WO2009122779A1 - テキストデータ処理装置、方法、プログラムが格納された記録媒体 - Google Patents
テキストデータ処理装置、方法、プログラムが格納された記録媒体 Download PDFInfo
- Publication number
- WO2009122779A1 WO2009122779A1 PCT/JP2009/052378 JP2009052378W WO2009122779A1 WO 2009122779 A1 WO2009122779 A1 WO 2009122779A1 JP 2009052378 W JP2009052378 W JP 2009052378W WO 2009122779 A1 WO2009122779 A1 WO 2009122779A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- symbol
- editing
- likelihood
- data processing
- text data
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/166—Editing, e.g. inserting or deleting
Definitions
- the present invention relates to a speech recognition device, a text data processing device, a text data processing method, and a text data processing program, and more particularly, a speech recognition device, a text data processing device, and a text data processing that edit symbols for text data.
- the present invention relates to a method and a text data processing program.
- the speech language processing unit converter includes a speech recognition device 450, a buffer memory 406, a processing unit converter 407, a statistical model memory 421, an empirical rule memory 422, and a Japanese-English translation. Part 408.
- the conventional speech language processing unit converter having such a configuration operates as follows.
- the voice recognition device 450 performs voice recognition. Then, the voice recognition device 450 writes the voice recognition result in the buffer memory 406. In the statistical model memory 421, a score representing the likelihood of a boundary corresponding to a phrase is learned.
- the processing unit conversion unit 407 uses the statistical model memory 421 to calculate a score representing the likelihood of a node boundary. Then, when the score exceeds the threshold value, the processing unit conversion unit 407 sets the position as a punctuation mark insertion candidate. Further, the processing unit conversion unit 407 finally determines the insertion of the punctuation according to the empirical rule extracted from the prosodic information of the punctuation insertion location in the empirical rule memory 422.
- the caption program production system shown in FIG. 8 includes a synchronization detection device 515, an integration device 517 including unit caption extraction 533, automatic caption generation 535, and timing information addition 537, a morpheme analysis 519, and a division rule 521. Has been.
- the conventional symbol insertion apparatus having such a configuration operates as follows.
- the integration device 517 obtains candidates for line breaks and page breaks to fit on one screen or one line.
- the synchronization detection device 515 evaluates the validity of the result, returns the result to the integration device 517 again, and finally performs line feed / page break and automatically divides.
- the integration device 517 passes the input sentence to the morpheme analysis 519 and analyzes the morpheme when there are more input sentences than the specified number of characters. Further, the integration device 517 presents a delimitable candidate according to the division rule 521 in consideration of the number of characters.
- the present invention has been made in view of such problems, and is a text data processing device, a text data processing method, a text data processing program, and a text data processing device capable of creating text with symbols attached at appropriate positions, It is another object of the present invention to provide a voice recognition device.
- a text data processing apparatus is a text data processing apparatus that edits a symbol with respect to input text, and is based on the symbol insertion frequency in a block composed of a plurality of divided texts.
- Symbol editing determination means for determining whether editing is necessary, and when the symbol editing determination means determines that symbol editing is necessary, the symbol editing likelihood is calculated based on the symbol insertion likelihood of the word and the inter-symbol distance.
- Symbol editing position calculating means for calculating and calculating a symbol editing position in the block from the symbol editing likelihood.
- a text data processing apparatus is a text data processing apparatus for editing a symbol with respect to input text, wherein a symbol is based on a symbol insertion frequency in a block composed of a plurality of divided texts. Based on the symbol insertion likelihood of the word and the symbol insertion history of the text that has already been inserted when the symbol editing determination unit determines that the symbol editing is necessary.
- Symbol editing position calculating means for calculating a symbol editing likelihood and calculating a symbol editing position in the block from the symbol editing likelihood.
- a text data processing method is a text data processing method for editing a symbol with respect to input text, wherein the symbol is based on a symbol insertion frequency in a block composed of a plurality of divided texts.
- a text data processing method is a text data processing method for editing a symbol with respect to input text, wherein the symbol is based on the frequency of symbol insertion in a block composed of a plurality of divided texts.
- a determination step for determining whether editing is necessary, and a symbol editing likelihood based on the symbol insertion likelihood of the word and the symbol insertion history of the text with the symbol inserted when it is determined that the symbol editing is necessary in the determination step And calculating a symbol editing position in the block from the symbol editing likelihood.
- a recording medium storing a text data processing program is a text data processing apparatus that edits a symbol for input text, from a plurality of divided texts to a computer. Determining whether or not symbol editing is necessary based on the frequency of symbol insertion in the block, and if the symbol editing determination means determines that symbol editing is necessary, the symbol insertion likelihood of the word and the distance between symbols Calculating a symbol edit likelihood based on the symbol edit likelihood and calculating a symbol edit position in the block from the symbol edit likelihood.
- a text data processing program is a text data processing program for editing a symbol with respect to input text, and inserting a symbol into a block composed of a plurality of divided texts to a computer.
- a determination step for determining whether or not symbol editing is necessary based on the frequency, and when it is determined that symbol editing is necessary in the determination step, the symbol insertion likelihood of the word and the symbol insertion history of the text with the symbol inserted Calculating a symbol editing likelihood based on the symbol editing likelihood, and calculating a symbol editing position in the block from the symbol editing likelihood.
- An object of the present invention is to provide a text data processing device, a text data processing method, a text data processing program, and a speech recognition device that can create text with symbols attached at appropriate positions.
- FIG. 1 is a block diagram showing a configuration of a text data processing apparatus according to a first embodiment
- 3 is a flowchart showing a text data processing method according to the first exemplary embodiment
- It is a block diagram which shows the structure of the text data processing apparatus concerning Embodiment 2.
- FIG. 10 is a flowchart showing a text data processing method according to the second embodiment;
- It is a block diagram which shows the structure of the text data processing apparatus concerning Embodiment 3.
- 1 is a diagram illustrating a configuration of an apparatus described in Patent Document 1.
- FIG. It is a figure which shows the structure of the apparatus of patent document 2.
