304 results about "Natural language processing" patented technology

The invention provides a context-based intelligent voice semantic understanding analysis method and system, and relates to the technical field of voice interaction, and the method comprises the steps: obtaining an input voice signal, extracting an acoustic feature, and decoding the acoustic feature to obtain a candidate text; constructing context semantic representation to carry out disambiguation processing; establishing a semantic dependency graph and carrying out multi-level correlation analysis; spreading context constraint information to perform multi-hop reasoning; and finally generating intention recognition and slot filling results and updating the session state. The semantic comprehension accuracy and the intelligent degree of the voice interaction system are improved by introducing the context information and the multi-level semantic dependency relationship.

The invention discloses an AI Agent-based low-code outbound verbal skill process configuration method and system, and the method comprises the steps: carrying out the word segmentation and semantic analysis of an original verbal skill text through a verbal skill configuration Agent, recognizing the type of a core business intention, extracting key entity information, carrying out the feature extraction and weight calculation of customer portrait data, and generating a customer feature model, fusing the key entity information with the customer feature model to generate an optimized verbal skill suggestion; performing keyword scanning, semantic rule matching and context association analysis on the optimized verbal skill suggestion through a compliance detection Agent to obtain a latest compliance standard update and generate a compliance detection result; carrying out automatic replacement and risk prompt supplement on violation contents through a dynamic correction function, and generating a compliant version verbal skill; and performing format standardization processing on the verbal skill of the compliant version through the outbound system interface, and then pushing the verbal skill to an outbound execution platform. According to the invention, the problems of low verbal skill configuration efficiency and insufficient compliance detection capability in the prior art are solved.

The invention provides a cue word optimization method and system based on a large language model, and the method comprises the steps: generating a theme corresponding to a question in a training set, the training set comprising a question-answer pair, and the question-answer pair comprising the question and a standard answer of the question; searching a first cue word corresponding to the theme in a knowledge base, wherein the knowledge base comprises a knowledge base theme and a cue word corresponding to the knowledge base theme; inputting the first cue word corresponding to the theme and the question into a large language model to obtain a response answer output by the large language model; and if the response answer is wrong compared with the standard answer, generating a new cue word based on the question, the theme, the response answer and the standard answer, and updating the first cue word corresponding to the theme into the new cue word in the knowledge base. The cue word optimization mode of the large language model is remarkably improved, and finally the overall accuracy and generalization ability of the model for answering based on the optimized cue words are improved.

The invention relates to a railway traffic field semantic recognition method and device based on few sample data and a medium, and the method comprises the steps: carrying out the preprocessing of the obtained railway traffic field data, and constructing a field dictionary and a rule base; converting the preprocessed domain data into enhanced voice, forming a training pair by the enhanced voice and a corresponding text, and performing domain fine tuning on the ASR model; inputting a voice instruction in the railway traffic field into the field fine-tuned ASR model, and outputting an ASR initial transcription text; on the three levels of sentence level, local word order level and semantic tag level, word legality judgment is performed on the ASR initial transliteration text through a multi-scale model, and a high-credibility anchor point set and uncertain content are recognized; and field semantic rewriting is carried out based on the legitimacy judgment result of the words, and finally a structured instruction text is output. Compared with the prior art, the method has the advantages of realizing accuracy and adaptability of semantic recognition in the field of railway traffic and the like.

The invention relates to the technical field of intelligent questions and answers, in particular to the technical field of artificial intelligence, and provides a data processing method, device and equipment and a storage medium, which are used for effectively reducing the storage overhead of KV data while ensuring the reply precision. The method comprises the following steps: splitting a problem sequence into a plurality of problem lexical elements, and extracting corresponding semantic key value vectors; performing multi-round iterative processing on the question sequence until a reply answer is output; in each round of iteration, based on the semantic key value vector of each historical lexical element, obtaining an answer lexical element of the round; taking the answer lexical elements as new historical lexical elements, and extracting semantic key value vectors of the new historical lexical elements for storage; when the stored semantic key value vector reaches a storage capacity threshold value, obtaining a lexical element compression quantity based on a problem lexical element quantity and a historical lexical element quantity; and based on the attention evaluation value of each historical lexical element for the answer lexical element of the current round, eliminating the historical lexical elements conforming to the lexical element compression quantity and the corresponding semantic key value vectors.

