Business data voice processing method and system based on artificial intelligence

Abstract

The invention provides a service data voice processing method and system based on artificial intelligence, and relates to the field. The service data voice processing method based on artificial intelligence comprises multi-mode voice preprocessing, domain enhancement voice recognition, knowledge graph driven semantic analysis, a dynamic dialogue management mechanism and a closed-loop feedback optimization system. According to the method, high-precision recognition, semantic understanding and structured processing of the service voice data are realized by fusing the domain knowledge graph, the dynamic model optimization and the context sensing technology, and the service data processing efficiency and accuracy are improved; secondly, the system comprises a voice acquisition and preprocessing module, an AI voice processing module, a business data processing module, a model optimization module and a data interface module, the efficiency and accuracy of business data voice processing can be remarkably improved, and the system is suitable for various professional scenes such as finance, customer service and medical treatment.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based business data voice processing method and system. Background Art

[0002] In modern information-based business scenarios, voice interaction has become an important way to input data. However, traditional business data voice processing solutions have the following shortcomings: 1. Speech recognition accuracy is significantly affected by business terminology, accents, and environmental noise. Keyword recognition errors are common, especially in professional fields (such as financial transactions and medical diagnosis). 2. The semantic understanding of speech content lacks business scenario specificity, making it difficult to convert unstructured speech into structured data that complies with business rules. 3. Unable to dynamically adapt to changes in business processes. When business rules are updated, the system needs to be redeployed, which lacks flexibility. 4. In multi-round voice interactions, the ability to process historical conversation context associations is weak, resulting in data processing gaps.

[0003] In existing technologies, some solutions achieve data processing through simple speech-to-text conversion combined with keyword matching. However, these solutions lack a deep understanding of business logic and cannot meet the needs of complex business scenarios. Therefore, there is an urgent need for a speech processing method and system that can efficiently and accurately process business data. Summary of the Invention

[0004] (1) Technical problems solved In response to the shortcomings of the existing technology, the present invention provides a business data voice processing method and system based on artificial intelligence. By integrating domain knowledge graphs, dynamic model optimization and context perception technology, it can achieve high-precision recognition, semantic understanding and structured processing of business voice data, improve the efficiency and accuracy of business data processing, and solve the problems raised by the above background technology.

[0005] (2) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: a business data voice processing method based on artificial intelligence, comprising the following steps: Step 1: Multimodal speech preprocessing: Speech signals are collected through a microphone array, and denoised and framed using the Mel-Frequency Cepstral Coefficients (MFCC) and Dynamic Time Warping (DTW) algorithms to extract multidimensional acoustic features including time domain, frequency domain, and energy distribution. Step 2: Domain-enhanced speech recognition: A pre-trained Transformer-based speech recognition model (ASR) is used. Business domain corpora (such as financial terminology dictionaries and medical diagnostic standards) are integrated during model training. Unsupervised pre-training (such as MPC masked predictive coding) is also used to improve the model's robustness in recognizing specialized terminology. Step 3: Knowledge graph-driven semantic analysis: Build a knowledge graph containing business entities, relationships, and rules, combine it with pre-trained language models such as BERT for semantic analysis, use slot filling technology to identify key business information (such as amount, time, customer ID), and use multi-hop reasoning to generate structured data frames; Step 4: Dynamic dialogue management mechanism: A multi-round conversation context model is built based on a finite state machine. State transitions are achieved through intent matching and menu triggering. The current semantic understanding results are corrected based on historical conversation records to ensure the coherence of the interaction logic. Step 5: Closed-loop feedback optimization system: An error correction model is built through reinforcement learning algorithms (such as PPO), and semantic understanding parameters are dynamically adjusted based on the verification results of the business rule engine (such as amount range verification and account format check). Federated learning technology is used to achieve cross-device model updates while protecting data privacy.

[0006] Preferably, the step 2 further comprises: Adversarial training and noise injection techniques are used to improve the robustness of the model in complex environments. Edge computing optimization strategies are combined to achieve lightweight model deployment and meet the real-time processing requirements of mobile devices. Through a multi-task learning framework, speech recognition, sentiment analysis, and speaker recognition are performed simultaneously to enhance the model's comprehensive understanding of speech data.

[0007] Preferably, the step three further comprises: Utilize entity linking technology to align identified business entities with external knowledge bases (such as industry standard databases), generate multi-dimensional semantic paths through relational reasoning, and improve parsing accuracy in complex business scenarios; Dynamically update the knowledge graph by combining user behavior data and historical interaction records to achieve adaptive expansion of business rules.