- Block division means 52 Symbol edit determination means 53 Symbol edit position calculation means 54 Symbol insertion model storage means 55 Symbol position determination means 56 Speech recognition means 60 Text data processing device 61 Symbol edit determination means 62 Symbol edit position calculation means 101 Temporary symbol insertion Position calculation means 102 Symbol insertion model storage means 103 Provisional symbol insertion result storage means 104 Block division means 105 Symbol edit determination means 106 Symbol edit position calculation means 107 Symbol position determination means 300 Input device 310 Data processing device 311 Provisional symbol insertion position calculation means 312 Block division means 313 Symbol edit determination means 314 Symbol edit position calculation means 315 Symbol position determination means 320 Data storage device 321 Symbol insertion model storage section 322 Provisional symbol insertion result storage section 323 Symbol insertion result storage section 406 Buffer memory 407 Processing unit conversion unit 408 Japanese-English translation unit 421 Statistical model memory 422 Empirical rule memory 450 Speech recognition device 515 Synchronization detection device 517 Integration device 519 Morphological analysis 521 Division rule 533 Unit
- the text data processing apparatus edits symbols for the input text.
- the text data processing apparatus may use symbols such as a punctuation mark “.”, A punctuation mark “,”, a question mark “?”, An exclamation mark “!”, A period “.”, A comma “,”, a line feed code, etc. Insert in position. Alternatively, remove the improperly positioned symbol from the input text.
- FIG. 1 is a block diagram showing the configuration of the text data processing apparatus.
- the text data processing device 60 determines whether or not symbol editing is necessary based on the frequency of symbol insertion in a block composed of a plurality of divided texts, and the symbol editing determination unit 61 determines that symbol editing is necessary.
- the symbol editing likelihood for the word included in the block is calculated based on the symbol insertion likelihood and the inter-symbol distance, and the symbol editing position in the block is calculated from the symbol editing likelihood for the word included in the block.
- Symbol editing position calculation means 62 is
- the symbol edit determining means 61 calculates the number of symbol insertions to be inserted into one block.
- the symbol editing position calculation means 62 calculates the symbol insertion likelihood for each word based on the symbol insertion model. Then, the symbol edit likelihood is calculated based on the symbol insertion likelihood and the inter-symbol distance. In the block, a position where the symbol editing likelihood is high is set as a symbol editing position, and a symbol is inserted at that position. The symbol editing likelihood calculated for each word in the block is compared, and the symbol editing position is determined based on the comparison result.
- symbol editing is necessary until the number of symbols in the block reaches the number of symbol insertions calculated by the symbol editing determination means 61. Therefore, symbols are inserted until the number of symbols in the block reaches the number of inserted symbols. Thereby, a symbol can be attached to an appropriate position.
- the symbol edit likelihood may be calculated using a symbol insertion history instead of the symbol distance. That is, it is also possible to calculate the symbol edit likelihood for the word included in the block based on the symbol insertion likelihood and the symbol insertion history corresponding to the insertion frequency of the text with the symbol inserted.
- FIG. 2 is a block diagram showing the configuration of the text data processing apparatus.
- the text data processing apparatus has block dividing means 51, symbol edit determining means 52, symbol edit position calculating means 53, symbol insertion model storage means 54, and symbol position determining means 55. Further, the text data processing apparatus includes voice recognition means 56.
- the text data processing device is an arithmetic processing device such as a personal computer capable of inputting and outputting data.
- a CPU Central Processing Unit
- ROM Read Only Memory
- RAM Random Access Memory
- a storage disk a storage disk, or the like serves as the symbol insertion model storage unit 54.
- a processor such as a CPU is a block dividing unit 51, a symbol editing determination unit 52, a symbol editing position calculation unit 53, a symbol insertion model storage unit 54, and a symbol position determination unit 55.
- these means perform each process by executing a computer program stored in advance. Therefore, these means may be composed of physically the same processor or the like. For example, when a calculation program stored in a storage disk or the like is executed by the CPU, various settings are performed by reading input settings and the like into a RAM or the like. Then, the data of the calculation process and the data of the calculation result are written into a RAM, a storage disk or the like. In this way, the following arithmetic processing is executed in accordance with the text data processing program recorded in advance in the arithmetic processing device.
- the voice recognition means 56 has a microphone or the like for receiving a human speech.
- the voice recognition unit 56 performs voice recognition processing and creates a text sentence based on the voice data acquired by the microphone.
- the text created by the speech recognition means 56 becomes the input text.
- the speech recognition means 56 may perform morphological analysis and divide the input sentence into words.
- the speaker may be identified based on the feature amount of the voice detected by the voice recognition unit 56.
- text may be input regardless of voice recognition.
- the input text may be created by voice transcription.
- the voice recognition means 56 can be divided and output in units of words.
- the morphological analysis may be performed by other than the voice recognition unit 56.
- the symbol editing determination means 52 described later performs morphological analysis on the morphological analysis input text. May be.
- the text recognized by the voice recognition means 56 is input to the block dividing means 51.
- the block dividing means 51 divides the input text into a plurality of blocks according to a predetermined standard. That is, the input text is divided into two or more blocks.
- a method of dividing the input text a topic, a speaker unit, x seconds (x> 0), n utterances (n> 0), or the like can be considered.
- pseudo symbols are inserted and divided into utterance units.
- a certain threshold that is, when a pose greater than the threshold is inserted, a symbol “pose” is inserted at that position.
- a pause symbol which is a pseudo symbol, is inserted into the block.
- the block dividing means 51 divides the input text into a plurality of blocks.
- One block is usually composed of a plurality of divided texts.
- the divided text refers to one lump when a character string in one block is divided by a certain symbol.
- the symbol is a punctuation mark
- the divided text corresponds to one sentence.
- one block is composed of a plurality of sentences.
- One block may be composed of one sentence.
- the symbol edit determination means 52 calculates the appropriate number of symbols to be inserted into each block. As a result, the number of inserted symbols is calculated for each of the I blocks. In calculating the number of inserted symbols, the following formula 1 is used as a conditional expression.
- ⁇ (C k ) min and ⁇ (C k ) max are threshold values for performing edit determination of the symbol C k .
- the threshold value may be determined empirically or may be determined experimentally.
- threshold values ⁇ (C k ) min and ⁇ (C k ) max are stored in advance.
- S i and Ck represent the symbol insertion frequency of the symbol C k of the i-th block. That is, the number of symbols to be inserted into each block is determined so that the symbol insertion frequency S i , Ck is between a preset upper limit value and lower limit value.
