JP3682922B2 - Real-time character correction device and real-time character correction program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a lapis lazuli to correct the resulting position of the erroneous character caret in real time real time character correction device and real-time text modified program.
[0002]
[Prior art]
Conventionally, audio spoken by a performer in a program such as a television program is displayed as subtitles. When the audio information is converted into text information and displayed, and the television program is a recorded broadcast, the audio information is converted into text data using a conversion software, and the error of the converted text data is corrected. Therefore, application software such as a word processor and an editor is usually used. Then, as a procedure for correcting text errors, use a pointing device such as a mouse (in the case of a touch panel, pointing pen or finger) to specify the wrongly recognized part, enter the correct word, The deleted part is deleted.
[0003]
On the other hand, when subtitles are produced in real time for the contents of a TV program or a lecture, a method is known in which the computer converts the text of the content of the TV program or lecture into text via a voice recognition means.
[0004]
At this time, if the correction is made while checking for typographical errors one by one, it may be significantly delayed from the progress of the television program or lecture, and it may be difficult to grasp the contents of the program or lecture with a fast tempo. In addition, when respecting real-time characteristics, there are many cases in which the character is left as it is without being corrected even if a typographical error occurs. Therefore, conventionally, a character data correction device has been proposed.
[0005]
This character data correction device converts voice data inputted by voice into character data, points out the error part of the converted character data, and corrects the error part of the indicated character data. By providing a point terminal device that finds and selects an error from text data output from a text data output device such as a speech recognition device, and a correction terminal device that corrects an erroneous word, The correction time is shortened.
[0006]
[Problems to be solved by the invention]
However, in the conventional character data correction device, the point terminal device and the correction terminal device share a role to increase the correction speed, but there is room for further improvement.
In other words, the character data correction device relies on the keyboard to input the location to be corrected by the correction terminal device, so that it takes time for the correction, and the correction that can further reduce the time can be performed in real time. It was desired.
[0007]
The present invention has been made in view of the above circumstances, and is deleted in units of sentences sandwiched between punctuation marks including a designated place where a typographical error is allowed or a place where a typographical error is allowed. by reproducing again, modifiable without interrupting the real-time property, the real-time character correction device, and to provide a and real-time character modification program.
[0008]
[Means for Solving the Problems]
In order to achieve the above object , the present invention is configured as follows. That is, in order to convert the speech input by the speaker into text by the speech recognition means and display the caption, the text screen generation / display means for generating a screen on the monitor screen is used to identify the place where the typographical error occurs. , A real-time character correction device that corrects in real time during voice input ,
Based on the designation by the operator with respect to the place where there is a typographical error in the converted text, the typographical error position detecting means for detecting the designated location where the typographical error is generated, and the typographical error detected by the typographical error position detecting means based on the location of a text, the beginning of a sentence containing the sections that cause the spelling until a comma or punctuation, or endnotes from comma or punctuation, or punctuation between sentences extracting means for extracting a sentence which is sandwiched by the comma or period And an erroneous sentence deleting means for deleting the sentence extracted by the sentence punctuation sentence extracting means, and a voice input of the speaker for the sentence deleted by the erroneous sentence deleting means is prompted, and the voice input of the sentence is performed. The voice is recognized and the result of the speech recognition is deleted. A re-recognition result reflecting means to be reflected, a switching signal transmitted from the erroneous sentence deleting means and the re-recognition result reflecting means, and a route from the voice recognition means to the re-recognition result reflecting means or the text screen generation display means. And a switching unit for switching.
[0009]
According to the first aspect of the present invention, the real-time character correcting device is based on a reading mark or a punctuation point including a portion where the typographical error is generated based on designation of a location where the typographical error is detected in the speech-recognized text. The sentence is extracted and deleted, and the real-time correction is performed by reflecting the result of the voice recognition again by prompting the voice input of the deleted sentence. Note that the real-time character correction device is configured to switch the route from the voice recognition unit to the re-recognition result reflection unit or the text screen generation display unit by a switching signal from the erroneous sentence deletion unit and the re-recognition result reflection unit. It has been switched to. A sentence sandwiched between reading marks or punctuation marks is a sentence sandwiched between punctuation marks and punctuation marks, punctuation marks and punctuation marks, punctuation marks and punctuation marks, and punctuation marks and punctuation marks. Furthermore, when specifying the typographical error generated in the sentence of the converted text, a checker (operator) for checking the utterance or the typographical error is performed on the monitor screen.
