JP2011170191A - Speech synthesis device, speech synthesis method and speech synthesis program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To easily and simply correcting a text for speech synthesis at a level of an input text, which is easily understood by an end user. <P>SOLUTION: The device is equipped with: a text editing section for editing a text which is an object of speech synthesis; a text editing determination section to determine whether or not text editing is performed in order to correct reading or accent of synthesis speech; an editing history data creating section of creating an editing history data to indicate an editing content of the text editing, when it is determined that text editing is performed in order to correct reading or accent of synthesis speech; and an editing history data storage section for storing the editing history data, when the created history data are registered. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to speech synthesis for generating synthesized speech from input text.

  Speech synthesizers and speech synthesizers synthesize speech from text input that is usually mixed with kanji and kana. At that time, the language processing unit morphologically analyzes the input text while referring to the word dictionary, and specifies words (or morphemes) constituting the input text. Further, the reading and accent of the word specified by the language processing unit can be obtained from the word dictionary. Japanese accents are "high and low" arrangements assigned to each Japanese sound, and are essential for ensuring intelligibility and naturalness of speech. The phonetic character strings for the input text are generated by connecting the readings and accents of each morpheme and appropriately performing accent transformation processing and processing for setting boundaries such as phrases and poses. The phonetic character string is a character string representing how to read text, and generally includes a katakana representing a reading, a symbol representing an accent, and a symbol representing a boundary between phrases and poses.

  Arbitrary Japanese texts are so diverse that the current language processor cannot analyze them with 100% accuracy. When reading or accenting errors occur in the language processing unit, it is common to manually correct the phonetic character string that is output from the language processing unit. As mentioned above, in the phonogram string, readings are often written in katakana, which is similar to the phonetics that we usually use. However, the long sound is indicated by “-” (for example, “Tokyo” is indicated as “Tokyo”), the particles “ha”, “to”, “to” are indicated by “wa”, “o”, “e”, and accents. In order to correct a phonetic character string, such as by showing an accent symbol at a certain place, it is necessary to fully understand and become familiar with the specification, and it cannot be easily corrected by general users.

  For the end user, the correction of the phonetic character string as described above is a high hurdle. On the other hand, it is possible to correct the input text so that the reading and accent in the phonetic character string are correct. Since the input text is a kana-kana mixed text that we usually use, the correction is easier to understand than the correction of the phonetic string, and it is considered less burdensome for the end user. However, it may be difficult to intuitively understand how to correct the synthesized speech by correcting the input text. Further, even after acquiring know-how for correcting the input text, it is troublesome and complicated to perform the same correction included in various texts every time the input text changes.

  For example, when editing the prosody of a synthesized speech (pitch pattern, phoneme length, power, etc.), the previous prosody modification information is retained, and the prosody modification information retained in the subsequent generation of the prosody information is retained. And a method of generating synthesized speech using the method is disclosed (for example, see Patent Document 1 and Patent Document 2).

JP 2004-309724 A JP 2005-345699 A

  However, editing the prosody of a synthesized speech is difficult, and is inappropriate for end users who want to make minimal corrections easily.

  Therefore, an object of the present invention is to make it possible to easily modify text for speech synthesis at the level of input text that is easily understood by the end user.

In order to achieve the above object, a speech synthesizer disclosed below includes a text editing unit that edits text to be synthesized, morphological analysis, and phonogram generation for the text before and after editing. To correct the reading of the synthesized speech or the accent based on at least the phonetic character string out of the morphological analysis result for each of the text before and after editing and the generated phonetic character string A text editing determination unit for determining whether or not the text editing is performed, and the text editing determination unit when the text editing determination unit determines that the text editing is performed to read a synthesized speech or correct an accent. There is an edit history data creation unit that creates edit history data indicating the edit contents of the edit, and an edit history data created by the edit history data creation unit. When the, and a editing history data storage unit for storing the edit history data.

Also, there is an error in the phonetic character string of the edited text edited by the text editing unit or the output unit that outputs the synthesized speech, and the reading of the phonetic character string or synthesized speech output by the output unit, or the accent. In this case, for the edited text, an error indication receiving unit that receives a reading of synthesized speech and an indication of an accent error from a user, and a reading of the synthesized speech for the edited text received by the error indication receiving unit, an accent error Edit history data search unit for retrieving edit history data indicating the edit contents of text editing to be performed in order to correct the reading of the synthesized speech or the accent for the indicated part of the edit history data storage unit And the text editing unit uses the editing history data searched and acquired by the editing history data searching unit. To re-edit the strike.