- the symbol insertion frequencies S i and Ck are calculated using the following Equation 2 or Equation 3.
- the symbol insertion frequencies S i and Ck are determined according to the appearance frequency of the symbol C k included in the i-th block.
- the value of the Si, Ck numerator satisfying Equation 1 is the number of symbols to be automatically inserted.
- S i, Ck is a symbol insertion frequency considering the number of words or characters for a block composed of one or more divided texts. Therefore, if the number of characters or the number of words is different, the number of inserted symbols is also different.
- Symbol insertion frequencies S i and Ck are calculated according to the number of characters or the number of words included in one block. By doing in this way, the number of symbol insertions inserted in a block can be calculated appropriately. That is, even when the length of text included in one block changes, the number of symbol insertions can be determined appropriately.
- the symbol editing determination means 52 may perform morphological analysis. However, processing is not performed on text that is already divided into word units, such as speech recognition results, and the information is used. That is, when the text is input together with information obtained by dividing the text into words, the information is used.
- the symbol editing position calculation means 53 calculates the symbol editing likelihood for each word and inserts the symbol at the optimal position within one block.
- the symbol editing position calculation unit 53 inserts symbols into the block by the number of symbol insertions obtained by the symbol editing determination unit 52.
- the symbol editing position calculation means 53 performs the following three processes.
- the first process is a process for obtaining the symbol insertion likelihood in units of words.
- the second process is a process for obtaining the symbol edit likelihood based on the symbol insertion likelihood.
- the third process is a process for obtaining the symbol insertion position in accordance with the symbol edit likelihood.
- the symbol insertion likelihood is calculated for each word.
- the symbol insertion model storage means 54 stores a symbol insertion model that models a symbol insertion likelihood that a symbol is inserted into a character string.
- a symbol insertion model that models a symbol insertion likelihood that a symbol is inserted into a character string.
- the symbol insertion model corresponding to these methods is stored in the symbol insertion model storage means 54.
- the symbol editing position calculation unit 53 uses the symbol insertion model stored in the symbol insertion model storage unit 54 to calculate a symbol insertion likelihood that is a likelihood that a symbol immediately after the focused word is inserted. In this way, the symbol insertion likelihood for all the words included in the block is calculated.
- the symbol insertion likelihood of a word may be calculated by a method other than Document 1 and Document 2.
- Such symbol insertion likelihood differs depending on the word. That is, the symbol insertion likelihood, which is the likelihood of being inserted immediately after the word, is determined by the word.
- the symbol edit likelihood is calculated based on the symbol insertion likelihood.
- the j-th word included in the block i is considered (i and j are natural numbers). If the symbol insertion likelihood of the symbol C k is p (i, j, C k ), the symbol edit likelihood Pr (i, j, C k ) of the symbol C k can be calculated by the following equation 4. .
- Equation 4 ⁇ , ⁇ , and ⁇ are predetermined values, and function as correction coefficients for symbol edit likelihood including L (i, j, C k ). Therefore, the values of the constants ⁇ , ⁇ , ⁇ are stored in advance in a memory or the like.
- L (i, j, C k ) can be expressed by the following equation 5 using a function f k (y) based on the intersymbol distance.
- the variable y is a distance from the j-th word included in the block i to any of the nearest symbols. That is, the number of words or characters up to the nearest symbol can be the distance y. Specifically, it is the distance to any symbol that has already been inserted, or a pseudo symbol, such as a “pause”. Thus, the distance from the word position to the previous or subsequent symbol is the inter-symbol distance. Therefore, the variable y can be expressed by the following formula 6 or 7.
- length (j, j ′) represents the number of characters between the j-th word and the j′-th word.
- L (i, j, C k ) since the term L (i, j, C k ) is included, the value of the symbol edit likelihood Pr (i, j, C k ) changes according to the inter-symbol distance. That is, as the distance between adjacent symbols increases, the value of symbol edit likelihood Pr (i, j, C k ) increases. Therefore, it becomes easier to insert a symbol in a section where the distance between symbols is long.
- the distance between symbols is the number of words or characters from the symbol editing position to the nearest symbol already inserted.
- equation 8 As another symbol edit likelihood calculation method, the following equation 8 can be used. Similarly, constants ⁇ , ⁇ , and ⁇ are used.
- Equation 8 is preset constants as in Equation 4.
- w j + n ′ j ⁇ n + 1 represents a word string from the immediately preceding n word to the immediately following n ′ word.
- N (w j + n ′ j ⁇ n + 1 , C k ) represents the frequency at which the symbol C k is inserted immediately after the i-th word w j in the word string w j + n ′ j ⁇ n + 1.
- N (w j + n ′ j ⁇ n + 1 , C k ) Calculate from insertion history.
- the symbol insertion history a symbol insertion result in a block whose symbol insertion has already been confirmed can be used. That is, the insertion result of the symbol inserted in the past can be used as the insertion history. Or you may use the symbol insertion determination result in the temporary symbol insertion position calculation means of Embodiment 2 mentioned later.
- the symbol edit likelihood reflecting the symbol insertion tendency of the evaluation target speaker is obtained.
- the symbol edit likelihood is calculated based on the symbol insertion history of text that has already been inserted. For example, it is calculated based on a symbol insertion history corresponding to the frequency of insertion of the symbol in other text already having a symbol.
- the symbol insertion history may be set according to the insertion frequency of the symbol in another text by the same speaker.
- the symbol insertion position is determined using the symbol editing likelihood. That is, for all words in one block, a word having the maximum symbol editing likelihood is obtained. That is, a word having the maximum symbol editing likelihood across one or more divided texts is obtained. Then, a symbol is inserted immediately after the word. This calculation formula is shown in Equation 9 below.
- Equation 9 means that the position with the highest symbol editing likelihood for the symbol C k is output. Repeat until the condition of Equation 1 is satisfied, and calculate the position where C k is inserted.
- the likelihood of symbol editing for all words in one block is compared.
- symbols are inserted in order from the word having the highest symbol editing likelihood.
- the number of symbols inserted according to Equation 1 is inserted into one block. That is, the symbol editing determination means 52 determines that symbol editing is necessary until the number of symbols inserted in the block reaches the number of symbol insertions.
