JP2006030326A - Speech synthesizer - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To realize readings of words and phrases that correspond to a document by more suitably reflecting correction results of readings, accents, and intonations by saving information regarding the relation of a word and a phrase together when the correct reading of the word changes depending upon whether there is another word or phrase is present nearby or with the field or category of document contents including the word. <P>SOLUTION: Disclosed is a speech synthesizer or speech data generating device characterized in that an editing means of editing an intermediate language of a document to be read aloud receives a correction instruction inputted for the intermediate language displayed by a display means and stores information on the correction instruction in a storage means, and the correction instruction includes at least specification of a word or phrase to be corrected and condition specification of correction reflection. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an audio data creation apparatus. In particular, the present invention relates to an intermediate language editing system that edits speech data called intermediate language that is input data to a speech synthesizer.

  Speech synthesizers that have been used only for reading out information such as bank balance inquiry and stock price information are now also installed in in-vehicle information terminals and mobile phones such as car navigation systems for automobiles. It has become like this. In these information providing devices, a speech synthesizer is used to read out information such as current news and guidance from a service company. In such applications, it is important that the information provider corrects the intermediate language data to be input to the speech synthesizer in advance so that reading errors and intonation unnaturalness are eliminated as much as possible.

  For this purpose, there is a technique for editing intermediate language data so that various reading styles can be specified using devices and tools for editing intermediate language data. As an example, if there is a word with multiple readings in the text to be read, the correct correct reading information is specified in the original document along with a function that allows the editor to specify the correct reading even while reading with the speech synthesizer. There is a technique for storing data in association with the data (see, for example, Patent Document 1).

Japanese Unexamined Patent Publication No. 7-134597

  In the technique described in Patent Document 1 described above, at the time of conversion to a phonetic symbol, a co-occurrence dictionary is used to read a word having a plurality of readings depending on what word is in the vicinity of the word. The function of changing is realized. On top of that, if the output reading is wrong, save the correct reading for the word with multiple readings and convert it to the corrected reading when the same word appears in another document Invented technology.

  However, this technique has a problem in that the word in which the corrected reading is stored is always corrected to the same reading because the correction of the reading cannot be stored under any conditions. Furthermore, unlike the reading of a word, the inflection of a voice is not determined by the word alone, but is known to vary depending on the presence and characteristics of nearby words and phrases. Does not include a mechanism that can properly handle the independence context dependency. In the present invention, when the correct reading of a certain word is close to another word or phrase or differs depending on the field / category of the document content including the word, By storing information related to the relationship as well, it is possible to reflect the correction results of reading, accent, and intonation more appropriately.

  A voice data generation apparatus or a voice synthesis apparatus including a generation unit that generates an intermediate language from a reading target document including a kanji and an editing unit that edits the intermediate language. In particular, the editing means receives a correction instruction input to the intermediate language displayed on the display means, and stores information on the correction instruction in the storage means. The correction instruction is at least a correction target. Including the phrase designation and the condition for reflecting the correction. A user interface in the editing means is also disclosed.

  According to the present invention, intermediate language data adjusted and edited so as to be correct reading / accenting for a certain word / phrase and easy to hear inflection is used as registered phrase data together with conditions for reusing the data. By recording it and inserting it automatically when it can be reused in the document that is currently being edited, it becomes possible to handle differences due to the context that cannot be correctly reflected by the conventional audio data creation apparatus.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a diagram for explaining the basic configuration of an audio data creation apparatus according to the present invention.
As shown in the figure, the basic configuration of the audio data creation device of the present invention is that the document data recording device 0010 in which the kanji-kana mixed text to be read is recorded, and the document data for inputting the document data to be edited therefrom is input. An input unit 0020, an intermediate language generation unit 0030 that gives reading / accent information to each word of the input document data, an intermediate language editing unit 0060 that provides an editing function of the generated intermediate language data, and Intermediate language display device 0050 for displaying the intermediate language data and the state of editing work, speech synthesis device 0070 for converting the generated intermediate language data into synthesized speech for viewing, and audio reproduction for reproducing the converted synthesized speech Device 0080, editing instruction input device 0090 for inputting a user's editing instruction in editing work, and edited intermediate language An audio data generation apparatus having an intermediate language output device 0100 outputs.

  Furthermore, registered phrase specifying means 0110 for creating registered phrase information in which correct reading / accent / intonation information of a certain word / phrase is associated with information of another word / phrase or document content category, by the user's specification, and the registered phrase Registered phrase storage means 0130 for storing the recorded phrase in the recording apparatus, registered phrase recording apparatus 0150 for recording and holding the registered phrase, registered phrase search means 0140 for searching for registered phrase information related to a certain word / phrase from the registered phrase recording apparatus, and search The registered phrase insertion means 0120 for inserting the correct reading, accent, and intonation of the original word / phrase into the intermediate language data based on the registered phrase information, and the language information output from the intermediate language generation means and the registered phrase insertion means Based on intermediate Having a reading accent candidate generating unit 0040 for generating a pattern that can be the individual readings for the word phrase accent intonation candidates in terms data.