  According to the above configuration, it is possible to easily modify the text for speech synthesis at the level of the input text that is easily understood by the end user.

1 is a block diagram showing an example of the overall configuration of a speech synthesizer according to Embodiment 1 of the present invention. The block diagram which shows an example of a structure of the speech synthesizer which concerns on Embodiment 1 of this invention. The block diagram which shows an example of a structure of the speech synthesizer which concerns on Embodiment 1 of this invention. The figure which shows an example of edit history data registration determination The figure which shows an example of edit history data registration determination The figure which shows an example of edit history data registration determination The figure which shows an example of edit history data registration determination Diagram showing an example of editing history data registration The figure which shows an example of search application of edit history data The figure which shows an example of search application of edit history data The figure which shows an example of search application of edit history data The figure which shows an example of search application of edit history data The flowchart which shows an example of operation | movement of the speech synthesizer which concerns on Embodiment 1 of this invention. The flowchart which shows an example of operation | movement of the speech synthesizer which concerns on Embodiment 1 of this invention. The flowchart which shows an example of operation | movement of the speech synthesizer which concerns on Embodiment 1 of this invention. The block diagram which shows an example of the whole structure of the speech synthesizer which concerns on Embodiment 2 of this invention.

[Embodiment 1]
The first embodiment shows an example in which speech synthesis software in which a user installs and uses a package product on his / her computer is applied to a speech synthesizer. Hereinafter, the speech synthesis apparatus 100 according to Embodiment 1 of the present invention will be described.

  FIG. 1 is an example of a block diagram showing an overall configuration of a speech synthesis apparatus 100 according to Embodiment 1 of the present invention. In FIG. 1, a speech synthesizer 100 includes a control unit 110, an editing unit 111, a language processing unit 120, a synthesized speech generation unit 130, a synthesized data management unit 140, an editing history registration / updating unit 150, an editing history search unit 160, and an editing unit. A history storage unit 170 is provided. The speech synthesizer 100 is connected to the display device 190, the input device 200, and the speech output device 210 via the interface unit 180. The input device 200 is an input device such as a mouse or a keyboard, for example.

  The control unit 110 is a module that controls the operation of the speech synthesizer 100. The interface unit 180 is a module that controls the input device 190 that inputs text and operations for the text, the display device 200 that outputs a system response to the input, and the audio output device 210 that outputs synthesized speech.

  The editing unit 111 is a module including a text editing unit 201, an error indication receiving unit 202, a text correction unit 203, and an output unit 204. The text editing unit 201 receives an input of a text to be subjected to speech synthesis from the user. The error indication receiving unit 202 receives an error location of the synthesized speech from the user. The text correction unit 203 corrects text to be subjected to speech synthesis.

  The language processing unit 120 is a module that performs morphological analysis of text and generates a phonetic character string. The language processing unit 120 is a module including a morphological analysis unit 121 and a phonetic character string generation unit 122.

  The synthesized speech generation unit 130 is a module that generates synthesized speech from a phonetic character string. Text being edited, text editing history, language analysis results, and phonetic character string results are temporarily stored and managed in the composite data management unit 140.

  The editing history registration / update unit 150 determines the text editing performed to correct the reading of the synthesized speech and the accent error based on the editing status of the input text, the morphological analysis result, and the phonetic character string. This is a module for registering editing history data in the history storage unit 170. The editing history storage unit 170 includes an editing history data storage unit 171. The editing history registration update unit 150 is a module including an editing history data determination unit 151, an editing history data creation unit 152, and an editing history data registration unit 153.

  The edit history search unit 160 is a module that searches the edit history data storage unit 171 for edit history data corresponding to a new input text when the user points out an inappropriate portion of the synthesized speech. is there. The edit history search unit 160 is a module that includes an edit history data search unit 161.

  FIG. 2 shows a first block diagram showing the configuration of the speech synthesizer according to the first embodiment of the present invention. This block diagram shows a configuration for determining whether or not the text editing is performed for correcting the reading accent error of the synthesized speech from the pre-edit text and the post-edit text, and storing it as an edit history. Show.

  The text editing unit 201 receives an input of a text to be subjected to speech synthesis from the user via the input device 200 and the interface unit 180. The input text is displayed on the display device 190, for example, and can be edited by the user. The text editing unit 201 can also accept text editing instructions from the user.