- the symbol edit likelihood for each symbol is obtained as in the following Equation 10, and the symbol C ⁇ having the highest likelihood excluding NULL is inserted.
- a restriction such that a symbol with a lower priority is not inserted at a place where a symbol with a higher priority is inserted.
- the insertion target symbol is a punctuation mark and a punctuation mark
- the punctuation mark has a higher priority than the punctuation mark
- no punctuation mark is selected at the place where the punctuation mark is inserted.
- the symbol position determination means 55 inserts a symbol into the input text based on the symbol insertion position information of the symbol editing position calculation means 53, and outputs the result. As a result, text data in which symbols are inserted at appropriate positions is output.
- symbols may be inserted by the above processing. For example, it is possible to calculate a symbol edit likelihood for a punctuation mark and insert a punctuation mark, and then calculate a symbol edit likelihood for a punctuation mark and insert a punctuation mark.
- FIG. 3 shows an example in which text is input without performing voice recognition.
- the input text is divided into blocks (step S501 in FIG. 3).
- the input text is divided into two or more blocks.
- the entire input text may be made into one block.
- step S502 an appropriate number of inserted symbols for each symbol is calculated for each divided block (step S502). That is, an appropriate number of symbol insertions is set according to the number of characters included in the block.
- step S503 the text in the block is divided into words (step S503). Thereby, the text in a block is divided
- the symbol edit likelihood is obtained using a value based on the symbol insertion likelihood and the inter-symbol distance, and the symbol edit position is calculated (step S504). That is, the symbol editing position calculation means 53 reads the symbol insertion model and calculates the symbol insertion likelihood for each word. Then, the symbol editing likelihood is calculated based on the symbol insertion likelihood and the inter-symbol distance.
- the symbol edit likelihood may be calculated by referring to the symbol insertion history instead of the inter-symbol distance. Further, the symbol edit likelihood may be calculated using both the inter-symbol distance and the symbol insertion history.
- step S505 the result of the previous step is reflected in the text (step S505), and this is executed for all blocks (step S506).
- step S506 it is determined that symbol editing is necessary until the number of symbols inserted in the block reaches the number of symbol insertions set in step S501. Then, symbols are inserted in order from a word position having a high symbol editing likelihood until the number of symbols inserted is reached. A series of processing may be executed for all symbols.
- the above text data processing device may be applied to a speech recognition device.
- the speech recognition processing result is used as the input text. That is, the voice recognition device is provided with a voice recognition unit and a text data processing device.
- the voice recognition unit inputs text resulting from the voice recognition process to the text data processing apparatus.
- the speech recognition unit may detect a pause position, and the symbol edit likelihood may be calculated based on the distance from the symbol edit position to the nearest pause position.
- an appropriate number of symbol insertions is calculated for each block including a plurality of divided texts. For this reason, whether or not to insert a symbol can be determined globally in units of blocks. That is, it is possible to insert an optimum amount at the optimum position in units of blocks, not locally the optimum position and amount in a short section such as several words or one sentence. Therefore, it is possible to prevent occurrence of a portion where symbols are excessively inserted in one block, and it is possible to add symbols at appropriate positions.
- the block may be composed of only one sentence. That is, the optimum symbol insertion position can be determined for each block including one or more divided texts.
- the symbol insertion position is specified in consideration of the distance between symbols in addition to the conventional symbol insertion likelihood. For this reason, it becomes possible to insert a symbol at a position where the number of symbols is small, in a portion of a word string in which symbols are easily inserted in terms of language. In addition, it is possible to insert a symbol at a position where there is no problem in terms of meaning or at an appropriate position without causing the sentence to become extremely long or being inserted into unnecessary portions and becoming shredded. In addition, since the inter-symbol distance is taken into account, the symbol insertion likelihood is low in the conventional symbol insertion method, so the symbol is not assigned, or a symbol is inserted at a place where another symbol is assigned. You can also. Therefore, a symbol can be inserted at an appropriate position.
- the symbol editing likelihood reflecting the symbol insertion tendency of the evaluation target speaker can be obtained. Therefore, it is possible to insert a symbol at a grammatically appropriate position according to the speaker's way of speaking.
- FIG. A second embodiment of the present invention will be described in detail with reference to the drawings.
- provisional symbol insertion position calculation means 101 As shown in FIG. 4, in the text data processing apparatus according to the second embodiment, provisional symbol insertion position calculation means 101, symbol insertion model storage means 102, provisional symbol insertion result storage means 103, block division means 104, , Symbol edit determining means 105, symbol edit position calculating means 106, and symbol position determining means 107. Note that the same contents as those in the first embodiment are omitted as appropriate.
- Text is input to the provisional symbol insertion position calculation means 101.
- speech transcription is used as input.
- the speech recognition result may be input text as in the first embodiment.
- the provisional symbol insertion position calculation unit 101 performs morphological analysis to segment the input text into units of words. Further, the likelihood that a symbol is inserted immediately after each word (symbol insertion likelihood) is calculated. Furthermore, symbol insertion determination is performed based on the symbol insertion likelihood. A temporary symbol is inserted into the input text based on the symbol insertion likelihood.
- the symbol insertion model storage unit 102 is the same as the symbol insertion model storage unit 54 in the first embodiment. Therefore, the symbol insertion model storage unit 102 stores a symbol insertion model.
- the temporary symbol insertion result storage means 103 stores the result of the temporary symbol insertion position calculation means 101. Specifically, in addition to the input sentence, the symbol insertion determination result and the symbol insertion likelihood of each symbol are stored.
- the block dividing unit 104 acquires a character string from the provisional symbol insertion result storage unit 103, and divides the text into blocks according to a predetermined standard. It is also possible to divide the text into blocks before inserting the provisional symbols.
- the division criteria are the same as those in the first embodiment. However, when n utterances are selected, it is calculated from the symbol of the provisional symbol insertion result.
- the symbol edit determination unit 105 determines whether the symbol insertion information of the block should be edited for each of the I blocks divided by the block division unit 104.
- the symbol edit determination means 105 obtains information on the words included in the block, information on the presence / absence of symbol insertion (symbol insertion information), and information on the likelihood of symbol insertion from the provisional symbol insertion result storage means 103, and uses these to make a determination. To do. Specifically, the symbol edit determination unit 105 determines whether to insert or delete a symbol. Equations 1 and 2 are used as the determination formula. Of course, equation 3 may be used instead of equation 2.
