JP2005044103A - Document creation device and method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a document creation device and method making it possible to efficiently select an optimal character string from a plurality of character string candidates corresponding to words recognized from input voice when a document is created through the recognition of the input voice. <P>SOLUTION: A plurality of character strings and another plurality of character strings stored in a storage means which stores readings corresponding to the former plurality of character strings are each set as either subjects to be learned or non-subjects to be learned. For a plurality of words recognized from the input voice, a plurality of candidate character strings with the same or similar readings to the words are selected from among the plurality of character strings stored in the storage means. The character strings which are the non-subjects to be learned are excluded from the plurality of candidate character strings corresponding to the words, and the character strings corresponding to the plurality of words recognized from the input voice are selected from the remaining candidate character strings to create a document. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a document creation device that recognizes input speech and creates a document.
[0002]
[Prior art]
In a conventional document creation device, a learning function is a table in which a word or phrase selected by an operator, or a word or phrase before or after the word or phrase is stored in a table, and the word or phrase appears in a candidate in a later input operation. The candidate ranking was adjusted in consideration of the time stored inside. Furthermore, in learning, specific parts of speech such as attached words, unconverted hiragana candidates, katakana candidates, and the like could be excluded from learning by setting.
[0003]
In general, the learning function in the document creation apparatus sequentially stores candidates (words or phrases) selected by the user from recognition or conversion candidates for input speech or character strings, and creates new speech or characters. When a string is input, if the candidate corresponding to the input speech or input character string includes a stored word or phrase, the recognition accuracy is improved by giving priority to the most recently learned candidate. Alternatively, it is a technique for improving the conversion accuracy.
[0004]
However, in the recognition candidate group or conversion candidate group corresponding to the input speech or character string, there are words that are generally used and words that are rarely used, and if the latter is selected, then the same recognition candidate group / When a conversion candidate group is obtained, in the conventional learning function, a learned word is given priority over a general word, and thus a candidate selection process is often required.
[0005]
In particular, in the case of Kana-Kanji conversion, depending on the reading of the input character string to be converted, the expression of the conversion result, the part of speech, etc., `` no learning '', `` temporary latest use learning '', `` temporary frequency learning '', There is a technique for changing to a learning method such as “permanent latest use learning” or “permanent frequency learning” and enabling selection of an optimal learning method according to the use situation (see, for example, Patent Document 1).
[0006]
[Patent Document 1]
Japanese Patent Laid-Open No. 9-62668
[0007]
[Problems to be solved by the invention]
When the input speech is recognized, when a word that is rarely used is selected from the candidate words for the input speech, the conventional learning function allows the candidate word for the speech to be input the next time the same speech is input again. As a result, the selected word is preferentially selected or displayed, and as a result, the optimum word must be selected again from the list of candidate words corresponding to the input speech, thereby efficiently creating a document. There was a problem that could not be performed.
[0008]
Therefore, in view of the above problems, the present invention efficiently creates an optimum character string from among a plurality of character string candidates corresponding to words recognized from input speech when creating a document by recognizing input speech. An object of the present invention is to provide a document creation apparatus and method that can be selected.
[0009]
[Means for Solving the Problems]
The present invention is for creating a document by recognizing input speech, and (a) stored in a storage means for storing a plurality of character strings and at least a reading corresponding to each of the plurality of character strings. In addition, each of the plurality of character strings is set as one of a learning target and a learning non-target, and (b) each of the plurality of words recognized from the input speech is stored in the storage unit. A plurality of candidate character strings having the same or similar reading as the word are selected from among the plurality of character strings, and (c) a character string corresponding to each of the plurality of words recognized from the input speech, Selecting from the remaining candidate character strings excluding the non-learning character string from the plurality of candidate character strings corresponding to a word, creating the document, and (d) one of the plurality of words The first word that is A character string corresponding to the first word in the document is replaced with a candidate character string selected by the user among the plurality of candidate character strings to be (e) a character string corresponding to each of the plurality of words Among the plurality of candidate character strings corresponding to each word, the learning target character string and the learning non-target character string, only the learning target character string corresponds to the word. By preferentially selecting from a plurality of candidate character strings, when creating a document by recognizing the input speech, it is best to select from among a plurality of character string candidates corresponding to the words recognized from the input speech. A character string can be selected efficiently.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[0011]
FIG. 1 shows a configuration example of a document creation apparatus 1 according to an embodiment of a document creation apparatus of the present invention, which is roughly composed of a document creation unit 11, an input unit 12, and an output unit 13. .
