JP2009187349A - Text correction support system, text correction support method and program for supporting text correction - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To correct character information output for a large amount by voice recognition processing into more easily understandable information in a short period of time and to provide it. <P>SOLUTION: A data processor to which character string information as the expression data made into texts of voice obtained by the voice recognition processing is input includes: a notation level correction means for generating a corrected text in which a word which may be an error at a notation level is corrected for the text indicated by the input character string information; a corrected text display means for causing the corrected text generated by the notation level correction means to be displayed on the screens of respective terminals for readers and respective terminals for operators; a correction instruction receiving means for receiving a correction instruction to the text under display by the operation of the operator from the terminal for the operator; and an image updating means for reflecting correction by the correction instruction received by the correction instruction receiving means on the screens of the respective terminals for the readers and the respective terminals for the operators specify a correction part. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention provides a sentence correction support system, a sentence correction support method for presenting a correction of a sentence indicated by character string information, which is voiced expression data obtained by voice recognition processing, as a sentence that can be easily understood by a reader, And a sentence correction support program.

  As an implementation method of voice recognition that provides information by voice that is exchanged at a meeting or voice from a medium such as a radio or television, for example, by analyzing voice information input via a microphone or the like, There is a method of outputting the result to an editor screen or the like. As a result, it is possible to provide more accurate information to hearing-impaired persons, elderly persons, and foreigners who are currently learning Japanese, who are difficult to acquire information by voice.

  By the way, in voice data processing of a computer, voice communication makes a lot of words uttered in a short time, and barriers such as surrounding noise make it difficult to identify voices.

  Thus, as an example of a technique for correcting a character string that has been erroneously recognized due to various causes to a correct character string, for example, Patent Document 1 discloses a character data correction device that performs correction with an operator interposed therebetween. Are listed. In the character data correction device described in Patent Document 1, the mismatched portion selecting means selects a mismatched portion with the voice by the operation of the operator for the character string obtained by converting the voice of the program contents into text data by voice recognition. I do. Then, the corrected text data that has been corrected in accordance with the content of the selected mismatched portion is input and added to the original text data to generate and output corrected additional text data.

  Further, for example, Patent Literature 2 describes a speech recognition method that corrects an error in a speech recognition result with reference to constraint information based on a semantic relationship between words. In the speech recognition method described in Patent Document 2, the semantic relationship between words and the degree of association indicating the degree are stored, and the most probable word string is selected based on the degree of association.

JP 2004-151614 A JP 2007-256836 A

  However, in the method of correcting by simply interposing an operator as described in Patent Document 1, the operator has to select a mismatched portion by following all the character strings to be entrusted, and the characters for a long time. In the case of chasing, the fluctuation of the sentence that occurs for that purpose becomes remarkable. Here, “sway” means a state in which an expression that is correct as a pronunciation but inappropriate as a written sentence is included, such as a homonym. As a result, it becomes difficult to grasp the main points of the information, and correct reading in real time is hindered. In this case, for example, even if a hearing-impaired person tries to use it for participating in a place for voice communication, there is no use.

  In addition, in the method of automatically correcting by computer analysis as described in Patent Document 2, there is a risk that a low accuracy result may be obtained. Moreover, if all spoken speech information is analyzed by computer analysis alone and is always output accurately, the time required for the analysis becomes longer, which also hinders correct reading in real time.

  In view of the above, an object of the present invention is to provide character information output in a large amount by voice recognition processing by correcting it to more easily understandable information in a short time.

  A sentence correction support system according to the present invention is a sentence correction support system for presenting a sentence indicated by character string information, which is voiced expression data obtained by voice recognition processing, as a sentence that can be easily understood by a reader. A data processing device for inputting character string information, which is voiced expression data, one or more reader terminals used by a reader, and one or more operator terminals used by an operator, The data processing device includes a notation level correcting unit that generates a corrected sentence by correcting a word that may be erroneous at a notation level for a sentence indicated by input character string information, and the notation level correcting unit. The corrected text display means for displaying the generated corrected text on the screen of each reader terminal and each operator terminal, and the operator terminal. A correction instruction receiving means for receiving a correction instruction for the displayed text by the operation of the data, and specifying a correction location on the screen of each reader terminal and each operator terminal for the correction by the correction instruction received by the correction instruction receiving means It is characterized by comprising a screen updating means for reflecting it as possible.

  Further, the sentence correction support method according to the present invention is a sentence correction support method for presenting a sentence indicated by character string information, which is voiced expression data obtained by voice recognition processing, as a sentence that can be easily understood by a reader. A data processing device to which character string information, which is voiced expression data of speech, is input, corrects a word that may be erroneous at the notation level for the text indicated by the input character string information. The corrected text is generated, and the generated corrected text is displayed on the screen of each reader terminal and each operator terminal. From the operator terminal, a correction instruction for the displayed text by the operation of the operator is received, and each reader The correction based on the correction instruction received from the operator terminal is reflected on the screen of the operator terminal and each operator terminal so that the correction location can be specified. To.

  In addition, the sentence correction support program according to the present invention provides a sentence correction support for presenting a sentence indicated by character string information, which is voiced expression data obtained by voice recognition processing, as a sentence that can be easily understood by a reader. Program for supporting correction of text applied to a data processing device to which character string information that is voiced text representation data is input, for a sentence indicated by character string information input to a computer , A process for generating a corrected sentence that corrects a word that may be erroneous at the notation level, a process for displaying the generated corrected sentence on the screen of each reader terminal and each operator terminal, from the operator terminal, Processing to accept correction instructions for the displayed text by the operator's operation, and for each reader terminal and each operator terminal The surface, the correction by the correction instruction received from the operator terminal, characterized in that to execute a process of identifiable reflect modifications locations.