[0010]
Further, in order to convert the voice input by the speaker into text by the voice recognition means and display the subtitles, the text screen generation / display means for generating the screen on the monitor screen is used, and the place where the typographical error has occurred is detected. In order to correct in real time during voice input , the computer is a real-time character correction program that functions as the following means. That is, the above means are a typographical error position detection means, a punctuation sentence extraction means , a false sentence deletion means , a re-recognition result reflection means , and a switching means .
[0011]
With this configuration, the real-time character correction program reads the punctuation marks from the beginning of the sentence including the location where the typographical error has occurred based on the designation of the location where the typographical error in the text has been converted by the inter-punctuation sentence extraction means. Alternatively, sentences up to a punctuation mark or a sentence sandwiched between punctuation marks or a punctuation mark are extracted. In addition, the sentence extracted by the sentence extraction means between the punctuation marks is deleted by the false sentence deletion means, and the speech input of the speaker for the sentence deleted by the mistake sentence deletion means is prompted by the re-recognition result reflection means. A voice input of the sentence is obtained and voice recognition is performed, and the result is reflected in the deleted sentence position. Further, the switching means switches the route from the voice recognition means to the re-recognition result reflecting means or the text screen generation / display means by a switching signal from the erroneous sentence deletion means and the re-recognition result reflecting means.
[0018]
DETAILED DESCRIPTION OF THE INVENTION
FIG. 1 is a block diagram cited for explaining the configuration and schematic operation of a system including a real-time character correcting device of the present invention.
The system of the present invention includes a speech recognition device 1 as a speech recognition means, a real-time character correction device 2, and a caption conversion device 3.
[0019]
The speech recognition device 1 receives speech from a speaker via a microphone or the like (1), converts it into text by a similarity comparison operation based on a built-in speech recognition dictionary, and the like to the real-time character correction device 2. Supply (2). At this time, the converted text may include a typographical error. A monitor screen 10 is installed at a position that can be recognized by the speaker so that the voice of the speaker can be converted into text and can be confirmed as a sentence.
[0020]
The real-time character correction device 2 specifies the location where the typographical error occurred (indicated by the hatched line in FIG. 4) based on the designation of the location where the typographical error in the speech-recognized text is present (designated by checking the touch panel screen by the operator). ), Or a sentence from the beginning of a sentence containing a typographical error to a punctuation mark or a punctuation mark, or a sentence from a punctuation mark or a punctuation mark to the end of a sentence, or a sentence between punctuation marks or a punctuation mark. Extracting and deleting ((3)), prompting voice input of the deleted sentence ((4)), reflecting the result of voice recognition again, correcting in real time, and delivering to subtitle conversion device 3 .
[0021]
The real-time character correction device 2 includes a touch panel 20, a confirmation screen is displayed via the touch panel 20, and a typographical error part is designated. The caption conversion device 3 connected to the real-time character correction device 2 synthesizes the text output from the real-time character correction device 3 and the broadcast program to produce and output the caption broadcast program.
[0022]
FIG. 2 is a block diagram showing an expanded function of the internal configuration of the real-time character correction apparatus 2 shown in FIG.
Each block shown below is specifically composed of a peripheral LSI including a CPU and a memory, and the CPU has a function stored in the block by reading and executing a program recorded in the memory. Here, the real-time character correction device configured to detect, extract, and delete text sentences will be described. However, even a real-time character correction device configured to detect, extract, and delete text characters may be used. I do not care.
[0023]
Real-time character correction device 2, the text screen generating display unit 21 (the text screen generating display means), spelling position detecting section 22 and (spelling position detecting means), punctuation between sentences extracting section 23 (punctuation between sentences extracted hand stage ) and, erroneous sentence deletion unit 24 (erroneous sentence delete hand stage), re-recognition result reflecting section (re-recognition result reflecting means) 25, and a switching unit 26.
[0024]
The text screen generation / display unit 21 has a function of generating a screen based on the text output by the voice recognition device 1 and displaying the screen on the touch panel 20. The content displayed here is a subtitle conversion device after the operator's check. 3 is supplied.
[0025]
The typographical error position detection unit 22 has a function of detecting and extracting a word at a designated location based on designation of a location where there is a typographical error in the speech-recognized text. Detected and extracted by a tablet installed in In addition, about the position location in the sentence designated by this typographical error position detection unit 22, the correct character conversion operation on the voice recognition device 1 side is performed by feeding back to the voice recognition device 1 that it is the position of the typographical error. It is possible to increase the reproduction probability for doing so.
