Intelligent dialogue scenario analysis method based on AI large model

Abstract

The application relates to the technical field of artificial intelligence, in particular to an intelligent dialogue scene analysis method based on an AI large model, which comprises the following steps: recognizing background noise and speech accent in user voice data, and converting the voice data into initial text information; standardizing and adjusting the initial text information based on a standard language database to obtain adjusted text information; analyzing the voice data to obtain emotional features, analyzing the adjusted text information to obtain lexical features, and determining user dialogue features according to the emotional features and the lexical features; constructing a dialogue scene model according to the dialogue features; generating initial answer information based on the dialogue scene model; receiving feedback data of the initial answer information from the user, and optimizing the dialogue scene model according to the feedback data, so that the target answer information is better than the initial answer information. The application improves the accuracy of intelligent dialogue results of the AI large model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to an intelligent dialogue scene analysis method based on an AI large model. BACKGROUND

[0002] With the development of artificial intelligence and Internet information technology, intelligent dialogue systems have been widely used in intelligent customer service, intelligent sound boxes, intelligent vehicles, intelligent Internet of Things, intelligent industrial services and many other scenarios.

[0003] A patent document with the Chinese patent application publication number CN116662511A discloses an intelligent dialogue scene analysis method using AI model recognition and an AI dialogue system. The method obtains service interaction dialogue text generated by a target user in a current intelligent dialogue scene, and uses an AI model recognition strategy to mine interest from the service interaction dialogue text, generating at least one interest dialogue sentence clustering cluster. The interest label corresponding to the interest dialogue sentence clustering cluster with an interest dialogue sentence number greater than a set number is determined as a target interest label. Based on the target interest label, the corresponding dialogue feedback page content is loaded to the intelligent dialogue scene page of the target user, thereby combining the interest dialogue sentence number of different interest labels in the intelligent dialogue scene to mine the interest of the target user.

[0004] Although the AI model recognition strategy can mine the interest of the user in the prior art, it is often difficult to accurately convert the voice for different user accents, resulting in loss or misunderstanding of dialogue information, and thus accurate answers cannot be given. SUMMARY

[0005] Therefore, the present application provides an intelligent dialogue scene analysis method based on an AI large model, which can solve the problem of inaccurate AI large model intelligent dialogue results by in-depth analysis of user voice accents and text information combined with the characteristics of the dialogue scene.

[0006] To achieve the above-mentioned purpose, the present application provides an intelligent dialogue scene analysis method based on an AI large model, which comprises:

[0007] Receiving initial voice data of a user, analyzing the initial voice data, and identifying background noise and voice accents according to the analysis results;

[0008] Processing the background noise to obtain processed voice data, analyzing the voice accents, and determining calibration parameters according to the analysis results;

[0009] Adjust initial conversion parameters in the initial speech conversion model based on the calibration parameters to obtain an adjusted speech conversion model, and convert the processed speech data based on the adjusted speech conversion model to obtain initial text information;

[0010] Standardize the initial text information based on a standard language database to obtain standard text information;

[0011] Analyze the standard text information, determine lexical features according to the text information analysis result, analyze the processed speech data, determine emotional features according to the speech data analysis result, and determine dialogue features based on the lexical features and the emotional features;

[0012] Analyze the dialogue features, determine scene labels according to the dialogue feature analysis result, and construct an initial dialogue scene model based on the scene labels;

[0013] Generate initial answer information based on the initial dialogue scene model and the processed speech data;

[0014] Collect feedback information on the initial answer information, judge the degree of model answer deviation based on the feedback information, and adjust the adjusted speech conversion model or the initial dialogue scene model based on the judgment result, so that the target answer information obtained after adjustment is better than the initial answer information.

[0015] Further, the step of identifying background noise and speech accent according to the analysis result comprises:

[0016] Identify the energy feature, prosodic feature and phoneme feature of the initial speech data;

[0017] Identify the background noise based on the energy feature;

[0018] Identify the speech accent based on the prosodic feature and the phoneme feature.

[0019] Further, the step of determining calibration parameters according to the analysis result comprises:

[0020] Determine the deviation value between the speech data and the standard speech based on the prosodic feature and the phoneme feature;

[0021] Determine the calibration parameters based on the deviation value.