- the symbol editing position calculation means 106 specifies the symbol editing position for the block that is determined to be edited by the symbol editing determination means 105. Specifically, the symbol edit likelihood is obtained based on the symbol insertion likelihood information acquired from the provisional symbol insertion result storage unit 103. The symbol editing position is specified based on the symbol editing likelihood. Then, the symbol insertion position information is updated, and the information is passed to the symbol position determination means 107.
- Equation 4 The symbol edit likelihood Pr (i, j, C k ) uses Equation 4 or Equation 8.
- y in equation (5) used in equation (4) can be obtained by the following equation (11) or equation (12) in addition to equations (6) and (7).
- the variable y may be calculated using Equation 11 or Equation 12. That is, the variable y may be a distance from the j-th word of the block i to the nearest symbol C k , or may be a distance to any symbol regardless of the type of the symbol. That is, since the symbol is inserted by the provisional symbol insertion position calculation unit 101, the inter-symbol distance of the same symbol Ck can be used. Specifically, the distance is the number of words or characters between symbols.
- Equation 9 the position where the symbol edit likelihood of the symbol C k is the highest among all word boundaries where the symbol C k of the block i is not inserted is obtained.
- a restriction such that a symbol with a lower priority is not inserted at a place where a symbol with a higher priority is inserted.
- the insertion target symbol is a punctuation mark and a punctuation mark
- the punctuation mark has a higher priority than the punctuation mark
- no punctuation mark is selected at the place where the punctuation mark is inserted.
- S i, Ck > ⁇ (C k ) max indicates that the number of insertions of symbol C k in block i is too large. Therefore, the symbol C k is deleted from the portion where the symbol C k of the block i is inserted until S i, Ck ⁇ ⁇ (C k ) max is satisfied .
- the deletion location, i.e., ci , m.noteq.Ck is obtained by the following equation (13).
- the above equation means that the position where the symbol edit likelihood is the lowest among the places where the symbol C k is inserted is output. Further, as a method for calculating another symbol deletion place, the following Expression 14 can be considered.
- the symbol position determination unit 107 inserts a symbol into the input text based on the symbol insertion position information output by the symbol editing determination unit 105 and the symbol editing position calculation unit 106, and outputs the result.
- the input text is divided into words (step S201 in FIG. 5).
- the information is used to divide into words.
- morphological analysis is performed to divide the input text into words.
- the symbol insertion likelihood is determined in units of words delimited in step S201, the symbol to be inserted immediately is determined, and the symbol insertion determination result (symbol insertion information) and information on the symbol insertion likelihood are added (Ste S202). This is performed for all input words (step S203).
- step S204 This makes symbol insertion determination for all words and inserts temporary symbols. Further, the input text is divided according to the above-mentioned predetermined criteria and divided into one or more blocks (step S204). Note that morphological analysis may be performed after the input text is divided into blocks. Note that step S204 may be performed before step S201 or before step S202.
- the symbol insertion frequency is calculated for each block (step S205), and it is determined whether or not symbol editing is necessary for the block based on the result (step S206).
- the editing position is calculated (step S207), and the symbol insertion information is updated (step S208). That is, the symbol is deleted when the number of inserted symbols exceeds the upper limit value, and the symbol is added when the lower limit value is smaller. Then, the symbol insertion position is determined based on the symbol insertion information, and the symbol is inserted into the input sentence (step S209). This series of processing is performed for all blocks (step S210).
- the intersymbol distance is calculated using the provisional symbol insertion result. Then, the symbol editing likelihood is obtained based on the inter-symbol distance, and the symbol editing is performed. Therefore, the function value based on the inter-symbol distance and the symbol edit likelihood can be obtained by using the symbol information having higher reliability than the pseudo inserted symbol. For this reason, it is possible to insert symbols with higher accuracy. In addition, since the number of symbols is determined based on the provisional symbol insertion result, it is possible to delete excessive symbols when symbols are excessively inserted. Furthermore, a symbol once inserted can be replaced with another more appropriate symbol in consideration of the distance between symbols and the symbol insertion history. Therefore, a symbol can be attached to an appropriate position.
- FIG. Embodiment 3 will be described in detail with reference to the drawings.
- FIG. 6 is a block diagram showing the configuration of the text data processing apparatus according to this embodiment.
- description is abbreviate
- the text data processing apparatus is a configuration diagram of a computer operated by the program when the first embodiment is configured by a program.
- the text data processing apparatus is an input device 300, a data processing device 310, and a data storage device 320. And an output device 330.
- the data processing device 310 includes provisional symbol insertion position calculation means 311, block division means 312, symbol edit determination means 313, symbol edit position calculation means 314, and symbol position determination means 315.
- the data storage device 320 includes a symbol insertion model storage unit 321, a provisional symbol insertion result storage unit 322, and a symbol insertion result storage unit 323. Each of these devices may be physically composed of a single device. That is, each device may be configured by one computer.
- the input device 300 inputs morpheme-analyzed text that does not include symbols, speech recognition results, and the like.
- the input device 300 may perform voice recognition processing.
- the input device 300 has a microphone or the like for acquiring audio data.
- the data processing device 310 includes a provisional symbol insertion position calculation unit 311, a block division unit 312, a symbol edit determination unit 313, a symbol edit position calculation unit 314, and a symbol position determination unit 315, and receives an input from the input device 300. Necessary word information and symbol insertion information are obtained from the data storage device 320, the symbol insertion position is calculated, the input character string is edited, and the result is sent to the output device 330.
- the data storage device 320 includes a symbol insertion model storage unit 321, a provisional symbol insertion result storage unit 322, and a symbol insertion result storage unit 323, and mainly includes word information, symbol insertion information, and symbol insertion likelihood information of an input character string.
- a symbol insertion model storage unit 321 a provisional symbol insertion result storage unit 322, and a symbol insertion result storage unit 323, and mainly includes word information, symbol insertion information, and symbol insertion likelihood information of an input character string.
- the symbol insertion model storage unit 321 stores a symbol insertion model.