  The morpheme analysis unit 121 performs morphological analysis of each text on the text before editing and the text after editing by the text editing unit 201, and obtains the analysis result of the text before editing and the analysis result of the text after editing. The phonetic character string generation unit 122 obtains a phonetic character string of the text before editing and a phonetic character string of the text after editing.

  The text is divided into a plurality of morphemes (for example, words) by morphological analysis. Reading, part of speech, and accent information are generated for each divided word. Examples of the morphological analysis method include the Viterbi algorithm and the longest match method, but the morphological analysis method used in the present invention is not limited to a specific one. The phonetic character string is intermediate notation data for speech synthesis that represents how to read text. The format of the phonetic character string in this embodiment is merely an example, and the way of expressing the phonetic character string is not limited to this.

  The editing history data determination unit 151 determines that the phonetic character string of the text has changed before and after editing, and compares the text before editing with the text after editing, or compares the text before editing with the text after editing. By examining the result of the morphological analysis with the text, it is possible to determine whether or not the text editing is performed in order to read the synthesized speech and correct the accent error.

  If the edit history data determination unit 151 determines that the text edit is a text edit that has been read and corrected to correct an accent error, the edit history data creation unit 152 registers the text edit as an edit history. Create edit history data. For example, the edit history data creation unit 152 generates edit history data indicating the edit contents of the portion where the phonetic character string is changed in the text.

  The edit history data registration unit 153 registers the edit history data created by the edit history data creation unit 152 in the edit history data storage unit 171.

  FIG. 3 shows a second block diagram showing the configuration of the speech synthesizer according to the first embodiment of the present invention. This block diagram shows a configuration in which the text editing unit 201 points out an inappropriate reading or accent in the input text, searches the editing history database for the editing history corresponding to the location, and applies it. Yes.

  When the user inputs text from the text editing unit 201, the morphological analysis unit 121 performs morphological analysis of the input text and obtains an analysis result. Note that each word resulting from the morphological analysis may be displayed with an underline. Thereby, the user can distinguish easily each word of the morphological analysis result of a text.

  Furthermore, the phonetic character string generation unit 122 generates a phonetic character string of the input text. A phonetic character string is output from the output unit 204, and an indication of an inappropriate part in the output input text is received from the error indication receiving unit 202. The output unit 204 may output synthesized speech synthesized from the phonetic character string. Thus, the user can accurately determine whether or not the synthesized speech is appropriate.

  Next, the edit history data search unit 161 searches the edit history data storage unit 171 based on the input text, the analysis result of the input text, and the error indication part.

  Next, if there is corresponding editing history data as a result of the search by the editing history data search unit 161, the text correction unit 203 corrects the text, and the corrected text is returned to the text editing unit 201.

  A specific example of a text editing determination method for correcting synthesized speech reading and accent errors will be described below. Reading of synthesized speech and text editing for correcting accent errors include (1) insertion or deletion of a sending kana, (2) substitution between hiragana and kanji, (3) giving a kana, (4) A punctuation mark or blank insertion is considered. In the editing history data determination unit 151, it is determined whether or not the text editing is a predetermined editing for reading the synthesized speech and correcting the accent error, and the editing history data creating unit 152 creates data to be registered as the editing history Examples of this are shown in FIGS.

  FIG. 4 is a diagram showing an example of (1) insertion or deletion of a sending pseudonym. When the input text is “This is Okonomiyaki”, when the user hears the synthesized voice, it is understood that “okonomiyaki” is misread as “Okonomisho”. In addition, if the input text is changed to “This is Okonomiyaki” (inserting the kana “ki”), it will be read correctly, so the user adopts this.

  At this time, the phonetic character string of “This is Okonomiyaki” is mistaken as “Koreha Okonomi 'Shodes.”, But the phonetic character string of “This is Okonomiyaki.” Is “Koreha Okonomiyakide's. Is correct (this phonetic character string is not necessarily displayed to the user). This is because “okonomiyaki” is registered in the word dictionary used for morphological analysis, but “okonomiyaki” is not registered. The reason is that a morphological analysis error occurred.