  In FIG. 1, there may be an embodiment in which the document data recording device 0010 is placed on a server and the document data input unit 0020 acquires document data via a network. In addition, there may be an embodiment in which the document data recording device 0010, the document data input unit 0020, and the intermediate language generation unit 0030 are placed on a server, and the intermediate language editing unit 0060 acquires intermediate language data via a network. There may also be an embodiment in which the registered phrase recording device 0150 is placed on a server, and the registered phrase storage unit 0130 and the registered phrase search unit 0140 exchange registered phrase information with the registered phrase recording device 0150 via a network.

Next, the basic configuration of the present invention will be described in detail with reference to FIG.
The document data recording apparatus 0010 is an apparatus in which a kanji / kana mixed text for editing reading, accent, intonation, etc. is recorded as an intermediate language, and is a recording medium such as a hard disk or a CDROM. The document data input unit 0020 is a unit for inputting document data from the document data recording device 0010, and is a program such as a hard disk device driver or a CDROM device driver. When the document data recording apparatus 0010 is placed on a server, the document data input unit 0020 is a program such as a network device driver that receives data via a network. The intermediate language generation means 0030 is a device that performs language analysis on document data input as kanji-kana mixed text, and at least performs morphological analysis processing that divides the input text into words and adds reading and accent information. Preferably, and also has a function of performing syntax / semantic analysis processing. Morphological analysis is a process of dividing kanji-kana mixed sentences into individual words. Based on connection data that defines the connectability between parts of speech and cost data that defines the cost of parts of speech, the minimum cost method and the number of phrases This is achieved by using a method such as the minimum method.

  In the syntax analysis, the dependency relationship with respect to the input sentence is output by using the grammatical data defining the part-of-speech relationship structure and the syntax analysis method such as LR parser or CYK analysis. The reading / accent candidate generation means 0040 uses a plurality of readings, accents, and intonations based on the language analysis data output from the intermediate language generation means 0030 and the registered phrase information output from the registered phrase search means 0120. Change candidate list data for each word / phrase having candidates is generated and output to intermediate language editing means. A method for realizing the reading / accent candidate generation means 0040 will be described in detail in the following embodiments. The intermediate language display device 0050 has a function of displaying the intermediate language data output from the intermediate language editing unit 0060 to the user, and can use a character / image display device such as a CRT.

  The intermediate language editing unit 0060 displays the intermediate language data output from the intermediate language generation unit 0030 on the intermediate language display device 0050 using the change candidate list data output from the reading / accent candidate generation unit 0040, and An editing instruction from the user is received from the editing instruction input device 0090, and an intermediate language editing function such as word / phrase reading / accent change in intermediate language data and adjustment of intonation is provided. A method for realizing the intermediate language editing unit 0060 will be described in detail in the following embodiments.

The speech synthesizer 0070 has a function of converting the intermediate language data output from the intermediate language generation unit 0030 and the intermediate language data edited by the intermediate language editing unit 0060 into speech waveform data. This can be realized by using a synthesis method.
The voice reproduction device 0080 is a speaker that reproduces the voice waveform data converted by the voice synthesis device 0070 as actual voice. The edit instruction input device 0090 is a device for a user to transmit an edit instruction to the intermediate language editing unit 0060, and can be realized using an information input device such as a mouse, a keyboard, or a tablet. The intermediate language output device 0100 outputs the intermediate language data adjusted to the correct reading / accent and appropriate intonation by the intermediate language editing unit 0060 to an external recording device, or transmits the intermediate language data to another system via a network. It has a function. It should be noted that when outputting to an external recording device, it can be realized as a program such as a hard disk device driver, and when transmitting via a network, it can be realized as a network transfer program.

Registered phrase designating unit 0110 uses intermediate language editing unit 0060 as a correct reading / accent by the user, intermediate language data obtained as a result of adjustment to appropriate intonation, or some words / phrases of intermediate language data in the vicinity thereof And the associated data is stored in the registered phrase recording device 0150 via the registered phrase storage unit 0130. The intermediate language editing means 0060 reads / accents candidates for words / phrases in the intermediate language data, which is the output result from the intermediate language generation means 0030, when displaying the intermediate language data on the intermediate language display device 0050. The change candidate list data, which is the output result of the generation unit 0040, and the registered phrase information retrieved from the registered phrase recording device 0150 via the registered phrase search unit 0120 are displayed in association with each other. The registered phrase insertion unit 0120 determines at which position in the intermediate language data the searched registered phrase data is to be inserted and displayed. The registered phrase specifying unit 0110, the registered phrase storage unit 0130, the registered phrase searching unit 0140, and the registered phrase inserting unit 0120 are each realized by reading a computer program by the processing device of the apparatus. Will be described.
Based on the basic configuration of the present invention as described above, an example implemented in a more specific form will be described below.