[0026]
The coordinate data detected by the typographical error position detection unit 22 is supplied to the sentence extraction unit 23 between punctuation marks. This punctuation sentence sentence extracting unit 23 has a function of extracting a punctuation mark including a part where a typographical error has occurred, or a subsequent sentence from a punctuation mark, or a sentence sandwiched between punctuation marks or punctuation marks. The sentence is supplied to the erroneous sentence deletion unit 24 so as to show the area in a state where the area is inverted or blinked, or a range is specified by parentheses. In addition, the sentence sandwiched between reading marks or phrases refers to a character group at a position sandwiched between reading marks and reading marks, reading and phrases, phrases and readings, and phrases and phrases.
[0027]
The erroneous sentence deletion unit 24 has a function of deleting an erroneous sentence based on the operation of the operator. The content deleted here is reflected in the text screen generation display unit 21, and the display of the sentence whose range is specified is displayed from the touch panel 20. Disappear. Incidentally, the erroneous sentence deletion section 24, the ruse Ntensu are displayed on the touch panel 20 is removed, and is configured to transmit a switching signal for switching the route switching unit 26.
[0028]
The re-recognition result reflection unit 25 obtains a voice input by prompting the voice input of the deleted sentence, performs re-speech recognition by the voice recognition device 1, and the result via the text screen generation display unit 21. Has a function to reflect. The re-recognition result reflecting unit 25 can determine the designated place by reversing the sentence so that the speaker can be recognized on the monitor screen 10 that can be recognized by the speaker as means for prompting voice input. It is good, and it may be performed by voice from the checker to the speaker via the earphone. The re-recognition result reflecting unit 25 is configured to transmit a switching signal for switching the route to the switching unit 26 when the result of re-recognition by the voice recognition device 1 is reflected on the text screen generation / display unit 21. Has been.
[0029]
The switching unit 26 is configured to operate according to a switching signal from the erroneous sentence deletion unit 24 and the re-recognition result reflection unit 25. The switching unit 26 normally controls connection so that the output from the speech recognition device 1 is supplied to the text screen generation / display unit 21, but the route is switched by the switching signal from the erroneous sentence deletion unit 24, and the speech recognition is performed. The connection control is performed so that the output from the device 1 is supplied to the re-recognition result reflecting unit 25, and the result re-recognized by the voice recognition device 1 is reflected in the text screen generation / display unit 21.
[0030]
Further, the switching means 26 uses a switching signal generated from the re-recognition result reflecting unit 25 when the sentence (word) is corrected by the re-recognition result reflecting unit 25 and the sentence is displayed on the text screen generation / display unit 21. It is configured to perform an operation of switching to the original route. Also requires modification congenital scan by re-recognition result reflection unit 25 again, sentence again if specified in the same position, the switching unit 26, re-recognition result reflects transmitted switching signal from the erroneous sentence deletion section 24 It will be switched again to the route on the part 25 side.
[0031]
Next, the operation of the system including the real-time character correction device will be described. FIGS. 3 and 4 are diagrams for explaining the operation of the embodiment of the present invention shown in FIGS. 1 and 2, and show an operation flowchart and an example of a screen configuration displayed on the touch panel, respectively. In addition, the operation | movement flowchart shown in FIG. 3 specifically shows the process sequence of the real-time character correction program of this invention.
The operation of the embodiment of the present invention shown in FIGS. 1 and 2 will be described in detail below with reference to FIGS.
[0032]
First, voice input is performed by a speaker such as an announcer or a caster (S31). As a result, the speech recognition apparatus 1 performs speech recognition based on the similarity calculation by the speech dictionary based on the input speech and converts it into text (S32). The recognized text is immediately displayed on the touch panel 20 via the text screen generation display unit 21 (S33).
[0033]
Here, when there is a typographical error in the text, the operator touches the display position (specified location) on the touch panel 20 corresponding to the typographical error. This is detected via the typographical error locating position detection unit 22 (S34), and the utterance of the sentence range based on the punctuation mark or punctuation point including the location where the typographical error punctuation extracted by the sentence punctuation sentence extraction unit 23 occurs. After the range is specified so that can be recognized, the erroneous sentence deletion unit 24 deletes all at once (S35, S36) . The innate scan including erroneous Ri point is deleted is reflected on the touch panel 20 via the text screen generating display unit 21. At the same time, a switching signal is sent from the erroneous sentence deletion unit 24 to the switching unit 26 to switch the route.