  In this example, the part corresponding to the common noun “Okonomi” and the unknown word “Yaki”, which is the analysis result of the text before editing, is the common noun “Okonomiyaki”, which is the analysis result after editing. The previous phonetic character string is “Koreha Okonomi's Shodes.”, But the edited phonetic character string is changed to “Koreha Okonomiyakide's.”. Then, when comparing the text before and after editing, the hiragana “ki” is inserted immediately after the kanji “yaki”, so that it can be determined that it is the insertion or deletion of the sending kana in (1). Further, as a result of the editing, the phonetic character string has changed and the synthesized speech has been adopted by the user, so that it can be determined that the synthesized speech has been read and the text has been changed to correct the accent error.

That is, whether or not these three conditions are satisfied is determined on condition that the unknown word becomes a common noun, the phonetic character string is changed, and the hiragana is inserted after the kanji. By doing so, it is determined whether or not the reading or accent error is edited. At this time, “yaki” is analyzed as an unknown word in the morphological analysis result of the pre-edit text. From the above, edit history data is
Index: “Yaki”
Applicable conditions: When "Yaki" is analyzed as an unknown word, the type of correction: Inserted insertion character string: "ki"
It becomes that. Note that, as in this example, a portion where the phonetic character string has changed and a portion that was an unknown word before editing can be used as an index. The index is a key when searching for editing history data. When registering the editing history, the application count “1” is set to indicate that the correction method has been used once.

  FIG. 5 is a diagram showing an example of replacement between (2) hiragana and kanji. When the input text is “I'm out of time now”, the user hears the synthesized speech and notices that the accent of “Now” is wrong. In addition, if the input text is changed to "Now, it's overtime", the accent will be correct and the user will adopt it.

  At this time, the phonetic character string of “Now,” is “Tadaima,…” and the accent is wrong, because “Now” has been analyzed as a moving verb (ie, “Now” in greetings). The phonetic character string “Now it's out of time.” Is correct, “Tada’Ima…”.

  This is because in the word dictionary used for morphological analysis, “Tadaima” is registered as an impression verb, and “Ima” is registered as a common noun. Accents are registered as type 0 (Tadaima) and type 2 (Tada'ima), respectively. ing. When the user modifies the text to change “Ima” to “Ima”, “Ima” and “Ima” are easily read by referring to the word dictionary. .

  Although the impression verb “Tadaima”, which is the analysis result of the text before editing, is the common noun “Ima”, which is the analysis result after editing, and the phonetic string before editing is “Tadaima,…” The phonetic character string after editing changes to “tada'ima, ...”. When the text before and after editing is compared, the hiragana “now” becomes the kanji “Tama”, so it can be determined that it corresponds to the replacement between the hiragana and kanji in (2). Further, as a result of the editing, the phonetic character string has changed and the synthesized speech has been adopted by the user, so that it can be determined that the synthesized speech has been read and the text has been changed to correct the accent error.

In other words, if the phonetic character string has changed and the hiragana character has been replaced with a kanji character, it is determined whether these two conditions are satisfied. It is judged whether or not. From the above, edit history data is
Index: “Now”
Applicable conditions: When "Imaima" is analyzed as a moving verb Type of correction: Hiragana and Kanji characters are replaced.
That's it. When registering the editing history, the application count “1” is set to indicate that the correction method has been used once.

  FIG. 6 is a diagram showing an example of (3) adding a pseudonym. The input text is “This is Okonomiyaki” as in FIG. 4, and “Okonomiyaki” is misread as “Okonomisho”. Here, in the input text, if the name “yaki” is added to the “yaki” part, the reading becomes correct, and the user adopts this.

  The phonetic string of “This is Okonomiyaki” is “Koreha Okonomi 'Shodes.”, But if the “yaki” kana is added to the “yaki” part of the input text, the phonetic string generator Can refer to this and generate the correct reading "Koreha Okonomiyakide '."

  The phonetic character string before editing is “Koreha Okonomi 'Shodes.”, But the phonetic character string after editing is changed to “Koreha Okonomiyakide'.”. Then, when comparing the text before and after editing, since the name “Yaki” is given to “Yaki” of “okonomiyaki”, it can be determined that (3) the name is given. Further, as a result of the editing, the phonetic character string has changed and the synthesized speech has been adopted by the user, so that it can be determined that the synthesized speech has been read and the text has been changed to correct the accent error.