As the first embodiment, a case will be described in which the present invention is implemented as an intermediate language editing system that adopts the basic configuration of FIG. 1 and has a basic GUI editing function.
FIG. 2 is an example of an edit screen layout according to the first embodiment. The document data list 1010 to be edited is displayed on the left side of the screen, the document content 1020 currently being edited is displayed on the upper right side of the screen, and the intermediate language data 1030 of the document currently being edited is displayed on the lower right side of the screen. ing.

  FIG. 3 is a flowchart of processing content processing in each device / means of the intermediate language editing system according to the first embodiment. First, when the intermediate language editing system is activated (step 2010), a plurality of document data to be edited is read from the document data recording device 0010, and the document titles are displayed in a list in the document data list area 1010 of the intermediate language display device 0050. (Step 2020). Next, the intermediate language editing system enters a loop for waiting for designation of the document to be edited from the user (step 2030). When the user selects document data to be edited from the document data list via the editing instruction input device 0090, the intermediate language editing system displays the contents of the editing target document (step 2040). This content display is a process of reading the content of the specified edited document from the document data recording device 0010 and displaying it in the document content area 1020. Next, the contents of the read designated document are input to the intermediate language generation means 0030 and converted into intermediate language data (step 2050). As described above, the intermediate language generation unit 0030 performs conversion into intermediate language data using a language analysis method such as morphological analysis or syntax analysis. The intermediate language generation means can use conventional morphological analysis methods, syntax analysis methods, and the like.

  The result of the morphological analysis performed by the intermediate language generation unit 0030 is shown in FIG. As a result of the morphological analysis processing, the input sentence is divided into words, and reading / accent candidates for each word are searched from the dictionary data. In the example of FIG. 4, for the 18th morpheme “highest value”, a reading / accenting candidate (first candidate) “Psycho'chi” and a second candidate “Saitaka'ne” are searched. It is shown that.

  FIG. 5 shows an example of the analysis result when the syntax analysis processing is performed by the intermediate language generation unit 0030. In the example of FIG. 5, the dependency relationship between clauses is analyzed, for example, the phrase “today” and the clause “stock” are both related to the clause “market conditions”. Further, the intermediate language generation means 0030 co-occurs data called co-occurrence data that indicates what word is nearby when the word is read in the input sentence. In some cases, the data dictionary is searched. An example of co-occurrence data is shown in FIG. In the example of FIG. 6, the morpheme “highest value” is often read and accented as “psycho” when there are words such as “temperature” and “experiment” nearby, and “stock price” is nearby. When there are words such as “stock” and “closing price”, it is often read and accented as “Saitaka'ne”.

The intermediate language generation means 0030 finally combines the analysis results as shown in FIGS. 4 to 6 to generate intermediate language data using the most probable reading / accent. 7 and 8 show examples of intermediate language data. FIG. 7 shows a case where co-occurrence data is not used, and FIG. 8 shows a case where appropriate co-occurrence data exists and the “highest value” reading / accent can be correctly given.
The intermediate language data and the language analysis data thus generated by the intermediate language generation means 0030 are then transferred to the reading / accent candidate generation means 0040 for reading / accent candidate generation processing (step 2060) and the registered phrase. The data is passed to the search means 0140, and a registered phrase search process (step 2070) is performed. Next, in the reading / accent candidate generation process (step 2060), the intermediate language data (FIGS. 7 and 8) output from the intermediate language generation means 0030 and the language analysis data (FIGS. 4, 5, and 6) are used. First, a word having a plurality of readings / accents is extracted, and a list of a plurality of readings / accents for the words is created.

  In the document data example here, as shown in FIG. 9, only the word “highest value” has a plurality of reading / accenting candidates. The example of FIG. 9 is a reading / accent candidate list when the intermediate language generation means 0030 performs the analysis process without using the co-occurrence data FIG. 6, and “Psycho'-chi” is the first candidate, “Saitaka ' “Ne” is the second candidate. If the intermediate language generation means 0030 performs the analysis process using the co-occurrence data in FIG. 6, the order of the first candidate and the second candidate in FIG. 9 is switched.

  On the other hand, the intermediate language data and language analysis data output from the intermediate language generation unit 0030 are input to the registered phrase search unit 0140, and the registered phrase data corresponding to the word / phrase in the sentence is stored in the registered phrase recording device 0150. It is searched whether it exists (step 2070). The registered phrase data in the present invention means data in which the result of adjustment to correct reading / accent and easy to hear inflection in the previous intermediate language editing work is stored together with the word / phrase.

