JP2003173197A - Text voice synthesizing apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a text voice synthesizing apparatus which reads a word just as desired by a user without caring about the operation mode of a system and without designating the format for agates (KANA, Japanese syllabary) alongside Chinese characters every time to the same notation word. <P>SOLUTION: A text voice conversion part 142 is provided with a text analyzing part 11 which analyzes words in an inputted text by using a main dictionary 144 and a user dictionary 143 to convert them into an intermediate language, a voice synthesizing part 12 for synthesizing a voice waveform from the intermediate language, an agate format processing part 13 which judges whether agate format data in which the words and their reading are written side by side is included or not in the inputted text associating with the text analyzing part and which judges whether a format as an agate format is satisfied or not, and a user word adding part 14 which receives the agate format data recognized to be the agate format from the agate format processing part, adds word information required for user dictionary registration and adds it to the user dictionary. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text-to-speech synthesizer for reading text data as synthesized speech, and more particularly to a text-to-speech synthesizer having a user dictionary automatic registration function.

[0002]

2. Description of the Related Art In recent years, with the rapid expansion of users of the Internet, a personal computer for accessing it, or a portable information terminal such as a mobile phone, a PDA, a pager (pager (registered trademark)), has been It is rapidly gaining popularity in homes and schools.

One of the reasons for this is a message exchange system for electronic mail, Internet news and the like. In addition, these messages (eg,
(E-mail) is converted to voice and transferred to the telephone, or converted to voice and read aloud on the terminal, or notification of incoming e-mail is output to the pager owned by the destination user, or sent from a facsimile. A new type of message exchange system that integrates various message systems and information terminals, such as sending image information as multimedia e-mail, has begun to appear, centering on these e-mail messages and voice synthesis. The service is spurring further user expansion.

In this message exchange system, "text analysis processing" is one of the necessary functions for reading aloud electronic mail and network news. The text analysis process is a process for analyzing an input Japanese sentence using grammatical information and word information prepared in advance and outputting a voice synthesis parameter.

However, since many new words and coined words are used in these electronic mails and network news,
In some cases, the text is not read exactly as the user intended.
Therefore, in order to solve this problem, in the conventional text-to-speech synthesizer, a user dictionary that the user can freely customize is provided separately from the system dictionary.
The above problem was solved.

According to this method, the user dictionary is edited in advance so that the user can read it as intended, and a means for registering the edited user dictionary in the text-to-speech synthesizer is newly provided. When the text-to-speech synthesizer analyzes the input text data, not only the main dictionary but also the user dictionary can be used.
It solves the above problem.

However, this conventional technique has the following problem (1). Problem (1): In order to edit the user dictionary, means other than the text dictionary synthesizer such as the user dictionary editing device or the user dictionary editing software is required.

To solve this problem, a method for directly inputting word data in a text format into a device has been disclosed in a document: (Text-to-speech converter: JIII Journal of Technical Disclosure (Kogi No. 94-1789).
8)). Hereinafter, this technique will be briefly described.

The text-to-speech conversion device described in this document has two operation modes, a "text-speech conversion mode" and a "user dictionary registration mode", and mode switching can be performed by inputting a mode switching command from the keyboard. Done.

FIG. 14 shows the overall structure of a general text-to-speech synthesizer having a main dictionary and a user dictionary. In the "text-speech conversion mode", text data is treated as "readout data". The control unit 141 sends the input text data to the text-to-speech conversion unit 142. The text-to-speech conversion unit 142 uses the main dictionary 144 in which the reading and accent of each word are registered.
And analyze the text data using the user dictionary 143,
Finally converted to voice waveform data. The generated voice waveform data is output from the speaker. In this way, the text data 146 is read aloud.

In the "user dictionary registration mode", the text data is treated as "registered word data", and the control unit 141 sends the input text data to the data conversion unit 149. The data conversion unit 149 checks whether the text data conforms to a predetermined format. Specifically, it is checked whether or not the word notation / reading / accent is correctly described, and if there is an error, the control unit 141 receives information about the error location / content from the data conversion unit 149 and displays it. It is displayed on 147 to notify the user. If there is no error, the data conversion unit 149 converts the text data into a predetermined binary format and stores it in the user dictionary 143. In this way, word registration is performed.

As described above, according to this conventional technique, since the user can directly input the word data in the text format to the device when registering the words in the user dictionary, the user can create the binary format data which is difficult for humans to understand. , No more typing. Further, it is possible to save the trouble of converting the text format data into the binary format in advance and then inputting the data into the apparatus. Further, it is not necessary to prepare a dedicated program (tool) for converting the text format into the binary format separately from the device. Therefore, the problem (1) can be solved.