[0022] Further, the step of standardizing the initial text information based on a standard language database comprises:

[0023] The standard words in the standard language database are compared with the initial text information one by one to determine abnormal words, the similarity between the abnormal words and the standard words is calculated, and the abnormal words with a similarity greater than a preset word similarity are replaced with the standard words;

[0024] The actual grammar rules in the initial text information are identified, the actual grammar rules are compared with the standard grammar rules in the standard language database, and errors in grammar are determined and corrected according to the comparison result.

[0025] Further, the step of determining the word feature according to the text information analysis result comprises:

[0026] The standard text information is subjected to word segmentation processing to obtain a plurality of segmented words.

[0027] The frequency of occurrence of the segmented words is identified, and the semantic feature corresponding to the segmented word with the highest frequency of occurrence is determined as the word feature.

[0028] Further, the step of determining the emotional feature according to the speech data analysis result comprises:

[0029] The timbre feature and the speech rate feature of the speech data are extracted.

[0030] A speech feature vector is constructed based on the timbre feature and the speech rate feature.

[0031] The emotional feature is determined based on the speech feature vector and an emotional analysis model.

[0032] Further, the step of determining the dialogue feature based on the word feature and the emotional feature comprises:

[0033] A feature vector is constructed based on the word feature and the emotional feature.

[0034] The dialogue feature is determined based on the feature vector.

[0035] Further, the step of establishing an initial dialogue scene model according to the dialogue feature analysis result comprises:

[0036] The dialogue feature is matched with a preset scene classification model to determine the scene label.

[0037] The initial dialogue scene model is constructed according to the scene label and a preset scene template.

[0038] Further, the step of generating initial answer information based on the initial dialogue scene model and the processed speech data comprises:

[0039] The target dialogue scene to which the processed speech data belongs is determined based on the initial dialogue scene model.

[0040] generate initial answer information according to the target dialogue scene and a preset scene reply strategy;

[0041] The preset scene reply strategy comprises a reply template and reply content in different dialogue scenes.

[0042] Further, the step of judging the deviation degree of the model answer based on the feedback information comprises:

[0043] collecting feedback data of the user for the initial answer information, including a satisfaction score and a reply content evaluation;

[0044] extracting keywords in the reply content evaluation, analyzing the keywords, and determining a text satisfaction degree according to the analysis result;

[0045] determining an actual satisfaction degree according to the satisfaction score and the text satisfaction degree;

[0046] determining the deviation degree based on the actual satisfaction degree and a preset satisfaction degree.

[0047] Compared with the prior art, the beneficial effects of the present application are that, through analysis of initial voice data, background noise can be accurately recognized and processed, effectively reducing the interference of noise on the accuracy of voice recognition, adjusting the parameters of the voice conversion model according to the analysis result of the voice accent, so that the model can better adapt to voice recognition of different accents, improve the recognition accuracy, and through standardization adjustment of the initial text information by the standard language database, the output text information is more standardized and accurate, the analysis of the lexical features and emotional features helps to better understand the user's intention and emotional state, so that the intelligent dialogue is more natural and smooth, based on the analysis of the dialogue features, the scene label can be automatically determined, and the initial dialogue scene model is constructed, so that the system can provide personalized dialogue service according to different scenes, through optimization of the scene model, the user's demand in different scenes can be more accurately recognized and understood, and the pertinence and effectiveness of the dialogue are improved, through collection of the feedback information for the initial answer information, and according to the feedback information, the deviation degree of the model answer is judged, which helps the system to understand the user feedback in real time, find potential problems, and based on the judgment result, the voice conversion model or the initial dialogue scene model is adjusted, so that the system can continuously learn and optimize, and the accuracy of the dialogue system is continuously improved.

[0048] Especially, the energy feature reflects the distribution and change of energy in the speech signal, and the background noise is accurately identified by identifying the energy feature, thereby effectively reducing the influence of noise on subsequent speech recognition and processing. The prosody feature and the phoneme feature reflect the pronunciation characteristics and accent differences of the speech. Based on these features, the accent recognition can more accurately determine the user's accent type, providing an accurate basis for the adjustment of the subsequent speech conversion model. By identifying the energy feature to identify the background noise, the system can adapt to different noise environments and improve the robustness of the system. Based on the prosody feature and the phoneme feature, the accent recognition can make the system better adapt to users with different accents and improve the adaptability and robustness of the system, providing more accurate and personalized services.