[0034]
Then, uttered is again prompting for its innate scan unit content in speaker, the utterance is re-input to the speech recognition apparatus 1 via the switching unit 26 of FIG. 2 (S37), the re-recognition result reflecting section 25 The voice recognition result is displayed on the touch panel 20 (monitor screen 10) via the text screen generation display unit 21. When the text screen generation / display unit 21 displays a sentence on the touch panel 20 (monitor screen 10), a switching signal is sent from the re-recognition result reflecting unit 25 to the switching unit 26. To the root of At this time, by correcting the erroneous sentence as described above, it is possible to correct the typographical error in real time without requiring much time for the correction. Here, when the typographical error is not corrected, the utterance is prompted again (S37: yes), and the processes after S32 are repeated.
[0035]
If the voice recognition is not properly performed, the real-time property cannot be secured, but if the correction by the manual input or the real-time property is important, it is left unattended (S39).
[0036]
Also, the speech recognition by the speech recognition device 1 is likely to cause misrecognition due to utterance problems such as "speaking" and "ambiguous pronunciation". The accuracy rate will increase. Also, because it is difficult to utter a difficult word and the word does not exist in the speech recognition dictionary, if a misrecognition occurs, it can be recognized by replacing the difficult word with a simple word when uttering again. The rate can be expected to improve. When generating by replacing a word, the typographical error position detector 22 designates the position of the typographical error so that the speaker can determine which character is difficult for the speech recognition apparatus 1 to recognize. Can do.
[0037]
When it is finally confirmed that there is no typographical error (S34: no), the caption conversion processing by the caption conversion device 3 is started (S38). It is converted into a subtitle broadcast format (NAB format: Japanese commercial broadcasting joint name “T027-1996”) by the subtitle conversion device 3, and the video is synthesized and broadcast as a subtitle broadcast.
[0038]
FIG. 4 shows a screen image displayed on the touch panel 20. Here, when the operator touches a place where a typographical error has occurred (“Herpus”), the range between punctuation marks including this word (for the first time, subtitle broadcasting is assumed to be Hyu) is deleted after being designated as a range. Then, the words corresponding to the deleted sentence are uttered again to realize real-time character correction. In addition, by designating which word a typographical error has occurred by the typographical error position detection unit 22, the information of the typographical character is fed back to the speech recognition apparatus 1, and the same conversion is not caused when the same sentence is read. It is convenient to do so.
[0039]
As described above, according to the present invention, by extracting and deleting a sentence sandwiched between punctuation marks including a portion where a typographical error is recognized, the correction can be made without impairing real-time performance.
In the above-described embodiment of the present invention, only correction in the case where the content of a TV program or a lecture is converted into text using speech recognition has been described. However, in addition, a television program or a lecture using electronic shorthand etc. It can be applied in the same way to correction when converting the contents of the text to electronic text. Furthermore, the present invention is effective not only in the field of producing captions for broadcasting in real time, but also in the field of providing captions for lectures for the hearing impaired or the field of producing captions for pre-recorded programs using voice recognition. is there.
[0040]
2 are executed by each of the text screen generation display unit 21, the typographical error position detection unit 22, the punctuation sentence extraction unit 23, the erroneous sentence deletion unit 24, the re-recognition result reflection unit 25, and the switching unit 26 shown in FIG. Is recorded on a computer-readable recording medium, and a program recorded on the recording medium is read into a computer system and executed, whereby the real-time character correction system of the present invention is realized. The computer system here includes an OS and hardware such as peripheral devices.
[0041]
Further, the “computer-readable recording medium” refers to a portable medium such as a flexible disk, a magneto-optical disk, and a CD-ROM, or a storage device such as a ROM and a hard disk built in the computer system. The “computer-readable recording medium” refers to a volatile memory (RAM) in a computer system serving as a system or client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. As described above, it is assumed that the program is held for a certain period of time.
[0042]
If the speaker can perform a voice input operation while confirming the sentence of the text on the monitor screen 10, it can be adjusted according to the timing of the speaker, and the typographical omission in the sentence of the text is checked. After making quick corrections together with the checker, you can provide sentences without mistakes. In addition, even if there is no checker, it is possible to provide sentences with few mistakes after making quick corrections.
In addition, the switching signal of the switching unit 26 has been described here as a configuration sent from the erroneous sentence deletion unit 24 and the re-recognition result reflection unit 25, but is not particularly limited as long as the switching operation of the switching unit 26 can be appropriately performed. Is not to be done. Furthermore, the real-time character correction device may be configured to include a storage unit such as a hard disk for storing data converted into text.
[0043]
【The invention's effect】
As described above, the real-time character correction device and the real-time character correction program according to the present invention have the following excellent effects.