However, this conventional technique has a problem (2).
There is. Problem (2): The user switches the mode in consideration of whether the current operation mode of the text-to-speech converter is the “text-speech conversion mode” or the “user dictionary registration mode”. Must be done, which is a burden on operation.

To solve this problem, in the text-to-speech conversion mode, by directly specifying the reading of a word in the text,
A method has been proposed in which the text is read as specified. The conventional technique will be briefly described below.

The overall configuration of the device is similar to that shown in FIG. In this case, the text-to-speech conversion unit 142
Is a text analysis unit (not shown) for receiving the external data 146 and converting a kanji / kana mixed sentence into a prosodic phonetic symbol (hereinafter referred to as an intermediate language) and a voice for generating a voice waveform from the intermediate language. The text data is received from the synthesis unit (not shown) and the text analysis unit, and it is determined whether or not the text data is in the ruby format. In the case of the ruby format (see FIG. 7), the word notation and the reading are acquired. , And a ruby format processing unit (not shown) for outputting to the text analysis unit.

The ruby format will be briefly described with reference to FIG. The ruby format is composed of notation data 71 and reading data 72, and is a designation method for inserting the notation data 71 into the text data and reading the notation data 71 as the reading data 72.

Next, the overall processing will be briefly described.
In the "text-to-speech conversion mode", the user sets the reading at a position where he / she wants to directly specify the reading of the word, and sends the text data 14 to the control unit 141 in the text reading mode.
Send 6. The control unit 141 sends the input text data 146 to the text-to-speech conversion unit 142.

The text analysis section in the text-to-speech conversion section 142 receives the above-mentioned text data, and the user dictionary 1
43, the main dictionary 144, and the ruby format processing unit are used to perform word cutout processing, and a plurality of cutout words are collected in phrase units based on information attached to the words to generate an intermediate language. To do. The speech synthesizer generates a speech waveform from the generated intermediate language using the main dictionary 144 and outputs it to the speaker 145.

As described above, according to this conventional technique, by additionally providing a means for processing the ruby format in the text-to-speech conversion unit, it is possible to use the text described by using the ruby format. As a result, the way of reading can be freely customized on the spot without preparing any special means other than the text-to-speech converter. Furthermore, it is not necessary to be aware of the current operation mode, and the problem (2) can be solved.

However, this conventional technique has the following problem (3). Problem (3): This is a technique to specify the reading only for the specified word, so if the word with the same notation appears multiple times in the text, it must be specified in the ruby format each time it appears. There is.

[0021]

As described above,
By using the user dictionary, new words and coined words frequently appearing in message data such as net news and electronic mail can be read aloud as intended by the user, but the above problem (2) occurs and Using the ruby format to avoid causes problem (3). At present, the technology to solve both of these at the same time is
Not yet proposed.

According to the present invention, by providing a means for automatically adding a ruby-format-specified word to the user dictionary, the ruby specification can be made every time for the same notation word without being aware of the operation mode of the system. An object of the present invention is to provide a text-to-speech synthesizer that reads out a word as intended by a user without performing the operation.

[0023]

Therefore, in the text-to-speech synthesizer of the present invention, a text-to-speech conversion unit for converting the input text into a speech waveform and a device for analyzing the words of the input text are provided in advance in the apparatus. In a text-to-speech synthesizer including a main dictionary provided and a user dictionary in which words can be registered by the user, the text-to-speech conversion unit analyzes the words in the input text using the main dictionary and the user dictionary and converts them into an intermediate language. Whether the text analysis unit for conversion, the voice synthesis unit for synthesizing a speech waveform from an intermediate language, and the text analysis unit cooperate with the input text to include ruby format data in which words and their readings are juxtaposed. A ruby format processing unit that makes a judgment and a judgment as to whether or not the ruby format is satisfied, and a ruby former The recognized ruby format data to be collected by by adding the word information necessary to the user dictionary registration receipt from the ruby format processing unit and a user word additional unit to be added to the user dictionary.

[0024]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of a text-to-speech synthesizer of the present invention will be described in detail below with reference to the drawings.

First Embodiment [Structure] FIG. 1 is a diagram showing the structure of the text-to-speech converter 142 in the text-to-speech synthesizer of FIG. 14 as the first embodiment. Part 14
Reference numeral 2 denotes a text analysis unit 11 for analyzing a word of an input text (see FIG. 8) by using the main dictionary 144 and the user dictionary 143, and a text analysis unit 1.
1. A speech synthesis unit 12 for generating a speech waveform from the intermediate language output from 1, a ruby format processing unit 13 for checking a ruby format (see FIG. 7) in cooperation with a text analysis unit, and a ruby format. The user word adding unit 14 receives data recognized as being present from the ruby format processing unit 13, adds word information necessary for registering the user dictionary such as grammatical information and pronunciation information, and adds it to the user dictionary 143. .

