1048 results about "Natural language processing" patented technology

The invention relates to dialect speech synthesis, in particular to a semi-supervised dialect emotion speech synthesis system based on hybrid experts, which comprises a text analysis module for preprocessing an input dialect text and generating a text representation vector through feature extraction and feature fusion; the mixed expert module is used for acquiring dialect acoustic features, rhythm features, emotional features and general acoustic features; the dynamic routing module is used for realizing intelligent cooperation among experts through a task-aware soft routing algorithm; the semi-supervised learning module is used for training supervised learning by using the marked dialect emotion voice data and training self-supervised learning by using the unmarked dialect emotion voice data; the acoustic parameter generation module is used for integrating the output of each expert to generate a complete acoustic parameter set; the neural vocoder is used for converting the acoustic parameter set into final dialect emotion voice; the method can effectively overcome the defect that the dialect emotional voice is difficult to accurately synthesize under the condition of lack of sample resources.

The disclosure herein describes using a deep learning model to identify topic segments of a communication transcript. A communication transcript including a set of utterances is obtained. The set of utterances is divided into a plurality of utterance windows, wherein each utterance window of the plurality of utterance windows includes a different subset of utterances of the set of utterances, and wherein each utterance of the set of utterances is included in at least one utterance window of the plurality of utterance windows. For each utterance window of the plurality of utterance windows, each utterance in the utterance window is classified as a topic boundary or a non-boundary using a deep learning model. Topic segments of the communication transcript are identified based on utterances of the set of utterances that are classified as topic boundaries. A communication transcript summary is generated using the communication transcript and the identified topic segments.

A large language model predicts a topic summarization of a long document given a set of segments from the long document that pertain to a topic of interest. The set of segments from the long document are selected by searching for similar segments from other documents that have been labeled to indicate whether or not the segment pertains to a topic of interest. The search is based on an embedding of a segment from the long document closely matching embeddings of the labeled segments. Each segment of the long document is scored based on the labels of the closest-matching similar segments. The segments from the long document are ranked by their respective score and the highest-scored segments are included in a prompt to the large language model for the model to generate a topic summarization of the long document.

A cyber security system that protects against cyber threats including a synthetic clone of a voice of a speaker can include several components. A deep learning model is trained to analyze an audio file and produce one or more embeddings of the audio file. One or more AI classifiers are trained to analyze the one or more embeddings of the audio file from the deep learning model to determine whether it is likely that the voice of the speaker engaging with a user is real or the synthetic clone of the voice of the speaker. The voice clone detection bot can be resident on a computing device of the user and can integrate with different sources of audio data on the computing device of the user in order to collect the audio file containing an attempt to synthetically clone the voice of the speaker protected by the cyber security system.

A device may receive, at a server system, a free-form query. The device may select two-way communication history records from a database associated with the free-form query to identify selected history records. The device may generate individual summaries of individual history records of the selected history records by processing a set of corresponding individual history records using a chunking algorithm, constructing language responses from outputs of the chunking algorithm using a large language model, and aggregating the language responses. The device may process the language responses using the large language model to generate an individual summary for corresponding individual history records.

Training a mixed language speech recognition model, and speech recognition performed by the model, can be enhanced. Model manager can control training and / or operation of the model. Mixed language data generation manager can generate an enhanced mixed language dataset that can enhance training of the model. Fine tuner can facilitate configuring hyperparameters of the model. Audio-based information and transcript that are representative of the dataset can be applied to the model to facilitate model training. FAL evaluator can determine fidelity, accuracy, and latency of performance of speech recognition on the audio-based information and transcript by the model. Based on such determination, mixed language data generation process and / or hyperparameters can be updated to enhance further training of the model to enhance fidelity, accuracy, and / or latency regarding performance of speech recognition by the model. Model manager can control one or more iterations of model training.