According to Claims 1 and 2 of the present invention, in the case of converting from text recorded in a recording medium such as a TV program, a lecture, or a VTR to text, it is possible to correct a portion where a typographical error occurred in real time, Since almost no time is required for the correction, it can be applied to an application in which real-time property is important for the correction.
[0044]
In particular, when subtitles are produced in real time for the contents of a TV program or lecture, typographical typographical errors may occur due to a voice recognition error, compared to the method of speech recognition by a computer based on the sound of the contents of the TV program or lecture. Even if it occurs, by recognizing or indicating the location where this typographical error is recognized, the sentence sandwiched between punctuation marks including the word with the typographical error is extracted and deleted, and the deleted part is spoken again. By repeatedly uttering and converting it into text, it is possible to reflect it in subtitles in real time by making quick corrections.
[0045]
Further, the present invention enables real-time correction in the same way as in the case of subtitle production by electronic shorthand in addition to the above-described subtitle production by voice recognition.
[0046]
Further, according to the first and second aspects of the present invention, since it is possible to indicate which character conversion is wrong in sentence units within the sentence of the text, the voice recognition means. Even if the speaker inputs a voice for the same designated portion, it is possible to make a quick correction by voice without repeating wrong conversion characters.
[Brief description of the drawings]
FIG. 1 is a block diagram cited for explaining the configuration and schematic operation of a real-time character correction system of the present invention.
FIG. 2 is a block diagram showing an expanded function of the internal configuration of the real-time character correction apparatus shown in FIG.
FIG. 3 is a diagram cited for explaining the operation of the embodiment of the present invention, and is a diagram showing an operation flowchart;
FIG. 4 is a diagram cited for explaining the operation of the embodiment of the present invention, and is a diagram showing an example of a screen configuration displayed on the touch panel.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Speech recognition apparatus 2 Real-time character correction apparatus 3 Subtitle conversion apparatus 21 Text screen generation | occurrence | production display part 22 Typing character position detection part 23 Punctuation sentence sentence extraction part 24 Error sentence deletion part 25 Re-recognition result reflection part 26 Switching part

Claims (2)

Voice input by speaker, for caption display is converted into text by a speech recognition unit, using a text screen generating display means for screen generation on the monitor screen, the spelling The resulting location, sound A real-time character correction device that corrects in real time during input ,
Based on the designation by the operator for the location where there is a typographical error in the converted text, a typographical error locating means for detecting the designated location where the typographical lexation has occurred,
Based on the text location with the typographical error detected by the typographical error location detection means, from the beginning of the sentence including the location where the typographical error occurs, to the reading or punctuation, or from the punctuation or punctuation to the end of the sentence, or the punctuation or punctuation Punctuation sentence extraction means for extracting sentences sandwiched between,
Erroneous sentence deletion means for deleting the sentence extracted by the sentence extraction means between punctuation marks;
Re-recognition result that prompts the voice input of the speaker to the sentence deleted by the erroneous sentence deletion means, obtains the voice input of the sentence, performs voice recognition, and reflects the result in the position of the deleted sentence Reflection means,
A switching unit that switches the route from the voice recognition unit to the re-recognition result reflection unit or the text screen generation display unit by a switching signal transmitted from the erroneous sentence deletion unit and the re-recognition result reflection unit;
A real-time character correction device comprising: Voice input by speaker, for caption display is converted into text by a speech recognition unit, using a text screen generating display means for screen generation on the monitor screen, the spelling The resulting location, sound To correct in real time while typing ,
A typographical error position detecting means for detecting a designated location where the typographical error has occurred based on designation by an operator for a location where there is a typographical error in the converted text;
Based on the text location with the typographical error detected by the typographical error location detection means, from the beginning of the sentence including the location where the typographical error occurs, to the reading or punctuation, or from the punctuation or punctuation to the end of the sentence, or the punctuation or punctuation Punctuation sentence extraction means for extracting sentences sandwiched between
Erroneous sentence deletion means for deleting the sentence extracted by the sentence extraction means between punctuation marks,
Re-recognition result that prompts the voice input of the speaker to the sentence deleted by the erroneous sentence deletion means, obtains the voice input of the sentence, performs voice recognition, and reflects the result in the position of the deleted sentence Reflection means,
Switching means for switching the route from the voice recognition means to the re-recognition result reflecting means or the text screen generation / display means by a switching signal transmitted from the erroneous sentence deletion means and the re-recognition result reflecting means;
Real-time character correction program characterized by functioning as

Images