A method and system for training a virtual agent is provided herein. The method and system comprises storing conversations between the virtual agent and a user in logs. The method and system further comprises mining the logs to retrieve utterances. The method and system further comprises computing regression for each of the plurality of the charging time segments. The method and system further comprises providing a score to the utterances. Further, the method ranking the utterances based on the score.

A method includes obtaining a plurality of training data sets each associated with a respective native language and includes a plurality of respective training data samples. For each respective training data sample of each training data set in the respective native language, the method includes transliterating the corresponding transcription in the respective native script into corresponding transliterated text representing the respective native language of the corresponding audio in a target script and associating the corresponding transliterated text in the target script with the corresponding audio in the respective native language to generate a respective normalized training data sample. The method also includes training, using the normalized training data samples, a multilingual end-to-end speech recognition model to predict speech recognition results in the target script for corresponding speech utterances spoken in any of the different native languages associated with the plurality of training data sets.

The invention discloses an English vocabulary follow-up pronunciation correction method based on English teaching, and relates to the technical field of voice processing, and the method comprises the following steps: S1, constructing a standard vocabulary pronunciation unit; s2, carrying out phoneme difference positioning by using a standard vocabulary pronunciation unit; s3, performing syllable dynamic comparison by using the phoneme deviation distribution of the learner; s4, performing vocabulary level comparison by using the learner phoneme deviation distribution and the syllable deviation mapping; and S5, performing difference mapping output by using the vocabulary pronunciation difference set. By setting the difference comparison between the actual pronunciation of the learner and the standard pronunciation, an accurate alignment relationship can be established at the phoneme level, and the difference between the actual pronunciation and the standard pronunciation in the aspects of acoustic characteristics, pronunciation duration, phoneme boundaries and the like is quantized, so that compared with the prior art, only the overall similarity or fuzziness judgment can be given, and the accuracy of pronunciation is improved. And a difference comparison mechanism can provide a clearer error correction reference, so that the learner can clearly recognize the position and degree of the problem when receiving the result.

The invention provides a vertical field corpus extraction method and device, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence, and the method comprises the following steps: obtaining a plurality of general corpora, and field keywords and sample corpora of a target vertical field; filtering the plurality of general corpora based on the field keyword to obtain a plurality of candidate corpora; and filtering the plurality of candidate corpora based on the semantic relevancy between the candidate corpora and the sample corpora to obtain the corpora of the target vertical field. According to the method and device provided by the invention, a multi-stage filtering mechanism is adopted, filtering is carried out through the keyword and semantic relevancy, and it is ensured that the finally screened corpus is not only related to the vocabulary level, but also highly matched with the target vertical field on the semantic level; related vertical field corpora are accurately and efficiently extracted from general large-scale corpora, and the performance and reliability of a large model in the vertical field are improved.

The embodiment of the invention provides an anonymization privacy protection method and system for voice information retention. The method comprises the following steps: extracting speaker embedding of an original audio, eliminating the tone of the speaker, and keeping semantic and rhythm speaker irrelevant features; embedding and inputting a speaker into a speaker anonymous module matched with a three-stage stream based on a U-Net architecture to obtain anonymous embedding; and combining irrelevant features of the speaker with anonymous embedding by using a pre-trained voice reconstruction model to generate anonymized voice with tone privacy. According to the embodiment of the invention, voice anonymization facing content privacy and tone privacy reserved by voice information is realized, the effectiveness of the anonymized voice generated by using the method in a downstream task is superior to that of a baseline model, meanwhile, the privacy of a speaker is also guaranteed, and safe use of data is realized.

The invention discloses a dialect speech enhancement method and device based on cross-modal and confrontation verification, and belongs to the technical field of speech recognition and enhancement. According to the method, the dialect voice and the lip movement video are jointly modeled, so that the accuracy and the naturalness of a dialect generation task are remarkably improved; a set of generation-adversarial-feedback closed-loop enhancement framework is constructed, and a multi-dimensional adversarial verification mechanism is introduced, so that the model is self-evolved, and enhancement data is ensured to be diversified and accord with real use habits of dialects. The dialect knowledge graph and the voice generation model are creatively fused, double enhancement of semantic and cultural levels is achieved, the generated dialect voice content better conforms to the real language environment and cultural background of the dialect voice content, and the method is particularly suitable for protection and inheritance of endangered or small dialects. The method can be directly applied to subsequent dialect recognition, synthesis or protection and the like.