The invention discloses a language learning assistance application system based on speech recognition, and belongs to the technical field of learning assistance. The system comprises an intelligent voice acquisition and enhancement module for realizing acquisition and preprocessing of high-quality voice signals in a complex environment; the multi-modal feature extraction module dynamically extracts and fuses multi-dimensional learning feature information, completes feature analysis and boundary segmentation of a phoneme level, and is responsible for feature normalization and optimization; the deep speech recognition module is used for realizing high-precision speech recognition based on the normalized features; the intelligent evaluation engine module is used for realizing real-time and multi-dimensional evaluation on the language performance of the learner; the personalized learning management module integrates the evaluation results for dynamically planning an optimal learning path for the learner, manages the learning progress, and provides learning effect prediction and intervention suggestions at the same time; and the learning interaction experience module is responsible for visual presentation of learning feedback, multi-modal man-machine interaction design and user interface optimization.

HyperText Markup Language (HTML) content analysis (HCA) using machine learning is described. A feature vector schema may be generated based on domain names corresponding to HTML webpages and corresponding indications of a status of the HTML webpage. The schema may map each position in a feature vector of a given HTML webpage to a resource identifier. Information may be processed using the schema to generate respective feature vectors. The feature vectors may be used to train a model to generate risk indicators for HTML webpages. A potentially parked domain webpage or a potentially malicious domain webpage may be received. A feature vector for the webpage may be generated and inputted to the model. The model may generate a risk indicator for the webpage. The risk indicator may be output and may cause responsive actions. The model may be updated based on a determination indicating whether the webpage was a parked domain webpage or a malicious domain webpage.

The invention discloses a multi-language translation method and system, and relates to the field of electrical digital data process.The method comprises the steps that cultural feature information and semantic structure information are obtained; splitting the source language text into a plurality of semantic units based on the semantic structure information; retrieving cultural connotation information of which the association degree with the semantic unit is greater than a preset threshold value; determining that the semantic unit is a culture sensitive semantic unit; extracting a target language expression mode based on the cultural connotation information; performing semantic combination on the target language expression mode and standard translation results of other semantic units except the culture sensitive semantic unit in the source language text to obtain a translation result; judging whether the semantic coherence of the translation result meets a preset condition or not; if the preset condition is met, outputting a translation result; and if the preset condition is not met, re-executing the step of semantic combination. The accuracy of the cultural connotation and the emotional color contained in the transmission source language is improved, and then the cultural adaptability of the translation result is improved.

The invention discloses a language transformation method and device based on retrieval enhancement and a storage medium, and relates to the technical field of language transformation.The method comprises the steps that in response to a query request, query information is determined based on the query request; performing intention recognition on the query information, and determining a target cue word template from a preset cue word template library; performing question recall on the query information to obtain target data; and performing language conversion through the target data and the target cue word template to generate a target structured query language. The problem that an original technical scheme is insufficient in generalization in a specific field is solved by retrieving business knowledge and industry common knowledge related to the problem through intention recognition, and the problem that the model effect cannot be improved by performing fine adjustment in the specific field due to the limitation of computing resources is solved through problem recall. And the language conversion efficiency and accuracy are improved.

Techniques are described herein for generating a semantically related search query that differs by at least one lexical token from document data provided as input. A machine-learning model may be trained using supervised and / or deep learning algorithms and a training data set including historical queries and the document data identified as being semantically related to those queries. The semantically-related search query may be associated with the document and utilized in subsequent searches to match the document to a subsequent query based on identifying a lexical match between the subsequent search query and the semantically-related search query associated with the document. These techniques enable semantically related search queries to be assigned offline but utilized to identify semantically related documents using lexical matching techniques. Identifying these types of matches create more diverse search results while maintaining a high degree of relevance between the query and the documents of the result set.

