89 results about "Character (computing)" patented technology

The invention discloses a prosodic hierarchy model training method for text-to-speech and a text-to-speech method and a text-to-speech device by a prosodic hierarchy model. The training method includes: training massive non-tagged corpus data to obtain character vectors of individual characters; obtaining textual features and tagging corresponding to the trained data according to the character vectors and prosodic tagging data; based on a deep neural network and a bidirectional LSTM (long-short term memory) neural network, training the prosodic hierarchy model according to the textual features and tagging of the trained data. The training method has the advantages that when a character granularity based dictionary in the training method is compared with a traditional word granularity based dictionary, entry scale is decreased effectively, requirements of the model and resource files on computing resources and memory spaces are lowered, and usability of a prosodic prediction model in embedded intelligent devices is guaranteed while performance of the prosodic prediction model is improved.

The embodiment of the invention discloses a text information processing method and device, and relates to the field of cloud computing. A specific embodiment of the method comprises the steps of identifying a to-be-processed text from an image comprising the to-be-processed text; inputting the to-be-processed text into a pre-trained recurrent neural network language model, and identifying wronglywritten characters in the to-be-processed text; inputting the wrongly written characters in the to-be-processed text into a pre-trained text error correction model to obtain similar characters corresponding to the wrongly written characters; and determining correct characters corresponding to the wrongly written characters in the similar characters by utilizing the text error correction model based on the coherence of the to-be-processed text, and replacing the wrongly written characters with the correct characters to obtain an error correction text of the to-be-processed text. The wrongly written characters are recognized through the pre-trained recurrent neural network language model, and the correct characters of the wrongly written characters are obtained through the pre-trained text error correction model, so that the error correction text is obtained, and the accuracy of a recognition result is improved.

The invention relates to an encryption method and an encryption device for a font library in the field of character information processing. The font library comprises internal code information and dot matrix information. The encryption method comprises the following steps of: performing distributed computing on the internal code information with a first key to generate a secondary key; encrypting the dot matrix information with the secondary key to generate dot matrix data cipher text information; and encrypting the dot matrix data cipher text information or the combination of the internal code information and the dot matrix data cipher text information with the secondary key to generate a verification code. The encryption method and the encryption device for the font library provided by the embodiment of the invention have the advantages of preventing illegal users from replacing the font library and thus effectively improving the confidentiality of the font library.

The invention discloses a real-time multi-modal language analysis system and method based on mobile edge intelligence, and the system comprises three types of mobile edge intelligent servers: a mobilebase station (MGS), an unmanned vehicle (UGV) and an unmanned aerial vehicle (UAV), and the computing resources of the three types of mobile edge intelligent servers are sequentially reduced, and themoving flexibility is sequentially improved. According to the real-time multi-modal language analysis system, language data of a user is divided into three modalities, namely characters, voice and images, and a calculation task is allocated to a proper MEI server to be executed according to the calculation and analysis difficulty and the size of required calculation resources. According to the method, a real-time multi-modal language analysis and calculation problem in a dynamic environment is constructed, then a task unloading matrix and a resource allocation matrix are generated through a deep learning online optimization method, and meanwhile, movement path planning of an MEI server is automatically implemented according to channel conditions and interference during communication. Thetrained DNN can be suitable for a dynamic scene in which the number of multi-modal computing tasks changes, and has very high practicability.

The embodiment of the invention discloses an error correction method and device and computing equipment. The method comprises the steps: according to pinyin features of characters in a to-be-corrected statement, detecting the first probability of each character in the to-be-corrected statement, the first probability being used for representing the possibility that the character is an error character; determining a first feature of each character in the statement to be corrected according to the first probability; according to the font characteristics of the characters in the to-be-corrected statement, detecting a second probability of each character in the to-be-corrected statement, the second probability being used for representing the possibility that the character is an error character; determining a second feature of each character in the statement to be corrected according to the second probability; fusing the first feature and the second feature to obtain a fused feature of each character in the statement to be corrected; and determining an error correction result of the to-be-corrected statement according to the fusion feature. According to the embodiment of the invention, the error correction effect can be improved.

The invention discloses a certificate identification method and device, a computing device and a storage medium. The method comprises the steps of detecting at least one character block in a to-be-identified certificate image; identifying each character of the at least one character block to obtain a character identification result of each character block; matching the at least one character blockand the character identification result of each character block with layout structure information corresponding to the certificate image; and based on a matching result, determining element content corresponding to each certificate element of the certificate image. Therefore, adverse effects of light rays, backgrounds, segmentation errors and the like can be avoided, and support is provided for guaranteeing certificate identification accuracy.