[0049] Especially, by analyzing the prosody feature and the phoneme feature of the speech data and comparing them with the standard speech, the deviation value between the speech data and the standard speech is accurately determined. Based on this deviation value, the parameters of the initial speech conversion model are accurately adjusted, so that the model can better adapt to and recognize different speech data, thereby improving the accuracy of speech recognition.

[0050] Especially, by comparing the standard words in the standard language database with the words in the initial text information one by one, abnormal words can be identified. By calculating the similarity and replacing abnormal words with a similarity higher than a preset threshold, the number of error words in the text is significantly reduced, and the accuracy of the text is improved. The replacement of abnormal words helps to ensure that the text follows the standard language specifications, making the text clearer and more accurate. By identifying the actual grammar rules in the initial text information and comparing them with the standard grammar rules in the standard language database, grammar errors can be accurately found and corrected, which helps to optimize the structure of the text and improve the grammatical correctness of the text. The initial text information is comprehensively standardized and adjusted, effectively improving the overall quality of the text and improving the processing efficiency and accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 The flowchart of the intelligent dialogue scene analysis method based on the AI large model provided by the embodiment of the application;

[0052] Figure 2 The flowchart of identifying background noise and speech accent in the intelligent dialogue scene analysis method based on the AI large model provided by the embodiment of the application;

[0053] Figure 3 The flowchart of standardizing the initial text information in the intelligent dialogue scene analysis method based on the AI large model provided by the embodiment of the application;

[0054] Figure 4 The flowchart of determining the word feature in the intelligent dialogue scene analysis method based on the AI large model provided by the embodiment of the application. DETAILED DESCRIPTION

[0055] In order to make the objects and advantages of the present application clearer, the following further describes the present application with reference to examples; it should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0056] The preferred embodiments of the present application are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that the embodiments are only used to explain the technical principles of the present application and are not intended to limit the protection scope of the present application.

[0057] It should be noted that, in the description of the present application, the terms indicating the direction or positional relationship such as "upper", "lower", "left", "right", "inner", "outer" and the like are based on the direction or positional relationship shown in the drawings, which is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application.

[0058] In addition, it should also be noted that, in the description of the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connection" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through an intermediate medium, or internal communication of two elements. Those skilled in the art can understand the specific meaning of the above terms in the present application according to the specific circumstances.

[0059] Please refer to Figure 1 The embodiment of the present application provides an intelligent dialogue scene analysis method based on an AI large model, which comprises the following steps:

[0060] In step S100, initial voice data of a user is received, the initial voice data is analyzed, background noise and voice accent are recognized according to the analysis result;

[0061] In step S200, the background noise is processed to obtain processed voice data, the voice accent is analyzed, and calibration parameters are determined according to the analysis result;

[0062] In step S300, the initial conversion parameters in the initial voice conversion model are adjusted based on the calibration parameters to obtain an adjusted voice conversion model, the processed voice data is converted based on the adjusted voice conversion model to obtain initial text information;

[0063] In step S400, the initial text information is standardized and adjusted based on a standard language database to obtain standard text information;

[0064] Step S500, analyzing the standard text information, determining a lexical feature according to an analysis result of the text information, analyzing the processing voice data, determining an emotional feature according to an analysis result of the voice data, and determining a dialogue feature based on the lexical feature and the emotional feature;

[0065] Step S600, analyzing the dialogue feature, determining a scene label according to an analysis result of the dialogue feature, and constructing an initial dialogue scene model based on the scene label;

[0066] Step S700, generating initial answer information based on the initial dialogue scene model and the processing voice data;

[0067] Step S800, collecting feedback information on the initial answer information, judging a model answer deviation degree based on the feedback information, and adjusting the voice conversion model or the initial dialogue scene model based on a judgment result, so that target answer information obtained after adjustment is better than the initial answer information.