The invention relates to the technical field of language learning, and discloses an English auxiliary pronunciation training method and system based on multi-task learning, and the method comprises the following steps: constructing a multi-task learning framework which comprises at least two subtask models related to pronunciation, comprising the steps of pronunciation accuracy scoring, pronunciation error type positioning and intonation fluency analysis. A voice signal, a lip movement video and a text content of a user are collected through a multi-modal data input module; and performing joint coding on the multi-modal data by using the shared feature extraction layer to generate fused acoustics. According to the English auxiliary pronunciation training method and system based on multi-task learning, the pronunciation training path of the user can be adaptively adjusted, the requirements of different users can be met, the English pronunciation level can be more effectively improved, and through strategies such as multi-modal data input, shared feature extraction, inter-task attention mechanism and adaptive training path adjustment, the English pronunciation training efficiency can be improved. Accurate evaluation and personalized training of English pronunciation are realized.

A method of training a transcoder algorithm using audio samples having high resource and low resource language components. The method includes evaluating batches of audio samples based on accent, dialect, accented and language specific domain, accented and language specific subdomain, dialect and language specific domain, dialect and language specific subdomain, or combinations thereof. The method includes up-sampling audio samples having at least one language component from the group and adding the up-sampled audio samples to the batches of audio samples. The method includes generating hidden representations using a pre-trained neural network, measuring differences of delta between predictions and true labels from metadata and predicting a score for each task. The method also includes comparing the predicted score to a true label from the metadata of each audio and tuning the parameter space of the transcoder algorithm based on the comparison between the predicted score and the known score.

System and techniques to facilitate the translation of a sign language into another language are described herein. A modular architecture may be used in which the output of different classifiers may be used to produce intermediate representations, or final translations, of the sign language. These classifiers may be trained on different types of signs to enhance accuracy while reduce training time and complexity.

The invention relates to a method for converting Lao characters into phonemes based on transfer learning, and aims to realize efficient phoneme conversion by utilizing a model pre-trained on other languages and finely adjusting limited Lao data. According to the method, the problem of data scarcity can be relieved, the training process of the model can be accelerated, and the generalization ability of the model on new tasks is improved. A more robust Lao character transphoneme model can be constructed through transfer learning, and a foundation is laid for subsequent speech synthesis. As Thai with high language similarity with Lao is also a low-resource language and cannot be used as a pre-training language, Chinese is selected as the pre-training language. A Chinese open source data set is firstly used for training a Chinese G2P model based on a Transform architecture, decoding end parameters of the model are initialized to a student model, fine tuning training is performed by using a Lao data set with the scale of ten thousand, and the accuracy of a Lao character-to-phoneme model is improved.

An interactive method based on a large model, a training method, and an intelligent agent, which relate to fields of artificial intelligence, speech recognition, speech interaction, deep learning, large models, and application scenarios of knowledge search, autonomous driving, intelligent customer service, intelligent speech control, smart e-commerce, AI healthcare. The interactive method includes: acquiring a request speech; performing a speech recognition on the request speech to obtain a speech recognition feature representing a request semantics; and processing the speech recognition feature using the large model to obtain a response text, where the response text includes response words arranged in sequence, a target response word among the response words is determined by processing the speech recognition feature and an associated response word feature using an attention fusion layer of the large model, and the associated response word feature is related to an associated response word arranged before the target response word.

Systems and methods for translating a first coding language into a second coding language train a machine learning (ML) model on a first coding language specific data set relating to the first coding language, in which the ML model is trained to translate one or more code sets of the first coding language to respective one or more code sets of the second coding language; using the ML model, generate various unit test cases, in which the unit test cases run the one or more code sets of the second coding language in parallel with the one or more code sets of the first coding language; iteratively test and refine the ML model until a maturity threshold is reached; and upon reaching the maturity threshold, containerize the one or more code sets of the second coding language into one or more applications.

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.

A method classifies feedback from transcripts. The method includes receiving an utterance from a transcript from a communication session and processing the utterance with a classifier model to identify a topic label for the utterance. The classifier model is trained to identify topic labels for training utterances. The topic labels correspond to topics of clusters of the training utterances. The training utterances are selected using attention values for the training utterances and clustered using encoder values for the utterances. The method further includes routing the communication session using the topic label for the utterance.