The invention discloses an electric power field Chinese text error correction method and device, a storage medium and computing equipment, and the method comprises the steps: inputting a sentence in an electric power field Chinese text needing error correction into a trained electric power field pre-training language model, and obtaining a prediction character sequence of each word in the sentence; screening the predicted character sequence of each character to obtain a semantic candidate set of each character in the sentence; respectively inputting the same sentence into a pinyin confusion dictionary, a font confusion dictionary and a power field user-defined confusion dictionary to obtain a pinyin confusion set, a font confusion set and a user-defined confusion set of each character in the sentence; and performing error correction on characters in the sentence based on the semantic candidate set, the pinyin confusion set, the font confusion set and the user-defined confusion set. According to the method, the pre-training language model is adopted to replace a statistical language model, the text error correction scheme for the power industry is constructed, and the text error correction effect can be effectively improved.

A language sequence labeling method and includes: reading a first embedding representation of a language sequence, the first embedding representation including a character-level word embedding representation, a pre-trained word embedding representation, and a global word embedding representation of the language sequence, the global word embedding representation referring to a global context representation of the language sequence; performing first depth transformation (DT) encoding on the first embedding representation based on a first DT recurrent neural network (RNN), to output a first hidden-layer state representation corresponding to each word in the language sequence; and decoding the first hidden-layer state representations of the language sequence, to obtain a labeling result of one or more elements in the language sequence.

The invention discloses an evaluating method for the Chinese character writing correctness. The evaluating method comprises the following steps: pre-storing coded information of strokes of standard Chinese characters, then collecting information of strokes of Chinese characters written by a user, coding the strokes of the Chinese characters written by the user, analyzing and comparing codes of all strokes, computing a score, analyzing and comparing the reasonability of Chinese character structures, computing a score, and finally grading the Chinese character writing correctness. By adopting the method, a method which can automatically evaluate whether Chinese character writing strokes, stroke orders and Chinese character structures are correct and timely and accurately remind the user to correct mistakes can be obtained.

The invention discloses a text classification method, computing equipment and a computer storage medium, and the method comprises the steps: carrying out the training of an unsupervised corpus, extracting the semantic features of all characters in the unsupervised corpus and the semantic features of all common words, and obtaining a corpus feature set; performing word segmentation processing on the labeled sample corpus to obtain a word segmentation processing result, and determining common words and non-common words contained in the word segmentation processing result; carrying out segmentation processing on the non-common words to obtain each character contained in the non-common words; semantic features corresponding to common words contained in the word segmentation processing result and semantic features corresponding to all characters contained in non-common words are obtained from the corpus feature set; training to obtain a violation classification model according to the obtained semantic features and the annotation information of the annotated sample corpus; and based on the violation classification model, performing classification processing on the to-be-classified text.According to the method, semantic-level content classification can be realized, and the text classification accuracy is improved.

A computing device is described that outputs for display at a presence-sensitive screen, a graphical keyboard having keys. The computing device receives an indication of a selection of one or more of the keys. Based on the selection the computing device determines a character string from which the computing device determines one or more candidate words. Based at least in part on the candidate words and a plurality of features, the computing device determines a spelling probability that the character string represents an incorrect spelling of at least one candidate word. The plurality of features includes a spatial model probability associated with at least one of the candidate words. If the spelling probability satisfies a threshold, the computing device outputs for display the at least one candidate word.

A computer-implemented method can include receiving, at a computing device including one or more processors, an input from a user. The input can include one or more characters in a first writing system. The method can further include segmenting the input to obtain one or more segmentations, where each segmentation can include at least one segment including at least one character in the first writing system. A fuzzy model can be applied to the segmentations to obtain potential formal representations for the segmentations. Each of the potential formal representations can be in the first writing system and represent text in a second writing system. A plurality of character candidates can be determined based on the potential formal representations. Each of the plurality of character candidates can be a possible appropriate representation of the user input in the second writing system.

The invention discloses a Chinese named entity recognition method based on multilevel residual convolution and an attention mechanism, and belongs to the field of natural language processing. According to the method, a multi-level residual convolutional network of a joint attention mechanism is adopted. In order to solve the problem that the model efficiency is low when a traditional recurrent neural network processes sequence information, multi-stage residual convolution is introduced to obtain local context information in different ranges, the computing power of hardware is fully utilized, and the model efficiency is remarkably improved. In addition, the recurrent neural network cannot effectively acquire global context information due to gradient disappearance and gradient explosion problems, so that the performance of the network is greatly influenced. According to the method, an attention mechanism is introduced into the network, and the importance weight of each character is calculated by constructing the relationship between each character and the sentence, so that global information is learned. Finally, the transition probability of the character tag is calculated by using the conditional random field to obtain a reasonable prediction result, and the robustness of the named entity recognition model is further improved.