That is, by determining whether or not these two conditions are satisfied, on the condition that the phonetic character string has changed and that the place where the phonetic character string has changed has been assigned a kana. It is determined whether reading or accent error editing is performed. At this time, editing history data is
Index: “Yaki”
Applicable conditions: When “Yaki” is analyzed as an unknown word, the type of correction: Giving a pseudonym Sending pseudonym: “Yaki”
That's it. When registering the editing history, the application count “1” is set to indicate that the correction method has been used once.

  FIG. 7 is a diagram showing an example of (4) insertion of a punctuation mark or a blank. When the input text is "Current company ...", "Current company" is read in a row and is unnatural. At this time, by inserting a reading point after “current”, it becomes natural reading, and the user adopts this.

The phonetic string of “Currently in the company…” is “Genzaiki's Gioni Oker…”, and “Current” and “Company” are combined into a single accent phrase, indicating that it is inappropriate. . Here, an accent phrase is a unit of a word that forms a group of accents in Japanese, and a combination of a plurality of words having individual accents to form a single accent phrase is referred to as accent combination.
This is a result of determining that “current company” is a compound word such as “current time” or “current location”. On the other hand, if the input text is “currently in a company…”, the phonetic character string becomes “Genzai, Kiyogoni Okelle…” and is divided into two accent phrases.

  The phonetic character string before the editing is “Genzaiki's Gyoni Oker ...”, but the phonetic character string after the editing is changed as “Genzai, Ki'Gyoni Oker ...”. Then, when comparing the text before and after editing, since the punctuation mark “,” is inserted immediately after “current”, it can be determined that (4) the punctuation mark or blank is inserted. Further, as a result of the editing, the phonetic character string has changed and the synthesized speech has been adopted by the user, so that it can be determined that the synthesized speech has been read and the text has been changed to correct the accent error.

That is, on the condition that a punctuation mark is inserted and the phonetic character string of the morpheme before the punctuation is changed, it is determined whether or not these two conditions are satisfied. It is determined whether or not the error has been edited. At this time, editing history data is
Index: “Current”
Applicable condition: When "present" is parsed as a common noun, type of correction: insert punctuation or white space, insert string: ","
That's it. When registering the editing history, the application count “1” is set to indicate that the correction method has been used once.

  The editing history data determination unit 151 determines whether or not it is a synthesized speech reading or text editing performed to correct an accent, insertion or deletion of a sending kana, conversion from kana to kanji, kanji to kana This can be done by determining whether it is conversion to, adding a kana, or inserting or deleting punctuation or white space.

  FIG. 8 shows an example of editing history data registration. According to the example of the editing history data registration determination shown in FIGS. 4 to 7, an editing history database as shown in this figure is constructed. That is, in the case of the index “yaki”, the part of speech “unknown word”, the type of correction “insert feed kana”, the insertion character string “ki”, and the application count “1” are registered. In the case of the index “yaki”, the part of speech “unknown word”, the correction type “giving furigana”, the kana “yaki”, and the application count “1” are registered. In the case of the index “Tadaima”, the part of speech “impression verb”, the type of correction “replacement between hiragana and kanji”, the changed character string “Imaima”, and the number of applications “1” are registered. In the case of the index “current”, the part of speech “noun”, the type of correction “insertion of punctuation or white space”, the insertion character string “,”, and the application count “1” are registered.

  FIG. 9A to FIG. 9D are diagrams illustrating an example in which the text is corrected by searching the editing history database for the composite text. Hereinafter, a description will be given with reference to FIGS. 9A to 9D.

  When the user listens to the synthesized voice of the input text “Takoyaki is now,” the first reading of “Takoyaki” is “Takosho”, which indicates that it is misread.

  When the user points out the “yaki” part as an inappropriate part, the editing history database is searched for editing history data that uses the “yaki” character as a search key and adds the name “yaki” to “yaki”. Come on. This editing history may be adopted as it is, but when the user searches for another correction method, editing history data to which the pseudonym “ki” is given after “baked” is searched. If the user adopts this editing history, the text will be “Now we have baked octopus”, and it will be confirmed that the reading of this part is correct by listening to the synthesized speech.

  If the user moves on to correction of other parts or determines that the whole sentence is correct, the edit that inserts “ki” after “fire” adopted earlier is confirmed, and the number of times the edit history data is applied is added. Updated.

  Each editing history data stored in the editing history data storage unit 171 has a function of counting the number of times of application applied to re-editing. The editing history data search unit 161 stores editing history data having a high number of applications. The search is preferentially performed, and the edit history data storage unit 171 can update the application count when the edit history data searched by the edit history data search unit 161 is applied to re-editing.