The invention discloses a literature classification method and system based on a large language model, and relates to the technical field of large language models.The literature classification method based on the large language model comprises the specific steps that an n-level category system is defined, naming is conducted through an upper concept, the category relation and unique identification are clarified, a tree structure is formed and coded, and a dimension set is defined; the method comprises the following steps of: binding core terms of fine-level categories and literature types classified according to text features, determining a sample format, performing progressive sampling according to the levels of the categories, constructing a user-defined rule base, judging whether the standard is reached or not through a consistency rate, calculating a rule conformity and a feature word coincidence rate, and further determining a labeling qualification rate. And according to the marked qualified rate, reasons of unqualified categories are positioned and processed. According to the method, the literature can be further distinguished according to the text features, and rich dimensions are provided for subsequent literature analysis, retrieval and management.

The invention discloses a text-to-SQL (Structured Query Language) generation and auditing method and device based on retrieval enhancement, and the method comprises the steps: analyzing a user intention by utilizing a generative Schema routing model, and automatically selecting a target database and a related table to form a query range; an example similar to the current question is retrieved from the historical question and answer corpus and serves as a Few-Shot prompt to be provided for the large language model; then, self-adaptively constructing a joint prompt according to the difficulty of the problem, guiding the LLM to generate a preliminary SQL query, and correcting an error through a cyclic feedback mechanism; an LLM-driven SQL script automatic auditing and optimization suggestion module is introduced, comprehensive auditing is conducted on the generated SQL from the aspects of grammar specifications, query performance, structure optimization, safety protection and the like, and optimization suggestions are given out and serve as filtering and feedback links before the SQL falls into a library. According to the method, retrieval enhancement and prompt generation technologies are fully combined, and the generation efficiency, reliability and safety are improved while the correctness of the query result is ensured.

The invention provides a retrieval enhancement generation method and system based on context semantics and hybrid reordering, and relates to the technical field of retrieval enhancement generation, and the method comprises the following steps: segmenting a knowledge base according to paragraphs to obtain a plurality of independent paragraphs, segmenting the knowledge base into a plurality of logic blocks according to a logic structure, and correspondingly generating summary information; enhancing the target paragraph according to the logic block associated with the target paragraph screened out by the user query statement and the summary information of the logic block to obtain an enhanced text; generating a context abstract for the corresponding target paragraph based on the enhanced text, vectorizing the enhanced text and the context abstract, storing the vectorized text and context abstract in a vector database, and synchronously constructing a keyword index database; performing mixed retrieval in a vector database and a keyword index database according to the user query statement to obtain a retrieval result; and the retrieval results are reordered according to the answer correlation in an inverted order to obtain an inverted sequence. The method has the beneficial effects that knowledge waste caused by text block isolation and position prejudice is avoided, and the retrieval precision is greatly improved.

The invention provides a voice data generation method based on a large model and a method for training the large model, and relates to the technical field of artificial intelligence, in particular to the technical fields of voice generation, intelligent customer service, video production and the like. The large model-based voice data generation method comprises the following steps of: receiving a rhythm description text and a voice text, wherein the rhythm description text describes a pronunciation rhythm intention of a plurality of text words in the voice text; semantic fusion is conducted on the rhythm description text and the voice text through a large model, rhythm fusion features are obtained, sub-features in the rhythm fusion features represent the pronunciation rhythm of the voice segments for the text words, and the to-be-generated target voice data comprise the voice segments; and according to the rhythm fusion feature and a specified pronunciation attribute related to the specified object, generating target voice data which represents that the specified object pronunciates according to the pronunciation rhythm intention and corresponds to the voice text.

An object of the system according to the embodiment is to develop a new language in which meanings and nuances can be shared and understood in common throughout the world.SOLUTION: A system includes an analysis unit, a generation unit, a creation unit, and a provision unit. The analysis unit analyzes multilingual data and analyzes the meaning and nuance of each language. The generation unit generates grammar and vocabulary of the new language based on the meaning and nuance extracted by the analysis unit. The creation unit creates a grammar document and a dictionary based on the grammar and the vocabulary generated by the generation unit. The providing unit provides an educational material for learning a new language based on the grammar document or the dictionary created by the creating unit.SELECTED DRAWING: Figure 1

The application relates to a song recommendation model training method, a song recommendation method, computer equipment and a storage medium, and relates to the technical field of artificial intelligence. The application can improve the adaptability of a model-recommended song to a user. Method: for each sample user's song listening information sequence sample, the historical and future song listening information sequences are divided, the historical song listening information sequence is input into a first feature extraction branch of a to-be-trained model to extract song feature representations of each historical time step, the future song listening information sequence is input into a second feature extraction branch to extract song feature representations of each future time step, the first and second branches of the model are trained based on the similarity of the feature representations of each historical time step and the feature representations of each future time step, and when the similarity of the feature representations of the historical time step and the feature representations of the future time step corresponding to the same user is greater than or equal to a first threshold value and the similarity of the above feature representations corresponding to different users is less than a second threshold value, a trained song recommendation model is obtained.