  An example of registered phrase data stored in the registered phrase recording device 0150 is shown in FIG. The meaning of the registered phrase data example in FIG. 10 is that the adjusted reading / accent / intonation data 9040 for the word “phrase 9010” at the closing price is “# 80 Owari'Nede | 1”, and the adjusted data is This indicates that if the related word / phrase “this year” is immediately after the word / phrase “at closing price” (+1), it can be reused. Here, data such as “# 80” and “| 1” are parameters for adjusting ease of listening such as intonation. Since the prosody / inflection adjustment parameters are written in the intermediate language text data as they are, if the adjusted intermediate language data is saved as it is, the adjusted prosody can be saved and used for further processing. it can. In the other data example of FIG. 10, the adjusted data “Saitaka'ne” for “highest price” is within 5 words before and after the word / phrase “stock price” or “close price” as seen from the original word. Indicates that it can be reused.

  In the following description, it is assumed that, as a result of the registered phrase search process (step 2070), two registered phrase data shown in FIG. 10 are searched from the registered phrase recording device 0150. Next, the registered phrase data (FIG. 10) searched in the registered phrase search process (step 2070) is transferred to the registered phrase insertion means 0120, and the registered phrase insertion process (step 2090) is performed. In the registered phrase insertion process (step 2090), a portion of the intermediate language data output from the intermediate language generation process (step 2050) that matches the searched registered phrase data is subjected to the adjusted reading / accent / inflation data. Perform the replacement process with. FIG. 11 shows the result of the registered phrase insertion process performed on the intermediate language data in FIG. 8 and the registered phrase data in FIG.

  Next, the intermediate language data (FIG. 11) output from the registered phrase inserting unit 0120 and the reading / accent candidate list (FIG. 9) output from the reading / accent candidate generating unit 0040 are sent to the intermediate language editing unit 0060. Then, the intermediate language display process (step 2080) of the edited document is performed. In the intermediate language display processing (step 2080) of the edited document, the intermediate language data (FIG. 11) and the reading / accent candidate list (FIG. 9) are easily readable and editable on the intermediate language display device 0050 by the user. Is displayed. An example of the intermediate language data displayed in the intermediate language display process (step 2080) of the edited document is shown in FIG. In this layout example, change candidates are shown in the menu format for each unit called an accent phrase. In this example, two changes are made for each of the two words "Close price" and "Maximum price". Candidates are shown. Further, as a result of the registered phrase insertion processing (step 2090), the adjusted reading / accent / intonation data “# 80 O'Nede | 1” recorded for “at the closing price” is inserted as the first candidate. , “Highest value” corresponds to “Saitaka'ne” that matches the retrieved registered phrase data in the reading / accent candidate list output from the reading / accent candidate generation process (step 2060). This indicates that the second candidate has already been selected.

  Thus, when the intermediate language data is displayed on the intermediate language display device 0050 by the intermediate language display processing (step 2080) of the edited document, the intermediate language editing means 0060 enters a loop (step 2100) that accepts an editing instruction from the user. enter. When an intermediate language editing instruction request from the user is generated via the editing instruction input device 0090, the intermediate language editing means 0060 executes the instructed editing process (step 2110).

  When the execution of the editing process is completed, the intermediate language editing unit 0060 returns to the loop for receiving the editing instruction from the user again (steps 2120 and 2100). If the editing instruction input from the user is a request to end editing of the document data being edited (step 2120), the system performs an intermediate language output process (step 2130) of the editing result, and then The process returns to a loop for waiting for designation of document data to be edited (step 2020). In the intermediate language output process (step 2130) of the editing result, the intermediate language data (such as FIG. 11) finally created as a result of executing the user's editing instruction is output to an external recording medium by the intermediate language output device 0100. Is stored or transmitted to a server device or the like via a network.

  Next, a case where there is an individual editing instruction from the user via the editing instruction input device 0090 will be described. The input method of editing instructions can be broadly divided into a method of specifying a word / phrase to be edited and then specifying what editing operation is to be performed on the target word / phrase. There may be a method of designating a word / phrase to be an operation target after designating an editing operation to be performed. The editing instruction input device 0090 according to the present invention corresponds to any designation method by passing the editing target phrase / phrase and the editing operation contents together to the intermediate language editing means 0060 after the editing instruction designation operation is completed. It is possible. In addition to specifying the target and the operation separately, there may be a case where the target and the operation are specified at the same time using different input devices such as a mouse and a keyboard or a mouse and a voice. The same applies to the designation method. Therefore, the intermediate language editing unit 0060 does not have to be based on either method.

As an example of the editing operation, first, reading / accent change will be described.
In the intermediate language layout displayed as shown in FIG. 12, it is assumed that the user designates “Saitaka'Neo” and an editing instruction is given to change its reading / accent to “Psycho'-Ochio”.

