A voice interaction analysis method and system for an intelligent voice robot

Abstract

The present invention relates to the field of voice interaction analysis technology, and in particular to a voice interaction analysis method and system for an intelligent voice robot. The method comprises the following steps: improving the clarity of user interaction voice through multi-scale noise decomposition and filtering enhancement, performing granular segmentation and long-distance semantic analysis on clear voice, generating a semantic structure, performing multi-level semantic recognition based on the structure, extracting specific meanings and making preliminary responses, generating personalized semantic graphs through independent graph structure mapping, predicting user interaction dialogues, building an advance response engine based on dialogue templates, preparing to respond to user needs, embedding the engine into an intelligent voice robot, and realizing efficient voice interaction analysis. The present invention realizes a more accurate and efficient voice interaction analysis method for an intelligent voice robot.

Description

Technical Field

[0001] The present invention relates to the field of voice interaction analysis technology, and in particular to a voice interaction analysis method and system for an intelligent voice robot. Background Art

[0002] Existing voice interaction technologies mainly focus on voice recognition and semantic understanding of single interactions. However, in actual applications, due to factors such as environmental noise interference, differences in user speaking habits, and the complexity of interaction scenarios, the accuracy of voice recognition is unstable, affecting the user interaction experience. Especially in complex noise environments, traditional voice enhancement methods often adopt a unified filtering strategy and cannot effectively handle noise interference in different frequency ranges, resulting in distortion or residual noise problems in the enhanced voice signal. Most current voice interaction systems adopt a fixed response mode and lack the ability to deeply understand and predict user personalized needs. Although some systems have introduced semantic analysis technology, they mainly stay at the shallow semantic understanding level and cannot accurately grasp the deep semantic information and behavior patterns of users in the long-term interaction process. At the same time, the response mechanism of existing systems is often passive, that is, processing and responding are started only after receiving user instructions. The interaction efficiency in this mode is low and cannot meet the user's requirements for real-time and fluency. Summary of the invention

[0003] Based on this, it is necessary to provide a voice interaction analysis method and system for an intelligent voice robot to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a voice interaction analysis method of an intelligent voice robot comprises the following steps:

[0005] Step S1: Identify the current user interaction voice; perform multi-scale noise decomposition on the current user interaction voice, and perform interval-scale filtering enhancement to obtain a filtering enhanced signal;

[0006] Step S2: granularly segmenting the filtered enhanced signal and performing long-distance semantic analysis to generate an interactive speech semantic structure;

[0007] Step S3: performing multi-level semantic recognition on the filtered enhanced signal based on the interactive speech semantic structure to obtain the interactive speech meaning representation; performing an initial dynamic response to the interactive speech meaning representation and recording the response operation data, and constructing a dialogue response template;

[0008] Step S4: Perform independent graph structure mapping on the interactive speech meaning representation to obtain a personalized semantic graph; perform interactive prediction on the response operation data according to the personalized semantic graph to generate a predicted user interactive dialogue;

[0009] Step S5: preparing an advance response for the predicted user interaction dialogue based on the dialogue response template, and constructing a response engine to obtain an advance response engine;

[0010] Step S6: Adaptively perform collaborative fusion processing on the intelligent voice robot based on the advance response engine to perform voice interaction analysis of the intelligent voice robot.

[0011] The present invention effectively distinguishes and removes background noise of different frequencies through a multi-scale noise decomposition method, and combines the filtering enhancement technology of distinguishing scales to ensure that the user interactive voice signal is still clear in a complex environment, significantly reduce the voice recognition errors caused by noise interference, and improve the system's adaptability in noisy environments. The granular segmentation technology is used in combination with long-distance semantic analysis to enable the system to understand user voice from different levels. The generated interactive voice semantic structure not only contains basic semantic information, but also covers multi-dimensional data such as context, emotion and intention, which improves the system's ability to understand complex contexts. The multi-level semantic recognition implemented based on the interactive voice semantic structure ensures a detailed analysis of the user's specific expression, accurately captures the user's true intention, reduces misunderstandings, and improves the relevance and accuracy of the response. Through the initial dynamic response mechanism, the system can immediately respond to the user's voice input. The constructed dialogue response template provides a structured response basis for subsequent interactions, ensuring consistency and fluency in multiple interactions, improving user experience, and the personalized semantic map generated by independent graph structure mapping. The system can make customized responses based on the user's historical interaction data and preferences. The personalized mechanism increases user participation and enhances user loyalty to the system. By using personalized semantic graphs for interaction prediction, the system can identify user potential needs in advance, reduce user waiting time, and improve the smoothness of real-time interaction, so that users can feel higher efficiency during use. Combining dialogue response templates with the advance response engine, it can realize demand prediction and corresponding preparation before the user clearly expresses his intention, improve the naturalness of interaction, and make the user experience smoother, as if having a conversation with a real person. By recording response operation data, the system continuously analyzes user feedback and realizes adaptive learning. The intelligent voice robot can gradually optimize the interaction strategy over time, improve the overall intelligence level, and adapt to the ever-changing user needs and preferences. The adaptive collaborative fusion processing of the advance response engine ensures the efficiency of the intelligent voice robot in voice interaction analysis, integrates multiple analysis algorithms and response strategies, so that the system can maintain efficient operation in complex interaction scenarios, and improve overall performance and user satisfaction.

[0012] Preferably, step S1 comprises the following steps:

[0013] Step S11: recognizing the current user interaction speech; performing fast Fourier transform on the current user interaction speech to obtain the interactive speech frequency domain features;

[0014] Step S12: performing frequency domain scale recognition on the interactive speech frequency domain features to generate a frequency scale set;

[0015] Step S13: dividing the current user interaction speech into frequency intervals based on the frequency scale set to obtain multiple frequency sub-interval speech;

[0016] Step S14: performing noise energy recognition on the frequency sub-interval speech based on a preset noise energy threshold, and generating noise features for each frequency interval;

[0017] Step S15: estimating the noise intensity of the speech in the frequency sub-interval according to the noise characteristics of each frequency interval to obtain a noise distribution spectrum;

[0018] Step S16: Calculating the filter coefficients of each frequency sub-interval for the speech using the noise distribution spectrum to obtain a set of optimal filter coefficients;

[0019] Step S17: performing filtering enhancement in each frequency sub-interval on the frequency sub-interval speech based on the preferred filter coefficient set to obtain a filtered enhanced signal.

[0020] The present invention effectively extracts the frequency domain features of user interactive speech through the fast Fourier transform technology, providing an accurate basis for subsequent frequency domain analysis. The frequency scale set generated by frequency domain scale recognition can provide multi-level frequency information for in-depth analysis of speech signals. The interactive speech is divided into frequency intervals based on the frequency scale set to ensure that different frequency features are fully utilized, making subsequent processing more targeted. The noise energy recognition technology analyzes each frequency sub-interval, and the generated noise features provide an important basis for subsequent noise processing. The noise intensity estimation technology combines the noise features of each frequency interval, and the generated noise distribution map provides a clear guiding direction for filtering processing. The preferred filter coefficient set calculated by the noise distribution map ensures the efficiency and accuracy of the filtering process. The frequency interval filtering enhancement based on the preferred filter coefficient set is used to process the interactive speech. The final filter enhancement signal significantly improves the speech quality, ensuring that the user interaction experience is smoother and more natural.

[0021] Preferably, step S2 comprises the following steps:

[0022] Step S21: performing phoneme boundary detection on the filtered enhanced signal to obtain a phoneme sequence table;

[0023] Step S22: segmenting the filtered enhanced signal into syntactic units according to the phoneme sequence table to generate multi-level language units;

[0024] Step S23: extracting semantic features from the multi-level language units to obtain semantic features of the language units;

[0025] Step S24: performing long-range association semantic inference on the semantic features of the language unit to generate a semantic association network;

[0026] Step S25: constructing a hierarchical structure of the multi-level language units based on the semantic association network to obtain an interactive speech semantic structure.

[0027] The present invention can accurately identify the phoneme boundaries in the filtered enhanced signal through the phoneme boundary detection technology, and the generated phoneme sequence table provides a clear basis for subsequent processing. The syntactic unit segmentation is performed according to the phoneme sequence table, and multi-level language units can be effectively generated to ensure the integrity and accuracy of the syntactic structure. The semantic features of the multi-level language units are extracted, and the generated language unit semantic features provide rich information for understanding the user's intention. The long-range association semantic inference technology analyzes the relationship between language units, and the generated semantic association network can reveal the deep semantic connection. The hierarchical structure construction based on the semantic association network ensures the systematization and orderliness of the interactive speech semantic structure, improves the organization and processing efficiency of information, and by constructing the interactive speech semantic structure, the system can understand the user's intention and needs more comprehensively, thereby providing a solid foundation for subsequent intelligent response.