[0012]
The input unit 12 includes, for example, a microphone 131 for inputting a voice such as a user's speech, a voice recognition unit 132 that recognizes a voice (input voice) input from the microphone 131 and converts it into a character code, and a user An editing instruction input unit 133 for inputting various editing instructions to the document creation apparatus 1 is provided. From the input unit 12, a character code obtained as a result of voice recognition of the input voice, a control signal (editing instruction for dictionary maintenance, a signal for instructing document creation, and instruction for candidate selection) indicating an editing instruction input by the user Signal etc.) is output and input to the document creation unit 11.
[0013]
The document creation unit 11 performs voice recognition at the input control unit 112, dictionary maintenance unit 114, candidate selection unit 115, word learning unit 116, language analysis unit 118, candidate expansion unit 119, output control unit 124, and input unit 12. Based on the obtained character code, an analysis rule storage unit 121 that stores an analysis rule used when language analysis is performed by the language analysis unit 118, a learning table storage unit 117 that stores a learning table, and a temporary storage of recognition candidates. A candidate buffer 120 for storing the word dictionary, a word dictionary storage unit 122 for storing the word dictionary, and a document buffer 125 for temporarily storing the document.
[0014]
In the following description, a word dictionary stored in the word dictionary storage unit 122 is simply referred to as a word dictionary 122, and a learning table stored in the learning table storage unit 117 is simply referred to as a learning table 117. The analysis rule stored in the analysis rule storage unit 121 is simply referred to as an analysis rule 121.
[0015]
The document creation unit 11 reads a character string corresponding to each of a plurality of words recognized from the input speech, and a plurality of candidate character strings whose readings are the same as or similar to the word among the plurality of character strings in the word dictionary 122. Are selected from the remaining candidate character strings excluding the candidate character strings that are not subject to learning, and a document is created. The created document is stored in the document buffer 125. Further, the first character string designated by the user in the document is replaced with a candidate character string selected by the user from among a plurality of candidate character strings corresponding to the first character string, and the document is edited. .
[0016]
The character code and control signal output from the input unit 12 are input to the input control unit 112. When a signal instructing maintenance of the word 122 is input, the dictionary maintenance unit 114 is activated. When a signal instructing candidate selection is input, the candidate selecting unit 115 is activated.
[0017]
The dictionary maintenance unit 114 stores each word stored in the word dictionary 122 in association with attribute information for each word in order to distinguish whether the word is a learning non-target or a learning target. Here, for example, among the learning target and non-learning target words, only the non-learning target word is stored in association with “1” as the learning non-target attribute information.
[0018]
In the initial state, all the words in the word dictionary 122 may be set as learning targets (that is, the learning non-target attribute information “1” is not registered), but are generally rarely used. The learning non-target may be set in advance for the word (that is, the learning non-target attribute information “1” is not registered).
[0019]
When the character code is input from the input unit 12, the input control unit 112 stores it in the input buffer 113 and activates the language analysis unit 118. Here, the character code may be uniquely determined or may have a plurality of candidates.
[0020]
The language analysis unit 118 refers to the word dictionary 122 and the analysis rule 121 when the input target is analyzed from the input time interval of the input speech or the number of characters stored in the input buffer 113, and the character code stored in the input buffer 113. Are sequentially combined to create a word sequence, and pass the word sequence to the candidate expansion unit 119.
[0021]
In the word dictionary 105, for each of a plurality of words, identification information for identifying each word (also referred to as a dictionary number or dictionary ID), its notation, reading, part of speech, frequency of use, etc. are registered in advance. When the word specified by the user is set from the learning target to the learning non-target, the dictionary maintenance unit 114 writes the learning non-target attribute information as information of the word, and the word specified by the user When setting the learning non-target as the learning target, the dictionary maintenance unit 114 deletes the learning non-target attribute associated with the word.
[0022]
The analysis rule 121 describes whether connection between parts of speech is possible, ease of connection between words, and the like.
[0023]
The candidate expansion unit 119 acquires a homophone from the word dictionary 120 for each word or phrase in the word sequence output from the language analysis unit 118, and stores the acquired homophone word dictionary number and notation in the candidate buffer 120. Store. Further, the candidate expansion unit 119 gives candidate ranks to the homophones stored in the candidate buffer 120 and corresponding to the words in the word series. At that time, the highest candidate ranking is assigned to the most recently used word stored in the learning table 117. After determining the candidate rank, the candidate expansion unit 119 stores the dictionary number and notation of the first candidate in the document buffer 125. In the document buffer 125, a link is established so that the homophone word stored in the candidate buffer 120 can be recognized from the first candidate.