[0008] Preferably, the step 5 further comprises: The Kalman filter algorithm is introduced to dynamically estimate the uncertainty of business data, providing high-precision state feedback for reinforcement learning and improving the real-time and accuracy of error correction decisions. Differential privacy technology is used to desensitize model update parameters to ensure the security of user data during distributed training.

[0009] The present invention further discloses an artificial intelligence-based business data voice processing system, which is based on the artificial intelligence-based business data voice processing method described above and includes a voice acquisition and preprocessing module, an AI voice processing module, a business data processing module, a model optimization module, and a data interface module.

[0010] Preferably, the voice acquisition and preprocessing module is used to acquire voice signals and perform preprocessing such as noise reduction and framing, and includes a microphone array interface and a digital signal processor (DSP).

[0011] Preferably, the AI ​​voice processing module integrates an ASR model that incorporates business corpus to achieve voice-to-text conversion and complete business semantic parsing and slot filling.

[0012] Preferably, the business data processing module has a built-in business rule library to verify the legitimacy of structured data, and triggers a counter-question mechanism based on the rule verification results to supplement information, while converting the semantic analysis results into a data format that complies with business specifications.

[0013] Preferably, the model optimization module collects business processing results as model optimization data, and uses reinforcement learning and transfer learning algorithms to dynamically update AI model parameters, while automatically updating the domain knowledge graph according to business changes.

[0014] Preferably, the data interface module provides a data interaction interface with a business system (such as CRM, ERP) to implement the storage and call of structured data.

[0015] (3) Beneficial effects The present invention provides a method and system for processing business data voice based on artificial intelligence, which has the following beneficial effects: 1. This method achieves high-precision recognition, semantic understanding and structured processing of business voice data by integrating domain knowledge graph, dynamic model optimization and context-aware technology, thereby improving the efficiency and accuracy of business data processing.

[0016] 2. This system includes a voice acquisition and preprocessing module, an AI voice processing module, a business data processing module, a model optimization module, and a data interface module. It can significantly improve the efficiency and accuracy of business data voice processing and is suitable for various professional scenarios such as finance, customer service, and medical care. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the processing method of the present invention; Figure 2 Schematic diagram of the processing system of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] Example 1: like Figure 1 As shown, an embodiment of the present invention provides a method for processing business data voice based on artificial intelligence, comprising the following steps: Step 1: Multimodal speech preprocessing: Speech signals are collected through a microphone array, and denoised and framed using the Mel-Frequency Cepstral Coefficients (MFCC) and Dynamic Time Warping (DTW) algorithms to extract multidimensional acoustic features including time domain, frequency domain, and energy distribution. Step 2: Domain-enhanced speech recognition: A pre-trained Transformer-based speech recognition model (ASR) is used, integrating business domain corpora (such as financial terminology dictionaries and medical diagnostic standards) during the model training phase, and improving the model's robustness in recognizing professional terminology through unsupervised pre-training (such as MPC masked predictive coding). (1) Using adversarial training and noise injection techniques to improve the robustness of the model in complex environments, combined with edge computing optimization strategies to achieve lightweight model deployment and meet the real-time processing requirements of mobile devices; (2) Through a multi-task learning framework, speech recognition, sentiment analysis, and speaker recognition are performed simultaneously to enhance the model's comprehensive understanding of speech data.

[0020] Step 3: Knowledge graph-driven semantic analysis: Build a knowledge graph containing business entities, relationships, and rules, combine it with pre-trained language models such as BERT for semantic analysis, identify key business information (such as amount, time, customer ID) through slot filling technology, and use multi-hop reasoning to generate structured data frames. (1) Utilize entity linking technology to align identified business entities with external knowledge bases (such as industry standard databases), generate multi-dimensional semantic paths through relational reasoning, and improve parsing accuracy in complex business scenarios; (2) Dynamically update the knowledge graph by combining user behavior data and historical interaction records to achieve adaptive expansion of business rules.

[0021] Step 4: Dynamic dialogue management mechanism: A multi-round dialogue context model is built based on a finite state machine, state transfer is achieved through intent matching and menu triggering, and the current semantic understanding results are corrected in combination with historical dialogue records to ensure the coherence of the interaction logic.

[0022] Step 5: Closed-loop feedback optimization system: By building an error correction model through reinforcement learning algorithms (such as PPO), the semantic understanding parameters are dynamically adjusted based on the verification results of the business rule engine (such as amount range verification and account format check), and federated learning technology is used to achieve cross-device model updates while protecting data privacy. (1) Introducing the Kalman filter algorithm to dynamically estimate the uncertainty of business data, providing high-precision state feedback for reinforcement learning, and improving the real-time and accuracy of error correction decisions; (2) Model update parameters are desensitized through differential privacy technology to ensure the security of user data during distributed training.

