subsequent data mining and business analysis. The user interaction module allows users to easily view, edit and download call records, improving user convenience and satisfaction.

[0060] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, equivalent substitution or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for capturing and processing a voice call group call, characterized in that: The following steps are involved: Step S1: establishing a multi-party voice call connection through a voice call system according to the multi-user voice call requirements; Step S2: Record the call voice and upload the recorded data to cloud storage; Step S3: using a speech recognition model to convert the recorded data in real time to generate a text record; Step S4: Analyze the text records in real time to extract keywords of key information; Step S5: Analyze the number and keyword information and determine whether it is the target group. If so, switch to manual connection according to the conversion words. Otherwise, extract the keywords of key information from the voice call system recording and make a real-time judgment. Step S6: The key words of the key information are associated with the original recording data and stored to form a complete record.

2. The method for capturing and processing a voice call group call according to claim 1, characterized in that: The speech recognition model is established in step S3 in the following manner: Step S31: data preparation, collecting a large amount of voice data from multiple sources; Step S32: data preprocessing, converting the audio file to a uniform sampling rate and format, and removing background noise; Step S33: feature extraction, performing feature extraction on the pre-processed audio file and extracting spectrum information; Step S34: extracting text features, converting the audio file into text, and converting the text into a sequence of characters or words; Step S35: training the speech recognition model according to the extracted audio features and text sequence, and enhancing the training data by adding noise or changing the speed; Step S36: performing sequence-to-sequence modeling on the speech recognition model through a loss function, and optimizing the speech recognition model through an optimization algorithm; Step S37: Test the speech recognition model through an independent test set until the speech recognition model is verified.

3. The method for capturing and processing a voice call group call according to claim 1, characterized in that: The step S4 is implemented in the following manner: Step S41: Through the microservice architecture, the processing logic is divided into multiple microservices, and the message queue technology is used to transmit the text records to the processing system in real time; Step S42: batch processing multiple text records and using an asynchronous processing model to increase processing speed; Step S43: Encrypt and store the text records and keyword results.

4. The method for capturing and processing a voice call group call according to claim 1, characterized in that: In step S5, whether the target group is determined is determined by: Step S51: First, identify the number and determine whether it is in the restricted list; Step S52: judging based on the tone, emotion and patience of the caller, and judging whether it is the target group based on the number of times the keywords and context of the text appear; Step S53: Conduct a multi-party voice call through the voice call system, and estimate the time required to end the call based on the needs of the callers.

5. The method for capturing and processing a voice call group call according to claim 4, characterized in that: The S52 target group triggering conditions are: Keyword matching: the extracted keywords are compared with the keyword standards of the preset target group. If they meet the keyword standards, manual docking is triggered; Model prediction: prediction is made through the target group prediction model of the voice call system. If the prediction is the target group, manual docking is triggered.

6. The method for capturing and processing a voice call group call according to claim 4, characterized in that: The multi-party voice call in step S53 is performed in the following manner: Step A1: Monitor the problems of multiple parties in real time through the voice call system, and display the problems in the form of text; Step A2: Determine whether the question is a routine question. If so, the voice call system extracts the question response script and simulates a human voice to respond. Otherwise, the voice system selects the caller who asked an unconventional question and responds with an artificial voice.

7. A voice call group call capture and processing system, using the voice call group call capture and processing method as claimed in any one of claims 1 to 6, characterized in that: It includes a multi-party call module, a recording module, a speech-to-text module, a key information extraction module and a data storage module; Multi-party call module, used for voice calls among multiple users and establishing multi-party voice call connections; The recording module is used to record voice calls in real time and upload the recorded data to the cloud for storage; The speech-to-text module is used to convert the recorded data in real time using speech recognition to generate text records; Keyword extraction module, which is used to analyze text records using natural language processing to extract key information; The data storage module is used to associate key information with the original recording data to form a complete record.

8. The voice call group call capture and processing system according to claim 7, characterized in that: The system also includes a user interaction module and an identity authentication module. User interaction module, which provides a tool interface for users to view, edit and download records; The authentication module is used for user authentication so that only authorized users can access the records corresponding to the user.

9. The voice call group call capture and processing system according to claim 7, characterized in that: The system also includes a data encryption module for encrypting the recording data uploaded to the cloud and generating text records.

10. The voice call group call capture and processing system according to claim 7, characterized in that: Applications of the keyword extraction module include but are not limited to sentiment analysis, entity recognition and keyword extraction.

Images