[0068] Specifically, through analysis of initial voice data, the embodiment of the application can accurately identify and process background noise, effectively reduce the interference of noise on the accuracy of voice recognition, adjust the parameters of the voice conversion model according to the analysis result of the voice accent, so that the model can better adapt to voice recognition of different accents, improve the recognition accuracy, standardize the initial text information through the standard language database, ensure that the output text information is more standardized and accurate, analyze the lexical feature and the emotional feature, which helps to better understand the user's intention and emotional state, so that the intelligent dialogue is more natural and smooth, based on the analysis of the dialogue feature, the scene label can be automatically determined, and the initial dialogue scene model can be constructed, so that the system can provide personalized dialogue service according to different scenes, through optimization of the scene model, the user's demand in different scenes can be more accurately recognized and understood, the pertinence and effectiveness of the dialogue are improved, through collection of feedback information on the initial answer information, and according to the feedback information, the model answer deviation degree is judged, which helps the system to understand the user feedback in real time, find potential problems, and adjust the voice conversion model or the initial dialogue scene model based on the judgment result, realize the continuous learning and optimization of the system, and continuously improve the accuracy of the dialogue system.

[0069] Referring to Figure 2 As shown in the figure, the step of identifying the background noise and the voice accent according to the analysis result comprises:

[0070] Step S110, identifying energy features, prosody features and phoneme features of the initial voice data;

[0071] Step S120, identifying the background noise based on the energy features;

[0072] Step S130, identifying the speech accent based on the prosody feature and the phoneme feature.

[0073] Specifically, the step of identifying the energy feature, the prosody feature and the phoneme feature of the initial speech data in the embodiment of the present application comprises:

[0074] Converting the initial speech data into an initial digital signal, pre-processing the initial digital signal, segmenting the pre-processed speech signal, obtaining a plurality of frames, calculating the sum of squares of a plurality of sampling points in any frame, and obtaining the energy feature;

[0075] Extracting the fundamental frequency of the initial speech data by the autocorrelation method, and determining the prosody feature based on the changes of a plurality of fundamental frequencies;

[0076] Converting the initial speech data into a phoneme sequence by a phoneme recognition algorithm, and extracting the phoneme feature based on the phoneme sequence.

[0077] Specifically, the pre-processing of the initial digital signal in the embodiment of the present application can include filtering, denoising, pre-emphasis and the like to improve the quality of the signal.

[0078] The segmentation of the pre-processed speech signal is usually based on a length of 20-30 milliseconds.

[0079] Specifically, the step of identifying the background noise based on the energy feature in the embodiment of the present application comprises:

[0080] Determining the frames corresponding to a greater preset energy based on the energy feature, and taking the initial speech data corresponding to a plurality of frames as the background noise;

[0081] Suppressing the background noise based on a noise suppression algorithm.

[0082] Specifically, the embodiment of the present application reflects the distribution and change of energy in the speech signal through the energy feature, accurately identifies the background noise through the energy feature, thereby effectively reducing the influence of noise on subsequent speech recognition and processing, and reflects the pronunciation characteristics and accent differences of speech through the prosody feature and the phoneme feature. The accent recognition based on these features can more accurately determine the accent type of the user, provide an accurate basis for the adjustment of the subsequent speech conversion model, identify the background noise through the energy feature, so that the system can adapt to different noise environments, improve the robustness of the system, and perform accent recognition based on the prosody feature and the phoneme feature, so that the system can better adapt to users with different accents, improve the adaptability and robustness of the system, and provide more accurate and personalized services.

[0083] Specifically, the step of determining the calibration parameter according to the analysis result comprises:

[0084] determine a deviation value between the speech data and the standard speech based on the prosody features and the phoneme features;

[0085] determine the calibration parameter based on the deviation value.

[0086] Specifically, the step of calculating the deviation value includes:

[0087] performing similarity comparison between the prosody features and standard prosody features of the standard speech to obtain prosody feature similarity;

[0088] performing similarity comparison between the phoneme features and standard phoneme features of the standard speech to obtain phoneme feature similarity;

[0089] calculating total similarity based on the prosody feature similarity and the phoneme feature similarity;

[0090] subtracting the total similarity from the complete similarity corresponding to the complete similarity to obtain the deviation value.

[0091] Specifically, the calibration parameter is the deviation value divided by the complete similarity.

[0092] Specifically, the standard speech is a speech sample without accent or with standard pronunciation.