[0028] Preferably, step S24 includes the following steps:

[0029] Contextually encode the semantic features of the language unit to obtain a contextual semantic matrix;

[0030] Based on the contextual semantic matrix, long-range dependency recognition is performed on the semantic features of the language unit to obtain a dependency tree;

[0031] Track the inter-sentence associations in the dependency tree to obtain the topic link relationship;

[0032] Simulate the topic link relationship and semantic propagation to generate simulated semantic flow data;

[0033] Based on the simulated semantic flow data, the cross-sentence relationship of the semantic features of the language unit is inferred to generate a semantic association network.

[0034] The present invention performs context encoding on the semantic features of language units, and the generated context semantic matrix can enrich semantic information and enhance the system's ability to understand context. The long-range dependency recognition implemented based on the context semantic matrix, and the obtained dependency tree effectively reveals the deep-level relationship between language units and improves the accuracy of semantic analysis. The inter-sentence association tracking technology of the dependency tree can identify the topic link relationship, ensuring that the system can maintain the theme consistency when processing multi-sentence interactions. By performing semantic propagation simulation on the topic link relationship, the simulated semantic flow data generated provides an intuitive basis for cross-sentence relationship reasoning and enhances the coherence of semantic understanding. The semantic association network generated based on the cross-sentence relationship reasoning performed based on the simulated semantic flow data provides the system with comprehensive semantic connections, ensuring the accurate capture and response of user intentions, thereby achieving a higher level of semantic understanding in complex interaction scenarios and supporting a more natural and accurate user interaction experience.

[0035] Preferably, step S3 comprises the following steps:

[0036] Step S31: performing multi-level recursive parsing on the filtered enhanced signal based on the interactive speech semantic structure to obtain a semantic representation matrix;

[0037] Step S32: identifying the core semantic components of the semantic representation matrix to obtain the interactive speech meaning representation;

[0038] Step S33: constructing an intention recognition model for the interactive speech meaning representation to generate a user intention recognition model;

[0039] Step S34: performing interactive response deduction on the interactive speech meaning representation according to the user intention recognition model to generate a predicted interactive response plan;

[0040] Step S35: Perform initial response processing according to the predicted interactive response plan and record the response operation data; construct a dialogue response template for the response operation data to generate a dialogue response template.

[0041] The present invention implements multi-level recursive parsing based on the semantic structure of interactive speech, and the generated semantic representation matrix provides comprehensive semantic information, enhances the system's deep understanding of speech content, identifies core semantic components of the semantic representation matrix, ensures that the system can accurately extract the meaning representation of interactive speech, and improves the accuracy of information extraction. By constructing an intention recognition model, the user's historical interaction data and semantic information can be effectively integrated. The generated model provides a reliable basis for subsequent responses. The interactive response deduction based on the user intention recognition model ensures that the system can quickly adapt to user needs and improve the timeliness and relevance of the interaction. The initial response processing performed according to the predicted interactive response plan can quickly feedback the user's intention, and at the same time record the response operation data to ensure the system's timely response to the user. The response operation data is analyzed, and the constructed dialogue response template provides a standardized response mechanism, which makes subsequent user interactions more consistent and natural, and achieves a higher level of intelligence in voice interaction analysis, meeting user needs while improving the overall interactive experience.

[0042] Preferably, step S4 comprises the following steps:

[0043] Step S41: performing semantic topic analysis on the interactive speech meaning representation to obtain core semantic categories;

[0044] Step S42: performing semantic space encoding on the core semantic categories to generate a mapping structure diagram;

[0045] Step S43: constructing node relationships for core semantic categories based on the mapping structure diagram to obtain a personalized semantic graph;

[0046] Step S44: performing interactive context simulation on the response operation data to obtain a current simulated interactive context;

[0047] Step S45: Based on the current simulation interaction context and the personalized semantic graph, interactive prediction is performed on the response operation data to generate a predicted user interaction dialogue.

[0048] The present invention can accurately identify core semantic categories by performing semantic topic analysis on the representation of interactive speech meaning, providing a clear direction and basis for subsequent analysis, and generates a mapping structure diagram by encoding the semantic space of the core semantic categories, ensuring that the relationship between different semantics is visualized and enhancing the system's understanding ability. Node relationship construction is performed based on the mapping structure diagram, which can reflect the user's specific semantic needs and improve the relevance and pertinence of the response. The interactive context simulation is performed on the response operation data to generate the current simulated interactive context, ensuring that the system can better understand the user's needs in real scenarios. The interactive prediction based on the current simulated interactive context and the personalized semantic map can generate a predicted user interaction dialogue that can identify the user's intention in advance and improve the timeliness of the response, and can provide a more personalized and intelligent interactive experience, meet user needs, and enhance user participation and satisfaction, thereby having the ability to make dynamic adjustments based on user feedback and interaction history, ensuring that efficient performance is maintained in a constantly changing interactive environment.

[0049] Preferably, step S45 includes the following steps:

[0050] Perform feature vectorization processing on the current simulation interaction context to obtain a context feature vector;

[0051] Quantify the interaction tendency of the personalized graph according to the context feature vector and generate the user interaction tendency response;

[0052] Conduct interactive prediction modeling on personalized semantic graphs to generate interactive behavior prediction models;

[0053] The interactive behavior prediction model is used to perform multiple rounds of interactive deduction on the response operation data to obtain an interactive dialogue branch tree;

[0054] Based on the user's interaction tendency response, the interactive dialogue branch tree is predicted to have coherent interactive behaviors and generate predicted user interaction dialogues.

[0055] The present invention performs feature vectorization processing on the current simulated interaction context, and the generated context feature vector can provide a structured data basis for subsequent analysis. The personalized graph is quantified for interaction tendency according to the context feature vector, and the generated user interaction tendency reaction can accurately reflect the user's preferences and needs. The personalized semantic graph is interactively predicted and modeled, and the generated interactive behavior prediction model provides the system with an in-depth understanding of user behavior. The interactive behavior prediction model is used to perform multiple rounds of interactive deduction on the response operation data, and the obtained interactive dialogue branch tree can display different interactive paths and enhance the flexibility of the system. The interactive dialogue branch tree is predicted to have coherent interactive behavior based on the user interaction tendency reaction, and the generated predicted user interactive dialogue can ensure that the system maintains consistency and naturalness in multiple rounds of dialogues, can provide a more personalized and smooth interactive experience, meet the specific needs of users, enhance the user's sense of participation and satisfaction, and can adapt to changes in complex interactive environments to ensure efficient response and continuous optimization.

[0056] Preferably, step S5 comprises the following steps:

[0057] Step S51: extracting dialogue elements from the predicted user interactive dialogue to obtain an interactive dialogue element set;

[0058] Step S52: formulating a response plan for the interactive dialogue element set based on the dialogue response template to generate an interactive response plan;

[0059] Step S53: integrating the functional modules of the interactive response scheme to obtain an interactive response module chain;

[0060] Step S54: construct a response engine for the interactive response module chain to obtain an advanced response engine.

[0061] The present invention can generate an interactive dialogue element set by extracting dialogue elements of predicted user interactive dialogues, provide structured information for subsequent response formulation, improve the pertinence and effectiveness of the response, formulate a response plan for the interactive dialogue element set based on the dialogue response template, ensure that the response plan can meet user needs, enhance the personalized service capability of the system, integrate functional modules of the interactive response plan, and form an interactive response module chain that can realize the collaborative work of multiple functions, improve the flexibility and expansibility of the system, construct a response engine for the interactive response module chain, and generate an advanced response engine that can achieve fast and intelligent response, enhance the responsiveness of the system in complex interactive scenarios, and through accurate element extraction and response plan formulation, the system can achieve higher response efficiency in user interaction, reduce user waiting time, ensure that users feel a more natural and smooth interactive experience during use, improve user satisfaction, and can dynamically adjust according to user behavior and preferences, continuously optimize response strategies, and adapt to changing interactive needs.

[0062] Preferably, step S6 comprises the following steps:

[0063] Step S61: performing feature matching on the intelligent voice robot based on the advance response engine to obtain mutually adapted features;

[0064] Step S62: mapping device instructions to the intelligent voice robot according to the mutual adaptation characteristics to generate a device operation instruction set;

[0065] Step S63: Perform intelligent interactive response on the intelligent voice robot hardware according to the device operation instruction set to perform voice interaction analysis of the intelligent voice robot.