  By storing the number of times of application in this way, it can be applied when searching for editing history data when there are multiple correction methods, such as inserting a kana and assigning a kana, such as “Yaki”. A search with a high frequency can be prioritized and the search can be made more efficient.

  In addition, when the user points out an inappropriate part of the accent of the “currently” part, edit history data that replaces “currently” with “currently” using the “currently” character string as a search key in the edit history database. Is searched. By adopting this, the text “Tako-yaki is now ready” will be generated and it will be the correct accent. If it is determined that the user is correct, the editing that converts the “currently adopted” to “immediately” is confirmed, and the number of times of application of the editing history data is added and updated.

  In the present embodiment, the editing history registered and searched in the editing history database is by the user himself, and the editing method performed by himself in the past rather than referring to the database when the editing method of the text is unknown. Is applied to new texts that require similar editing, thereby improving the efficiency of editing.

  A voice output unit that outputs a synthesized voice of the edited text edited by the text editing unit, and a reading of the synthesized voice output by the voice output unit, or a reading of the synthesized voice with respect to the edited text when there is an error in the accent, An error indication receiving unit that receives an indication of an accent error from the user, and a reading of the synthesized speech with respect to the edited text received by the error indication receiving unit, and reading of the synthesized speech for an accent error indication point or correcting the accent The edit history data search unit retrieves the edit history data indicating the edit contents of the text editing to be performed from the edit history data storage unit. The edit text can be re-edited using the edit history data acquired in this way.

  This makes it possible for even unskilled users to easily modify synthesized speech by editing text, and even users familiar with speech synthesizers can edit text simply by pointing out where they want to be corrected. Efficient operation is possible.

  In the example shown below, the process determines the predetermined text editing performed to correct the reading accent error in the synthesized speech, stores it as an editing history, and reads the synthesized speech and searches the editing history where there is an accent error. In this case, the processing to be corrected in this case functions in one embodiment. 10A to 10C are flowcharts showing the flow of processing in the first embodiment of the present invention. Hereinafter, it demonstrates according to the flowchart of FIG. 10A-FIG. 10C.

  First, the composite text editing area is initialized (step S1001), and user text editing is started. The user's text editing is monitored by the system at an appropriate timing. The editing history data determination unit 151 determines whether or not the user has edited the text (step S1002).

  Next, the editing history data determination unit 151 may perform the text editing performed by the user in order to correct the predetermined text editing described in (1) to (4), that is, the reading accent error of the synthesized speech. It is determined whether or not the text editing is performed (step S1003).

  In step S1003, if the text editing is likely to be performed to correct the reading accent error in the synthesized speech, the editing history data is subjected to language processing to generate a phonetic character string, and the phonetic character string. If there is a change, the editing history data creation unit 152 creates editing history data of the text editing content, and the editing history data registration unit 153 temporarily stores the editing history data (step S1004 → S1005 → S1006).

  On the other hand, the edit history data search unit 161 determines whether an indication of an inappropriate portion of the synthesized speech has been received on the text from the user (step S1007).

  In step S1007, when an inappropriate part of the synthesized speech is received from the user as an indication on the text, the editing history data search unit 161 performs the language processing of the input text (step S1008), It is searched whether there is any edit history data corresponding to the indicated location in the data storage unit 171. Note that an inappropriate part may be pointed out by pointing out an inappropriate word that is underlined. At this time, the edit history data search unit 161 determines whether there is change history data to be applied to the indicated location in the edit history data storage unit 171 (step S1009).

  If there is no editing history data to be applied in step S1009, the editing history data search unit 161 displays that fact and returns to text editing (step S1010). On the other hand, if there is editing history data to be applied in step S1009, the editing history data search unit 161 asks the user whether to apply the editing history data, and searches until the editing history data to be applied is determined. Repeat (steps S1011 to S1013). If the application history is determined, the editing history data search unit 161 temporarily stores the editing history data (step S1014).

  Further, when the user instructs execution of speech synthesis in order to confirm the synthesized speech, language processing, generation of the synthesized speech, and output of the synthesized speech are performed to prompt the user to confirm (steps S1015 to S1018). .

  When the user moves to editing other than the part corrected immediately before or when it is determined that there is no problem with the entire composition, the editing history data registration unit 153 can determine that the correction of the corresponding part is correct. It is determined that the text correction and history data application (temporarily stored) performed are appropriate, and registration and update in the editing history database are performed (steps S1019 → S1020).