  ADVANTAGE OF THE INVENTION According to this invention, the character information output in large quantities by voice recognition processing can be corrected and provided in a short time to more easily understandable information.

  Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a configuration example of a text correction support system according to the present invention. The text correction support system shown in FIG. 1 is a text correction support system for correcting and presenting a text indicated by character string information, which is voiced expression data obtained by voice recognition processing, into a text that is easy for a reader to understand. The system is a data processing device 1 to which character string information, which is voiced expression data, is input, one or more reader terminals 2-1 used by a reader, and one or more terminals used by an operator And an operator terminal 2-2.

  The data processing apparatus 1 also includes a notation level correction unit 11, a corrected text display unit 12, a correction instruction reception unit 13, and a screen update unit 14.

  The notation level correcting means 11 generates a corrected sentence in which a word that may be erroneous at the notation level is corrected for the sentence indicated by the input character string information. For example, the notation level correcting unit 11 compares each word included in the sentence indicated by the character string information with a word dictionary registered in advance, and if it is not registered in the word dictionary, the notation level is corrected. Since there is a possibility of an error in the level, the corrected sentence may be generated by converting the word into hiragana. In addition, the notation level correcting unit 11 collates each word included in the sentence indicated by the character string information with a word dictionary registered in advance, and words having different notations having the same reading in the word dictionary are obtained. If the word is registered, the corrected document may be generated by converting the word into hiragana on the assumption that the word may have an error at the notation level.

  The corrected text display means 12 displays the corrected text generated by the notation level correction means 11 on the screen of each reader terminal and each operator terminal.

  The correction instruction receiving means 13 receives from the operator terminal 2-2 a correction instruction for the displayed text by the operator operation.

  The screen updating unit 14 reflects the correction based on the correction instruction received by the correction instruction receiving unit 13 on the screen of each reader terminal 2-1 and each operator terminal 2-2 so that the correction location can be specified.

  FIG. 2 is a block diagram showing another configuration example of the text correction support system according to the present invention. As shown in FIG. 2, the data processing apparatus 1 may further include a sentence processing unit 15, a processed document display unit 16, and a mode setting unit 17.

  The sentence processing unit 15 generates a processed sentence obtained by processing the corrected sentence generated by the notation level correcting unit 11 according to a predetermined display mode.

  The processed document display means 16 associates the corrected text being displayed with the processed text corresponding to the display mode set for the reader on the screen of the reader terminal 2-1 used by each reader. At the same time, at least the processed text corresponding to the display mode displayed on the reader terminal is displayed on the screen of the operator terminal 2-2.

  In the case of this example, the correction instruction receiving means 13 receives a correction instruction for the processed text by the operator's operation from the operator terminal. Further, the screen updating unit 14 reflects the correction by the correction instruction on the processed text received by the correction instruction receiving unit 13 on the screens of the reader terminals and the operator terminals so that the correction part can be specified.

  The mode setting means 17 receives a display mode switching instruction by the reader's operation from the reader terminal 2-1. The mode setting means 17 is, for example, a detailed mode that presents almost the same content as the utterance content, a subtitle mode that eliminates redundancy from the utterance content, and a simple mode that presents the main points of the utterance content in a list as display modes. And may be switched according to the level of the reader.

  In addition, when the mode setting unit 17 accepts a display mode switching instruction, the processed text display unit 16 is processed at least in the requesting terminal for the reader according to the display mode set for the reader. May be displayed.

  Hereinafter, more specific embodiments will be described. FIG. 3 is a block diagram illustrating a more specific configuration example of the text correction support system. A text correction support system 1000 shown in FIG. 3 includes a speech recognition device 100, a data processing device 200, and one or more character input / output devices 300.

  The voice recognition device 100 is based on existing technology, and is a microphone device that inputs voice, an analog-digital converter device that converts voice, which is an analog signal, into a digital signal, and software that implements voice recognition processing (more specifically, And a processor such as a CPU that operates in accordance with the software.

  Note that the speech recognition apparatus 100 according to the present embodiment converts at least input speech into a character string and outputs it to the data processing apparatus 200. Note that the speech recognition result output by the speech recognition apparatus 100 may include not only a character string indicating a word expressed by the input speech but also grammatical information such as a word break in the character string.

  The data processing device 200 performs control for outputting to the character input / output devices 300 while correcting the character string of the speech recognition result input from the speech recognition device 100 as necessary. The data processing device 200 includes a recognition result correction unit 210, a mode setting unit 220, a text processing unit 230, a text editing unit 240, and a screen control unit 250.

  The recognition result correction unit 210 corrects an error at a notation level such as erroneous conversion on the character string information obtained by the speech recognition apparatus 100. The recognition result correction unit 210 is a processing unit that realizes the function of the above-described notation level correction unit 11.

  The mode setting means 220 sets a text display mode for each reader. In the present embodiment, a detailed mode, a caption (summary) mode, and a simple mode are prepared as text display modes. The detailed mode is a mode in which a sentence expressed by input speech is displayed as it is. The subtitle mode is a mode in which a sentence expressed by input speech is displayed in a summarized form like a subtitle. The simple mode is a mode in which keywords that are points of information acquisition are listed in a sentence expressed by input speech. For example, the mode setting means 220 sets a text display mode for the reader in response to a request from the reader. The mode setting unit 220 is a processing unit that realizes the function of the mode setting unit 17 described above.