[0066] The present invention performs feature matching on the intelligent voice robot based on the advance response engine, can identify mutually compatible features, ensure a high degree of consistency between the system and user needs, perform device instruction mapping according to the mutually compatible features, and the generated device operation instruction set provides a clear guide for the specific operation of the intelligent voice robot. The device operation instruction set is used for the intelligent voice robot hardware to ensure that the system can efficiently perform voice interaction analysis and improve the accuracy and timeliness of the response. Through feature matching and instruction mapping, it can flexibly respond to various interaction scenarios and enhance the adaptability to different user needs. Accurate device operation instruction execution ensures that the interaction between the user and the intelligent voice robot is more natural and smooth, and improves the overall user experience. It can quickly generate and execute device operation instructions, improve the response efficiency of the intelligent voice robot during the interaction process, and enable the intelligent voice robot to demonstrate a higher level of intelligence in complex environments and meet the diverse interaction needs of users.

[0067] The present invention also provides a voice interaction analysis system of an intelligent voice robot, which is used to execute the voice interaction analysis method of the intelligent voice robot as described above. The voice interaction analysis system of the intelligent voice robot comprises:

[0068] The speech enhancement module is used to recognize the current user interaction speech; perform multi-scale noise decomposition on the current user interaction speech, and perform interval-scale filtering enhancement to obtain a filtering enhancement signal;

[0069] The semantic structure analysis module is used to segment the filtered enhanced signal into granularity and perform long-distance semantic analysis to generate the interactive speech semantic structure;

[0070] The initial response module is used to perform multi-level semantic recognition on the filtered enhanced signal based on the interactive speech semantic structure to obtain the interactive speech meaning representation; perform the initial dynamic response to the interactive speech meaning representation and record the response operation data, and at the same time build a dialogue response template;

[0071] The interactive prediction module is used to map the interactive speech meaning representation to an independent graph structure to obtain a personalized semantic graph; interactively predict the response operation data based on the personalized semantic graph to generate a predicted user interactive dialogue;

[0072] An advance response preparation module is used to prepare an advance response for the predicted user interaction dialogue based on the dialogue response template, and to construct a response engine to obtain an advance response engine;

[0073] The robot embedding module is used to perform adaptive collaborative fusion processing on the intelligent voice robot based on the advance response engine to perform voice interaction analysis of the intelligent voice robot.

[0074] The invention uses a speech enhancement module to effectively reduce background noise interference through multi-scale noise decomposition and filtering enhancement technology, ensures that the user interaction speech signal is clear and discernible, and significantly improves recognition accuracy. The semantic structure analysis module generates rich interactive speech semantic structures through granular segmentation and long-distance semantic analysis, and enhances the system's comprehensive understanding of user intentions. The initial response module realizes multi-level semantic recognition of the filter enhancement signal, can quickly feedback user needs, and records response operation data at the same time, improves the real-time and efficiency of the interaction. The interaction prediction module generates a personalized semantic map through independent graph structure mapping, and predicts responses based on user historical interaction data, enhances user participation, and improves satisfaction. The advance response preparation module prepares the predicted user interaction dialogue based on the dialogue response template, ensuring that the system can make a natural response in advance when the user does not clearly express his intention. Each module records and analyzes user interaction data to achieve adaptive learning. The system can gradually optimize the interaction strategy, improve the overall intelligence level, and adapt to changing user needs. The robot embedding module embeds the advance response engine into the intelligent voice robot to ensure efficient execution of voice interaction analysis in complex interaction scenarios, and improve overall performance and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 A schematic diagram of the steps of a voice interaction analysis method for an intelligent voice robot;

[0076] Figure 2 Detailed implementation flow chart of step S2;

[0077] Figure 3 is a schematic diagram of a detailed implementation process of step S3;

[0078] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0079] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are 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 technicians in this field without creative work are within the scope of protection of the present invention.

[0080] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0081] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0082] To achieve this, please refer to Figures 1 to 3 , a voice interaction analysis method of an intelligent voice robot, comprising the following steps:

[0083] Step S1: Identify the current user interaction voice; perform multi-scale noise decomposition on the current user interaction voice, and perform interval-scale filtering enhancement to obtain a filtering enhanced signal;

[0084] Step S2: granularly segmenting the filtered enhanced signal and performing long-distance semantic analysis to generate an interactive speech semantic structure;

[0085] Step S3: performing multi-level semantic recognition on the filtered enhanced signal based on the interactive speech semantic structure to obtain the interactive speech meaning representation; performing an initial dynamic response to the interactive speech meaning representation and recording the response operation data, and constructing a dialogue response template;

[0086] Step S4: Perform independent graph structure mapping on the interactive speech meaning representation to obtain a personalized semantic graph; perform interactive prediction on the response operation data according to the personalized semantic graph to generate a predicted user interactive dialogue;

[0087] Step S5: preparing an advance response for the predicted user interaction dialogue based on the dialogue response template, and constructing a response engine to obtain an advance response engine;

[0088] Step S6: Adaptively perform collaborative fusion processing on the intelligent voice robot based on the advance response engine to perform voice interaction analysis of the intelligent voice robot.

[0089] The present invention effectively distinguishes and removes background noise of different frequencies through a multi-scale noise decomposition method, and combines the filtering enhancement technology of distinguishing scales to ensure that the user interactive voice signal is still clear in a complex environment, significantly reduce the voice recognition errors caused by noise interference, and improve the system's adaptability in noisy environments. The granular segmentation technology is used in combination with long-distance semantic analysis to enable the system to understand user voice from different levels. The generated interactive voice semantic structure not only contains basic semantic information, but also covers multi-dimensional data such as context, emotion and intention, which improves the system's ability to understand complex contexts. The multi-level semantic recognition implemented based on the interactive voice semantic structure ensures a detailed analysis of the user's specific expression, accurately captures the user's true intention, reduces misunderstandings, and improves the relevance and accuracy of the response. Through the initial dynamic response mechanism, the system can immediately respond to the user's voice input. The constructed dialogue response template provides a structured response basis for subsequent interactions, ensuring consistency and fluency in multiple interactions, improving user experience, and the personalized semantic map generated by independent graph structure mapping. The system can make customized responses based on the user's historical interaction data and preferences. The personalized mechanism increases user participation and enhances user loyalty to the system. By using personalized semantic graphs for interaction prediction, the system can identify user potential needs in advance, reduce user waiting time, and improve the smoothness of real-time interaction, so that users can feel higher efficiency during use. Combining dialogue response templates with the advance response engine, it can realize demand prediction and corresponding preparation before the user clearly expresses his intention, improve the naturalness of interaction, and make the user experience smoother, as if having a conversation with a real person. By recording response operation data, the system continuously analyzes user feedback and realizes adaptive learning. The intelligent voice robot can gradually optimize the interaction strategy over time, improve the overall intelligence level, and adapt to the ever-changing user needs and preferences. The adaptive collaborative fusion processing of the advance response engine ensures the efficiency of the intelligent voice robot in voice interaction analysis, integrates multiple analysis algorithms and response strategies, so that the system can maintain efficient operation in complex interaction scenarios, and improve overall performance and user satisfaction.

[0090] In the embodiment of the present invention, refer to Figure 1 , is a schematic diagram of the steps of a voice interaction analysis method of an intelligent voice robot of the present invention. In this example, the voice interaction analysis method of an intelligent voice robot includes the following steps:

[0091] Step S1: Identify the current user interaction voice; perform multi-scale noise decomposition on the current user interaction voice, and perform interval-scale filtering enhancement to obtain a filtering enhanced signal;

[0092] In this embodiment, a high-sensitivity microphone array is used to capture audio signals, and digital signal processing technology is combined to ensure that the user's voice is effectively captured in a complex environment. The audio signal is preprocessed to remove silent segments and background noise to improve the signal quality. The system uses an acoustic model for real-time speech recognition, combined with a hidden Markov model (HMM) and a deep neural network (DNN) to ensure adaptability to a variety of dialects and changes in speech speed, performs multi-scale noise decomposition on the collected voice signal, uses wavelet transform to decompose the signal into multiple frequency bands, extracts signal information of different frequencies, applies an adaptive filtering algorithm to each frequency band for selective enhancement, sets different thresholds to remove background noise, ensures that the amplitude of the effective voice signal is improved, and the processed frequency band signals are synthesized again to generate a clear interactive voice signal.

[0093] Step S2: granularly segmenting the filtered enhanced signal and performing long-distance semantic analysis to generate an interactive speech semantic structure;

[0094] In this embodiment, short-time Fourier transform (STFT) is used to convert the signal into the time-frequency domain to ensure that each segmented segment can retain sufficient semantic information. The entire speech signal is divided into multiple overlapping small segments, and the window size is set to 20 to 50 milliseconds. The signal is processed by a sliding window technology to obtain the time-frequency characteristics of each small segment, and acoustic features are extracted. Features such as Mel-frequency cepstral coefficients (MFCC) and linear predictive coding (LPC) are used to extract semantically related information. The extracted features are then input into a trained long short-term memory network (LSTM) model for long-distance semantic analysis. The LSTM model can effectively capture the contextual information in the speech segment. The model performance is evaluated by cross-validation of the training set and the validation set. The generated interactive speech semantic structure contains the core semantic information of each segment and its position in the conversation.