- the temporary symbol insertion result storage unit 322 stores the calculation result of the temporary symbol insertion position calculating unit 311 and sends necessary information to the block dividing unit 312.
- the symbol insertion result storage unit 323 stores the results of the symbol editing determination unit 313 and the symbol editing position calculation unit 314, and sends the results to the symbol position determination unit 315.
- the present invention can be applied to applications such as a voice recognition device that converts a voice signal into text and a program for realizing the voice recognition device on a computer.
- the text data processing program according to the present embodiment is installed in the speech recognition apparatus.
- applications such as content playback devices and content search devices that display, play back, and search content in units divided by dividing audio and video content into appropriate units, and transcription support devices for recorded audio data Is also applicable.
- Symbols can be inserted at appropriate positions for character strings obtained by converting speech into text.
- a symbol is inserted into a sentence in which spoken speech is converted into a text, one sentence is often long or the sentence is shredded, so that it can be appropriately edited.
- an appropriate insertion position is determined for each block including a plurality of sentences. Therefore, an appropriate amount of symbols can be inserted at an appropriate position.
- the present invention relates to a speech recognition device, a text data processing device, a text data processing method, and a text data processing program, and more particularly, a speech recognition device, a text data processing device, and a text data processing that edit symbols for text data.
- the present invention can be applied to a method and a text data processing program.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
- Document Processing Apparatus (AREA)
Abstract
Description
52 記号編集判定手段
53 記号編集位置算出手段
54 記号挿入モデル記憶手段
55 記号位置確定手段
56 音声認識手段
60 テキストデータ処理装置
61 記号編集判定手段
62 記号編集位置算出手段
101 仮記号挿入位置算出手段
102 記号挿入モデル記憶手段
103 仮記号挿入結果記憶手段
104 ブロック分割手段
105 記号編集判定手段
106 記号編集位置算出手段
107 記号位置確定手段
300 入力装置
310 データ処理装置
311 仮記号挿入位置算出手段
312 ブロック分割手段
313 記号編集判定手段
314 記号編集位置算出手段
315 記号位置確定手段
320 データ記憶装置
321 記号挿入モデル記憶部
322 仮記号挿入結果記憶部
323 記号挿入結果記憶部
406 バッファメモリ
407 処理単位変換部
408 日英翻訳部
421 統計モデルメモリ
422 経験的規則メモリ
450 音声認識装置
515 同期検出装置
517 統合化装置
519 形態素解析
521 分割ルール
533 単位字幕抽出
535 自動字幕生成
537 タイミング情報付与
本発明の第2実施形態について図面を参照して詳細に説明する。
実施形態3について図面を参照して詳細に説明する。図6は、本実施の形態にかかるテキストデータ処理装置の構成を示すブロック図である。なお、実施の形態1,2と同様の内容については、説明を省略する。
Claims (17)
- 複数の分割テキストからなるブロック中の記号挿入頻度に基づいて記号編集の要否を判定する記号編集判定手段と、
前記記号編集判定手段において前記記号編集が要と判定された場合に、単語の記号挿入尤度と記号間距離とに基づいて記号編集尤度を算出し、前記記号編集尤度から前記ブロック中の記号編集位置を算出する記号編集位置算出手段と、を備えるテキストデータ処理装置。 - 前記記号間距離が前記記号編集位置から既に挿入済みの最近傍の記号までの単語数、又は文字数に応じて決定されている請求項1に記載のテキストデータ処理装置。
- 複数の分割テキストからなるブロック中の記号挿入頻度に基づいて記号編集の要否を判定する記号編集判定手段と、
前記記号編集判定手段において前記記号編集が要と判定された場合に、単語の記号挿入尤度と記号挿入済みのテキストの記号挿入履歴とに基づいて記号編集尤度を算出し、前記記号編集尤度から前記ブロック中の記号編集位置を算出する記号編集位置算出手段と、を備えるテキストデータ処理装置。 - 前記記号挿入履歴が同一話者によるテキストにおける当該記号の挿入頻度に応じて設定されている請求項3に記載のテキストデータ処理装置。
- 前記ブロック中に挿入される記号挿入数を算出し、前記ブロック内の記号数が前記記号挿入数になるまで、前記記号編集を要と判定する請求項1乃至4のいずれか1項に記載のテキストデータ処理装置。
- 請求項1乃至5のいずれか1項に記載のテキストデータ処理装置と、
前記テキストデータ処理装置に対して音声認識処理の結果によるテキストを出力する音声認識部とを備える音声認識装置。 - 前記音声認識部がポーズ箇所を検出し、
前記記号編集尤度が、前記記号編集位置から最近傍の前記ポーズ箇所までの距離に基づいて算出されている請求項6に記載の音声認識装置。 - 複数の分割テキストからなるブロック中の記号挿入頻度に基づいて記号編集の要否を判定する判定ステップと、
前記判定ステップにおいて前記記号編集が要と判定された場合に、単語の記号挿入尤度と記号間距離とに基づいて記号編集尤度を算出し、前記記号編集尤度から前記ブロック中の記号編集位置を算出するステップと、を備えるテキストデータ処理方法。 - 前記記号間距離が前記記号編集位置から既に挿入済みの最近傍の記号までの単語数、又は文字数に応じて決定されている請求項8に記載のテキストデータ処理方法。
- 複数の分割テキストからなるブロック中の記号挿入頻度に基づいて記号編集の要否を判定する判定ステップと、
前記判定ステップにおいて前記記号編集が要と判定された場合に、単語の記号挿入尤度と記号挿入済みのテキストの記号挿入履歴とに基づいて記号編集尤度を算出し、前記記号編集尤度から前記ブロック中の記号編集位置を算出するステップと、を備えるテキストデータ処理方法。 - 前記記号挿入履歴が同一話者によるテキストにおける当該記号の挿入頻度に応じて設定されている請求項10に記載のテキストデータ処理方法。
- 前記ブロック中に挿入される記号挿入数を算出し、前記ブロック内の記号数が前記記号挿入数になるまで、前記記号編集を要と判定する請求項8乃至11のいずれか1項に記載のテキストデータ処理方法。
- 入力されたテキストに対して記号を編集するテキストデータ処理プログラムが格納された記録媒体であって、
コンピュータに対して、
複数の分割テキストからなるブロック中の記号挿入頻度に基づいて記号編集の要否を判定させる判定ステップと、
前記判定ステップにおいて前記記号編集が要と判定された場合に、単語の記号挿入尤度と記号間距離とに基づいて記号編集尤度を算出させ、前記記号編集尤度から前記ブロック中の記号編集位置を算出させるステップと、を備えるテキストデータ処理プログラムが格納された記録媒体。 - 前記記号間距離が前記記号編集位置から既に挿入済みの最近傍の記号までの単語数、又は文字数に応じて決定されている請求項13に記載のテキストデータ処理プログラム。
