49 results about "Text alignment" patented technology

The invention relates to an alignment system of an on-line speech text and a method thereof. The system comprises a text processing module, an error detection module, an error recovery module and a compulsory alignment module. The compulsory alignment module comprises a characteristics extraction module, a search space construction module and an alignment decoding module. The error recovery modulecomprises a language model estimation module, a language model interpolation module, a speech recognition module and a text alignment and similarity calculation module. Method for the system and themethod to detect end of a sentence is an improvement of a conventional method based on Viterbi alignment, information of a search space by beam search is used for estimating activity degree A (t, somegae) of the search space at the end of the sentence, and estimating sentence end time in partial meaning *<*>. The system and the method have the function for automatically detecting and jumping overunmatched segment errors in a text and a speech; can generate online input speech current and corresponding text alignment result at real time and can process a long text with errors.

The invention provides an automatic oral English marking method based on feature fusion. A multi-feature fusion method is used for marking intonation tests in a test of oral English in large scale. A continuous speech signal and a speech recognition text are used as research objects, and the intonation tests in the test of oral English are analyzed from two different angles consisting of speech and text. On the one hand, input speech is analyzed, and a speech feature is extracted and matched and compared with a reference standard; on the other hand, the speech recognition text is compared with the reading text via a text alignment method, and a marking mechanism gives assessment and a mark according to a similarity level. Experimental results show that the method is relatively low in algorithm complexity, and a marking result accords with subjective sensation of people.

The present invention relates to a speech recognition error correction method and device. The method includes the following steps that: historical recognition results obtained by a current speech recognition model are acquired, wherein the historical recognition results include a plurality of pieces of text information; logical analysis is performed on the plurality of pieces of text information, and target text information with repeat logic is extracted; the target text information with repeat logic is subjected to wrong channel statistics, so that a text alignment model can be generated; speech information inputted by a user is received; and the speech information is decoded according to the current speech recognition model and the text alignment model, so that a target speech recognition result is obtained. With the speech recognition error correction method and device provided by the technical schemes of the invention adopted, the obtained speech recognition result can be more accurate and better meet the requirements of the user, and therefore, the use experience of the user can be enhanced.

The multi-level vision-text semantic alignment model MSAVT used for image-text matching is provided, the news event retrieval method based on the multi-level vision-text semantic alignment model MSAVT used for image-text matching is provided, news event cross-modal image-text search is achieved, and the current news retrieval requirement is met. The image-text alignment precision of the cross-modal retrieval model provided by the invention is higher, and when the cross-modal retrieval model is applied to news event cross-modal image-text retrieval, indexes such as recall rates of multiple levels, average accuracy and the like are remarkably improved. And meanwhile, a pre-trained BERT model is introduced to extract text features, so that the generalization performance of the algorithm is improved. The model adopts a public space feature learning method, vector representations of images and texts can be independently obtained, namely, vector representations of retrieval results can be stored in advance, retrieval time is short, and the method can be applied to actual scenes.

The invention belongs to the technical field of language translation, and discloses a term translation method and device, and the method comprises the steps: converting a term library into dictionaries in key and value forms, and storing the dictionaries in a database; Inputting the normal to-be-translated text into a machine translation terminal, and automatically generating a translated text andcorresponding text alignment information through machine translation; And according to the alignment information, replacing the corresponding term with a specified explanation. According to the method, machine translation and term translation in manual translation are combined, so that more results summarized by human experience are received by machine translation, the translation quality is further improved, and the expectation of translation workers is achieved; According to the method, the user can be allowed to use the pre-defined term library in the using process and apply the term library to machine translation, so that the term translation in the machine translation better conforms to the expected effect.

The invention discloses a data entity relationship extraction method based on deep learning. The method comprises the following steps: 1, obtaining training data by adopting an open entity relationship extraction method; mapping the data entity relationship instances to a large number of texts in an entity knowledge base by means of a DBPedia, OpenCyc, YAGO or FreeBase entity knowledge base, obtaining training data through a text alignment method, and obtaining training corpora with noise annotations; re-annotating the noise annotation by adopting a supervised entity relationship extraction method, and training a machine learning model on the basis of annotated training data; and extracting a data entity relationship corresponding to the entity pair combination. According to the method, the data entity relationship is extracted by combining the open entity relationship extraction method and the supervised entity relationship extraction method. The training data acquisition efficiency of the open entity relationship extraction method is high. The training data acquired by the supervised entity relationship extraction method is high in accuracy. The extraction efficiency and the accuracy of the entity relationship are improved.