The invention discloses a judicial field semantic abstracting method, system and device based on a large model and a storage medium. The method comprises the steps that voice recognition is conducted on a court trial video, voice is converted into a court trial text, the court trial text is labeled, and a labeled text set is obtained to serve as a training set; training the large model by using the labeled text set, and finely adjusting the large model after each round of training; performing voice recognition on the actual court trial video, converting the voice into an actual court trial text, and extracting a timestamp; utilizing the trained large model to identify the court hearing process of the actual court hearing text, and reasoning a court hearing process text; matching the reasoned court hearing process text to timestamps by using a matching algorithm, and generating video anchor points when the timestamps at the starting positions of all the processes in the court hearing process text correspond to the anchor points of the processes of the court hearing video; abstracting the reasoned court hearing process text; and combining the abstract text with the video anchor points to generate structured court hearing information. According to the invention, the court trial video processing efficiency is improved.

One example method includes enabling a user to define a query that comprises a hybrid search syntax (HSS), and the HSS is in the form of a single string that comprises a lexical component and a semantic component, receiving the query from the user, parsing the lexical component and the semantic component into respective groups of one or more sub-components, performing a lexical search based on the sub-components of the lexical component, and performing a semantic search based on the sub-components of the semantic component, ranking, separately from each other, results of the lexical search and results of the semantic search, combining the results of the lexical search and the results of the semantic search to create aggregate results, and re-ranking the aggregate results.

The invention provides a cue word determination method and device based on a large model, equipment and a medium, and relates to the field of data processing, in particular to the technical fields of cloud computing, artificial intelligence, intelligent search, big data, large models and the like. According to the specific implementation scheme, a first cue word is obtained; inputting the first cue word into the large language model to obtain an intention recognition result of the first cue word; and outputting the first cue word as a second cue word in response to the correctness of the intention recognition result. By applying the technical scheme provided by the embodiment of the invention, the cue word can be determined according to the intention recognition result obtained by performing intention recognition on the cue word through the large language model, and the accuracy of the cue word is improved.

A system for emotionally enhancing dialogue includes a computing platform having processing hardware and a system memory storing a software code including a predictive model. The processing hardware is configured to execute the software code to receive dialogue data identifying an utterance for use by a digital character in a conversation, analyze, using the dialogue data, an emotionality of the utterance at multiple structural levels of the utterance, and supplement the utterance with one or more emotional attributions, using the predictive model and the emotionality of the utterance at the multiple structural levels, to provide one or more candidate emotionally enhanced utterance(s). The processing hardware further executes the software code to perform an audio validation of the candidate emotionally enhanced utterance(s) to provide a validated emotionally enhanced utterance, and output an emotionally attributed dialogue data providing the validated emotionally enhanced utterance for use by the digital character in the conversation.

The embodiment of the invention discloses a question and answer retrieval method and device, a storage medium and a terminal. Receiving input data corresponding to the question and answer request, and generating an intention recognition cue word according to the input data; inputting the intention recognition cue word into a large language model, and controlling the large language model to recognize a target intention category corresponding to the question and answer request based on the intention recognition cue word; and determining a target retrieval scheme corresponding to the target intention category from the at least one retrieval scheme, performing retrieval on the input data according to the target retrieval scheme to obtain a corresponding retrieval result, and outputting the retrieval result as answer content corresponding to the question and answer request. During query, prompt words are generated for input data and input into the large language model, and the large language model identifies a target intention category implied in the input data. And after the intention category of the query is identified, a matched retrieval scheme is determined according to the information, and after retrieval is executed according to the determined retrieval scheme, a retrieval result conforming to the query intention of the user is obtained.

The present disclosure provides a real-time language translation system embodied in a physical device. Further, the real-time language translation system may include an input device which may be configured for receiving generating a user input data representing a linguistic input from a user. Further, the linguistic input corresponds to a user language. Further, the real-time language translation system may include a processing device which may be configured for generating a translation data based on the user input data. Further, the translation data represents a translation of the linguistic input. Further, the generating may be based on an AI module. Further, the processing device may be communicatively coupled to the input device. Further, the real-time language translation system may include a presentation device which may be configured for presenting the translation data. Further, the presentation device may be communicatively coupled to the processing device.