[0095] Step S3: performing multi-level semantic recognition on the filtered enhanced signal based on the interactive speech semantic structure to obtain the interactive speech meaning representation; performing an initial dynamic response to the interactive speech meaning representation and recording the response operation data, and constructing a dialogue response template;

[0096] In this embodiment, a convolutional neural network (CNN) is used to perform deep learning processing on each semantic segment, and the extracted features are input into the trained model. The high-level features of the signal are extracted through multi-layer convolution and pooling operations. Finally, the CNN model is classified through a fully connected layer to identify the specific interactive speech meaning. Feature normalization and data enhancement technology are used to improve the generalization ability of the model. Each recognition result is matched with a preset dialogue response template to ensure that the generated response is targeted, and the data of the response operation is recorded, including the response time, content and user feedback information.

[0097] Step S4: Perform independent graph structure mapping on the interactive speech meaning representation to obtain a personalized semantic graph; perform interactive prediction on the response operation data according to the personalized semantic graph to generate a predicted user interactive dialogue;

[0098] In this embodiment, an independent graph structure mapping is performed on the interactive speech meaning representation, a graph database containing user interests, historical interaction data and recognition results is constructed, and the data is modeled using graph neural network (GNN) technology to generate a personalized semantic graph. Each node represents the user's preferences and interests, and the edges represent the relationship between different semantics. By analyzing the graph structure and capturing the user's interaction pattern, the user's needs in different situations can be understood more effectively. According to the personalized semantic graph, the response operation data is interactively predicted, the user's next needs are identified, and a predicted user interaction dialogue that meets the user's needs is generated. This process not only improves the intelligence level of the system, but also makes the interaction process more natural and smooth, and the user's personalized experience is enhanced. By continuously updating the graph structure, it is ensured that the system can adapt to the user's changing needs in real time.

[0099] Step S5: preparing an advance response for the predicted user interaction dialogue based on the dialogue response template, and constructing a response engine to obtain an advance response engine;

[0100] In this embodiment, based on the dialogue response template, advance response preparation is performed for the predicted user interaction dialogue, and natural language generation technology is used to automatically generate appropriate response text in combination with the user's historical interaction data and personalized semantic graph. The construction of the response engine is a key step to ensure that it can respond quickly when the user initiates a dialogue and improve the smoothness of the interaction. The engine continuously optimizes the response strategy through a real-time feedback mechanism, adjusts the response content according to the user's input, and ensures the accuracy and adaptability of the response.

[0101] Step S6: Adaptively perform collaborative fusion processing on the intelligent voice robot based on the advance response engine to perform voice interaction analysis of the intelligent voice robot.

[0102] In this embodiment, based on the constructed advanced response engine, the intelligent voice robot is adaptively coordinated and integrated, and the response engine is integrated into the voice robot using an embedded system to ensure that the system can process the user's voice interaction in real time. Combined with the voice recognition module and the semantic analysis module, the intelligent voice robot can efficiently perform voice interaction analysis, understand the user's intentions in real time, and make rapid feedback. The system processes the user's input voice signal in real time and combines historical interaction data to optimize the accuracy and fluency of voice recognition, thereby improving the naturalness and humanity of user interaction. The robot can work effectively in complex environments, adapt to various user needs and situations, and provide users with personalized services.

[0103] Preferably, step S1 comprises the following steps:

[0104] Step S11: recognizing the current user interaction speech; performing fast Fourier transform on the current user interaction speech to obtain the interactive speech frequency domain features;

[0105] Step S12: performing frequency domain scale recognition on the interactive speech frequency domain features to generate a frequency scale set;

[0106] Step S13: dividing the current user interaction speech into frequency intervals based on the frequency scale set to obtain multiple frequency sub-interval speech;

[0107] Step S14: performing noise energy recognition on the frequency sub-interval speech based on a preset noise energy threshold, and generating noise features for each frequency interval;

[0108] Step S15: estimating the noise intensity of the speech in the frequency sub-interval according to the noise characteristics of each frequency interval to obtain a noise distribution spectrum;

[0109] Step S16: Calculating the filter coefficients of each frequency sub-interval for the speech using the noise distribution spectrum to obtain a set of optimal filter coefficients;

[0110] Step S17: performing filtering enhancement in each frequency sub-interval on the frequency sub-interval speech based on the preferred filter coefficient set to obtain a filtered enhanced signal.

[0111] In this embodiment, a speech recognition module is used to extract the audio signal emitted by the user from the system noise and other non-speech audio, and the fast Fourier transform (FFT) method is used to convert the speech signal from the time domain to the frequency domain. The amplitude and phase characteristics of the speech signal are represented in the frequency domain through FFT processing to implement the frequency domain feature extraction operation. The frequency domain characteristics reflect the energy distribution of speech at different frequencies and provide basic data support for subsequent frequency domain analysis. The frequency domain feature set generated by FFT contains multiple frequency components and can display the characteristic information of the speech signal at each frequency point in detail. The frequency domain scale of the interactive speech frequency domain features is identified, and the frequency domain scale division algorithm is applied to the common frequency range of human speech (such as 20Hz to 20kHz). ), classify the frequency domain features according to the preset scale, generate a frequency scale set, the frequency scale set is divided into several scale segments according to the distribution law of the speech frequency, each scale segment contains the frequency domain feature value of a specific frequency range, the divided frequency scale set can effectively distinguish the main frequency components in the speech signal, so that the subsequent frequency interval division processing has targeted and hierarchical data support, the frequency interval division of the current user interaction speech is performed based on the frequency scale set, the frequency domain feature data is divided into multiple different frequency sub-intervals by using the frequency interval division algorithm, different frequency interval ranges are set according to each frequency segment in the frequency scale set, and multiple frequency sub-intervals are generated, each sub-interval represents a speech component within a specific frequency range, and the frequency interval division makes The subsequent noise energy recognition and intensity estimation are more refined, and the signal characteristics of each frequency sub-interval can be analyzed independently, so as to obtain more accurate noise distribution information. The noise energy of the frequency sub-interval speech is recognized based on the preset noise energy threshold, and the energy value of each frequency sub-interval is compared with the noise energy threshold. The frequency segment exceeding the threshold is identified as the speech signal noise area, the noise area is marked and the noise characteristic data of each frequency interval is generated. The noise characteristic data includes the frequency range, energy level and occurrence frequency of the noise. The noise information of each frequency interval is extracted through noise energy recognition to provide basic data for the subsequent noise intensity estimation. The noise intensity of the frequency sub-interval speech is estimated according to the noise characteristics of each frequency interval. Estimation: Noise intensity estimation uses a weighted average algorithm or other statistical methods to generate noise intensity distribution information by weighted calculation of noise energy data in each frequency interval, and obtain a noise distribution spectrum. The noise distribution spectrum provides noise intensity and spatial distribution in each frequency interval, providing data support for filter design and parameter optimization. The noise intensity estimation can clearly understand the noise intensity change trend of each frequency segment, so that different filtering processes can be applied in a targeted manner in subsequent processing steps. The noise distribution spectrum is used to calculate the filter coefficients of each frequency sub-interval speech, and the noise distribution spectrum is compared with the noise intensity data of each frequency sub-interval. The adaptive filter algorithm is applied to assign corresponding filter coefficients to each frequency sub-interval.The adaptive filter can dynamically adjust the filter coefficient according to the noise distribution to make the filtering effect more accurate and improve the noise suppression effect in different frequency intervals to an ideal level. The preferred filter coefficient set ensures that the filtering operation has efficient noise suppression ability and low distortion effect. The calculated preferred filter coefficient set can maximize the integrity of the voice signal. Based on the preferred filter coefficient set, the voice in the frequency sub-interval is filtered and enhanced in each frequency interval. The method of combining the bandpass filter and the adaptive filter is used to perform independent filtering on each frequency sub-interval. The coefficient value in the preferred filter coefficient set is used to suppress the noise and enhance the voice in the corresponding frequency segment of each sub-interval, so that the noise components of the filtered voice signal in each frequency segment are effectively eliminated, and the enhanced voice signal has higher clarity, ensuring the balance and stability of the final output interactive voice signal in each frequency segment. ,

[0112] Preferably, step S2 comprises the following steps:

[0113] Step S21: performing phoneme boundary detection on the filtered enhanced signal to obtain a phoneme sequence table;

[0114] Step S22: segmenting the filtered enhanced signal into syntactic units according to the phoneme sequence table to generate multi-level language units;

[0115] Step S23: extracting semantic features from the multi-level language units to obtain semantic features of the language units;

[0116] Step S24: performing long-range association semantic inference on the semantic features of the language unit to generate a semantic association network;

[0117] Step S25: constructing a hierarchical structure of the multi-level language units based on the semantic association network to obtain an interactive speech semantic structure.