  The sentence processing unit 230 processes a sentence in accordance with each mode. The sentence processing unit 230 is a processing unit that realizes the function of the mode setting unit 17 described above.

  The text editing unit 240 edits the text processed by the text processing unit 230 in accordance with a correction operation from the operator. The text editing unit 240 is a processing unit that realizes the function of the above-described correction instruction receiving unit.

  The screen control unit 250 performs processing for causing each character input / output device 300 to display a sentence generated from the character string of the voice recognition result at a predetermined timing. The processing performed by the screen control unit 250 includes processing for causing each character input / output device 300 to update the portion edited (corrected) by the text editing unit 240. In the present embodiment, the screen control unit 250 causes each character input / output device 300 to display a sentence when the correction by the recognition result correction unit 210 is completed, and then when the display mode of the sentence is set for each reader. When the text editing unit 240 edits the text, the text editing section 240 redisplays the text so that the fact and the location can be understood. The screen control unit 250 is a processing unit that realizes the functions of the corrected text display unit 12, the screen update unit 14, and the processed text display unit 16 described above.

  The character input / output device 300 is an input / output device for displaying a sentence corrected, processed, and edited by the data processing apparatus 200 and inputting an operation on the sentence. The character input / output device 300 is realized by, for example, a personal computer (PC) having a network communication function including an output device such as a display device and an input device such as a mouse and a keyboard. The embodiment will be described using an example in which two PCs are arranged. Here, it is assumed that an operator supports character editing on one of the two units, and a user (reader) who cannot easily obtain information by voice browses the corrected text on the remaining one unit. It is possible to arrange two or more units for the operator or two or more units for the reader. Further, a configuration in which one of the CPUs of the PC that implements the character input / output device 300 implements the data processing device 200 is also possible.

  Next, the operation of this embodiment will be described. FIG. 4 is a flowchart showing an operation example of the text correction support system 1000. In the example shown in FIG. 4, the data processing device 200 has a series of character string information (for example, a character string for a sentence divided into lengths until the next time when a person speaks in front of a person. It is a loop structure in which the processing of steps S10 to S70 is repeated for each unit of information. Steps S60 and S70 may be processed at a timing according to the operation from the operator, away from the loop structure.

  Although not shown in the figure, first, speech uttered by a human is input to the speech recognition apparatus 100. The voice recognition device 100 converts the voice input by the voice recognition processing into character string information that is textual expression data of the voice, and outputs it to the data processing device 200. Note that the character string information includes not only information on the character string when the speech is written in text, but also information on readings and phrase breaks.

  In the data processing device 200, the recognition result correction unit 210 uses the character string information output from the speech recognition device 100 as input information, and corrects the text that is the recognition result of the speech recognition device 100 indicated by the character string information ( Step S10). Note that the character string information indicating the sentence corrected by the recognition result correcting unit 210 is input to the sentence processing unit 230 as original data for displaying the processed sentence.

  The recognition result correcting unit 210 corrects an error at a notation level such as erroneous conversion for the sentence indicated by the character string information. Note that an error at the notation level means a word that exists as Japanese (word), such as a homonym, but is an error as a written sentence.

  In this embodiment, the recognition result correcting unit 210 has a word dictionary and corrects an error at the notation level by collating with the word dictionary. In this embodiment, only correction within a range that can be collated with the word dictionary is performed, and correction that is determined from the context before and after is not performed. Such correction is supplemented by utilizing the existing invention in the text processing means 230 or by a correction instruction by the operator.

  When the correction by the recognition result correction unit 210 is completed for a series of character string information, the screen control unit 250 sets each corrected text obtained by the recognition result correction unit 210 as a primary sentence and outputs each character input / output device. It is displayed on the screen 300 (step S20). For example, each character input / output device 300 is made to access the data processing device 200 in advance, and the screen control means 250 includes corrected character string information for the terminal application software downloaded at that time. The corrected text may be displayed on the screen of each character input / output device 300 by transmitting a screen update request message. In addition, for example, screen information of a Web page that incorporates a process of performing automatic update periodically is prepared in advance, and each character input / output device 300 is allowed to access the Web page to thereby input / output each character. The corrected text may be reflected (published) on the Web page in response to an automatic update request from the device 300 and displayed.

  Here, since the subsequent processing of the mode setting means 220 and the text editing means 240 requires the intervention (operation) of the reader or the operator, there is a possibility that a time loss of about several tens of seconds may occur in terms of information acquisition. Considering the fact that it is expensive, it is displayed as the primary sentence without waiting for the operation from the reader or the operator in order to make it possible to grasp the situation (the main points of the content being spoken) as soon as possible. FIG. 5 is an explanatory diagram illustrating an example of a screen displayed on the character input / output device 300. For example, as shown in FIG. 5, the screen control unit 250 displays the primary sentence on the upper part of the screen, and then displays the processed sentence or the edited sentence on the lower part of the screen as the secondary sentence. It is displayed in a configuration that does not impair the real-time property and allows easy understanding of correct information.