The invention discloses a difference description statement generation method and device, equipment and a medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: carrying out the feature splicing of image coding features and text coding features, inputting the spliced coding features into a preset image-text alignment unit constructed based on a preset self-attention mechanism to obtain spliced alignment features, using a preset noise monitoring unit constructed based on a preset self-attention mechanism and a preset cross-attention mechanism to process image alignment features and text alignment features obtained after splitting the text coding features and the spliced alignment features so as to extract difference signals; the difference description statement is generated by utilizing the preset difference description generation algorithm and based on the difference signal, and therefore, the part, which cannot be aligned with the image, in the human language text is positioned based on the preset cross-attention mechanism, and the corresponding interpretation description is given, so that the problem that a computer cannot perform normal reasoning due to human language errors is solved.

The invention discloses a video detection method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring audio information and a subtitle text corresponding to a target video clip in a to-be-detected video, wherein the subtitle text comprises at least one text unit; performing voice recognition on the audio information to obtain a target text corresponding to the audio information, the target text comprising at least one text unit; according to the text unit of the target text and the text unit of the subtitle text, performing text alignment on the target text and the subtitle text to obtain a text alignment result; and detecting the integrity of the target video clip according to the text alignment result. According to the embodiment of the invention, the video content integrity detection can be performed based on the audio and subtitles of the target video clip, the video integrity detection efficiency is improved, and the detection cost is reduced.

The invention discloses an original translated text alignment method and apparatus, and belongs to the technical field of translation. The method comprises the steps of generating a two-dimensional table of original text sentence identifiers and translated text sentence identifiers; determining a sentence matching rate between each original text sentence and each translated text sentence, and filling a corresponding position in the two-dimensional table with the sentence matching rate; according to at least one set matching rate, gradually scanning the two-dimensional table, and in each row and each column of the two-dimensional table, at most determining a first sentence matching rate matched with the set matching rate; and according to the determined first sentence matching rate, aligning the corresponding original text sentences and translated text sentences. Based on the sentence matching rate among all the sentences participating in alignment, original translated text alignment isperformed, so that the original translated text alignment accuracy is improved.

The invention provides a character segmentation and recognition method based on a CTC deep neural network. The method comprises the following steps: a1, extracting features of an input image by usinga CNN; a2, carrying out the CELL segmentation on the features extracted in a1, fixing the height and width of CELL, and determining the number of CELL by the length of the image; a3, directly segmenting and classifying each CELL of the determined features, and outputting segmentation signals; a4, calculating the loss between the real segmentation signal and the segmentation signal output by the model by using CTCLOSS, feeding back the loss condition and training the whole model; a5, segmenting the text by using the segmentation signal output in the step a3, carrying out CNN + softmax classification identification on a single character, mapping a real segmentation signal from the annotated text, and automatically solving the text alignment problem by using the CTCLOSS. According to the invention, the OCR recognition speed is improved, and the recognition optimization is targeted after the characters are cut into single characters, so the final precision is improved; a recognition framework is improved, and a recognition process is separated into character segmentation and single character recognition, so optimization can be separately carried out in a targeted manner.

The invention provides a sound and text realignment and information presentation method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a target audio, a pre-editing recognition text and a post-editing recognition text, wherein the pre-editing recognition text is a recognition text obtained through the automatic voice recognition of the target audio, and the edited text is a text obtained by editing the recognition text before editing; performing forced alignment on the target audio and the recognition text before editing to determine audio starting and ending time corresponding to each character in the recognition text before editing; performing text alignment on the pre-edited recognition text and the post-edited recognition text to determine a character corresponding to each character in the post-edited recognition text in the pre-edited recognition text; and for each character in the edited recognition text, determining the audio starting and ending time of the character corresponding to the character in the pre-edited recognition text as the audio starting and ending time of the character. According to the invention, high-precision sound and text re-alignment between the target audio and the edited recognition text is realized.