[0093] Specifically, the speech processing technology such as Dynamic Time Warping (DTW) or acoustic model can be used to calculate the similarity between the input speech and the standard speech.

[0094] Specifically, the prosody features and the phoneme features of the speech data are analyzed and compared with the standard speech to accurately determine the deviation value between the speech data and the standard speech. Based on the deviation value, the parameters of the initial speech conversion model are accurately adjusted so that the model can better adapt to and recognize different speech data, thereby improving the accuracy of speech recognition.

[0095] Referring to Figure 3 The step of standardizing and adjusting the initial character information based on the standard language database includes:

[0096] In step S410, the standard words in the standard language database are compared with the initial character information one by one to determine abnormal words, the similarity between the abnormal words and the standard words is calculated, and the abnormal words with a similarity greater than a preset word similarity are replaced with standard words.

[0097] Step S420, identifying the actual grammar rule in the initial character information, comparing the actual grammar rule with the standard grammar rule in the standard language database, determining the error grammar and correcting according to the comparison result.

[0098] Specifically, the grammar error in the initial character information is corrected, which may include tense consistency, subject-verb consistency, and use of articles.

[0099] Specifically, the preset vocabulary similarity is 80%, and based on the historical comparison data, it is determined whether the abnormal vocabulary needs to be replaced by the standard vocabulary when the preset vocabulary similarity is set to 80%, so as to reduce the subsequent influence.

[0100] Specifically, the embodiment of the present application can identify abnormal vocabulary by comparing the standard vocabulary in the standard language database with the vocabulary in the initial character information one by one, significantly reduce the error vocabulary in the text by calculating the similarity and replacing the abnormal vocabulary with a similarity higher than the preset threshold, improve the accuracy of the text, and the replacement of the abnormal vocabulary helps to ensure that the text follows the standard language specification, so that the text is more clear and accurate. By identifying the actual grammar rule in the initial character information and comparing it with the standard grammar rule in the standard language database, the grammar error can be accurately found and corrected, which helps to optimize the structure of the text and improve the grammatical correctness of the text. The initial character information is comprehensively standardized and adjusted, which effectively improves the overall quality of the text and improves the processing efficiency and accuracy.

[0101] Referring to Figure 4 As shown, the step of determining the vocabulary feature according to the character information analysis result comprises:

[0102] Step S510, performing word segmentation processing on the standard character information to obtain a plurality of segmented words;

[0103] Step S520, identifying the frequency of the segmented words, and determining the semantic feature corresponding to the segmented word with the highest frequency as the vocabulary feature.

[0104] Specifically, the semantic feature can be realized by querying a dictionary, using a word vector (such as Word2Vec, GloVe, etc.) or combining context information.

[0105] Specifically, the text is decomposed into meaningful word units by word segmentation processing, which helps to more accurately capture the key information in the text, and the segmented word with the highest frequency is selected as the vocabulary feature, which can highlight the core vocabulary in the text, thereby improving the accuracy of subsequent text analysis. The model trained based on these features can better adapt to different text data and improve the generalization performance of the model.

[0106] Specifically, the step of determining the emotional feature according to the voice data analysis result comprises:

[0107] extracting a timbre feature and a speech rate feature of the voice data;

[0108] constructing a voice feature vector based on the timbre feature and the speech rate feature;

[0109] determining the emotional feature based on the voice feature vector and an emotional analysis model.

[0110] Specifically, the emotional analysis model of the embodiment of the present application is constructed based on a plurality of voice feature vectors and corresponding emotional features, and the emotional analysis model is generated by training the plurality of voice feature vectors and corresponding emotional features through a machine learning algorithm. The algorithm can classify voice data and determine its emotional label, such as positive, negative, neutral, etc.

[0111] Specifically, the timbre feature can include pitch, timbre loudness, etc., and can be extracted using methods such as short-time Fourier transform (STFT), mel-frequency cepstral coefficient (MFCC), etc. The speech rate feature can include speech rate fast or slow, pause frequency, etc., and can be obtained by calculating the number of syllables or words per unit time in the voice signal.