[0118] As an example of the present invention, see Figure 2 In this example, step S2 includes:

[0119] Step S21: performing phoneme boundary detection on the filtered enhanced signal to obtain a phoneme sequence table;

[0120] In this embodiment, phoneme boundary detection is performed on the filtered enhanced signal. The speech signal is divided into phoneme units by adopting a dynamic time warping (DTW) algorithm combined with a short-time energy and short-time zero-crossing rate boundary detection method. The filtered enhanced signal is decomposed into multiple frames, and its characteristic parameters are calculated for each frame. Different phonemes appearing in the clear speech signal are analyzed frame by frame through dynamic time warping processing, and the start and end boundaries of the phonemes are located. Accurate detection of phoneme boundaries can eliminate invalid parts in the speech signal, ensure that the boundaries of each phoneme unit are clear, and avoid confusion between phonemes. The generated phoneme sequence table includes the start time, end time and phoneme category information of each phoneme unit, forming the basic data support for subsequent language unit segmentation.

[0121] Step S22: segmenting the filtered enhanced signal into syntactic units according to the phoneme sequence table to generate multi-level language units;

[0122] In this embodiment, the filtered enhanced signal is segmented into syntactic units according to the phoneme sequence table, and the phoneme boundary information in the phoneme sequence table is used for analysis in combination with the syntactic rules in the statistical language model. A layer-by-layer segmentation algorithm is used to divide the speech signal into different syntactic units. By establishing a mapping relationship between phoneme sequences and vocabulary units, the phoneme sequences are converted into vocabulary and phrases. The vocabulary units are then parsed according to syntactic rules to generate multi-level language units. The multi-level language units include language structures at the phoneme level, vocabulary level, phrase level and sentence level. The multi-level segmentation ensures the integrity of the speech signal content, so that the language units at each level are interrelated in syntactic structure, which lays a solid structural foundation for subsequent semantic extraction.

[0123] Step S23: extracting semantic features from the multi-level language units to obtain semantic features of the language units;

[0124] In this embodiment, semantic features are extracted from multi-level language units. By using the word embedding model in natural language processing, feature extraction is performed on each language unit to obtain a feature vector containing semantic information. During the semantic feature extraction process, each language unit is associated with its context to enhance its semantic accuracy. Combined with the context semantic vector calculation, the meaning of words can be accurately expressed in different contexts. By performing feature extraction on the vocabulary layer, phrase layer and sentence layer layer by layer, it is ensured that the semantic information at each level can be captured and stored in the semantic feature set, thereby establishing a high-quality semantic data foundation for long-range semantic association analysis.

[0125] Step S24: performing long-range association semantic inference on the semantic features of the language unit to generate a semantic association network;

[0126] In this embodiment, long-range association semantic inference is performed on the semantic features of language units, and the long-range semantic relationship between each language unit is calculated layer by layer through a deep semantic network to generate a semantic association network. The long-range association analysis performs global association inference on the semantic features of each language unit to identify the implicit semantic relationship between the semantic units. Through iterative calculation of the semantic network, the accuracy of the semantic units in long-range semantics is improved. The generated semantic association network contains multi-layer association paths, reflecting the multi-level semantic dependency relationship between language units, realizing the deep association of the semantic structure and the complete representation of the semantic information flow, and providing effective semantic association support for the construction of the semantic structure.

[0127] Step S25: constructing a hierarchical structure of the multi-level language units based on the semantic association network to obtain an interactive speech semantic structure.

[0128] In this embodiment, a hierarchical structure is constructed for multi-level language units based on a semantic association network, and a hierarchical semantic structure model is constructed using a graph neural network (GNN). The multi-level language units are constructed into a hierarchical semantic structure according to the path of the semantic association network. During the structure construction process, the association path of each language unit in the semantic association network is converted into the form of graph nodes and edges. Through node connection and weight setting, the language units at each level are organized into an orderly semantic structure. The constructed interactive speech semantic structure can fully represent the semantic levels and structural levels contained in the speech signal, provide a basis for the intelligent voice robot to achieve multi-level semantic understanding and response, and realize the semantic depth analysis of user interactive voice signals.

[0129] Preferably, step S24 includes the following steps:

[0130] Contextually encode the semantic features of the language unit to obtain a contextual semantic matrix;

[0131] Based on the contextual semantic matrix, long-range dependency recognition is performed on the semantic features of the language unit to obtain a dependency tree;

[0132] Track the inter-sentence associations in the dependency tree to obtain the topic link relationship;

[0133] Simulate the topic link relationship and semantic propagation to generate simulated semantic flow data;

[0134] Based on the simulated semantic flow data, the cross-sentence relationship of the semantic features of the language unit is inferred to generate a semantic association network.

[0135] In this embodiment, the context of the semantic features of the language units is encoded. By using a bidirectional long short-term memory network combined with a context window method, the semantic features of each language unit are encoded together with its context information to generate a context semantic matrix, which can effectively capture the positional relationship and semantic interaction of each language unit in the text before and after the sentence, so as to ensure the comprehensive encoding of the context. For the semantic features of language units at all levels, the context is recursively processed layer by layer through a multi-layer neural network, and the relevant context information is combined in the context window to generate a context semantic matrix containing multi-dimensional semantic information. Each row of the context semantic matrix represents the semantic feature of the current language unit in a specific context. The language unit is encoded based on the context semantic matrix. Meta-semantic features are used for long-range dependency recognition. A dependency analysis algorithm (such as a dependency parsing method based on the BERT model (bidirectional transformer model)) is used to calculate the semantic dependencies between language units in the context semantic matrix to construct a dependency tree. The dependency analysis algorithm analyzes each language unit pair by pair based on semantic features and contextual relationships, converts the dependency structure of each language unit into a tree graph structure, and obtains a complete dependency tree by calculating the semantic similarity and dependency weight between each node. The nodes of the dependency tree represent the semantic features of each language unit, and the connected edges indicate the dependency paths between language units. This tree structure lays the foundation for the path tracking of inter-sentence topics in subsequent semantic associations. Inter-sentence association tracking, using the deep tracking algorithm in the recursive neural network (RNN) to perform topic tracking analysis on each dependency path, extract inter-sentence dependency paths with the same or similar topics, and generate topic link relationships. The inter-sentence association tracking algorithm combines the path data in the dependency tree, calculates the semantic clustering of each language unit at the sentence level, connects the topic-related nodes, and thus builds a chain of inter-sentence topic associations. The establishment of topic link relationships is through multi-layer recursive tracking, so that the sentence content of the same topic in the semantic structure can be effectively associated to ensure a comprehensive analysis of cross-sentence semantic associations, and simulates the semantic propagation of topic link relationships. By establishing a semantic propagation model (such as the propagation based on the graph convolutional network GCN), the topic link relationship can be effectively associated. Model), multi-step propagation processing is performed on the semantic features in the topic link relationship, the semantic flow and transfer features of language units in different contexts are simulated, and simulated semantic flow data are generated. The semantic propagation model adopts a recursive propagation method, so that each topic link relationship continuously transmits semantic information during the propagation process. The semantic flow data contains the semantic change trend after multiple propagations between language units, ensuring the comprehensive association of multi-layer semantic structures. The process of simulating semantic flow records the changes in semantic features of language units in multi-level propagation, providing a basis for cross-sentence relationship reasoning. According to the simulated semantic flow data, cross-sentence relationship reasoning is performed on the semantic features of language units. The reasoning method based on Bayesian network is used, combined with the association features in the simulated semantic flow.By reasoning and analyzing the semantic dependency relationship between language units, a semantic association network is generated. The Bayesian network algorithm can perform dependency reasoning on multi-level propagation semantic data to build a cross-sentence semantic association network. By analyzing the association path and semantic flow relationship of each language unit, the semantic association level of each language unit is determined, so that the semantic information of each language unit in the semantic association network is accurately reflected in the multi-level relationship, and finally the deep semantic association between language units is realized.

[0136] Preferably, step S3 comprises the following steps:

[0137] Step S31: performing multi-level recursive parsing on the filtered enhanced signal based on the interactive speech semantic structure to obtain a semantic representation matrix;

[0138] Step S32: identifying the core semantic components of the semantic representation matrix to obtain the interactive speech meaning representation;

[0139] Step S33: constructing an intention recognition model for the interactive speech meaning representation to generate a user intention recognition model;

[0140] Step S34: performing interactive response deduction on the interactive speech meaning representation according to the user intention recognition model to generate a predicted interactive response plan;

[0141] Step S35: Perform initial response processing according to the predicted interactive response plan and record the response operation data; construct a dialogue response template for the response operation data to generate a dialogue response template.