  Next, the mode setting means 220 sets a text display mode for each reader in response to a request from the reader (step S30). In this example, as the text display mode, we prepare a “simple mode” that provides rough information, a “subtitle mode” that is the same level as subtitles, and a “detail mode” that refines the recognition results. The configuration can be selected from the three types. The mode setting means 220 receives the sentence display mode designated by the reader in the character input / output device 300 as an event, and sets or sets the reader-specific sentence display mode managed (held) by the data processing apparatus 200. Or change it. And the process for outputting the processed text according to the set mode is performed. For example, the text processing unit 230 is notified that a certain character input / output device 300 has set or changed the mode.

  A plurality of modes may be set for the character input / output device 300 used as an operator terminal. For example, when editing a sentence on one operator terminal, all the processed sentences corresponding to each mode are displayed on the one operator terminal. It is also possible to have the operator check only the detailed mode and reflect the correction to the detailed mode in other modes.

  Next, the sentence processing unit 230 processes the sentence (here, the sentence corrected by the recognition result correcting unit 210) in accordance with each reader mode set by the mode setting unit 220 (step S40). In the case of the “simple mode”, for example, by applying the keyword extraction method disclosed in Japanese Patent Application Laid-Open No. 2005-107793, character string information expressed in a format for enumerating keywords that can be information acquisition points is displayed. It may be generated. In the case of “subtitle mode”, for example, an automatic summarization device disclosed in Japanese Patent Application Laid-Open No. 2007-233823 is applied to generate character string information that is expressed in a condensed form like subtitles. May be. In the case of “detail mode”, for example, a sentence processing device disclosed in Japanese Patent Application Laid-Open No. 11-171058 is applied, and the recognition result is expressed in a format that is refined so that it can be read as correct Japanese. Character string information to be generated may be generated.

  It should be noted that the process of generating the processed text corresponding to each mode is performed in advance without waiting for the notification from the mode setting unit 220, and when the notification from the mode setting unit 220 is received, the screen control unit 250 is notified. The processed text corresponding to the mode set for the reader may be output. Further, for example, an initial value of the mode may be set in advance, and the process may be performed in the currently set mode until a switching request (event) is received from the reader. If there is a switching request from the reader, the processing after step S30 may be performed again.

  When the text processing unit 230 completes the text processing according to the mode set by a certain reader for a series of character string information, the screen control unit 250 displays the reader's text display mode obtained by the text processing unit 230. The processed text corresponding to the text is displayed as a secondary text on the screen of the character input / output device 300 used by the reader (step S50). In addition, the process which concerns on the screen display of a secondary sentence may be the same as that of step S20. For example, the screen control unit 250 transmits the screen update request message including the character string information of the processed text to the terminal application software installed in the target character input / output device 300, thereby The processed text may be additionally displayed on the screen of the character input / output device 300. Further, for example, the processed text may be additionally displayed by reflecting (posting) the processed text on the Web page in response to an automatic update request made according to the screen information of the Web page from the target character input / output device 300. Good. When displaying the secondary text, it is preferable to display the secondary text in association with the primary text that is already displayed.

  Next, the text editing unit 240 receives a text correction instruction from the character input / output device 300 used as an operator terminal, and edits the text according to the instruction (step S60). For example, when the text editing unit 240 receives information indicating the setting of the range for the displayed text, which is input in response to an operator operation, the text editing unit 240 stores the information of the setting range and corrects characters for the character string within the setting range. A column may be input. And the information of the correction character string within the input setting range may be received. For example, the text editing unit 240 incorporates a function for accepting a range setting operation for a displayed text and inputting a corrected character string in the screen information to be displayed on the operator terminal, and information necessary according to the screen operation. May be transmitted to the data processing apparatus 200 and received. Note that the text editing unit 240 may perform the screen control processing in steps S20 and S50 for the operator terminal.

  In addition, the text editing unit 240 instructs the screen control unit 250 to update the screen of the text editing portion by the correction when the correction is confirmed.

  The screen control means 250 updates the screen display of each character input / output device 300 so that the part edited by the text editing means 240 can be known by the correction instruction from the operator (step S70). For example, the screen control unit 250 transmits a screen update request message including information on the edited portion (for example, the content after correction, the start position of the correction range, the length, the transmission counter of the corresponding sentence, etc.) The edited text may be displayed on the screen of the input / output device 300 so that the part can be understood. The transmission counter is information for identifying a sentence in a unit processed as a series of sentences. For example, each time character string information is transmitted in step S20, the character string information is identified. It is a number assigned for the purpose. Each character input / output device 300 tracks the processed text to be updated based on the transmission counter transmitted from the data processing device 200, and then, within the range indicated by the transmitted start point position and length. The character string may be overwritten with the correction contents by the operator. At that time, an expression (color etc. different from other character strings) distinguished from other character strings is used.

  Further, for example, the screen control unit 250 edits the latest edited text on the Web page in response to an automatic update request made according to the screen information of the Web page displayed on each character input / output device 300. The screen display may be updated by reflecting (posting) with emphasis.

  The operation of this embodiment will be described below using a specific example. FIG. 6 is an explanatory diagram showing an example of an implementation image of the text correction support system 1000. In the example illustrated in FIG. 6, a server connected to a microphone is used as the voice recognition device 100 and the data processing device 200. A PC connected to the server via a network such as a LAN is used as the character input / output device 300 (reader terminal) for the reader. Further, as the operator character input / output device 300 (operator terminal), a PC connected to the server via a network is used. In this example, the speech recognition apparatus 100 is realized by software (more specifically, a CPU that operates according to the software) having an existing speech word processor function (a function for converting speech into a character string) that the server has. . The data processing device 200 is realized by software having a sentence correction function (function of each processing means of the data processing device 200) of the server. Note that the software having the sentence correction function is implemented as plug-in software of software having the voice word processor function.