  In this editing instruction input, for example, assuming that the editing instruction input device 0090 is a mouse, an edit target word / phrase is specified by clicking “Saitaka 'Neo” which is currently valid. You can enter it into the system by clicking on '-thio' and specifying editing instructions to change to another reading / accent candidate. In response to this editing instruction, the intermediate language editing unit 0060 changes the reading / accent data corresponding to “highest value” from the current “Saitaka 'Neo” to “Psycho' thio”. This is performed in the process execution step (step 2110). The editing result is also updated on the intermediate language layout of the intermediate language display device 0050 by the intermediate language editing means 0060.

Next, an editing instruction called registration phrase designation will be described.
The registered phrase specifying process is executed when the user gives an editing instruction for specifying a registered phrase to the intermediate language editing unit 0060 via the editing instruction input device 0090. This registered phrase designating process is a process for newly recording and storing registered phrase data as shown in FIG. 10, and the registered phrase data stored here is used in another document data editing operation. Searched by the search means 0140 and reused.

FIG. 13 shows a flowchart of the registered phrase specifying process.
FIG. 13 is a flowchart in the case where the designation of the editing instruction in the editing instruction input device 0090 is the order in which the word / phrase to be edited is designated after the editing operation is designated. In some cases, it may be reversed, and in other cases, it may be specified at the same time. In the following, description will be made assuming that the co-occurrence data shown in FIG. 6 and the registered phrase data shown in FIG. 10 are not stored in advance. In this case, as the intermediate language data for the sentence example displayed in the document content display area 1020 of FIG. 2, the data including the error shown in FIG. 8 is output, and the intermediate language shown in FIG. The data layout will be displayed. After that, the user performs an intermediate language editing process on the intermediate language data in this state, selects the correct reading / accent for “highest value”, and “at the closing price”. Suppose that the intonation easy-to-listen is specified for “” and then the intermediate language data shown in FIG. 13 is edited.

  In this state, first, when a registered phrase specifying operation is instructed by the user via the editing instruction input device 0090, the intermediate language editing device 0060 activates the registered phrase specifying means 0110 (step 12010). Next, the registered phrase specifying unit 0110 executes a registered word / phrase specifying process (step 12020), and receives a specified input of a word / phrase to be reused which is the main body of the registered phrase data. For example, the designation input may be performed by clicking the position of the word / phrase displayed on the intermediate language display device 0050 with the mouse or by moving the cursor displayed in the intermediate language display device 0050 intermediate language data to the word / phrase. For example, an input method such as performing a special key input such as a return key after moving to a position with a keyboard or the like can be employed. Here, it is assumed that the adjusted word “# 80 O'line'de | 1” in FIG. 12 is selected by this registered phrase specifying process (step 12020).

  Next, the registered phrase designating unit 0110 executes a related word / phrase designating process (step 12030), and becomes a context data for the adjustment of the previously input registered word / phrase “# 80 Oline'de | 1”. Accepts specified word / phrase input. This designation input can be similarly designated using a mouse or a keyboard. However, since a plurality of words / phrases are selected and displayed on the intermediate language display device 0050, the color of the selected words / phrases, the size of characters, It is effective to change the display method such as changing the typeface.

  Next, the registered phrase designating unit 0110 executes a related word position adjustment input process (step 12040), and inputs information about the positional relationship of the related word / phrase from the registered word / phrase. For this input, the system automatically counts the number of morphemes between two words / phrases, or presents the value to the user and inputs the adjusted value. You may take the method of making it. Next, the registered phrase designating unit 0110 generates registered phrase data (step 12050) based on the registered word / phrase data, the related word / phrase data, and the related word position data input in steps 12020 to 12040. ). This process converts the three types of information input in the previous steps into a data format that can be stored in the registered phrase storage device 0150. Specifically, the four field values shown in FIG. It is possible to use a structure such as a database record.

  Next, the generated registered phrase data is transferred to the registered phrase storage unit 0130 and recorded / saved in the registered phrase recording device 0150. Registered phrase information newly recorded and stored in this way is automatically reused in editing other document data as described in the first half of the first embodiment. As a result, correct reading / accenting and listening are obtained. Easy inflection can be given more appropriately.

As a second embodiment, an example of an intermediate language editing tool that allows more intuitive designation using a GUI when the registered phrase designation unit 0110 according to the present invention is implemented will be described. Since the processing flow of the second embodiment is the same as that of the first embodiment except for the processing in the registered phrase specifying unit 0110 (the processing shown in the flowchart of FIG. 13), the description thereof is omitted here. .
Hereinafter, in the processing flow of the intermediate language editing tool of FIG. 3, the designation of the registered phrase is selected as the editing instruction, the registered phrase specifying means 0110 is started, and the registered phrase specifying process shown in FIG. explain.