The invention relates to the technical field of language teaching, and discloses an intelligent system applied to Chinese language teaching, and the system comprises a voice recognition module which is used for collecting a voice signal of a learner and converting the voice signal into a phoneme sequence; the pronunciation difference analysis module is used for performing alignment analysis on the dialect phoneme sequence and the mandarin phoneme sequence of the learner to generate pronunciation difference data; the knowledge graph module is used for constructing a multi-layer knowledge graph of dialects and mandarin and describing the relationship among phonemes, grammar rules and vocabularies; and the personalized learning path generation module is used for dynamically generating a personalized mandarin learning path based on the pronunciation difference data and the knowledge graph of the learner. A multi-task learning framework based on a pre-training model Wav2Vec 2.0 is adopted, pronunciation difference analysis is performed through joint recognition of mandarin and dialect voices in combination with a dynamic time warping algorithm, and the technical effect of accurately capturing pronunciation characteristics and differences of learners is achieved.

Computing units provided at local sites or edge locations are programmed to execute conversational tools that generate pertinent, domain-specific responses to queries received from workers at such sites or locations. The conversational tools are large language models that are trained with domain-specific knowledge documents. Data representing queries are received from workers at such sites or locations and provided as inputs to the conversational tools along with text representing nearest knowledge documents from a knowledge base associated with the domain, as well as contextual data. Responses identified based on outputs received from the conversational tools in response to the inputs are provided to the workers that generated the queries. Where subsequent queries are received from the workers, responses to the subsequent queries are identified based on the subsequent queries, nearest knowledge documents, contextual data, and conversational histories including previously received queries and responses to such queries.

The invention discloses a voice emotion recognition method based on supervised comparative learning, and the method comprises the following steps: obtaining a large-scale voice data set containing a plurality of emotion categories, and carrying out the systematic preprocessing and emotion labeling of voice data, so as to construct a high-quality training data set; constructing a speech emotion recognition model based on an emotion anchor point generation module, a speech emotion representation module, a contrast learning module and an emotion anchor point self-adaption module; training the emotion recognition model by using the data set constructed in the step 1; and inputting a to-be-recognized target voice into the trained model, and obtaining a final emotion prediction result by calculating the similarity between the utterance emotion representation and the closest emotion anchor point. According to the method provided by the invention, a supervised comparative learning method is adopted, so that the problem that similar emotions are difficult to distinguish is solved; meanwhile, the emotion recognition accuracy can be improved, and the generalization ability of the model is enhanced.

The invention discloses a knowledge distillation-based Hainan dialect speech recognition optimization system, which comprises a data preprocessing module for processing Hainan dialect speech data and Hainan dialect text data, and outputting to obtain an MFCC speech feature sequence, a labeled text label and Hainan dialect text data; the teacher model module comprises an RNN (Recurrent Neural Network) language model, a CNN (Convolutional Neural Network) language model and a Transform language model, dialect text data are respectively input into the RNN language model, the CNN language model and the Transform language model, and a soft label and a middle layer feature are obtained through dynamic temperature regulation; the student model training module is used for performing knowledge distillation and parameter adjustment and optimization on a student model according to the MFCC voice feature sequence, the marked text tag, the soft tag and the middle layer feature; the output module is used for performing voice recognition on the Hainan dialect by using the student model after knowledge distillation and parameter optimization to obtain a recognition result; moreover, the trained student model is small in size and low in calculation complexity, and an accurate recognition result can be obtained finally.

A method and system for automatically analyzing and / or summarizing communications using language operators, the method receiving a definition of a language operator, the definition comprising one or more phrases indicative of a match to a concept associated with a communication; surfacing, in a UI, one or more results of applying the language operator to the communication, the surfacing of each result of the one or more results using an interactive UI element linked to one or more communication phrases, the interactive UI element highlighting a match between the concept and the one or more communication phrases; and training a machine learned (ML) model to improve an accuracy of the result based on one or more inputs received via the UI.