[0023] like Figure 2 As shown, an embodiment of the present invention further provides an artificial intelligence-based business data voice processing system, based on the artificial intelligence-based business data voice processing method described above, including: The voice acquisition and preprocessing module is used to collect voice signals and perform noise reduction and framing preprocessing, including a microphone array interface and a digital signal processor (DSP); The AI ​​voice processing module integrates an ASR model with business corpus to achieve speech-to-text conversion, complete business semantic analysis and slot filling; The business data processing module has a built-in business rule library to verify the legitimacy of structured data. It also triggers a query mechanism based on the rule verification results to supplement information and converts the semantic analysis results into a data format that complies with business specifications. The model optimization module collects business processing results as model optimization data, and uses reinforcement learning and transfer learning algorithms to dynamically update AI model parameters, while automatically updating the domain knowledge graph based on business changes; The data interface module provides a data interaction interface with business systems (such as CRM and ERP) to realize the storage and call of structured data.

[0024] Example 2: This embodiment differs from the first embodiment in that it is an artificial intelligence-based service data voice processing system, including: (1) Voice acquisition and preprocessing module: Integrates a microphone array and a digital signal processor (DSP) to achieve noise reduction, framing, and multi-dimensional feature extraction of voice signals, and supports low-latency data transmission with edge devices; (2) AI voice processing module: Speech recognition unit: Deploys a Transformer ASR model that incorporates business corpus, supports unsupervised pre-training and multi-task learning, and outputs text sequences with confidence levels. Semantic Understanding Unit: This includes the BERT language model and domain knowledge graph, and generates structured business data through entity recognition, relationship extraction, and multi-hop reasoning. Dialogue Management Unit: This unit builds a context model based on a finite state machine and machine learning, supporting state transition logic for menu triggering, intent matching, and historical dialogue association. Multimodal fusion subunit: Integrates visual information (such as lip recognition) and speech signal processing modules, and improves recognition accuracy in noisy environments through a cross-modal attention mechanism; Sentiment analysis subunit: Builds a sentiment classification model based on convolutional neural networks (CNN) and bidirectional LSTM, outputs semantic analysis results containing sentiment tendencies, and assists business decision-making (3) Business data processing module: Rule engine: Built-in dynamic business rule library, supports legality verification such as amount verification and process sequence check, and outputs verification results with risk levels; Interactive error correction unit: Combines reinforcement learning models to generate counter-question strategies and implements multimodal interaction through rich media cards and floating menus; (4) Model optimization module: Feedback collection unit: collects business processing results and user feedback to build a labeled optimized data set; Incremental learning unit: uses a federated learning framework to implement cross-device model updates and combines transfer learning technology to quickly adapt to new business scenarios; Knowledge graph update unit: automatically expands the entities and relationships of the knowledge graph based on user behavior data and external data sources; Adversarial training subunit: This subunit uses a generative adversarial network (GAN) to simulate malicious voice inputs and improve the model's ability to identify fraudulent commands. Explainability Enhancement Subunit: Uses attention visualization technology to generate semantic parsing paths, providing decision-making basis for business personnel; (5) Data interface module: provides RESTful API and message queue interface, supports real-time data interaction with business systems such as CRM and ERP, and is compatible with multiple data formats such as JSON and Protobuf.

[0025] Example 3: Taking the financial customer service scenario as an example, the implementation process of the present invention is described in detail: Scenario application: Bank customers use voice to transfer funds 1. Voice collection and preprocessing The customer inputs the following voice message through the mobile banking app: "I want to transfer 5,000 yuan to Zhang San. The account number is 6222XXXX and the transfer will be made at 10 am tomorrow." The system collects the voice message through the mobile phone microphone and extracts the valid voice segment after noise reduction processing.

[0026] 2. Feature extraction and speech recognition MFCC features are extracted from the speech clip and fed into an ASR model that incorporates banking terminology (such as "transfer," "account number," and "current account") to convert the resulting text into the following: "I want to transfer 5,000 yuan to Zhang San. Account number is 6222XXXX. The transfer will be made at 10:00 AM tomorrow." 3. Business semantic understanding and data structuring (1) The semantic understanding unit combines the bank transfer business knowledge graph to identify slot information: (2) Payee: Zhang San (3) Amount: RMB 5,000 (4) Account number: 6222XXXX (5) Time: 10:00 AM tomorrow (6) The conversation management unit checks the historical conversation (if the customer has logged into the account before), automatically fills in the payer information (current account), and generates a transfer data frame.