[0024]
The output control unit 124 displays each character string in the document stored in the document buffer 125 on the output unit 13 and displays the word linked to the word in the character string stored in the candidate buffer 120. A list of homophone candidates is displayed on the output unit 13.
[0025]
On the other hand, the input control unit 112 activates the candidate selection unit 115 when the control signal input from the input unit 12 instructs the selection of the next candidate for the word in the character string displayed on the output unit 13. . The candidate selection unit 115 displays a list of homophones of the word linked to the word from the candidate buffer 108 on the output unit 13.
[0026]
When the user selects a desired word from the list, the candidate selection unit 115 refers to the word dictionary 122 and determines whether or not attribute information indicating that the selected word is a learning non-target word is attached to the selected word. When the attribute information is not attached (only when it is not a learning non-target word), the word learning unit 116 is notified that the selected word is confirmed.
[0027]
In response to the notification, the word learning unit 116 stores, in the learning table 117, information on the confirmed word (learning target word that is not a learning non-target word) and information on the immediately preceding and following words as candidate selection history information. To do.
[0028]
In addition, the candidate selection unit 115 retrieves the selected word from the candidate buffer 120, replaces the word in the document buffer 112 displaying the list of candidates currently with the selected word, and updates the screen display. To do. In other words, the candidate selection unit 115 edits the document in the document buffer 125.
[0029]
FIG. 2 is a flowchart for explaining the processing operation of the document creation apparatus 1 in FIG. 1. The learning target word registered in the word dictionary 122 is changed to a learning non-target, or a learning non-target is selected. The processing operation when a word is changed to a learning target and a document is created according to the presence or absence of learning non-target attribute information of the word in the word dictionary 122 is shown.
[0030]
The input control unit 112 receives a signal for instructing maintenance of the word dictionary from the input unit 12 (step S1), and then receives information for designating a word to be processed by the user (step S2). ), The dictionary maintenance unit 114 acquires a word that matches the input information from the word dictionary 122 (step S3). Here, the information for designating a word may be all or a part of information such as word notation, reading, part of speech. In step S3, a list of acquired words is displayed. From this list, a word to be newly learned or a word to be learned is selected from words that are already not to be learned (step S4). .
[0031]
The dictionary maintenance unit 114 temporarily stores a temporary file in order to provide attribute information indicating that the word selected for learning non-target (the learning target) has a learning non-target attribute (step S5). In order to delete the selected word from the table 117, the word learning unit 116 is notified of the selected word. When the notified word is stored in the learning table 117, the word learning unit 116 deletes the notified word from the learning table 117 (step S6). Further, the word selected to be a learning target (non-learning target) is temporarily stored in a temporary file in order to delete the attribute information indicating that it is a learning non-target attribute (step S7).
[0032]
Steps S2 to S8 are repeated until the operation for adding or deleting learning non-target attribute information to the words registered in the word dictionary 122 is completed. The dictionary maintenance unit 114 updates the word dictionary 122 based on the contents recorded in the temporary file at the end of the maintenance (step S9). That is, the learning non-target attribute information is recorded in the word dictionary 122 for the word selected to be the learning non-target in the word dictionary 122, and the word selected to be the learning target in the word dictionary 122 is stored in the word dictionary 122. The learning non-target attribute information is deleted from the word dictionary 122.
[0033]
In step S1, after a signal for instructing document creation is input from the input unit 12, when a character code as a result of speech recognition of the input speech is input (step S10), the input control unit 112 displays the input character code. The data are sequentially stored in the input buffer 103 (step S11).
[0034]
The language analysis unit 118 monitors the character code stored in the input buffer 113, and if it is determined that the character is to be analyzed from the input time interval, the number of characters, etc. (step S12), the language analysis unit 118 refers to the word dictionary 122 and the analysis rule 121 Language analysis is performed (step S13).
[0035]
Here, the linguistic analysis is to combine the character codes to be analyzed in time series, and create a word series that is the most linguistically correct from the information in the word dictionary 122 and the analysis rule 121. The language analysis unit 118 stores the word sequence of the analysis result in the candidate buffer 120 via the candidate expansion unit 119 (step S14).