  In addition, each PC that implements the character input / output device 300 can perform communication based on a communication protocol such as TCP / IP. Further, client software may be installed that communicates with the server to update the screen display or notify a correction instruction. It is also possible to use existing software such as a browser without preparing special client software.

  In the server, voice information output from a speaker is input via a microphone, and software having a voice word processor function for realizing the voice recognition apparatus 100 converts the information into character string information by performing voice recognition processing. The converted character string information is sequentially input to software having a sentence correction function for realizing the data processing device 200.

  When the character string information is input, the recognition result correction unit 210 realized by software having a sentence correction function performs correction processing. Here, “As I already have a lot of money at Tsuchibun, etc., I'll start with the Nendoroid Nendoroids, starting from the first half of the year.” Character string information indicating the sentence that "We will conduct disaster prevention drill for your company building in 2007 from 10:05 am on Monday, June 4th" in response to what was spoken. Is described as an example.

  FIG. 7 is a flowchart illustrating an example of the correction process of the recognition result correction unit 210. The recognition result correction unit 210 performs a series of processes according to the flowchart shown in FIG. 7 for each word included in the character string information output from the speech recognition apparatus 100, so that the first sentence which is a corrected sentence is obtained. The next sentence (more specifically, character string information indicating the first sentence) is generated. Here, each word included in the character string information output from the speech recognition apparatus 100 is checked against the word dictionary to determine whether or not Japanese is correctly used as a document. A primary sentence is generated by temporarily converting a phrase or phrase that cannot be determined to be erroneous into hiragana.

  FIG. 8 is an explanatory diagram showing an example of a word dictionary format. In the word dictionary, for each word, a reading and a written character string are associated with each other and registered. In the example shown in FIG. 8, the information of each word is registered in the format of word [] [] []. Note that a number for identifying the first character (reading) is assigned to the first array subscript of word [] [] []. For example, a word starting with “a” is registered as word [11] [] []. A word starting with “I” is registered as word [12] [] []. A word starting with “o” is registered as word [15] [] []. A word beginning with “ka” is registered as word [21] [] []. A word starting from “sa” is registered as word [31] [] []. That is, in the example shown in FIG. 8, the tenth digit of the first array subscript number corresponds to the order of consonants in the order of 50 tones, and the first digit corresponds to the order of vowels. Instead of separating consonants and vowels, numbers may be assigned in order to the sounds in the order of the 50 tones as array subscript numbers.

  The next array subscript is assigned a number for identifying the first written character of a word starting from a predetermined character (reading). For example, in word [11] [1] [], a word whose first written character is “love” is registered. Also, a word whose first written character is “indigo” is registered in word [11] [2] []. It should be noted that the first subscript may be registered in the order of the Japanese syllabary without using an array subscript.

  Further, in the word dictionary, as a misuse phrase, a word that is read differently between a word to be spoken and a word to be written may be registered in association with the reading of the utterance word and the written character string. . In the example shown in FIG. 8, an example is shown in which the utterance word reading “Touri” and the written character string “Dori” are registered in association with each other as the misuse phrase registration word.

  In addition, the word dictionary may include information on words and expressions with high combination probabilities such as catching and receiving each word.

  First, as shown in FIG. 7, the recognition result correction unit 210 fetches information in the input buffer in order to use the text (character string information) output by the speech recognition apparatus 100 as input information (step S101). At the time of fetching, the sentence is broken down into several words, and the following scan is performed for each broken word. That is, the recognition result correction unit 210 loops the scanning process of steps S103 to S108 by the number of decomposed words (step S102). Hereinafter, the word currently being scanned is referred to as word X.

  As the scanning process in the loop, the recognition result correction unit 210 first identifies what the first character of the word X is (step S103). Next, collation with the word dictionary is performed based on the identified character (step S104). For example, when the word X “bare hand” is scanned, the first character is identified as “su” in step S103, and collation in the word dictionary may be performed based on the “su”. In the example shown in FIG. 8, it is only necessary to scan the word [33] [] [] portions as the collation destination that is the registered word of the first character “su”.

  If a word that matches the first character is registered (Yes in step S105), the registered word starting with the first character is scanned in order to perform the next process (step S106). Here, the recognition result correction unit 210 checks whether or not the word X to be collated and the registered word that is the collation destination match (step S107). The same processing is repeated by moving to a word and returning to step S107 (step S108).

  On the other hand, if the word X to be collated matches the registered word that is the collation destination, it is determined whether or not the homonym of the registered word exists in the word dictionary (step S109). If there is a homonym, the word X is considered to contain ambiguity as Japanese, and the word X is converted to hiragana notation (step S110). On the other hand, when the homonym does not exist, the registered word is used as it is (step S111).

  For example, the recognition result correction unit 210 determines whether or not there is an array in which the same word as the word X is registered in the array group (word [33] [] []) as a collation destination. If it exists, it may be determined whether or not there is another sequence that reads the same way. In the example illustrated in FIG. 8, when the word X is “bare hand”, it is determined that the “bare hand” registered as word [33] [1] [0] matches the word X, and further, word [ 33] It is determined that “already” registered as [2] [0] is a homonym of the word X. In such a case, a process of replacing the recognition result “bare hand” with a hiragana “sude” is performed.