[0142] As an example of the present invention, see Figure 3 In this example, step S3 includes:

[0143] Step S31: performing multi-level recursive parsing on the filtered enhanced signal based on the interactive speech semantic structure to obtain a semantic representation matrix;

[0144] In this embodiment, a multi-level recursive analysis is performed on the filtered enhanced signal based on the semantic structure of the interactive speech, and a multi-level recursive neural network (RNN) is used to recursively analyze the semantic units in the interactive speech signal layer by layer to form a semantic representation matrix. The semantic structure of the interactive speech is recursively processed layer by layer according to the levels of words, phrases, sentences, etc., so that each layer of the recursive neural network can capture the semantic relationship and meaning of the corresponding level, thereby generating a semantic representation matrix at each level. Each row in the semantic representation matrix represents the deep semantic information of a semantic unit of interactive speech. At the same time, each semantic unit is encoded in the matrix and its contextual relationship is recorded, so as to ensure that in the multi-level analysis process, the semantic representation matrix can fully express the multi-dimensional semantic content of the interactive speech signal.

[0145] Step S32: identifying the core semantic components of the semantic representation matrix to obtain the interactive speech meaning representation;

[0146] In this embodiment, the core semantic components of the semantic representation matrix are identified, and the semantic features in the matrix are extracted using a convolutional neural network (CNN). The representation matrix containing multi-level information is input into the feature extraction layer of the CNN so as to identify the core semantic components therefrom. The core semantic component extraction gradually narrows the scope of the semantic matrix through convolution kernel screening and pooling operations, and focuses on the part that expresses key information. The identified core semantic components represent the key semantic elements of the user's intention in the interactive speech, such as "query", "reservation" and other operation behaviors or objects. The meaning of the specific interactive speech is obtained by identifying the semantic representation matrix, which provides a specific semantic information basis for subsequent intent recognition.

[0147] Step S33: constructing an intention recognition model for the interactive speech meaning representation to generate a user intention recognition model;

[0148] In this embodiment, an intention recognition model is built for the representation of the meaning of the interactive speech, and a classifier model based on a long short-term memory network (LSTM) is used to build the intention recognition model. The LSTM network has the advantage of effectively capturing long-distance dependencies in time series, and can accurately identify the intention features of user speech in intention recognition. In the process of building the intention recognition model, the extracted semantic components are used as input training data for supervised learning training, so that the model can automatically classify and recognize the user intention of the interactive speech. After the intention recognition model is generated, it can be classified and processed based on the feature information of the user's interactive speech meaning representation. The generated user intention recognition model will become the main tool for the voice robot to analyze interactive intentions.

[0149] Step S34: performing interactive response deduction on the interactive speech meaning representation according to the user intention recognition model to generate a predicted interactive response plan;

[0150] In this embodiment, interactive response deduction is performed on the interactive speech meaning representation according to the user intention recognition model. Combined with the predictive ability of the user intention recognition model, deduction is performed through an interactive prediction algorithm such as a deduction model based on the Markov decision process. The Markov decision process can predict the response operation caused by the user intention based on the historical response pattern. The interactive response deduction combines semantic information for multi-step prediction processing to ensure that the predicted interactive response plan can cover different response options. During the deduction process, the output data of each response prediction will be used to evaluate the optimal response option, and finally generate a predicted interactive response plan for the interactive speech. The plan records the feasible response mode of the interactive speech generated by the system according to the intention recognition model and the predicted interaction flow.

[0151] Step S35: Perform initial response processing according to the predicted interactive response plan and record the response operation data; construct a dialogue response template for the response operation data to generate a dialogue response template.

[0152] In this embodiment, the initial response processing is performed according to the predicted interactive response plan and the response operation data is recorded. The response module of the interactive voice system is used to generate a response operation according to the predicted interactive response plan. In the actual response process, each content in the response plan is executed step by step in sequence, and the output result corresponding to each execution step is recorded in real time in the response data log. After the initial response processing is completed, the recorded response operation data is used as a reference for subsequent dialogue templates, and the response operation data is further templated. Cluster analysis is performed on multiple interaction record data to determine dialogue patterns with similar intentions, construct dialogue response templates, and generate dialogue response templates.

[0153] Preferably, step S4 comprises the following steps:

[0154] Step S41: performing semantic topic analysis on the interactive speech meaning representation to obtain core semantic categories;

[0155] Step S42: performing semantic space encoding on the core semantic categories to generate a mapping structure diagram;

[0156] Step S43: constructing node relationships for core semantic categories based on the mapping structure diagram to obtain a personalized semantic graph;

[0157] Step S44: performing interactive context simulation on the response operation data to obtain a current simulated interactive context;

[0158] Step S45: Based on the current simulation interaction context and the personalized semantic graph, interactive prediction is performed on the response operation data to generate a predicted user interaction dialogue.

[0159] In this embodiment, semantic topic analysis is performed on the meaning representation of the interactive speech, and the core semantic components in the filtered enhanced signal are analyzed using a semantic parsing tool to identify the main intention and information category expressed by the speech content. The semantic parsing tool locates the core semantic topic of the interactive content by analyzing the lexical structure, syntactic structure and semantic correlation between words in the interactive speech. The topic represents the main category of communication content in the interactive speech. The system maps the analysis results to core semantic categories so that subsequent steps can carry out specific analysis and processing of the category. For example, when a user asks a question about the weather, the semantic parsing tool will identify "weather" as a core semantic category and assign a corresponding label. The generated core semantic category is used as the input for the subsequent semantic space encoding step. The system performs semantic space encoding on the core semantic categories. Based on semantic classification and multidimensional space projection technology, the core semantic categories are projected into a structured semantic space. The system extracts the main features of the core semantic categories through the encoding module and generates a mapping structure diagram of the category using the spatial mapping algorithm. The mapping structure diagram encodes the core semantic categories with their potential contextual semantic categories and associated labels to achieve mutual position mapping and relationship expression between categories. The system constructs the spatial relationship of categories with the help of a high-dimensional semantic matrix. This relationship will show the position of the core semantic categories relative to other categories in the multidimensional space, thereby providing a data basis for the generation of personalized semantic maps for the core semantic categories. The node relationship of the core semantic categories is analyzed based on the mapping structure diagram. Construct, use graph theory algorithm to construct the node relationship network between semantic categories, the system regards each semantic category in the mapping structure diagram as a node, calculates the association weight and co-occurrence frequency between categories, and uses clustering algorithm and hierarchical aggregation method to construct the relationship between nodes to form a personalized semantic map. The map presents the semantic categories frequently used by users in voice interaction and their logical hierarchical relationships. By constructing the map, the system can capture the user's unique expression preferences and word usage patterns, and provide reference data for subsequent interactive context simulation and personalized response. For example, in the map, nodes such as "weather", "temperature", and "precipitation" are strongly connected, and interactive context simulation is performed on the response operation data. The simulation model module is used to calculate the current interactive context information and the voice data input by the user. The simulation model converts the user's historical interaction mode into simulation data through the analysis of multiple rounds of interaction records, and puts the response operation data into the simulation environment for context verification. The interaction model in the simulation environment can generate a representative interaction context based on the user's communication habits and expressions. The context data can reflect the user's emotional expression and information needs for the voice robot. The simulation results will be presented in the form of a simulated interaction context, providing a more intuitive contextual basis for the personalized semantic map. The response operation data is interactively predicted based on the current simulated interaction context and personalized semantic map. The regression analysis and prediction model in the machine learning algorithm are used to predict the user's future interaction content by combining the simulated interaction context and the semantic map.The system generates predicted user interaction dialogues by classifying, aggregating and probabilistically analyzing user voice data. The prediction model infers the semantic categories and semantic needs that users will mention in subsequent dialogues based on the semantic category node relationships in the personalized semantic graph. Through the execution of the prediction module, it lays the foundation for the personalized interaction response plan generated by the user.

[0160] Preferably, step S45 includes the following steps:

[0161] Perform feature vectorization processing on the current simulation interaction context to obtain a context feature vector;

[0162] Quantify the interaction tendency of the personalized graph according to the context feature vector and generate the user interaction tendency response;

[0163] Conduct interactive prediction modeling on personalized semantic graphs to generate interactive behavior prediction models;

[0164] The interactive behavior prediction model is used to perform multiple rounds of interactive deduction on the response operation data to obtain an interactive dialogue branch tree;

[0165] Based on the user's interaction tendency response, the interactive dialogue branch tree is predicted to have coherent interactive behaviors and generate predicted user interaction dialogues.