[0036]
For each word stored in the candidate buffer 120, the candidate expansion unit 119 obtains, from the word dictionary 122, a word that is a homophone of the word (a word having the same or similar reading) and is a conversion candidate for the word, It adds to the candidate buffer 120 (step S15). Further, the candidate expansion unit 119 checks whether each word stored in the candidate buffer 120 is recorded in the learning table 117.
[0037]
Since the history information about the recently selected word (for example, the word ID of the word) is recorded in the learning table 117, it is recorded as a history in the learning table 117 in the words stored in the candidate buffer 120. If there is a word, the word candidate rank stored in the candidate buffer 120 is determined so that the word rank corresponding to the latest history recorded in the learning table 117 is the highest (step S16). The candidate expansion unit 119 stores in the document buffer 125 a word sequence obtained by replacing each word of the word sequence obtained by the language analysis unit 118 with the word having the first candidate rank (step S17).
[0038]
The output control unit 124 displays on the output unit 13 the first candidate consisting of the first candidate word stored in the document buffer 125 (step S18). The user confirms the first candidate displayed on the screen, and if it is different from the expected notation, selects the next candidate.
[0039]
The next candidate selection is started by instructing which word candidate is selected from the editing instruction input unit 123 of the input unit 12 by means of, for example, a mouse click (step S19). The candidate selection unit 115 is a list of homophones (candidate words stored in the candidate buffer 120) of the designated word linked to the word designated by the user in the character string displayed on the output unit 13. Is displayed on the output unit 13 (step S20).
[0040]
When the user selects a desired candidate from the list of candidates (step 21), the candidate selecting unit 115 refers to the word dictionary 122 to check whether learning non-target attribute information is given to the selected word. If not, information on the selected word (such as the word ID of the selected word and the word IDs of the words before and after the word) is notified to the word learning unit 116 (step S22). ).
[0041]
The word learning unit 116 records information on the selected word in the learning table 117 as candidate selection history information (step S23). Further, the candidate selection unit 115 rewrites the word before selection stored in the document buffer 125 with the selected word (step S24), and updates the screen display (step S18). That is, the result of replacement with the word selected by the next candidate selection operation is displayed on the screen.
[0042]
Note that the input control unit 112 receives a control signal input from the input unit 12 when it is neither a control signal for instructing dictionary maintenance, a control signal for instructing document creation, or a control signal for instructing candidate selection. Performs processing instructed by the control signal and waits for input (step S25).
[0043]
FIG. 3 shows an example of the word dictionary 122. The word dictionary 122 stores, for each word, a word ID that is a key when searching for the word, and information such as reading / notation / part of speech / frequency necessary for language analysis by the language analysis unit 118. ing. This information is appropriately increased or decreased depending on the type of input. Further, attribute information indicating whether or not the word is a non-learning object is also stored in association with each word.
[0044]
When the user performs the processing from step S1 to step S9 in FIG. 2, for example, the words with the word IDs “xx08” and “xx0a” are set as learning non-target words among the words in FIG. The dictionary maintenance unit 114 writes “1” as learning non-target attribute information corresponding to the words having the word IDs “xx08” and “xx0a” in the word dictionary 122 (see FIG. 3B).
[0045]
In step S22 of FIG. 2, the candidate selection unit 115 refers to the word dictionary 122 as shown in FIG. 3B and writes “1” as learning non-target attribute information of the word selected as the next candidate. The selected word is not recorded in the learning table 117. That is, the selected word is not learned.
[0046]
Next, a character string corresponding to the input speech (a character string output from the document creation unit 11 and displayed from the output unit 13 in a case where a learning non-target attribute is not given to a certain word in the word dictionary 122) ) With a specific example.
[0047]
As a result of voice recognition of the input voice from the input unit 12, as shown in FIG. 4A, a character code string such as “Meet Hirosawa” is obtained and stored in the input buffer 113. (Step S1, Step S10 to Step S11 in FIG. 2) As a result of the language analysis by the language analysis unit 118, a word series such as “Hirosawa / san / ni / o / ai / shi / masu” is output. In FIG. 4, “/” indicates a break between words. This word sequence is stored in the candidate buffer 120 (steps S12 to S14). The candidate expansion unit 119 uses the word dictionary 122 to obtain a homophone for each word in this word series. In an initial state (a state in which learning is not performed), homophones are ranked based on the frequency of use of each word stored in advance in the word dictionary 122 (steps S15 and S16), and FIG. As shown in the figure, a word series consisting of the words with the highest candidate ranking for each word, such as “Hirosawa / san / ni / o / meet / shi / masu” is obtained (steps S17 to S18). When a list of next candidates for “Hirosawa” is displayed (step S19), a list of conversion candidates as shown in FIG. 4C is obtained (step S20). From this list, the second candidate “Serizawa” is selected (step S21).