- 入力されたテキストに対して記号を編集するテキストデータ処理プログラムが格納された記録媒体であって、
コンピュータに対して、
複数の分割テキストからなるブロック中の記号挿入頻度に基づいて記号編集の要否を判定させる判定ステップと、
前記判定ステップにおいて前記記号編集が要と判定された場合に、単語の記号挿入尤度と記号挿入済みのテキストの記号挿入履歴とに基づいて記号編集尤度を算出させ、前記記号編集尤度から前記ブロック中の記号編集位置を算出させるステップと、を備えるテキストデータ処理プログラムが格納された記録媒体。 - 前記記号挿入履歴が同一話者によるテキストにおける当該記号の挿入頻度に応じて設定されている請求項15に記載のテキストデータ処理プログラムが格納された記録媒体。
- 前記ブロック中に挿入される記号挿入数を算出し、前記ブロック内の記号数が前記記号挿入数になるまで、前記記号編集を要と判定する請求項13乃至16のいずれか1項に記載のテキストデータ処理プログラムが格納された記録媒体。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US12/922,046 US8892435B2 (en) | 2008-04-03 | 2009-02-13 | Text data processing apparatus, text data processing method, and recording medium storing text data processing program |
JP2010505435A JPWO2009122779A1 (ja) | 2008-04-03 | 2009-02-13 | テキストデータ処理装置、方法、プログラム |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2008097350 | 2008-04-03 | ||
JP2008-097350 | 2008-04-03 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2009122779A1 true WO2009122779A1 (ja) | 2009-10-08 |
Family
ID=41135178
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2009/052378 WO2009122779A1 (ja) | 2008-04-03 | 2009-02-13 | テキストデータ処理装置、方法、プログラムが格納された記録媒体 |
Country Status (3)
Country | Link |
---|---|
US (1) | US8892435B2 (ja) |
JP (1) | JPWO2009122779A1 (ja) |
WO (1) | WO2009122779A1 (ja) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2011248538A (ja) * | 2010-05-25 | 2011-12-08 | Nippon Telegr & Teleph Corp <Ntt> | 記号入力支援装置、記号入力支援方法、及びプログラム |
JP2015219480A (ja) * | 2014-05-21 | 2015-12-07 | 日本電信電話株式会社 | 対話状況特徴計算装置、文末記号推定装置、これらの方法及びプログラム |
US10606940B2 (en) | 2013-09-20 | 2020-03-31 | Kabushiki Kaisha Toshiba | Annotation sharing method, annotation sharing apparatus, and computer program product |
JP2020160782A (ja) * | 2019-03-26 | 2020-10-01 | 日本放送協会 | 自然言語データ処理装置およびプログラム |
WO2021215262A1 (ja) * | 2020-04-20 | 2021-10-28 | 株式会社Nttドコモ | 句点削除モデル学習装置、句点削除モデル及び判定装置 |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2013089130A (ja) * | 2011-10-20 | 2013-05-13 | Sony Corp | 情報処理装置、情報処理方法、プログラム、及び記録媒体 |
CN104143331B (zh) | 2013-05-24 | 2015-12-09 | 腾讯科技(深圳)有限公司 | 一种添加标点的方法和系统 |
CN104142915B (zh) * | 2013-05-24 | 2016-02-24 | 腾讯科技(深圳)有限公司 | 一种添加标点的方法和系统 |
CN104156344B (zh) * | 2014-07-21 | 2016-09-28 | 小米科技有限责任公司 | 文本编辑方法和装置 |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0417059A (ja) * | 1990-05-10 | 1992-01-21 | Canon Inc | 文書作成装置 |
JPH04259065A (ja) * | 1991-02-14 | 1992-09-14 | Toshiba Corp | 言語生成装置 |
JP2004070634A (ja) * | 2002-08-06 | 2004-03-04 | Seiko Epson Corp | 読点挿入位置決定方法および読点挿入位置決定装置ならびに読点挿入位置決定処理プログラム |
Family Cites Families (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3009642B2 (ja) | 1997-10-22 | 2000-02-14 | 株式会社エイ・ティ・アール音声翻訳通信研究所 | 音声言語処理単位変換装置 |
CN1159662C (zh) * | 1998-05-13 | 2004-07-28 | 国际商业机器公司 | 连续语音识别中的标点符号自动生成装置及方法 |
US6067514A (en) * | 1998-06-23 | 2000-05-23 | International Business Machines Corporation | Method for automatically punctuating a speech utterance in a continuous speech recognition system |
US7881936B2 (en) * | 1998-12-04 | 2011-02-01 | Tegic Communications, Inc. | Multimodal disambiguation of speech recognition |
US7490092B2 (en) * | 2000-07-06 | 2009-02-10 | Streamsage, Inc. | Method and system for indexing and searching timed media information based upon relevance intervals |
JP3782943B2 (ja) * | 2001-02-20 | 2006-06-07 | インターナショナル・ビジネス・マシーンズ・コーポレーション | 音声認識装置、コンピュータ・システム、音声認識方法、プログラムおよび記録媒体 |
US7035804B2 (en) * | 2001-04-26 | 2006-04-25 | Stenograph, L.L.C. | Systems and methods for automated audio transcription, translation, and transfer |
JP4538618B2 (ja) | 2001-05-17 | 2010-09-08 | 独立行政法人情報通信研究機構 | 字幕番組制作システムにおける表示単位字幕文の自動生成方法 |
US20040163034A1 (en) * | 2002-10-17 | 2004-08-19 | Sean Colbath | Systems and methods for labeling clusters of documents |
US7580838B2 (en) * | 2002-11-22 | 2009-08-25 | Scansoft, Inc. | Automatic insertion of non-verbalized punctuation |
US20040148170A1 (en) * | 2003-01-23 | 2004-07-29 | Alejandro Acero | Statistical classifiers for spoken language understanding and command/control scenarios |
JP2005157494A (ja) * | 2003-11-20 | 2005-06-16 | Aruze Corp | 会話制御装置及び会話制御方法 |
US20070118372A1 (en) * | 2005-11-23 | 2007-05-24 | General Electric Company | System and method for generating closed captions |