  In the first embodiment, the embodiment in which the designated word / phrase designation (step 12020) and the related word / phrase designation (step 12030) are designated by a keyboard, a mouse, or voice input has been described. It is often difficult to specify what phrases should be specified unless there is a certain level of knowledge about speech, language, and grammar. On the other hand, in the second embodiment, a designation method in a menu format is realized.

  FIG. 15 shows a screen layout example of this embodiment. FIG. 15 shows a registered word / phrase designating process (steps) when the registered phrase designating unit 0110 is activated and the registered phrase designating process is started according to the flowchart of FIG. 13 and the mouse cursor is clicked on the word / phrase to be registered. 12020) is completed, and as a response of the system to that, the related word / phrase and the related word position candidates are displayed in a menu format.

  FIG. 16 shows a flowchart of registered phrase designation processing in the present embodiment. From the flowchart of FIG. 13, the related word / phrase designation process (step 12030) is changed to the related word / phrase candidate list generation process (step 15030), and the related word position adjustment input process (step 12040) is the candidate from the user. The process is changed to a selection waiting process (step 15040). Further, FIG. 17 shows a flowchart for explaining the internal detailed processing of the related word / phrase candidate list generation processing (step 15030) of FIG.

  Hereinafter, the operation of the registered phrase specifying unit 0110 in the present embodiment will be described with reference to the flowcharts of FIGS. 16 and 17. First, the editing instruction input device 0090 sends an editing instruction for specifying a registered phrase from the user to the intermediate language editing unit 0060. Next, the intermediate language editing unit 0060 activates the registered phrase specifying unit 0110, and proceeds to a registered phrase specifying process. Here, the control shifts to the flowchart of FIG. 16, and first, a registered word / phrase designation process (step 15020) is executed.

  In this embodiment, the registered word / phrase designation process (step 15020) is performed by clicking a word to be registered with the mouse cursor or selecting a plurality of words by dragging or the like to determine a phrase to be registered and then selecting the phrase. Click to complete. Information indicating which word or phrase is selected is passed to the next related word / phrase candidate generation process (step 15030). In the related word / phrase candidate generation process (step 15030), the word / phrase deeply related to the word / phrase based on the information about the word / phrase to be registered passed from the registered word / phrase specifying process (step 15020). And processing that creates a list sorted according to the depth of association.

  Next, the process moves to the flowchart of FIG. 17, and the related word / phrase candidate list generation process is started. First, the part-of-speech estimation process (step 16030) of the registered word / phrase passed from the registered word / phrase designation process (step 15020), the extraction of words immediately before and after the registered word / phrase, the part-of-speech estimation, and the presence / absence of punctuation marks The word / part of speech / punctuation mark determination process immediately before and after the determination (step 16040), and the characteristic word extraction process from the document for extracting a characteristic word / phrase in the document including the registered word / phrase (step 16040) 16020) is executed. These three processes may be executed in parallel, or may be executed in order one by one.

  First, in the feature word extraction process (step 16020) from the document, for example, in the case of the example sentence shown in FIG. 2, it is determined that this document relates to economy / stock, and “stock” and “economy” are identified from this document. Extract feature words such as “highest value”. For this feature phrase extraction process, for example, using information such as the appearance frequency (appearance probability) information of words calculated from a large amount of text corpus, or the simultaneous appearance frequency (coincidence probability) in a set of multiple words, Can be realized.

  FIG. 18 shows an example of a word appearance probability dictionary. When the appearance probability of each word included in the document is determined by referring to dictionary data describing the appearance probability in a text corpus for each word as shown in FIG. Can be regarded as having high importance, that is, a characteristic phrase. Next, the registered word / phrase part-of-speech estimation process (step 16030) may be performed using a word dictionary used in morphological analysis, or language information such as part of speech or dependency is obtained from the intermediate language generation unit 0030 together with intermediate language data. You may get it and use it. Next, the word / part of speech / punctuation mark determination processing (step 16040) immediately before and after the position takes out the words immediately before and immediately after the registered word / phrase, estimates the part of speech using the above-described method, and further punctuation marks immediately before and after. Since it is sufficient to check whether or not exists, details are omitted.

  Part-of-speech estimation processing (step 16030) of the registered word / phrase passed from the registered word / phrase designation processing (step 15020), extraction of words immediately before and after the registered word / phrase, estimation of the part-of-speech, presence / absence of punctuation marks, etc. Word / part-of-speech / punctuation determination processing immediately before and after the determination (step 16040), and feature word extraction processing from the document for extracting characteristic words / phrases from the document including the registered word / phrase (step 16020) As a result of the process (), for example, data as shown in FIG. 19 is obtained.