[0166] In this embodiment, feature vectorization is performed on the current simulated interaction context. Through feature extraction algorithms and natural language processing modules, information features and behavioral expressions in the current simulated interaction context are analyzed, and each context feature is encoded as vector data for numerical analysis and processing. During the feature vectorization process, deep learning models such as convolutional neural networks are used to perform multi-dimensional encoding of multiple features such as voice emotion, language style, and information density of the interaction context. Each feature vector is recorded as an independent element in the context feature matrix. Through feature vectorization processing, a quantitative description of the interaction context is achieved, which provides data support for the subsequent quantification of the interaction tendency of the user's personalized map. The personalized map is interactively analyzed according to the context feature vector. Quantification of mutual tendency: Through the quantitative analysis module, the relationship between the current context feature vector and the node of the user's personalized semantic map is calculated, and the structure of the personalized map is simplified by combining principal component analysis to highlight the user's interaction tendency in this context. Through the quantitative calculation of interaction tendency, the system can identify the language preference, information demand tendency and potential emotional flow shown by the user in the interaction. The distribution of each semantic node in the user's interaction tendency will be mapped to the user's interaction tendency response. The quantitative result presents the user's interaction tendency in this specific context as a numerical expression, and the personalized semantic map is interactively predicted and modeled. The multi-layer neural network model (MLP, Multi-Layer Perceptron) to predict interactive behavior, and iterate the user's interactive behavior pattern through supervised learning and unsupervised learning to identify the user's behavior and reaction path in different contexts. In the process of interactive prediction modeling, the system takes the nodes with higher user tendencies in the semantic map as the target, and generates the optimal parameters of the prediction model through the feedback information of historical interaction records and context feature vectors in the training data. The interactive behavior prediction model can predict the user's potential interactive behavior and needs based on the input interactive context information, and provide model support for multiple rounds of interactive deduction. Based on the user behavior tendency information and historical interaction records obtained in the prediction model, the user's interaction path is gradually generated in the form of interactive nodes. In each round of deduction, the system According to the prediction results output by the model, the interactive dialogue content is deduced into a binary branch structure. By analyzing the response operation data in each branch node, an interactive dialogue branch tree containing multiple user interactions is generated. The branch of each node represents the user's interactive selection path. The path in the branch structure is gradually expanded in a multi-round deduction method to complete the generation of the user's interactive prediction path tree in different contexts. The generated interactive dialogue branch tree contains multiple rounds of user interactions and their corresponding response operations. Based on the user's interactive tendency response, the interactive dialogue branch tree is predicted to have coherent interactive behavior. Through multiple interactive tendency matching and branch path calculation, the interactive tendency response data is matched with the node content of the dialogue branch tree to identify the user's main interactive preference. In the matching process,The system uses a matching function to calculate the similarity between the feature vector of the tendency response and the node content in the branch tree, and screens out the interaction paths with higher similarity. The system generates a coherent interaction behavior prediction path for the user by predicting similar paths, and finally predicts the user's interactive dialogue.

[0167] Preferably, step S5 comprises the following steps:

[0168] Step S51: extracting dialogue elements from the predicted user interactive dialogue to obtain an interactive dialogue element set;

[0169] Step S52: formulating a response plan for the interactive dialogue element set based on the dialogue response template to generate an interactive response plan;

[0170] Step S53: integrating the functional modules of the interactive response scheme to obtain an interactive response module chain;

[0171] Step S54: construct a response engine for the interactive response module chain to obtain an advanced response engine.

[0172] In this embodiment, the predicted user interaction dialogue is extracted for dialogue elements. The predicted user interaction dialogue text is parsed for vocabulary, syntax and semantic structure by natural language processing (NLP) technology. The key components such as intent, theme and emotion contained in the dialogue are marked as dialogue elements by using syntax analysis and dependency extraction methods. The Named Entity Recognition (NER) method based on the knowledge base is used to identify the dialogue elements. Entity Recognition) and keyword extraction technology are used to extract content closely related to user needs and form an interactive dialogue element set. Through a pre-set dialogue response template, the information in the element set is matched with the predefined response format in the template. The rule matching algorithm is used to classify and process different user intentions and topics. Combined with dialogue scenario analysis and contextual logic judgment, targeted response plans are generated for the content in the interactive element set. The generated response plan will cover the content of the user's question, the feedback content provided by the system and the corresponding operation steps. By formulating a detailed interactive response plan that meets user needs, the response process is converted into an operational step sequence. Through modular construction, the operation corresponding to each response step is decomposed into independent functional modules, and the modules are integrated according to the order and dependency relationship in the interactive response plan to generate a module chain containing various functional modules. By adopting a module integration framework (such as service bus ESB, Enterprise Service Bus, etc.), the system can generate a module chain containing various functional modules. Bus) is used to connect different response function modules to form a modular response chain for sequential execution. Each module in the module chain completes a specific interactive function, such as sentiment analysis module, information feedback module, knowledge graph query module, etc. The module chain will gradually realize all the response functions required for user interaction. The response engine is constructed for the interactive response module chain. The integrated module chain is embedded in the response engine framework through the response engine construction tool. The functional modules in the module chain are automatically managed through engine configuration and task scheduling logic to generate an advanced response engine. By adopting multi-threaded parallel processing or task priority scheduling algorithm, it is ensured that each module can respond quickly under specific conditions to achieve the goal of advanced response. The configuration of the response engine includes the priority setting of each module, the embedding of data flow and error handling mechanism.

[0173] Preferably, step S6 comprises the following steps:

[0174] Step S61: performing feature matching on the intelligent voice robot based on the advance response engine to obtain mutually adapted features;

[0175] Step S62: mapping device instructions to the intelligent voice robot according to the mutual adaptation characteristics to generate a device operation instruction set;

[0176] Step S63: Perform intelligent interactive response on the intelligent voice robot hardware according to the device operation instruction set to perform voice interaction analysis of the intelligent voice robot.

[0177] In this embodiment, the actual hardware functional features of the intelligent voice robot are compared in detail with the response feature set generated by the advance response engine, and the command requirements output by various response function modules and the robot hardware functions are matched one by one using a matching algorithm (such as a collaborative filtering algorithm) and a feature mapping table to ensure that the hardware system can accurately support the required functional requirements. According to the module chain features in the response engine, the input and output parameters of the robot hardware are adapted to the corresponding functional requirements one by one to obtain a mutually adapted feature data set including the robot function modules and the response modules. Through the device instruction mapping framework, a corresponding device operation instruction set is generated for the feature matching results, and an instruction mapping algorithm (such as a hash mapping algorithm Hash) is used. The input and output requirements in the mutual adaptation features are mapped to actual hardware instructions by the device operation instruction set, and the generated device operation instruction set contains specific instruction codes and parameter information for each physical component or operation module of the robot. Each instruction code represents a specific action performed by the robot (such as steering gear rotation, microphone activation, speaker volume adjustment, etc.). The corresponding instructions are verified and optimized through parameter transmission and data processing mechanisms to ensure the consistency and accuracy of robot instruction execution. The hardware implementation of the interactive operation is completed through the construction of the operation instruction set, so that the instruction generation meets the requirements of the interactive solution. According to the device operation instruction set, the intelligent voice robot hardware is intelligently interactively responded to to perform the voice interaction analysis of the intelligent voice robot. Through the robot control framework (such as the robot operating system ROS, Robot Operating System), the generated device operation instruction set is sent to each hardware module through the communication interface, and the real-time scheduling and parallel control of the device instruction set are realized through the operating system kernel. In the control process, according to the parameter setting of the instruction set, the response operation of each hardware component is gradually triggered through multi-thread management and signal synchronization mechanism. When performing interactive analysis, the robot is controlled to respond to the user's voice input through multi-modal perception modules such as speakers, microphones, and cameras according to the solution set by the advance response engine.

[0178] The present invention also provides a voice interaction analysis system of an intelligent voice robot, which is used to execute the voice interaction analysis method of the intelligent voice robot as described above. The voice interaction analysis system of the intelligent voice robot comprises:

[0179] The speech enhancement module is used to recognize the current user interaction speech; perform multi-scale noise decomposition on the current user interaction speech, and perform interval-scale filtering enhancement to obtain a filtering enhancement signal;

[0180] The semantic structure analysis module is used to segment the filtered enhanced signal into granularity and perform long-distance semantic analysis to generate the interactive speech semantic structure;

[0181] The initial response module is used to perform multi-level semantic recognition on the filtered enhanced signal based on the interactive speech semantic structure to obtain the interactive speech meaning representation; perform the initial dynamic response to the interactive speech meaning representation and record the response operation data, and at the same time build a dialogue response template;

[0182] The interactive prediction module is used to map the interactive speech meaning representation to an independent graph structure to obtain a personalized semantic graph; interactively predict the response operation data based on the personalized semantic graph to generate a predicted user interactive dialogue;

[0183] An advance response preparation module is used to prepare an advance response for the predicted user interaction dialogue based on the dialogue response template, and to construct a response engine to obtain an advance response engine;

[0184] The robot embedding module is used to perform adaptive collaborative fusion processing on the intelligent voice robot based on the advance response engine to perform voice interaction analysis of the intelligent voice robot.