The embodiment of the invention discloses a wrongly written character processing method, device and equipment, and the method comprises the steps: receiving text data of a target domain input by a user, the text data comprising special words of the target domain; performing word segmentation processing on the text data through a preset word segmentation strategy aiming at a target field, and performing identification processing aiming at special words of the target field on segmented words corresponding to the obtained text data to obtain target special words in the segmented words corresponding to the text data; carrying out wrongly written character recognition processing on the target proprietary word through a proprietary word error recognition model for the target field to obtain a first recognition result, and carrying out wrongly written character recognition processing on segmented words, except the target proprietary word, in the segmented words corresponding to the text data through a general word error recognition model to obtain a second recognition result; and based on the first recognition result and the second recognition result, determining wrongly written character information contained in the text data through fusion processing of the two different recognition results.

The embodiment of the invention provides a dialogue data set generation method and system and a model training method and system of a large model. According to the method, a generation model obtains an image data set; for each dialogue interface in the image data set, dialogue box detection is carried out on the dialogue interface through the detection sub-model, a dialogue box list is obtained, and the dialogue box list comprises a plurality of dialogue boxes and a category label corresponding to each dialogue box; identifying text information of each dialog box in the dialog box list through a character identification sub-model to obtain a text list; the text list comprises text information corresponding to each dialog box in the plurality of dialog boxes; through a category discrimination sub-model, based on the text list and the category label corresponding to each dialog box, and in combination with the target query corresponding to each dialog interface, generating answer information for the target query; and pairing each target query with the corresponding answer information, and constructing a target data set comprising a plurality of question and answer pairs.

System and methods for generating a cross-domain multilingual model are disclosed. In some embodiments, a disclosed method includes: storing, in a database, a plurality of first utterances associated with a first language, training a first model using the plurality of first utterances, the first model being associated with the first language, generating, using the first model, a plurality of first representations associated with the plurality of first utterances, training a second model, using the plurality of first representations, the second model being associated with a plurality of second languages, receiving, using the second model, a second utterance in the second language, and generating, using the second model, a response in one or more languages of the plurality of second languages.

A method and apparatus for selecting a sample corpus used to optimize a translation model, an electronic device, a computer readable storage medium, and a computer program product are provided. The method includes: after acquiring a first corpus, translating the first corpus by using a to-be-optimized translation model to acquire a second corpus with different types of languages, then translating the second corpus by using the to-be-optimized translation model to acquire a third corpus, then determining a difficulty level of the first corpus based on a similarity between the first corpus and the third corpus, and finally determining the first corpus as a sample corpus used to perform optimization training on the to-be-optimized translation model in response to the difficulty level satisfying requirements of a difficulty level threshold.

The present disclosure provides a text translation method, a server, a storage medium, and a program product. The method relates to the field of artificial intelligence, and comprises: acquiring an original text to be translated and a target language; searching a translation database for a translation text pair matching the original text, the translation text pair comprising a source language text and a target language text; and inputting the original text, the target language, and the translation text pair into a text translation model, translating, by means of the text translation model, a source language text appearing in the original text into a corresponding target language text, and generating a translated text in the target language corresponding to the original text. By performing external knowledge intervention on a text translation model on the basis of a pre-customized translation database, incorrect translations of custom vocabularies are effectively avoided without needing to retrain parameters of the text translation model, and the accuracy of translation of the text translation model for custom vocabularies such as proper nouns, terminology, and new terms can be improved, thereby improving the accuracy of text translation results.