[0027] 4. Business logic verification and data error correction (1) Rule engine verification: Amount > 0, account format is correct, and time is in the future, all of which comply with the rules; (2) If the system finds that the account number does not match the payee's name (for example, Zhang San corresponds to another account number in historical data), it triggers an error correction interaction: "The account number you provided does not completely match Zhang San's name. Do you confirm to continue the transfer?" 5. Data storage and feedback optimization After the transfer request is confirmed by the customer, the structured data is transferred to the bank's core system for processing, and the results are recorded. If it is subsequently determined that the account does belong to Zhang San, the model optimization module uses this case as a positive example to update the knowledge graph, improving recognition accuracy for similar scenarios in the future.

[0028] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for processing business data voice based on artificial intelligence, characterized by: The following steps are involved: Step 1: Multimodal speech preprocessing: Speech signals are collected through a microphone array, and denoised and framed using a combination of Mel-frequency cepstral coefficients and a dynamic time warping algorithm to extract multidimensional acoustic features encompassing time domain, frequency domain, and energy distribution. Step 2: Domain-enhanced speech recognition: A pre-trained Transformer-based speech recognition model is used. The business domain corpus is integrated during model training, and unsupervised pre-training is used to improve the model's robustness in recognizing professional terminology. Step 3: Knowledge graph-driven semantic analysis: Build a knowledge graph containing business entities, relationships, and rules, combine it with a pre-trained language model for semantic parsing, identify key business information through slot filling technology, and use multi-hop reasoning to generate structured data frames; Step 4: Dynamic dialogue management mechanism: A multi-round conversation context model is built based on a finite state machine. State transitions are achieved through intent matching and menu triggering. The current semantic understanding results are corrected based on historical conversation records to ensure the coherence of the interaction logic. Step 5: Closed-loop feedback optimization system: An error correction model is built through reinforcement learning algorithms, semantic understanding parameters are dynamically adjusted based on the verification results of the business rule engine, and federated learning technology is used to achieve cross-device model updates while protecting data privacy.

2. The method for processing business data voice based on artificial intelligence according to claim 1, characterized in that: The step 2 further includes: (1) Using adversarial training and noise injection techniques to improve the robustness of the model in complex environments, combined with edge computing optimization strategies to achieve lightweight model deployment and meet the real-time processing requirements of mobile devices; (2) Through a multi-task learning framework, speech recognition, sentiment analysis, and speaker recognition are performed simultaneously to enhance the model's comprehensive understanding of speech data.

3. The method for processing business data voice based on artificial intelligence according to claim 1, characterized in that: The step three further includes: (1) Utilize entity linking technology to align identified business entities with external knowledge bases, generate multi-dimensional semantic paths through relational reasoning, and improve parsing accuracy in complex business scenarios; (2) Dynamically update the knowledge graph by combining user behavior data and historical interaction records to achieve adaptive expansion of business rules.

4. The method for processing business data voice based on artificial intelligence according to claim 1, characterized in that: The step 5 further includes: (1) Introducing the Kalman filter algorithm to dynamically estimate the uncertainty of business data, providing high-precision state feedback for reinforcement learning, and improving the real-time and accuracy of error correction decisions; (2) Model update parameters are desensitized through differential privacy technology to ensure the security of user data during distributed training.

5. An artificial intelligence-based business data voice processing system, based on the artificial intelligence-based business data voice processing method according to any one of claims 1 to 4, characterized in that: It includes voice collection and preprocessing module, AI voice processing module, business data processing module, model optimization module, and data interface module.

6. The artificial intelligence-based business data voice processing system according to claim 5, characterized in that: The voice acquisition and preprocessing module is used to acquire voice signals and perform preprocessing such as noise reduction and framing, and includes a microphone array interface and a digital signal processor.

7. The artificial intelligence-based business data voice processing system according to claim 5, characterized in that: The AI ​​voice processing module integrates an ASR model with business corpus to achieve voice-to-text conversion and complete business semantic parsing and slot filling.

8. The artificial intelligence-based business data voice processing system according to claim 5, characterized in that: The business data processing module has a built-in business rule library to verify the legitimacy of structured data, and triggers a questioning mechanism based on the rule verification results to supplement information, while converting the semantic analysis results into a data format that complies with business specifications.

9. The artificial intelligence-based business data voice processing system according to claim 5, characterized in that: The model optimization module collects business processing results as model optimization data, and uses reinforcement learning and transfer learning algorithms to dynamically update AI model parameters, while automatically updating the domain knowledge graph according to business changes.

10. The artificial intelligence-based business data voice processing system according to claim 5, characterized in that: The data interface module provides a data interaction interface with the business system to realize the storage and call of structured data.

Images