  In step S104, if there is no match in the collation by the first character of the word X (No in step S105), it is regarded as not correct Japanese in the first place, and if the word X is expressed in kanji, the sentence is difficult to read. In this case, it is converted into a notation in hiragana (step S112). In addition, even if it is accurate Japanese, if it is difficult kanji, it is possible not to register it in the word dictionary and to convert it into notation in hiragana.

  In addition, while repeating the scanning process of step S106 to step S108, the scanning process is finished without the matching of the word X in the array group (a word group of which the first character matches the word X) as the collation destination. Also, the word X is regarded as not being in correct Japanese, and is converted into the hiragana notation (step S112).

  When the above-described series of processing is completed for the word X, the verification target (word X) is moved to the next decomposed word, and the same processing is repeated by returning to step S103 (step S113).

  Here, it is assumed that the sentence that is the result of the speech recognition by the speech recognition apparatus 100 is expressed in a kanji-kana mixed format, and that the internal information also includes information of only the hiragana text. . Therefore, in the processing of steps S110 and S112, it may be converted by extracting to which part of the sentence only the hiragana the kanji corresponds by pattern matching. If the speech recognition result is expressed in the hiragana format, it is sufficient to perform collation with the reading of the registered word in step S107.

  In addition, for an idiom such as “Touri”, it is only necessary to register in advance in the word dictionary that “Touri” is correctly written as “Dori” and to collate it.

  In this example, the parts that were originally written as “already”, “street”, and “disaster prevention” are recognized as “bare hands”, “touri”, and “hidden wife”, respectively, and the meaning is difficult to pass in Japanese It has become. The recognition result correcting unit 210 confirms that the phrase “bare hand” can be converted into “bare hand” and “already” by the word dictionary, and both of them are erroneous as Japanese unless the context is confirmed. It is recognized that it has a property that it cannot be determined whether or not it exists, and here, the hiragana is converted to “already” once.

  In addition, the word “Touri” is confirmed as a word that is not registered in the word dictionary as a general word, and is recognized as “not” in the word dictionary. Convert.

  The “lost wife” is converted to “disaster prevention” by recognizing that the next phrase is “training” and collating with the word dictionary. Eventually, the recognition result correction means 210 will correct the sentence to “Fiscal 2007 disaster prevention drill will be held from 10:05 am on Monday, June 4th, as we have already communicated with the notice.” Will be. At this stage, the point where “head office” is recognized as “your company” is not corrected.

  When the correction by the recognition result correction unit 210 is completed, the screen control unit 250 causes each character input / output device 300 to display the corrected sentence as a primary sentence. As shown in FIG. 9, when the speech recognition result by the speech recognition apparatus 100 is obtained, the speech recognition result (0th sentence) is displayed as it is, and the corrected sentence is immediately set as the detailed mode (default mode). It can also be displayed. In such a case, the processing for the detailed mode of the text processing means 230 is omitted.

  In the display screen, when the text is long and overflows from the display area at the top and bottom of the screen, the screen is scrolled upward so that the new text is displayed from the bottom to make it easier to see.

  Here, it is assumed that the reader selects “simple mode” or “caption mode” from the mode list provided in the center of the screen. The character input / output device 300 as a reader terminal used by the reader notifies the server (data processing device 200) of a mode change event in response to a screen operation from the reader. In the data processing device 200, the mode setting means 220 sets the mode in response to the mode change event notification, and the text corresponding to the changed mode is displayed on the character input / output device 300.

  The sentence processing unit 230 may receive the mode setting by the mode setting unit 220, recognize the state of the set mode, and output a sentence processed according to the mode.

  FIG. 10 is an explanatory diagram showing an example of screen display in the character input / output device 300 in a state where the subtitle mode is set. FIG. 11 is an explanatory diagram showing an example of a screen display in the character input / output device 300 in a state where the simple mode is set.

  The sentence editing unit 240 starts the operation with the output of the processed document by the sentence processing unit 230 as a trigger. In this example, the reader terminal, the operator terminal, and the server are ready to communicate with each other by a LAN connection or the like in advance, so that each time the processed text is output to the reader terminal screen, the operator terminal Also, it is assumed that the processed text with the same text is displayed. In addition, a sentence in which a series of processing is performed is transmitted in units, and a transmission counter is incremented by one each time transmission is performed.

  The text editing means 240 or the screen control means 250 causes the operator terminal to display the processed text portion displayed at the bottom of the screen in an editable manner. Thereafter, the text editing unit 240 performs editing processing according to the processing flow shown in FIG.

  FIG. 12 is a flowchart illustrating an example of editing processing by the text editing unit 240. Here, when the operator looks at the processed text displayed on the screen of the operator terminal and determines that there is a portion to be corrected, the range is set with the mouse or the like. In the operator terminal, a range setting notification event is notified to the server (data processing apparatus 200) in accordance with a screen operation from the operator. For example, a range setting notification event including the start position (character unit), the length, and the processed text transmission counter of the processed region is transmitted to the data processing device 200. FIG. 13 is an explanatory diagram illustrating an example of a screen image related to a range setting operation on the operator terminal.

  In the data processing device 200, the text editing unit 240 receives the range setting notification event (step S201). The text editing unit 240 returns a response that accepted the range setting to the operator terminal, and causes the operator terminal to display a screen for inputting a correct text within the range (step S202).

  Then, the operator terminal receives an input of a correct sentence for the set range from the operator, and transmits the received corrected sentence to the server. FIG. 14 is an explanatory diagram showing an example of a screen image related to a corrected text input operation on the operator terminal.