[Operation] First, the operation of the entire apparatus will be described with reference to FIGS. 1 and 14. This text-to-speech synthesizer handles text data as "readout data" in the "text-speech conversion mode". Control unit 141
Is a text-to-speech conversion unit 1 for input text data.
Send to 42. The text-to-speech conversion unit 142 analyzes the text data and converts it into an intermediate language. At this time, if the text data includes ruby format data, the ruby format processing unit 13 and the user word adding unit 1
4 is used to update the user dictionary 143. The generated intermediate language is sent to the voice synthesizer 12, converted into voice waveform data, and output from the speaker 145.

Next, the operation of the text analysis unit 11 will be described in detail. When the text analysis unit 11 receives the text data, the user dictionary 143, the main dictionary 144,
Using the ruby format processing unit 13, word cutout processing is performed, and based on the information attached to each of the cutout words, the intermediate language is output in a phrase unit (see FIG. 12). . At this time, if the ruby format designation data is present in the input text, the word cutout process is performed on the corresponding portion so that the designated reading is made without collating with the main dictionary or the user dictionary. Further, at this time, the notation, the pronunciation, the accent type, the number of mora, and the part of speech are acquired from the ruby designation as a set of word data,
Add it to the user dictionary.

In the present embodiment, the part-of-speech data is information not included in the ruby format data, but the system automatically assigns it as a common noun. As a result, the information extracted from the ruby format data and the additional information are registered in the user dictionary as word data as shown in FIG. 13. Therefore, when a word of the same notation appears thereafter, the reading in the user dictionary will be performed in the intermediate language. Can be converted to.

Next, referring to FIG. 2, the text analysis unit 11
The word cutout process, particularly the process of ruby format data, will be described in detail.

When the text is input, the word starting from the first character of the text is cut out. This process is a process executed recursively, and the extracted words are stored in the work buffer and the initial state is empty.

(S1) If the beginning is the end of a sentence, it is determined that the word extraction has succeeded, and the process proceeds to S9. If it is not the end of the sentence, proceed to S2.

(S2) The ruby format processing unit 13
It is determined whether or not ruby format data exists at the beginning of the input sentence. The ruby format processing section collates with a predetermined ruby format rule to determine whether or not the format is satisfied, and returns the result to the text analysis section 11. In the text analysis unit, the determination result of the ruby format processing unit is Yes.
If s (matches the format), the process proceeds to S11, and if No, the process proceeds to S3.

(S3) The main dictionary 144 and the user dictionary 143 are searched to acquire all the words starting from the beginning of the sentence, and these are used as a word candidate group.

(S4) If the number of word candidates in the acquired word candidate group is 0, it is determined that the word cutting has failed, and the process proceeds to S10 to end the word cutting. If the number of word candidates is one or more, the process proceeds to S5.

(S5) One word candidate is selected from the acquired word candidate group.

(S6) Check the grammatical connection relation between the selected word candidate and the previous word stored in the work buffer. As a result of the check, if the grammatical connection relation is allowed, the word candidate is stored in the work buffer and the process proceeds to S7. If the combination is not allowed, the process returns to S4 to select another word candidate. Perform processing.

(S7) In order to cut out a sentence starting from the next character of the word candidate (if the word is extracted from the ruby format data, the word in the ruby format), the word cutting process is recursively performed.

(S8) If the result of step S7 is successful, the process proceeds to S9. If the result is unsuccessful, the selected word candidate is discarded, and the process returns to S4 to process the next word candidate.

(S9) This process is terminated on the assumption that the words have been cut out successfully.

(S10) This process is terminated assuming that the cutting of the word has failed.

(S11) This process is performed when it is determined that the result is ruby format data as a result of the determination in step S2. Word information such as notation, reading, accent type, number of mora, and part-of-speech is extracted from the ruby format data as one set of word data, which is added to the work buffer and the process proceeds to S12. However, a flag is set in the word information to indicate that it has been extracted from the ruby format data, so that the grammatical connection check required in the latter stage will always succeed.

(S12) Word information extracted from the ruby format data is added to the user dictionary 143, and the process proceeds to S7. Note that the registered word based on the ruby format data becomes valid immediately after the processing of S12 is completed.