These three processes extract conditions from different viewpoints in order to enumerate related condition types for creating a related word / phrase candidate list. It is expected that the accuracy in the following determination is improved as the number of related condition types increases, and it is desirable that the conditions extracted here are particularly deeply related to differences such as intonation and reading.
Part-of-speech estimation processing of registered words / phrases (step 16030), determination processing of words / parts-of-speech / punctuation marks immediately before and immediately after (step 16040), feature word extraction processing from a document (step 16020) Is an example thereof, and it is not always necessary that these three are used, and these three are not necessarily executed simultaneously.

  Next, the determination result data shown in FIG. 19 is passed to the adjustment case corpus search process (step 16050). In the adjustment case corpus search process (step 16050), the adjustment result of reading, accenting, and inflection on the word / phrase performed so far, and the phrase / phrase determined to be related to the adjustment of the word / phrase. Is an adjustment quality prediction value that is a numerical value indicating how much of each related word / phrase candidate in FIG. 19 is related to the adjustment result to be registered. Is granted.

  For example, when the adjustment example corpus is configured as shown in FIG. 20, the predicted value assignment process is performed by first registering words / parts of words / phrases of words / phrases to be registered and registered words / phrases in the adjustment example corpus. And rearranging the corresponding related condition types based on the matching degree, and then outputting the matching degree as an adjusted quality prediction value. The degree of coincidence can be calculated by adding +10 points if one word matches or +5 points if one part of speech matches, and adding the values or performing some calculation. This adjustment example corpus data may be data obtained by converting and accumulating previously registered phrase information using the intermediate language editing tool of the present invention. It may be registered.

Next, the candidate list (FIG. 19) to which the adjustment quality predicted value is assigned in the adjustment example corpus search process (step 16050) is arranged in descending order of the predicted value in the rearrangement process using the adjustment quality predicted value (step 16060). Be replaced. The result is shown in FIG.
Next, this result is passed to the candidate selection waiting process (step 15040) from the user in the flowchart of FIG. 16, and is displayed in the form of a related word / position candidate menu for the registered word / phrase “# 80 Owari'Nede | 1”. It is converted and displayed on the intermediate language display device 0050. When the user selects the one related phrase / position candidate from the displayed menu, the registered phrase data generation process (step 15050) is executed and the registered phrase storage process (step 15060) is executed and designated. The registered words / phrases are stored together with appropriate related words / positions. Since the registered phrase data generation process (step 15050) and the registered phrase storage process (step 15060) are the same as those described in the first embodiment, a description thereof will be omitted.

  According to the second embodiment, even a user who lacks expertise such as speech, language, and grammar may select from related word candidates presented by the system. The adjustment result can be easily stored and reused as registered phrase information together with appropriate related phrases. Further, as another screen layout example according to the second embodiment, as shown in FIG. 22, the condition relating to the immediately preceding phrase is on the left side, the condition relating to the immediately following phrase is on the right side, the condition relating to the document feature word is on the upper side, etc. It is also conceivable to display a candidate menu corresponding to a position associated with words such as “immediately before”, “immediately after”, and “document”.

1 is a diagram illustrating a basic configuration of the present invention and a system configuration of Embodiment 1. FIG. The figure which shows the example of a screen layout of the intermediate language editing system of this invention. The figure which shows the flowchart of the process in Example 1 of this invention. FIG. 3 is a diagram illustrating an example of morpheme analysis data that is internally analyzed by an intermediate language generation unit in the first embodiment of the present invention. FIG. 5 is a diagram illustrating an example of syntax analysis data internally analyzed by an intermediate language generation unit in the first embodiment of the present invention. FIG. 5 is a diagram illustrating an example of co-occurrence analysis data internally analyzed by an intermediate language generation unit in the first embodiment of the present invention. FIG. 5 is a diagram illustrating an example of intermediate language data output from the intermediate language generation unit according to the first embodiment of the present invention. FIG. 5 is a diagram illustrating an example of intermediate language data output from the intermediate language generation unit according to the first embodiment of the present invention. The figure which shows an example of the reading / accent candidate list data output from a reading / accent candidate production | generation means in Example 1 of this invention. The figure which shows an example of the registration phrase data searched by the registration phrase search means from the registration phrase recording device in Example 1 of this invention. The figure which shows an example of intermediate language data after applying the registration phrase data output from a registration phrase insertion means in Example 1 of this invention. FIG. 6 is a diagram illustrating an example of an intermediate language data layout displayed on the intermediate language display device by the intermediate language display process of the edited document in the first embodiment of the present invention. The figure which shows the flowchart of the designation | designated process of a registration phrase in Example 1 of this invention. FIG. 6 is a diagram illustrating an example of an intermediate language data layout displayed on an intermediate language display device by intermediate language display processing of an edited document when registered phrase data does not exist in Embodiment 1 of the present invention. The figure which shows an example of the screen layout in the middle of performing the registration phrase designation | designated process in Example 2 of this invention. The figure which shows the flowchart of the designation | designated process of a registration phrase in Example 2 of this invention. The figure which shows the flowchart of a related word and phrase candidate list production | generation process in Example 2 of this invention. The figure which shows the structure of the word appearance probability dictionary used for the characteristic word extraction process from the document in Example 2 of this invention. In Example 2 of this invention, The figure which shows an example of the related phrase / phrase candidate list produced | generated in the middle of a related word / phrase candidate list production | generation process. The figure which shows the structure of the adjustment example corpus in Example 2 of this invention. In Example 2 of this invention, the figure which shows an example of the related phrase / phrase candidate list of the result rearranged by the adjustment quality prediction value. FIG. 10 is a diagram illustrating an example of another screen layout according to the second embodiment of the present invention.