[0112] Specifically, the embodiment of the present application captures key information closely related to the emotional state in the voice signal by extracting the timbre feature and the speech rate feature, thereby improving the accuracy of emotion recognition. The use of the emotional analysis model further classifies and identifies the extracted features, improving the precision and efficiency of emotion recognition. By adjusting the model parameters and feature extraction methods, different application scenarios can be flexibly adapted.

[0113] Specifically, the step of determining the dialogue feature based on the lexical feature and the emotional feature comprises:

[0114] constructing a feature vector based on the lexical feature and the emotional feature;

[0115] determining the dialogue feature based on the feature vector.

[0116] Specifically, for the lexical feature, methods such as TF-IDF (term frequency-inverse document frequency), word embedding (such as Word2Vec), etc. are used to convert the words in the text into numerical vectors. For example, the word vectors obtained by using the Word2Vec model are used to represent each segmented word, and these word vectors are aggregated (such as averaging, summing, etc.) to form a vector representation of the lexical feature.

[0117] Specifically, for emotional features, the embodiment of the present application uses an emotional analysis model (such as a deep learning-based model) to perform emotional classification on the text, and converts the classification results (such as positive, negative, neutral, etc.) into numerical representations, which are directly used as vectors of emotional features, or are further encoded (such as independent encoding, embedding encoding, etc.) to adapt to the construction of feature vectors.

[0118] Specifically, based on the constructed feature vectors, various machine learning methods (such as classification, clustering, regression, etc.) are used to determine the features of the dialogue, for example, if our goal is to identify the topic of the dialogue, we can train a classifier to predict the topic category of the dialogue according to the feature vectors.

[0119] Specifically, the embodiment of the present application combines lexical features and emotional features to more comprehensively capture information in the dialogue, thereby improving the accuracy of dialogue understanding. The combination of lexical features and emotional features enables more accurate understanding of the intent and meaning of the dialogue.

[0120] Specifically, the step of establishing an initial dialogue scene model according to the dialogue feature analysis result includes:

[0121] Based on the dialogue features and a preset scene classification model, the scene label is determined;

[0122] According to the scene label and a preset scene template, the initial dialogue scene model is constructed.

[0123] Specifically, the preset scene classification model of the embodiment of the present application is a model obtained by training a large amount of scene data through a machine learning algorithm. The model can identify the corresponding scene category according to the input dialogue features. In the matching process, the embodiment of the present application calculates the similarity of the dialogue features and each scene category in the preset scene classification model, and selects the scene category with the highest similarity as the scene label.

[0124] Specifically, by matching the dialogue features with the preset scene classification model, the embodiment of the present application can accurately identify the scene category of the dialogue, providing an important basis for subsequent dialogue understanding and processing. Using the preset scene template to construct the initial dialogue scene model can quickly generate dialogue content that meets the scene requirements, improving the processing efficiency and accuracy of the dialogue system. The establishment of the initial dialogue scene model provides a basic framework for subsequent dialogue generation and response, which helps to achieve a more natural and smooth dialogue interaction experience.

[0125] Specifically, the step of generating initial answer information based on the initial dialogue scene model and the processed voice data includes:

[0126] determine a target dialogue scene to which the processed voice data belongs based on the initial dialogue scene model;

[0127] generate initial answer information according to the target dialogue scene and a preset scene reply strategy;

[0128] The preset scene reply strategy comprises a reply template and reply content in different dialogue scenes.

[0129] Specifically, the preset scene reply strategy can comprise some common dialogue scenes and corresponding reply templates and reply content, for example, in a shopping scene, the reply template can comprise product introduction, price inquiry, purchase suggestion, etc., and the reply content is filled according to specific product information and user demand.

[0130] Specifically, the embodiment of the present application utilizes the initial dialogue scene model to perform scene recognition on the processed voice data, determines the target dialogue scene to which the voice data belongs, can quickly and accurately identify the scene category to which the voice data belongs, and after determining the target dialogue scene, generates initial answer information according to the preset scene reply strategy, thereby providing a basis for subsequent dialogue generation and response, and at the same time, since the preset scene reply strategy is configurable, it can be adjusted and optimized according to actual needs, thereby adapting to different application scenarios and needs.