  That is, the edit history data registration unit 153 newly registers in the edit history database if it is a text edit (step S1021). Further, the editing history data registration unit 153 adds “1” to the number of times of application and updates the existing history data (step S1022). In this way, this processing is repeated until the user is correct over the entire synthesized speech (step S1023).

As described above, according to the present invention, when the user himself / herself accumulates a predetermined editing history of input text that has been performed in the past and performs speech synthesis of new input text, the editing history is searched. By applying, it is possible to easily perform appropriate text correction at the input text level that is easy for the end user to understand.
[Embodiment 2]
In recent years, SaaS (Software as a service) type services are spreading as a form of software provision. Conventionally, software has been sold as licenses to users as packaged products, and users have to run on their own computers. On the other hand, in the SaaS type service, the software is operated on the server on the provider side, and the user uses the software as a service via the network and pays the service fee to the provider side.

  Using this form, it is conceivable to use the SaaS type speech synthesis service in speech synthesis. That is, the user uses the speech synthesis software on the server via the network. Therefore, the end user can easily use the speech synthesis software to convert the text into speech, or use the provided editing function to adjust to the desired synthesized speech.

  The second embodiment is an example of a SaaS type speech synthesis service, in which a speech synthesis server operates according to speech synthesis software. Each user uses his / her user terminal 300, 310,... 3N0 to use the speech synthesis service via the network. On the user terminal side, a browser for transmitting and receiving data to and from the speech synthesis server is prepared. In the speech synthesis server 500 (speech synthesizer) shown in FIG. 11, the same numbers are assigned to the same functional blocks (modules) as in FIG.

  In the present embodiment, data transmission / reception with each user terminal is performed in a format corresponding to the browser (for example, HTML, XML, etc.). The speech synthesis server 500 further includes a transmission data creation unit 400, a reception data analysis unit 410, and a data transmission / reception unit 420.

  The transmission data creation unit 400 converts data processed inside the speech synthesis server 500 into transmission data to the user terminal. The reception data analysis unit 410 converts data received from the user terminal into data processed inside the speech synthesis server 500. The data transmitter / receiver 420 transmits / receives these data to / from the user terminal.

  The data transmitting / receiving unit 420 receives input text data and error indication data input by the editing unit (browser) of the user terminal from the user terminal. The reception data analysis unit 410 converts the input text data and error indication data received by the data transmission / reception unit 420 into data processed inside the speech synthesis server 100.

  The editing history accumulation unit 170 accumulates editing history data indicating editing contents performed by the editing unit 111 based on the converted input text data.

  In addition, when text data is input from the user terminal, the editing unit 111 generates a phonogram string or synthesized speech data of the text data and outputs it to the user terminal. At this time, the transmission data creation unit 400 converts the phonetic character string or the synthesized speech into transmission data for the user terminal. The user terminal accepts the error indication and transmits the error indication data to the speech synthesis server. Based on the error indication data, the editing unit 111 refers to the editing history data stored in the editing history storage unit 170 and re-edits the input text data input by the editing unit of the user terminal. The transmission data creation unit 400 converts the re-edited text data re-edited by the speech synthesis server 500 into transmission data to the user terminal. The data transmission / reception unit 420 transmits the transmission data created by the transmission data creation unit 400 to the user terminal.

  In the second embodiment, the editing history data of other users is stored and can be searched, so even an unfamiliar user can easily correct the synthesized speech by editing the text by using the editing history of the other users. Can be done. In addition, even a user who is familiar with the speech synthesizer can perform editing simply by pointing out the part to be corrected.

  As described above, according to the present invention, it is software that is placed on a server like the SaaS type speech synthesis service and has a framework in which editing histories of other users can be accumulated and searched. However, it is possible to easily correct synthesized speech by editing text. In addition, it is possible to effectively adjust the synthesized speech based on the input text.

  The configuration described in the above embodiment merely shows a specific example, and does not limit the technical scope of the present invention. Any configuration can be employed within the scope of the effects of the present invention.

  The configuration of the speech synthesis server is not limited to the above example. For example, it is good also as a structure with which a user terminal or another server equips a part of function with which the speech synthesis server 100 is provided.