Explanation of symbols

0010 ... Document data recording device, 0020 ... Document data input means, 0030 ... Intermediate language generation means, 0040 ... Reading / accent candidate generation means, 0050 ... Intermediate language display device, 0060 ... Intermediate language editing means, 0070 ... Speech synthesizer, 0080 ... Audio playback device, 0090 ... Editing instruction input device, 0100 ... Intermediate language output device, 0110 ... Registered phrase designation means, 0120 ... Registered phrase insertion means, 0130 ... Registered phrase storage means, 0140 ... Registered phrase search means, 0150 ... Registered phrase recording device, 1010 ... Document data list display area, 1020 ... Document content display area, 1030 ... Intermediate language data display area, 2010 ... Start processing step, 2020 ... Document data reading / display processing step, 2030 ... Designation of edited document Yes judgment step, 2 40 ... Edited document content display processing step, 2050 ... Intermediate language generation processing step, 2060 ... Reading / accent candidate generation processing step, 2070 ... Registered phrase search processing step, 2080 ... Edited document intermediate language display processing step, 2090 ... Registration Phrase insertion processing step, 2100 ... Determination instruction input determination step, 2110 ... Instructed editing processing execution processing step, 2120 ... Editing end determination step, 2130 ... Editing result intermediate language output processing step, 3010 ... Morphological number, 3020 ... Morphological notation, 3030 ... Part of speech, 3040 ... Reading / accent (first candidate), 3050 ... Reading / accenting (second candidate), 5010 ... Morphological, 5020 ... Reading / accent, 5030 ... Co-occurrence word list, 8010 ... Morpheme, 8020 ... reading Cent (first candidate), 8030 ... reading / accent (second candidate), 9010 ... word / phrase, 9020 ... related word / phrase, 9030 ... position of related word, 9040 ... adjusted reading / accent / inflection, 12010 ... Registration phrase designation start step, 12020 ... Registered word / phrase designation processing step, 12030 ... Related word / phrase designation processing step, 12040 ... Related word position adjustment input processing step, 12050 ... Registered phrase data generation processing step, 12060 ... Registered phrase Storage processing step, 12070 ... Registered phrase designation end step, 15010 ... Registered phrase designation start step, 15020 ... Registered word / phrase designation processing step, 15030 ... Related word / phrase candidate list generation processing step, 15040 ... User? Candidate selection waiting processing step, 15050 ... registered phrase data generation processing step, 15060 ... registration phrase storage processing step, 15070 ... registration phrase designation end step, 16010 ... related word / phrase candidate list generation processing start step, 16020 ... in document Feature word extraction processing step from 16030 ... Registered word / phrase part-of-speech estimation processing step, 16040 ... Word / part-of-speech / punctuation determination step immediately before / immediately after position, 16050 ... Adjustment example corpus search processing step, 16060 ... Adjustment quality Sorting process step based on predicted values, 16070 ... related word / phrase candidate list generation process ending step.

Claims (5)

A means for inputting a document to be read aloud including kanji,
Generating means for generating an intermediate language from the reading target document;
Editing means for editing the intermediate language, speech synthesis means for performing speech synthesis based on the edited intermediate language,
An output means for the synthesized voice,
The editing means receives a correction instruction input to the intermediate language displayed on the display means, stores information on the correction instruction in the storage means,
The speech synthesizer according to claim 1, wherein the correction instruction includes at least specification of a word to be corrected and specification of conditions for reflecting the correction. 2. The speech synthesizer according to claim 1, wherein the modification reflection condition designation includes information on a co-occurrence phrase of the modification target phrase. The speech synthesizer according to claim 1 or 2, wherein the correction reflection condition designation further includes information on a position condition of the co-occurrence word / phrase. 4. The voice according to claim 1, wherein the editing unit presents a plurality of candidate conditions together with the intermediate language on the display unit, and accepts a selection instruction from the input unit. Synthesizer. The editing means estimates the part of speech of the input correction target phrase, estimates the part of speech of the word before and after the correction target phrase, extracts characteristic words in the target document, and extracts these results in the past 5. The speech synthesizer according to claim 1, wherein a correction condition determined by comparison with the correction record is displayed on the display means.

Images