WO2007091096A1 (en) * | 2006-02-10 | 2007-08-16 | Spinvox Limited | A mass-scale, user-independent, device-independent, voice message to text conversion system |
US20080070205A1 (en) * | 2006-06-16 | 2008-03-20 | Understanding Corporation, Inc. | Methods, systems, and computer program products for adjusting readability of reading material to a target readability level |
US20090037171A1 (en) * | 2007-08-03 | 2009-02-05 | Mcfarland Tim J | Real-time voice transcription system |
US20100138221A1 (en) * | 2008-12-02 | 2010-06-03 | Boys Donald R | Dedicated hardware/software voice-to-text system |
-
2009
- 2009-02-13 WO PCT/JP2009/052378 patent/WO2009122779A1/ja active Application Filing
- 2009-02-13 US US12/922,046 patent/US8892435B2/en not_active Expired - Fee Related
- 2009-02-13 JP JP2010505435A patent/JPWO2009122779A1/ja active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0417059A (ja) * | 1990-05-10 | 1992-01-21 | Canon Inc | 文書作成装置 |
JPH04259065A (ja) * | 1991-02-14 | 1992-09-14 | Toshiba Corp | 言語生成装置 |
JP2004070634A (ja) * | 2002-08-06 | 2004-03-04 | Seiko Epson Corp | 読点挿入位置決定方法および読点挿入位置決定装置ならびに読点挿入位置決定処理プログラム |
Non-Patent Citations (1)
Title |
---|
SATOSHI NAKAZAWA ET AL.: "Danwa Shihyo to Text-cho o Mochiita Koen Onsei to Presentation Shiryo no Taiozuke", JOURNAL OF NATURAL LANGUAGE PROCESSING, vol. 12, no. 2, 31 March 2005 (2005-03-31), pages 133 - 156 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2011248538A (ja) * | 2010-05-25 | 2011-12-08 | Nippon Telegr & Teleph Corp <Ntt> | 記号入力支援装置、記号入力支援方法、及びプログラム |
US10606940B2 (en) | 2013-09-20 | 2020-03-31 | Kabushiki Kaisha Toshiba | Annotation sharing method, annotation sharing apparatus, and computer program product |
JP2015219480A (ja) * | 2014-05-21 | 2015-12-07 | 日本電信電話株式会社 | 対話状況特徴計算装置、文末記号推定装置、これらの方法及びプログラム |
JP2020160782A (ja) * | 2019-03-26 | 2020-10-01 | 日本放送協会 | 自然言語データ処理装置およびプログラム |
JP7253951B2 (ja) | 2019-03-26 | 2023-04-07 | 日本放送協会 | 自然言語データ処理装置およびプログラム |
WO2021215262A1 (ja) * | 2020-04-20 | 2021-10-28 | 株式会社Nttドコモ | 句点削除モデル学習装置、句点削除モデル及び判定装置 |
Also Published As
Publication number | Publication date |
---|---|
US20110010175A1 (en) | 2011-01-13 |
US8892435B2 (en) | 2014-11-18 |
JPWO2009122779A1 (ja) | 2011-07-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2009122779A1 (ja) | テキストデータ処理装置、方法、プログラムが格納された記録媒体 | |
JP5104762B2 (ja) | コンテンツ要約システムと方法とプログラム | |
JP3782943B2 (ja) | 音声認識装置、コンピュータ・システム、音声認識方法、プログラムおよび記録媒体 | |
US7480612B2 (en) | Word predicting method, voice recognition method, and voice recognition apparatus and program using the same methods | |
JP5141695B2 (ja) | 記号挿入装置および記号挿入方法 | |
US9588967B2 (en) | Interpretation apparatus and method | |
EP1422692A2 (en) | Automatic insertion of non-verbalized punctuation in speech recognition | |
CN105336322A (zh) | 多音字模型训练方法、语音合成方法及装置 | |
JP2008134475A (ja) | 入力された音声のアクセントを認識する技術 | |
JPWO2005069171A1 (ja) | 文書対応付け装置、および文書対応付け方法 | |
JP2009139862A (ja) | 音声認識装置及びコンピュータプログラム | |
JP2010230695A (ja) | 音声の境界推定装置及び方法 | |
JP2012181358A (ja) | テキスト表示時間決定装置、テキスト表示システム、方法およびプログラム | |
US10304457B2 (en) | Transcription support system and transcription support method | |
JP2015212732A (ja) | 音喩認識装置、及びプログラム | |
JP2010139745A (ja) | 統計的発音変異モデルを記憶する記録媒体、自動音声認識システム及びコンピュータプログラム | |
Levy et al. | The effect of pitch, intensity and pause duration in punctuation detection | |
KR101747873B1 (ko) | 음성인식을 위한 언어모델 생성 장치 및 방법 | |
JP2021009253A (ja) | プログラム、情報処理装置、及び情報処理方法 | |
JP4758758B2 (ja) | 辞書作成装置および辞書作成プログラム | |
JP2000259176A (ja) | 音声認識装置およびその記録媒体 | |
JP2017215555A (ja) | 音声翻訳装置及び音声翻訳システム | |
JP3958908B2 (ja) | 書き起こしテキスト自動生成装置、音声認識装置および記録媒体 | |
KR100883649B1 (ko) | 텍스트/음성 변환 장치 및 방법 | |
JP5044791B2 (ja) | 字幕ずれ推定装置、補正装置および再生装置 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 09728669 Country of ref document: EP Kind code of ref document: A1 |
|
DPE2 | Request for preliminary examination filed before expiration of 19th month from priority date (pct application filed from 20040101) | ||
WWE | Wipo information: entry into national phase |
Ref document number: 12922046 Country of ref document: US |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2010505435 Country of ref document: JP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 09728669 Country of ref document: EP Kind code of ref document: A1 |