[0048]
When the word is a learning target, the word is recorded in the learning table (steps S22 and S23). At this time, the output control unit 124 replaces “Hirosawa” with the selected candidate “Serizawa” and displays it as shown in FIG. 4D (steps S24 and S18).
[0049]
After that, as shown in FIG. 4E, when there is a voice input such as “There was a lotus from Hirosawa” from the input unit 12, when this is stored in the input buffer 113 in the same manner as described above. As a result of the language analysis by the language analysis unit 118, a word sequence such as “Hirosawa / san / kara / renraku / ga / arashi / masashi / ta” is output, and this word sequence is stored in the candidate buffer 120. (Step S12 to Step S14). The candidate expansion unit 119 uses the word dictionary 122 to obtain a homophone for each word in this word series. At that time, since “Serizawa” selected as the previous candidate is recorded in the learning table 117, the candidate ranking of this word is the highest among the homophones corresponding to “Hirosawa”, and FIG. As shown in f), a word sequence consisting of the words having the first candidate rank for each word, such as “Serizawa / san / kara / contact / gai / yes / masashi / ta” is obtained (steps S16 to S18). ).
[0050]
On the other hand, when the word “Serizawa” is non-learning (for example, when the value of learning non-target attribute information is “1”), even if the word is selected from the candidate list in step S21, step S23 is skipped and the history is not recorded in the learning table 117. Therefore, in step S16, “Hirosawa”, which is the most frequently used among the homophones corresponding to “Hirosawa”, has the highest candidate ranking. As shown in FIG. The word sequence of “/ from / contact / ga // yes / better / ta” is obtained (steps S16 to S18).
[0051]
In the document creation device 1 of the above embodiment, when the next candidate is selected in step S19 and subsequent steps, the dictionary maintenance unit 114 is activated at the same time, and the word specified by the user in the character string displayed on the output unit 13 is displayed. When a list of conversion candidates corresponding to is displayed (step S20), a learning non-target attribute may be assigned to the words in the list (set as learning non-target). That is, for example, when a word sequence as shown in FIG. 4B is displayed from the output unit 13, the word “Hirosawa” in the word sequence is designated and the candidate selection unit 115 is activated, A list of word conversion candidates is displayed as shown in FIG. 5A (step S20). In the list of conversion candidates, as shown in FIG. 5A, character strings of options are arranged in descending order of candidate rank. Buttons B1 to B4 on the right side of each candidate character string instruct to set the word as a learning non-object, or instruct to change to a learning object if it is already set as a learning non-object. Is for. Further, each button B1 to B4 also serves as an indicator for displaying the fact that learning non-target attribute information is already given to the word. However, since the character string “Hirosawa” of the fifth candidate is a non-conversion candidate, there is no button.
[0052]
When the third candidate “Serizawa” is selected in the list shown in FIG. 5A, when the button B3 is selected (step S21), the learning non-target attribute information is added simultaneously with the candidate selection instruction (learning non-target). Although the notation “Hirosawa” currently displayed is replaced with “Serizawa”, step S23 is skipped, and the character string “Serizawa” is not stored in the learning table 117. Not recorded (step S24, step S18). In this case, the dictionary maintenance unit 114 writes “1” in the word dictionary 122 as learning non-target attribute information corresponding to the word “Serizawa” (see FIG. 3B).
[0053]
Next, when a list of conversion candidates for the input of “Hirosawa” is displayed (steps S19 and S20), “1” is included in the learning non-target attribute information corresponding to the word “Serizawa” in the word dictionary 122. Since it has been written, as shown in FIG. 5B, a check mark indicating that learning is not set is displayed on the button B3 corresponding to “Serizawa” in the conversion candidate list. .
[0054]
On the other hand, when the third candidate “Serizawa” is selected in the state where the candidate list as shown in FIG. 5A is displayed, the portion excluding the button B3 (for example, the character string “Serizawa” When the number “3” representing the candidate rank attached to the character string is selected (step S21), it means only the instruction for candidate selection (not including the instruction for giving the learning non-target attribute) and is currently displayed. The notation “Hirosawa” is replaced with “Serizawa”, and in step S23, the character string “Serizawa” is recorded in the learning table 117 (steps S24 and S18).