In one example, a computing device may include at least one processor that outputs a graphical keyboard. The at least one processor may determine a selection of one or more keys; and determine a character string based on the one or more keys in the selection, wherein the character string is not included in a vocabulary used by the computing device to identify character strings. The at least one processor may, in response to determining that the character string is not included in the vocabulary used by the computing device to identify character strings, associate the character string with an out-of-vocabulary language model score; and determine, based at least in part on a spatial model score associated with the gesture and the out-of-vocabulary language model score, a combined score that indicates a probability that the character string corresponds to the gesture; and output for display the character string.

The embodiment of the invention provides a data processing method and device, an identification method and device and computing equipment. A plurality of elements obtained by segmenting a training text are respectively expressed as nodes; wherein the element is composed of a single character or multiple characters; the plurality of different types of dictionaries are represented as nodes respectively; edges between the nodes are represented by using an incidence relation between the nodes, and a first graph is constructed; a text recognition model is trained by utilizing the first graph and training labels respectively marked for the plurality of elements; feature words in the text to be processed can be recognized and obtained by utilizing the text recognition model. According to the technical scheme provided by the embodiment of the invention, the text expression accuracy is improved, the model training accuracy is improved, and the text recognition accuracy is improved.

The invention discloses an audio and text synchronization method, computing equipment and a storage medium, and the method comprises the steps: obtaining an audio to be matched and a first text, and segmenting the first text to obtain a first statement set; segmenting the audio to obtain an audio fragment set, performing voice recognition on each audio fragment in the audio fragment set to obtaineach fragment statement, combining the fragment statements to obtain a second text, and obtaining a character sequence corresponding to the second text; sequentially extracting a first statement fromthe first statement set, obtaining a first character sequence corresponding to the first statement, extracting a second character sequence from a character sequence corresponding to the second text according to a preset window, matching the first character sequence with the second character sequence, and determining a third character sequence matched with the first character sequence, and establishing a synchronization relationship between the audio fragment corresponding to the third character sequence and the first statement. According to the scheme, accurate determination of the synchronization relationship between the audio fragments and the statements is realized.

Detection of typed and/or pasted text, insertion symbol tracking, and active element detection for computing systems are disclosed. In some embodiments, locations on a screen associated with a computing system at which a user always enters or pastes text may be identified, the text potentially including hot keys or other keys that do not cause the appearance of visible characters, and the physical location on the screen where the typing or pasting has occurred may be provided based on the location where the one or more characters have occurred, the location where the cursor is flickering, or the current resolution of both. This may be accomplished by identifying locations on the screen where changes have occurred and performing text recognition and/or inserted symbol detection on these locations. The physical location of the typing or pasting activity allows for the determination of active or focus elements in an application displayed on the screen.

The invention discloses a question and answer task downstream task processing method and model, and the method comprises the steps: obtaining a context representation HCKey perceived by key information and a question representation HQKey perceived by key information, and generating a context representation G perceived by a question; calculating an update vector z and a memory weight g based on G, and updating G to obtain an output vector Gg; generating a context granularity vector GC and a sequence granularity vector GCLS, generating an output vector Cout, using softmax to calculate the probability that each character in the context serves as an answer starting and ending position, and extracting a continuous subsequence with the maximum probability to serve as an answer. The invention provides a two-way cascading attention mechanism, and constructs a mechanism taking fine reading and slight reading as a whole and a multi-granularity module based on a granular computing thought, so that the model effectively pays attention to and screens effective information, better understands texts under various granularities, gives out more accurate answers, and improves the accuracy of text analysis. And the performance is improved on the basis of a baseline model.

The invention discloses a text processing method and device and computing equipment. The method comprises the steps that a stroke vector matrix is generated based on stroke information of a single character in a word, the single character is an ideographic character, and strokes are minimum continuous stroke units of the single character; inputting the stroke vector matrix into a first neural network for processing to obtain word vectors of words; word vectors of all words in the word sequence are obtained and input into a second neural network to be processed, hidden layer representation of the word sequence is obtained, and the hidden layer representation is used for representing the distribution probability of the next word of the word sequence.

The invention discloses a script verification method, which is executed in an operating system and comprises the following steps of executing a script file; judging whether the script file is a to-be-verified script file or not; if the script file is the to-be-verified script file, acquiring a script file path based on the script file, and writing the script file path into the character device file; reading the character device file so as to obtain a script file to be verified based on a script file path in the character device file; and performing signature verification on the to-be-verified script file, and if verification succeeds, continuing to execute the script file. The invention also discloses a corresponding script signature method and computing equipment. According to the script verification method, the running safety of the script can be fully guaranteed.