  In the example shown in FIG. 14, the operator confirms that there is no error in the input sentence, and presses the Enter key on the keyboard, so that the corrected sentence at the range designated portion is transmitted to the data processing device 200. In this case, an instruction is given to correct the location labeled “your company” to “head office”.

  When the text editing unit 240 of the data processing device 200 receives the corrected text of the range designation portion (step S203), the text editing means 240 edits the text by reflecting the corrected text on the processed text (step S204). Note that the text editing unit 240 also edits the corresponding part in the other mode when the correction part includes the character displayed in the other mode. Then, the screen control means 250 is caused to update the screen corresponding to the edited text (step S205).

  For example, the screen control means 250 tracks the processed text to be updated based on the content related to the correction part edited by the text editing means 240, the start position of the range, the length, and the text transmission counter. Above, the portion corresponding to the start position and length information of each character transmitted to each character input / output device 300 is sought. Here, the correction contents by the operator are overwritten with a different color (for example, red) from the sealed position.

  The editing operation by the operator is performed in a state where it cannot be seen at all on the screen of the reader terminal. By doing so, an environment is formed in which the reader can concentrate only on reading the currently displayed text. Then, when the editing operation by the operator is confirmed, the portion to be corrected is updated and displayed in a color-coded format. In the example shown in FIG. 14, an example of reverse display is shown, but the screen image after correction is also applied to the reader terminal by the screen update processing of the screen control means 250.

  In addition, on the screen of the reader's terminal, if the updated part is scrolled and is no longer visible, there will be an updated sentence, such as displaying a "*" mark on the right side of the corrected sentence display field. May be informed to the reader.

  In this example, the operator arranges a person who can listen to the original voice input to the voice recognition apparatus 100 and can understand the meaning, background, and situation with only the voice.

  Further, for example, an operator may be arranged according to each mode of simple / caption (summary) / detail, and editing work may be performed in each mode. For example, as shown in FIG. 15, an operator in charge of the subtitle mode may consider an implementation method of confirming the expression of text as subtitles. In the example shown in FIG. 15, an instruction to correct “I will do” at the end of “I will do” to “I will” is given.

  As described above, according to the present embodiment, when voice recognition is performed, even if it is easy to perform erroneous conversion due to an environment in which noise other than a sound source is likely to be input to the microphone, the “sway” can be performed. As long as the text editing support process is performed, the text can be output as a more accurate text. In the present invention, a minimum human interface through an operator for support is also proposed as one configuration. By including support by hand, it has a configuration that makes it easy to identify sentences that are clearly mistaken, and performs display control to present without impairing real-time performance. As a result, even in an environment where information can be obtained only by voice, it is easy to make a quick determination of the situation, and in particular, a soil is formed that makes it easier for a hearing impaired person to communicate independently.

  For example, in an environment where information can only be obtained by voice, for the hearing impaired, for example, what is happening now, who is laughing at what is wrong now, and who is now I don't even know what I'm saying about it, and I have to take action so that people around me can explain the situation in writing, etc. Information necessary for situation determination can be obtained without impairing real-time performance.

  It should be noted that if it is attempted to determine whether or not the error is accurate only by computer processing, there are problems that a sentence of a certain length is required and processing time is required, but in this embodiment, it is ambiguous. Is converted to hiragana and then receives correction instructions from the operator, so it is possible to output a sentence that eliminates the conversion result that may reduce recognition in a shorter span. .

  In addition, ambiguity, redundancy, etc. can be eliminated from character string information that is output in large quantities by voice recognition processing, and information can be provided in a more easily understandable form. The level at which the spoken text is displayed can be set at the discretion of the party, so that it can be reduced to the subtitle level of the movie or more easily according to the request, skill, etc. of the party. Since it is possible to determine whether to reduce the level, etc., it is possible to form an environment in which information is acquired in a very natural manner even in an environment where only the voice flows without stressing the parties.

  The present invention can be suitably applied to any system that is not limited to the text indicated by the character string information output as the speech recognition result but is a text that is updated in real time.

It is a block which shows the structural example of the text correction assistance system by this invention. It is a block which shows the other structural example of the text correction assistance system by this invention. It is a block diagram which shows the structural example of a more concrete text correction assistance system. 10 is a flowchart illustrating an operation example of the text correction support system 1000. It is explanatory drawing which shows an example of the screen displayed on the character input / output device. It is explanatory drawing which shows an example of the implementation image of the text correction assistance system. It is a flowchart which shows an example of the correction process of the recognition result correction | amendment means 210. FIG. It is explanatory drawing which shows an example of the format of a word dictionary. It is explanatory drawing which shows an example of the screen displayed on the character input / output device. It is explanatory drawing which shows the example of the screen display in the character input / output device 300 in the state by which the subtitle mode was set. It is explanatory drawing which shows an example of the screen display in the character input / output device 300 in the state in which the simple mode was set. It is a flowchart which shows an example of the edit process by the text edit means 240. It is explanatory drawing which shows the example of the screen image concerning range setting operation in an operator terminal. It is explanatory drawing which shows the example of the screen image concerning the correction text input operation in an operator terminal. It is explanatory drawing which shows the example of the screen image concerning the sentence correction in an operator terminal.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Data processing apparatus 11 Notation level correction means 12 Correction text display means 13 Correction instruction reception means 14 Screen update means 15 Text processing means 16 Processed text display means 17 Mode setting means 100 Speech recognition apparatus 200 Data processing apparatus 210 Recognition result correction means 220 Mode setting means 230 Text processing means 240 Text editing means 250 Screen control means 300 Character input / output device