[0131] Specifically, the step of determining the degree of reply deviation of the model based on the feedback information comprises:

[0132] Collecting feedback data of the user for the initial answer information, comprising a satisfaction score and a reply content evaluation;

[0133] Extracting keywords in the reply content evaluation, analyzing the keywords, and determining the text satisfaction degree according to the analysis result;

[0134] Determining the actual satisfaction degree according to the satisfaction score and the text satisfaction degree;

[0135] Determining the degree of deviation based on the actual satisfaction degree and a preset satisfaction degree.

[0136] Specifically, the preset satisfaction degree is a satisfaction threshold set according to historical data.

[0137] Specifically, the embodiment of the present application collects feedback data of the user for the initial answer information, understands the satisfaction degree and specific evaluation of the user for the model reply, extracts keywords in the reply content evaluation, and performs in-depth analysis, which is helpful to find specific opinions and suggestions of the user for the model reply, determines the text satisfaction degree, i.e., the satisfaction degree of the user for the reply content of the model, by analyzing the keywords, and further combines the satisfaction score to comprehensively obtain the actual satisfaction degree, thereby more comprehensively evaluating the quality of the model reply.

[0138] Specifically, the step of adjusting the speech conversion model or the initial dialogue scene model based on the judgment result comprises:

[0139] If the offset degree is greater than the preset offset degree, the speech conversion model is adjusted, including optimizing parameters of a speech-to-text algorithm or updating training data of the speech conversion model;

[0140] If the offset degree is less than or equal to the preset offset degree, the initial dialogue scene model is adjusted, including adjusting parameters of a scene classification model or updating a scene template and a preset reply strategy.

[0141] The preset reply strategy comprises a length limit of an answer and a keyword replacement rule.

[0142] Specifically, when the answer is too long, the length limit is used for trimming, the key information points in the answer are intelligently analyzed, the most critical part for understanding the question is preferentially retained, and redundant or secondary information is removed, and the preset length is the average of historical answer lengths; when some words in the answer are inaccurate, the keyword replacement rule is used for replacement, the keyword replacement rule is to establish a keyword replacement list, and the incorrect forms and corresponding correct forms are listed, when the system detects that the incorrect forms appear in the answer, the replacement is automatically performed, so as to ensure the accuracy and authority of the answer, and a context analysis mechanism is also integrated, the system can replace the keyword more accurately according to the overall content and context of the answer.

[0143] Specifically, the embodiment of the application adopts different adjustment strategies according to different offset degrees, when the offset degree is greater than the preset offset degree, it indicates that there is a large error in the process of converting speech into text by the speech conversion model, which leads to a large deviation of the selected dialogue features, thereby causing a large error in the construction of the initial dialogue scene model, so that the initial answer information generated has a large deviation from the real intention of the user, the speech conversion model is adjusted, including optimizing parameters of a speech-to-text algorithm or updating training data of the speech conversion model, improving the accuracy of speech conversion, thereby improving the construction accuracy of the initial dialogue scene model, so that the target answer information obtained is better than the initial answer information.

[0144] When the offset degree is less than the preset offset degree, the initial dialogue scene model is adjusted, including adjusting parameters of a scene classification model or updating a scene template and a preset reply strategy, so as to improve the accuracy of scene recognition and the effectiveness of the reply strategy, further reduce the degree of answer offset, and make the target answer information obtained better than the initial answer information.

[0145] So far, the technical solutions of the present application have been described in combination with the preferred embodiments shown in the drawings, but it is easy for those skilled in the art to understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical solutions after the changes or replacements will all fall within the protection scope of the present application.