[0055]
Further, when the third candidate “Serizawa” is selected in the state where the candidate list shown in FIG. 5B is displayed, a portion excluding the button B3 (for example, a character string “Serizawa” or the character When the number “3” representing the candidate rank attached to the column is selected (step S21), it means that only the candidate selection instruction is set while being set as a learning non-target, and is currently displayed as “Hirosawa”. Although the notation is replaced with “Serizawa”, step S23 is skipped and the character string “Serizawa” is not recorded in the learning table 117 (step S24, step S18).
[0056]
Further, in the state where the candidate list shown in FIG. 5B is displayed, when the third candidate “Serizawa” is selected, when the button B3 on which a check mark is displayed is selected (step S21), , Meaning deletion of candidate selection and learning non-target attribute information (setting as a learning target), the currently displayed “Hirosawa” is replaced with “Serizawa”, and “Serizawa” is displayed in step S23. A character string is recorded in the learning table 117 (step S24, step S18). That is, thereafter, the character string “Serizawa” is a learning target.
[0057]
Next, a character string corresponding to the input speech (a character string output from the document creation unit 11 and displayed from the output unit 13 in a case where a learning non-target attribute is not given to a certain word in the word dictionary 122) ) Will be described with another specific example.
[0058]
For example, as a result of speech recognition of the input speech in a political discussion, a character code string “Sefu” is obtained. On the other hand, the candidate expansion unit 119 uses the word dictionary 122 to search for this word sequence. As a result of obtaining the homophone of each word, when “safe” is output as the word having the first candidate rank for the word (step S17 to step S18), the character “sefu” is obtained from the next input voice. When the code is obtained, “safe” is output again. However, since this is a place for political debate, "Sef" should be "Government". In such a case, first, the dictionary maintenance unit 114 is activated and the learning non-target attribute is assigned to the character string “safe”, so that “safe” is not output, and the candidate ranking is always second place. Can output "government".
[0059]
As described above, according to the above embodiment, a character string is recorded in the learning table 117 by selection from a list of candidate character strings corresponding to the input speech, and the learning table is next input when similar speech is input. The function of preferentially outputting the character string recorded in 117 can be turned on / off for each word in the word dictionary at the user's will. When the user selects a rarely used word from the candidate list, the user selects another candidate that is frequently used because the word is output as the first candidate during the next document creation operation. There is no need. In addition, for example, by giving the learning non-target attribute to a word whose output is to be limited in advance, such as a discriminatory term, control is performed so that the word is only displayed when the word is intentionally selected. Is also possible.
[0060]
Even when a learning non-target word is selected from a list of a plurality of candidates for a word recognized from the input speech by giving an attribute to be non-learning for each word stored in the word dictionary, Next time, another candidate that is easier to use is preferentially displayed, and a document can be efficiently created without performing an unnecessary selection operation. Furthermore, learning non-target attribute information is given to words stored in the word dictionary in advance by setting a desired word as a learning non-target from a list of candidates as shown in FIG. 5 during document creation. The work to do can be omitted.
[0061]
The method of the present invention described in the embodiment of the present invention (for example, the processing operation shown in the flowchart of FIG. 2) is a program that can be executed by a computer, such as a magnetic disk (flexible disk, hard disk, etc.), optical disk (CD). -ROM, DVD, etc.) and storage media such as semiconductor memory can be distributed.
[0062]
Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of components disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.
[0063]
【The invention's effect】
As described above, according to the present invention, when a document is created by recognizing input speech, an optimum character string is efficiently selected from among a plurality of character string candidates corresponding to words recognized from the input speech. You can choose well.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration example of a document creation apparatus according to an embodiment of the present invention.
FIG. 2 is a flowchart for explaining a processing operation of the document creation apparatus.
FIG. 3 is a diagram showing an example of a word dictionary.
4 is a diagram for specifically explaining the operation of the document creation apparatus in FIG. 1; FIG.