Claims (10)

A sentence correction support system for presenting a sentence indicated by character string information, which is voiced expression data obtained by voice recognition processing, as a sentence that is easy for a reader to understand,
A data processing device for inputting character string information that is voiced expression data, one or more reader terminals used by a reader, and one or more operator terminals used by an operator,
The data processing device includes:
A notation level correcting means for generating a corrected sentence by correcting a word having a possibility of an error at the notation level with respect to a sentence indicated by input character string information;
Correction text display means for displaying the correction text generated by the notation level correction means on the screen of each reader terminal and each operator terminal,
A correction instruction receiving means for receiving a correction instruction for the text being displayed by the operator's operation from the operator terminal;
Sentence correction support, characterized in that the screen of each reader terminal and each operator terminal comprises screen updating means for reflecting the correction by the correction instruction received by the correction instruction receiving means so that the correction location can be specified. system. A text processing means for generating a processed text in which the corrected text is processed according to a predetermined display mode;
In correspondence with the correction text being displayed, the processed text corresponding to the display mode set for the reader is displayed on the screen of the reader terminal used by each reader, and the screen of the operator terminal is displayed. A processed text display means for displaying at least the processed text in accordance with the display mode displayed on the reader terminal;
The correction instruction accepting unit accepts a correction instruction for the processed text by an operator's operation from the operator terminal,
The screen update means reflects the correction by the correction instruction for the processed sentence received by the correction instruction receiving means on the screen of each reader terminal and each operator terminal so that the correction portion can be specified. Sentence correction support system. The notation level correction means collates each word included in the sentence indicated by the character string information with a word dictionary registered in advance, and if the word is not registered in the word dictionary, The sentence correction support system according to claim 1, wherein the corrected sentence is generated by converting the word into hiragana as a possibility of an error. The notation level correcting means collates each word included in the sentence indicated by the character string information with a word dictionary registered in advance, and when a word having a different notation having the same reading is registered in the word dictionary The sentence correction according to any one of claims 1 to 3, wherein a correction document is generated by converting the word into hiragana, assuming that the word may have an error at a notation level. Support system. A mode setting means for receiving a display mode switching instruction by a reader operation from a reader terminal,
The processed text display means, when the mode setting means accepts a display mode switching instruction, displays at least the processed text corresponding to the display mode set for the reader on the reader terminal of the request source. The sentence correction support system according to claim 1. The display mode includes a detailed mode for presenting almost the same content as the utterance content, a subtitle mode in which redundancy is eliminated from the utterance content, and a simple mode for presenting the main points of the utterance content in bulleted form. Item 6. The sentence correction support system according to any one of items 5 to 6. A sentence correction support method for presenting a sentence indicated by character string information, which is voiced expression data obtained by voice recognition processing, as a sentence that can be easily understood by a reader,
A data processing apparatus to which character string information that is voiced expression data is input,
For the sentence indicated by the input character string information, generate a corrected sentence that corrects a word that may be erroneous at the notation level,
Display the generated corrected text on the screen of each reader terminal and each operator terminal,
From the operator terminal, accept correction instructions for the displayed text by the operator operation,
A sentence correction support method characterized in that the correction by the correction instruction received from the operator terminal is reflected on the screen of each reader terminal and each operator terminal so that the correction part can be specified. Generate a modified sentence that has been processed according to a predetermined display mode,
In correspondence with the correction text being displayed, the processed text corresponding to the display mode set for the reader is displayed on the screen of the reader terminal used by each reader, and the screen of the operator terminal is displayed. , Display at least the processed text that matches the display mode displayed on the reader's terminal,
From the operator terminal, accept correction instructions for the processed text by the operator's operation,
The text correction support method according to claim 7, wherein the correction by the correction instruction for the processed text received by the correction instruction receiving means is reflected on the screen of each reader terminal and each operator terminal so that the correction portion can be specified. A sentence correction support program for presenting a sentence indicated by character string information, which is voiced expression data obtained by voice recognition processing, as a sentence that can be easily understood by a reader,
In a sentence correction support program applied to a data processing device to which character string information which is voiced expression data is input,
On the computer,
A process for generating a corrected sentence by correcting a word that may be erroneous at the notation level for the sentence indicated by the input character string information,
Processing to display the generated corrected text on the screen of each reader terminal and each operator terminal;
Processing to accept correction instructions for the displayed text by the operator's operation from the operator's terminal, and correction points received from the operator's terminal can be specified on the screen of each reader's terminal and each operator's terminal. Sentence correction support program to execute the process to be reflected in the document. On the computer,
A process of generating a secondary sentence obtained by processing the corrected sentence according to a predetermined display mode;
In correspondence with the correction text being displayed, the processed text corresponding to the display mode set for the reader is displayed on the screen of the reader terminal used by each reader, and the screen of the operator terminal is displayed. , Processing to display at least the processed text in accordance with the display mode displayed on the reader terminal,
Processing for receiving a correction instruction for the processed text by the operator's operation from the operator terminal, and correction by the correction instruction for the processed text received by the correction instruction receiving means on the screen of each reader terminal and each operator terminal The program for supporting sentence correction according to claim 9, wherein a process for reflecting the corrected part is specified.

Images