[0146] The above description is only the preferred embodiments of the present application and is not intended to limit the present application; for those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1.An AI large model-based intelligent dialogue scene analysis method, characterized in that, The method comprises the following steps: receiving initial voice data of a user, analyzing the initial voice data, and identifying background noise and voice accent based on the analysis result; processing the background noise to obtain processed voice data, analyzing the voice accent, and determining calibration parameters based on the analysis result; adjusting initial conversion parameters in an initial voice conversion model based on the calibration parameters to obtain an adjusted voice conversion model, and converting and processing the processed voice data based on the adjusted voice conversion model to obtain initial text information; standardizing and adjusting the initial text information based on a standard language database to obtain standard text information; analyzing the standard text information to determine lexical features based on the text information analysis result, analyzing the processed voice data to determine emotional features based on the voice data analysis result, and determining dialogue features based on the lexical features and the emotional features; analyzing the dialogue features to determine scene labels based on the dialogue feature analysis result, and constructing an initial dialogue scene model based on the scene labels; generating initial answer information based on the initial dialogue scene model and the processed voice data; collecting feedback information on the initial answer information, determining the degree of model answer deviation based on the feedback information, and adjusting the adjusted voice conversion model or the initial dialogue scene model based on the determination result, so that the target answer information obtained after adjustment is better than the initial answer information; The step of identifying the background noise and the voice accent based on the analysis result comprises: identifying the energy features, prosodic features, and phoneme features of the initial voice data; identifying the background noise based on the energy features; identifying the voice accent based on the prosodic features and the phoneme features; The step of determining calibration parameters based on the analysis result comprises: determining the deviation value between the voice data and the standard voice based on the prosodic features and the phoneme features; determining the calibration parameters based on the deviation value; The step of adjusting the adjusted voice conversion model or the initial dialogue scene model based on the determination result comprises: if the deviation degree is greater than a preset deviation degree, adjusting the adjusted voice conversion model, including optimizing the parameters of the voice-to-text algorithm or updating the training data of the voice conversion model; if the deviation degree is less than or equal to the preset deviation degree, adjusting the initial dialogue scene model, including adjusting the parameters of the scene classification model or updating the scene template and the preset reply strategy; The preset reply strategy includes answer length limit and keyword replacement rule. 2.The AI large model-based intelligent dialog scenario analysis method of claim 1, wherein, The step of standardizing and adjusting the initial text information based on the standard language database comprises: comparing the standard words in the standard language database with the initial text information one by one to determine abnormal words, calculating the similarity between the abnormal words and the standard words, and replacing the abnormal words with a similarity greater than a preset word similarity with standard words; identifying the actual grammar rules in the initial text information, comparing the actual grammar rules with the standard grammar rules in the standard language database, and determining and correcting the error grammar based on the comparison result. 3.The AI large model-based intelligent dialog scenario analysis method of claim 2, wherein, The step of determining the lexical feature according to the analysis result of the text information comprises: performing word segmentation on the standard text information to obtain a plurality of segmented words; identifying the occurrence frequency of the segmented words, and determining the semantic feature corresponding to the segmented word with the highest occurrence frequency as the lexical feature. 4.The AI large model-based intelligent dialog scenario analysis method of claim 3, wherein, The step of determining the emotional feature according to the analysis result of the voice data comprises: extracting the timbre feature and the speech rate feature of the voice data; constructing a voice feature vector based on the timbre feature and the speech rate feature; determining the emotional feature based on the voice feature vector and an emotional analysis model. 5.The AI large model-based intelligent dialog scenario analysis method of claim 4, wherein, The step of determining the dialogue feature based on the lexical feature and the emotional feature comprises: constructing a feature vector based on the lexical feature and the emotional feature; determining the dialogue feature based on the feature vector. 6.The AI large model-based intelligent dialog scenario analysis method of claim 5, wherein, The step of determining the scene label based on the analysis result of the dialogue feature to establish an initial dialogue scene model comprises: matching the dialogue feature with a preset scene classification model to determine the scene label; constructing the initial dialogue scene model according to the scene label and a preset scene template. 7.The AI large model-based intelligent dialog scenario analysis method of claim 6, wherein, The step of generating initial answer information based on the initial dialogue scene model and the processed voice data comprises: determining the target dialogue scene to which the processed voice data belongs based on the initial dialogue scene model; generating initial answer information according to the target dialogue scene and a preset scene reply strategy; wherein the preset scene reply strategy comprises reply templates and reply contents under different dialogue scenes. 8.The AI large model-based intelligent dialog scenario analysis method of claim 7, wherein, The step of determining the deviation degree of the model answer based on the feedback information comprises: collecting feedback data of the user for the initial answer information, including a satisfaction score and a reply content evaluation; extracting keywords in the reply content evaluation, analyzing the keywords, and determining a text satisfaction degree according to the analysis result; determining an actual satisfaction degree according to the satisfaction score and the text satisfaction degree; determining the deviation degree based on the actual satisfaction degree and a preset satisfaction degree.

Images