In various examples, determining emotion sequences for speech in conversational AI systems and applications is described herein. Systems and methods are disclosed that use one or more first machine learning models to determine a sequence of emotional states associated with audio data representing speech. To use the first machine learning model(s), the systems and methods may train the first machine learning model(s) using one or more second machine learning models, where the second machine learning model(s) is trained to determine scores indicating accuracies associated with sequences of emotional states. For instance, the second machine learning model(s) may be trained to determine the scores using audio data representing speech, sequences of emotional states associated with the speech, and indications of which sequences of emotional states better represent the speech as compared to other sequences of emotional states.

The invention relates to an English dictation intelligent scoring system based on a multi-modal large model. The system comprises a multi-modal dictation input processing module, a dictation content analysis module, a fine-grained error detection module, a multi-dimensional scoring module and a personalized feedback generation module. The system realizes full coverage of three dictation forms of text, handwritten form and voice, and adapts to different teaching scenes. According to the method, multi-dimensional errors such as spelling, form, writing specification, voice and the like can be accurately recognized, the single dimension of'correct rate 'is broken through, comprehensive scoring is performed from the four aspects of'accuracy, normalization, fluency and context adaptability', and the real dictation quality of students is reflected.

The application provides a quick retrieval method for language structure big data, according to the type of language structure data, the basic index unit to be constructed and the index content corresponding to the index unit are determined, when retrieving, the retrieval expression input by the user is analyzed, the retrieval unit corresponding to the index unit and the relationship between the retrieval units are analyzed, the result of each retrieval unit is obtained in the way of single chain binary tree postorder traversal and is combined according to the relationship between the analyzed retrieval units. In the application, the design of the index unit can utilize the lexical and syntactic information of the language data itself, realize the retrieval requirement of multi-directional language knowledge of the language structure data, the design of the structure retrieval language can enable the user to learn the retrieval language at a low cost and write the retrieval expression which is standard and meets the retrieval requirement.

The application belongs to the technical field of Chinese character image recognition, and particularly discloses a Chinese character image recognition method based on a four-corner number dictionary and related components. The method comprises the following steps: pre-processing an input Chinese character image to obtain skeleton information of a Chinese character to be recognized; performing four-corner number dictionary recognition on the stroke shape of the skeleton information to obtain corresponding four-corner numbers; calling the four-corner number dictionary to read homograph Chinese characters with the same four-corner numbers to obtain a homograph Chinese character set; when the number of Chinese characters in the homograph Chinese character set is one, outputting the unique Chinese character as a recognition result; when the number of Chinese characters in the homograph Chinese character set is within a first preset range, outputting a Chinese character with the highest confidence in the homograph Chinese character set as the recognition result; and when the number of Chinese characters in the homograph Chinese character set is within a second preset range, eliminating Chinese characters with inconsistent features in the homograph Chinese character set until the number of Chinese characters in the homograph Chinese character set is one or within the first preset range. The application has the advantage of quickly finding a correctly recognized target Chinese character from a homograph set of the four-corner number dictionary.

A feature library construction method, the method comprising: obtaining a first file sample set, the first file sample set comprising a plurality of file samples and a cleaning level of each file sample; extracting a feature word that can affect the cleaning level of each file sample from each file sample in the first file sample set; filtering the extracted feature words to obtain a plurality of effective feature words; setting a weight of each effective feature word, and constructing a feature library for recommending the cleaning level of a file according to each effective feature word and the weight of each effective feature word.

The embodiment of the invention provides a model training method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a vocabulary of a first model, the vocabulary comprises N1 first lexical elements, each first lexical element represents a tool identifier, and N1 is greater than or equal to 1; and training the first model based on the word list. Therefore, the tool knowledge (namely the tool identifier) is embedded into the word list of the first model in advance, and a cue word does not need to be filled with a tedious tool description text, so that lexical element consumption introduced by tool description in the cue word can be avoided. In this way, even if the tool scale is greatly increased, the length of the cue word can still be kept stable, so that the problem that the tool calling efficiency is reduced due to insufficient lexical element capacity in the cue word can be avoided, and the adaptability and expansibility of the model to a multi-tool scene are improved.

The purpose is to generate system utterances to be output to the user. [Solution] A dialogue system comprising: a specifying means for specifying a next system utterance to be output to a user based on a user utterance input by the user and a system utterance output immediately before the user utterance; a storage means for storing a user utterance and a system utterance output in response to the user utterance in association with each other as a dialogue history, or a storage means for storing a system utterance and a user utterance input in response to the system utterance in association with each other as a dialogue history, and a specifying means for specifying a next system utterance to be output to the user based on the user utterance input by the user and the dialogue history stored in the storage means.