FIG. 5 is a view showing a display example of a candidate list.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Document creation apparatus, 11 ... Document creation part, 12 ... Input part, 13 ... Output part, 112 ... Input control part, 113 ... Input buffer, 114 ... Dictionary maintenance part, 115 ... Candidate selection part, 116 ... Word learning part DESCRIPTION OF SYMBOLS 117 ... Learning table memory | storage part, 118 ... Language analysis part, 119 ... Candidate expansion | deployment part, 120 ... Candidate buffer, 121 ... Analysis rule memory | storage part, 122 ... Word dictionary memory | storage part, 124 ... Output control part, 125 ... Document buffer, 131: Microphone, 132: Voice recognition unit, 133: Editing instruction input unit.

Claims (8)

A document creation device that recognizes input speech and creates a document,
Storage means for storing a plurality of character strings and at least readings corresponding to each of the plurality of character strings;
Means for setting each of the plurality of character strings stored in the storage means as one of a learning target and a learning non-target;
For each of a plurality of words recognized from the input speech, a means for selecting a plurality of candidate character strings having the same or similar reading as the word from a plurality of character strings stored in the storage means;
A character string corresponding to each of a plurality of words recognized from the input speech is selected from the remaining candidate character strings excluding the non-learning character string from the plurality of candidate character strings corresponding to each word. Creating means for creating the document;
A candidate character string selected by the user among the plurality of candidate character strings corresponding to the first word that is one of the plurality of words, and a character string corresponding to the first word in the document Editing means for editing the document by replacing
Of the learning target character strings and the learning non-target character strings selected as character strings corresponding to the plurality of words by the creating means and the editing means, only the learning target character strings, Means for preferentially selecting from a plurality of candidate character strings corresponding to the word;
A document creating apparatus comprising: Display means for displaying a list of the plurality of candidate character strings corresponding to the first word when a character string in the document corresponding to the first word is designated by a user;
Means for setting the candidate character string to be learned among the plurality of candidate character strings displayed in the list as the learning non-target;
The document creation apparatus according to claim 1, further comprising: 3. The document creation apparatus according to claim 2, further comprising means for setting, as the learning target, the learning non-target candidate character string among the plurality of candidate character strings displayed in the list. 2. The document creation according to claim 1, wherein the storage unit stores attribute information representing the learning non-target in association with the learning non-target character string among the plurality of character strings. apparatus. A document creation method for recognizing input speech to create a document,
Each of the plurality of character strings stored in a storage unit that stores a plurality of character strings and at least a reading corresponding to each of the plurality of character strings is set as one of a learning target and a learning non-target. A first step;
For each of a plurality of words recognized from the input speech, a second candidate character string having the same or similar reading as the word is selected from the plurality of character strings stored in the storage means. Steps,
A character string corresponding to each of a plurality of words recognized from the input speech is selected from the remaining candidate character strings excluding the non-learning character string from the plurality of candidate character strings corresponding to each word. A third step of creating the document;
A candidate character string selected by the user among the plurality of candidate character strings corresponding to the first word that is one of the plurality of words, and a character string corresponding to the first word in the document A fourth step to replace
The learning target character string selected from the learning target character string and the non-learning character string selected as character strings corresponding to the plurality of words in the third step and the fourth step, respectively. Only, a fifth step for preferentially selecting from a plurality of candidate character strings corresponding to each word;
A document creation method characterized by comprising: 6. The document according to claim 5, further comprising a sixth step of setting the learning target candidate character string as the learning non-target among the plurality of candidate character strings corresponding to the first word. How to make. 6. The document according to claim 5, further comprising a seventh step of setting the candidate character string not to be learned as the learning target among the plurality of candidate character strings corresponding to the first word. How to make. A program for recognizing input speech to create a document,
On the computer,
Each of the plurality of character strings stored in a storage unit that stores a plurality of character strings and at least a reading corresponding to each of the plurality of character strings is set as one of a learning target and a learning non-target. A first step;
For each of a plurality of words recognized from the input speech, a second candidate character string having the same or similar reading as the word is selected from the plurality of character strings stored in the storage means. Steps,
A character string corresponding to each of a plurality of words recognized from the input speech is selected from the remaining candidate character strings excluding the non-learning character string from the plurality of candidate character strings corresponding to each word. A third step of creating the document;
A candidate character string selected by the user among the plurality of candidate character strings corresponding to the first word that is one of the plurality of words, and a character string corresponding to the first word in the document A fourth step to replace
The learning target character strings respectively selected as character strings corresponding to the plurality of words in the third step and the fourth step are selected from among a plurality of candidate character strings corresponding to the respective words. A fifth step for preferential selection;
A program that executes

Images