[0185] The invention uses a speech enhancement module to effectively reduce background noise interference through multi-scale noise decomposition and filtering enhancement technology, ensures that the user interaction speech signal is clear and discernible, and significantly improves recognition accuracy. The semantic structure analysis module generates rich interactive speech semantic structures through granular segmentation and long-distance semantic analysis, and enhances the system's comprehensive understanding of user intentions. The initial response module realizes multi-level semantic recognition of the filter enhancement signal, can quickly feedback user needs, and records response operation data at the same time, improves the real-time and efficiency of the interaction. The interaction prediction module generates a personalized semantic map through independent graph structure mapping, and predicts responses based on user historical interaction data, enhances user participation, and improves satisfaction. The advance response preparation module prepares the predicted user interaction dialogue based on the dialogue response template, ensuring that the system can make a natural response in advance when the user does not clearly express his intention. Each module records and analyzes user interaction data to achieve adaptive learning. The system can gradually optimize the interaction strategy, improve the overall intelligence level, and adapt to changing user needs. The robot embedding module embeds the advance response engine into the intelligent voice robot to ensure efficient execution of voice interaction analysis in complex interaction scenarios, and improve overall performance and user experience.

[0186] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0187] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A voice interaction analysis method for an intelligent voice robot, characterized in that: The following steps are involved: Step S1: Identify the current user interaction voice; perform multi-scale noise decomposition on the current user interaction voice, and perform interval-scale filtering enhancement to obtain a filtering enhanced signal; Step S2: granularly segmenting the filtered enhanced signal and performing long-distance semantic analysis to generate an interactive speech semantic structure; Step S3: performing multi-level semantic recognition on the filtered enhanced signal based on the interactive speech semantic structure to obtain the interactive speech meaning representation; Make an initial dynamic response to the interactive speech meaning representation and record the response operation data, while building a dialogue response template; Step S4: Perform independent graph structure mapping on the interactive speech meaning representation to obtain a personalized semantic graph; perform interactive prediction on the response operation data according to the personalized semantic graph to generate a predicted user interactive dialogue; Step S5: preparing an advance response for the predicted user interaction dialogue based on the dialogue response template, and constructing a response engine to obtain an advance response engine; Step S6: Adaptively perform collaborative fusion processing on the intelligent voice robot based on the advance response engine to perform voice interaction analysis of the intelligent voice robot.

2. The voice interaction analysis method of the intelligent voice robot according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: recognizing the current user interaction speech; performing fast Fourier transform on the current user interaction speech to obtain the interactive speech frequency domain features; Step S12: performing frequency domain scale recognition on the interactive speech frequency domain features to generate a frequency scale set; Step S13: dividing the current user interaction speech into frequency intervals based on the frequency scale set to obtain multiple frequency sub-interval speech; Step S14: performing noise energy recognition on the frequency sub-interval speech based on a preset noise energy threshold, and generating noise features for each frequency interval; Step S15: estimating the noise intensity of the speech in the frequency sub-interval according to the noise characteristics of each frequency interval to obtain a noise distribution spectrum; Step S16: Calculating the filter coefficients of each frequency sub-interval for the speech using the noise distribution spectrum to obtain a set of optimal filter coefficients; Step S17: performing filtering enhancement in each frequency sub-interval on the frequency sub-interval speech based on the preferred filter coefficient set to obtain a filtered enhanced signal.

3. The voice interaction analysis method of the intelligent voice robot according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: performing phoneme boundary detection on the filtered enhanced signal to obtain a phoneme sequence table; Step S22: segmenting the filtered enhanced signal into syntactic units according to the phoneme sequence table to generate multi-level language units; Step S23: extracting semantic features from the multi-level language units to obtain semantic features of the language units; Step S24: performing long-range association semantic inference on the semantic features of the language unit to generate a semantic association network; Step S25: constructing a hierarchical structure of the multi-level language units based on the semantic association network to obtain an interactive speech semantic structure.

4. The voice interaction analysis method of the intelligent voice robot according to claim 3 is characterized in that: Step S24 includes the following steps: Contextually encode the semantic features of the language unit to obtain a contextual semantic matrix; Based on the contextual semantic matrix, long-range dependency recognition is performed on the semantic features of the language unit to obtain a dependency tree; Track the inter-sentence associations in the dependency tree to obtain the topic link relationship; Simulate the topic link relationship and semantic propagation to generate simulated semantic flow data; Based on the simulated semantic flow data, the cross-sentence relationship of the semantic features of the language unit is inferred to generate a semantic association network.

5. The voice interaction analysis method of the intelligent voice robot according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: performing multi-level recursive parsing on the filtered enhanced signal based on the interactive speech semantic structure to obtain a semantic representation matrix; Step S32: identifying the core semantic components of the semantic representation matrix to obtain the interactive speech meaning representation; Step S33: constructing an intention recognition model for the interactive speech meaning representation to generate a user intention recognition model; Step S34: performing interactive response deduction on the interactive speech meaning representation according to the user intention recognition model to generate a predicted interactive response plan; Step S35: Perform initial response processing according to the predicted interactive response plan and record the response operation data; construct a dialogue response template for the response operation data to generate a dialogue response template.

6. The voice interaction analysis method of the intelligent voice robot according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: performing semantic topic analysis on the interactive speech meaning representation to obtain core semantic categories; Step S42: performing semantic space encoding on the core semantic categories to generate a mapping structure diagram; Step S43: constructing node relationships for core semantic categories based on the mapping structure diagram to obtain a personalized semantic graph; Step S44: performing interactive context simulation on the response operation data to obtain a current simulated interactive context; Step S45: Based on the current simulation interaction context and the personalized semantic graph, interactive prediction is performed on the response operation data to generate a predicted user interaction dialogue.

7. The method for analyzing speech interaction of an intelligent speech robot according to claim 6, characterized in that: Step S45 includes the following steps: Perform feature vectorization processing on the current simulation interaction context to obtain a context feature vector; Quantify the interaction tendency of the personalized graph according to the context feature vector and generate the user interaction tendency response; Conduct interactive prediction modeling on personalized semantic graphs to generate interactive behavior prediction models; The interactive behavior prediction model is used to perform multiple rounds of interactive deduction on the response operation data to obtain an interactive dialogue branch tree; Based on the user's interaction tendency response, the interactive dialogue branch tree is predicted to have coherent interactive behaviors and generate predicted user interaction dialogues.

8. The voice interaction analysis method of the intelligent voice robot according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: extracting dialogue elements from the predicted user interactive dialogue to obtain an interactive dialogue element set; Step S52: formulating a response plan for the interactive dialogue element set based on the dialogue response template to generate an interactive response plan; Step S53: integrating the functional modules of the interactive response scheme to obtain an interactive response module chain; Step S54: construct a response engine for the interactive response module chain to obtain an advanced response engine.

9. The voice interaction analysis method of the intelligent voice robot according to claim 1, characterized in that: Step S6 includes the following steps: Step S61: performing feature matching on the intelligent voice robot based on the advance response engine to obtain mutually adapted features; Step S62: mapping device instructions to the intelligent voice robot according to the mutual adaptation characteristics to generate a device operation instruction set; Step S63: Perform intelligent interactive response on the intelligent voice robot hardware according to the device operation instruction set to perform voice interaction analysis of the intelligent voice robot.

10. A voice interaction analysis system for an intelligent voice robot, characterized in that: The method for analyzing the speech interaction of the intelligent speech robot according to claim 1 is used to implement the speech interaction analysis method of the intelligent speech robot, and the speech interaction analysis system of the intelligent speech robot comprises: The speech enhancement module is used to recognize the current user interaction speech; perform multi-scale noise decomposition on the current user interaction speech, and perform interval-scale filtering enhancement to obtain a filtering enhancement signal; The semantic structure analysis module is used to segment the filtered enhanced signal into granularity and perform long-distance semantic analysis to generate the interactive speech semantic structure; The initial response module is used to perform multi-level semantic recognition on the filtered enhanced signal based on the interactive speech semantic structure to obtain the interactive speech meaning representation; perform the initial dynamic response to the interactive speech meaning representation and record the response operation data, and at the same time build a dialogue response template; The interactive prediction module is used to map the interactive speech meaning representation to an independent graph structure to obtain a personalized semantic graph; Interactively predict the response operation data based on the personalized semantic graph to generate predicted user interaction dialogues; An advance response preparation module is used to prepare an advance response for the predicted user interaction dialogue based on the dialogue response template, and to construct a response engine to obtain an advance response engine; The robot embedding module is used to perform adaptive collaborative fusion processing on the intelligent voice robot based on the advance response engine to perform voice interaction analysis of the intelligent voice robot.

Images