  In the embodiment of the present invention, a software program for realizing the above-described embodiment (in the embodiment, a program corresponding to the flowchart shown in the figure) is supplied to the apparatus, and the computer of the apparatus reads the supplied program. Including cases that can also be achieved by execution. Therefore, in order to realize the functional processing described in this embodiment by a computer, the program itself installed in the computer is also an embodiment of the present invention. That is, a program for realizing the functional processing of the present invention is also included in one aspect of the embodiment. A medium recording a program for realizing the functional processing of the present invention is also included in one aspect of the embodiment.

100 Speech Synthesizer 110 Control Unit 111 Editing Unit 120 Language Processing Unit 121 Morphological Analysis Unit 122 Phonetic Character String Generation Unit 130 Synthetic Speech Generation Unit 140 Synthetic Data Management Unit 150 Editing History Registration Update Unit 151 Editing History Data Determination Unit 152 Editing History Data creation unit 153 Editing history data registration unit 160 Editing history search unit 170 Editing history storage unit 171 Editing history data storage unit 180 Interface unit 190 Display device 200 Input device 201 Text editing unit 202 Error indication receiving unit 203 Text correction unit 210 Audio output Device 400 Transmission data creation unit 410 Reception data analysis unit 420 Data transmission / reception unit 500 Speech synthesis server

Claims (6)

A text editing section for editing the text to be synthesized,
A language processing unit that performs morphological analysis and phonogram generation on the text before and after editing,
Whether the text editing was performed to correct the reading of the synthesized speech or the accent based on at least the phonetic character string among the morphological analysis results for the text before and after editing and the generated phonetic character string A text editing determination unit for determining whether or not,
An editing history data creation unit for creating editing history data indicating editing contents of the text editing when the text editing judgment unit determines that the text editing is performed to read the synthesized speech or correct the accent; ,
A speech synthesizer comprising: an editing history data storage unit that stores editing history data when the editing history data created by the editing history data creation unit is registered. A phonetic character string of the edited text edited by the text editing unit, or an output unit for outputting synthesized speech;
When the phonetic character string output from the output unit or the synthesized speech is read or the accent is incorrect, the synthesized text is read from the user, and the error indication is received from the user. And
Editing history indicating the edited content of text editing to be performed in order to read the synthesized speech or correct the accent for the point where the accent error is pointed out with respect to the edited text received by the error indication receiving unit An edit history data search unit for searching for and acquiring data from the edit history data storage unit;
The text editing unit
The speech synthesizer according to claim 1, wherein the edit text is re-edited using the edit history data searched and acquired by the edit history data search unit. Each editing history data stored in the editing history data storage unit has a function of counting the number of times applied to re-editing,
The editing history data search unit
Search with priority given to the editing history data with a large number of application times,
The edit history data storage unit
The speech synthesizer according to claim 1 or 2, wherein when the edit history data searched by the edit history data search unit is applied to re-editing, the application count is updated.   The text editing determination unit determines whether or not the text editing is performed to read the synthesized speech or to correct the accent, insertion or deletion of the sending kana, replacement between hiragana and kanji, The speech synthesizer according to any one of claims 1 to 3, wherein the speech synthesizer is performed by determining whether it is at least one of giving, reading, blank insertion or deletion. A text editing step for editing the text to be synthesized,
Language processing steps for performing morphological analysis and phonetic character string generation on the text before and after editing,
Whether the text editing was performed to correct the reading of the synthesized speech or the accent based on at least the phonetic character string among the morphological analysis results for the text before and after editing and the generated phonetic character string A text editing judgment step for judging whether or not,
An editing history data creation step for creating editing history data indicating the editing content of the text editing when it is determined that the text editing is performed in order to read the synthesized speech or correct the accent in the text editing determination step; ,
A speech synthesis method comprising: an editing history data storage step of storing the editing history data when the editing history data created in the editing history data creation step is registered. On the computer,
A text editing step for editing the text to be synthesized,
Language processing steps for performing morphological analysis and phonetic character string generation on the text before and after editing,
Whether the text editing was performed to correct the reading of the synthesized speech or the accent based on at least the phonetic character string among the morphological analysis results for the text before and after editing and the generated phonetic character string A text editing judgment step for judging whether or not,
An editing history data creation step for creating editing history data indicating the editing content of the text editing when it is determined that the text editing is performed in order to read the synthesized speech or correct the accent in the text editing determination step; ,
A speech synthesis program for executing an editing history data storing step for storing the editing history data when the editing history data created in the editing history data creating step is registered.

Images