According to the above processing, the input text including the ruby format data is divided into words, the reading and accent information of each word is acquired, and finally converted into an intermediate language. Also, at the same time as processing the ruby format data, the word specified in the ruby format can be automatically registered in the user dictionary, so that the user can register the user word without having to be aware of the operation mode of the device. become.

Second Embodiment [Structure] As shown in FIG. 3, the text-to-speech conversion unit 142 in the second embodiment is almost the same as the text-speech conversion unit 142 in the first embodiment. Is different in that a ruby selection unit 24 is added.

The text-to-speech conversion unit 142 in the second embodiment uses the main dictionary 144 and the user dictionary 143 to extract a word from the input text and converts it into an intermediate language, and a text analysis unit 21. Whether the ruby format data is included in the input text in cooperation with the voice synthesis unit 22 for converting the intermediate language output from 21 into a voice waveform and the text analysis unit 21.
Further, the ruby format processing unit 23 for checking whether or not the format as the ruby format is satisfied.
And to add to the user dictionary 143 the ruby format data recognized as being in the ruby format, from the ruby format processing unit 23, added with word information necessary for user dictionary registration such as grammatical information and pronunciation information. Of the user word adding unit 25 and the user dictionary adding designation information from the ruby format data, and according to this information, the ruby selecting unit 24 that determines whether to add the word information created from the ruby format data to the user dictionary. Is provided.

[Operation] First, the operation of the entire apparatus will be described with reference to FIGS. 3 and 14. Similar to the first embodiment,
In the "text-to-speech conversion mode", text data is treated as "readout data". The control unit 141 sends the input text data (see FIG. 10) to the text-to-speech conversion unit 142. The text-to-speech conversion unit 142 analyzes the text data and converts it into an intermediate language. At this time, the ruby format data (see FIG. 9) is added to the text data.
Reference) is included, the ruby format processing unit 23
Using the ruby selection unit 24 and the user word addition unit 25,
The user dictionary 143 is updated. At this time, only the ruby format data designated for registration by the ruby selection unit 24 is registered in the user dictionary. The generated intermediate language is sent to the voice synthesizer 22, converted into voice waveform data, and output from the speaker 145.

Next, the operation of the text analysis unit 21 will be described in detail. Upon receiving the text data (see FIG. 10), the text analysis unit 21 uses the user dictionary 143, the main dictionary 144, and the ruby format processing unit 23 to
Performs word segmentation processing. At this time, if the ruby format data (see FIG. 9) is present in the input text, the word cut-out process is performed so that the corresponding reading is performed without matching with the main dictionary or the user dictionary. . At this time, if the user dictionary registration designation (93 in FIG. 9) is set to 1 in the ruby format data, the notation, the reading, the accent type, the number of mora, and the part of speech are acquired from the ruby format data, and one set is obtained. Add to the user dictionary as word data of. As a result, only for the ruby format data for which the user dictionary registration has been specified, the reading by specifying the ruby format will be enabled, so that if a word with the same notation appears, it can be converted into an intermediate language with the reading registered in the user dictionary. Becomes

Next, referring to FIG. 4, the text analysis unit 21
The word cutout process, particularly the ruby format data process, will be described in detail.

When the text data is input, the process of cutting out the word starting from the first character of the text is started. This process is a process executed recursively, and the extracted words are stored in the work buffer and the initial state is empty.

(S21) If the beginning is the end of a sentence, it is determined that the word extraction has succeeded, and the process proceeds to S29. If it is not the end of the sentence, proceed to S22.

(S22) Ruby format processing unit 23
Determines whether ruby format data exists at the beginning of the input sentence. The ruby format processing unit collates with a predetermined ruby format rule, determines whether or not the format is satisfied, and returns the result to the text analysis unit 21. If the determination result of the ruby format processing unit is Yes (matches with the format), the text analysis unit advances the processing to S31,
If No, the process proceeds to S23.

(S23) The main dictionary 144 and the user dictionary 143 are searched to acquire all the words starting from the beginning of the sentence, and these are set as a word candidate group.

(S24) If the number of word candidates in the acquired word candidate group is 0, it is determined that the word cutting has failed, and the word cutting process is terminated, so that the process proceeds to S30. Number of word candidates is 1
If the number is greater than or equal to the number, the process proceeds to S25.

(S25) One word candidate is selected from the acquired word candidate group.

(S26) The grammatical connection relation between the selected word candidate and the previous word stored in the work buffer is checked. As a result of the check, if the grammatical connection relation is allowed, the word candidate is stored in the work buffer and the process proceeds to S27. If the combination is not allowed, the process returns to S24 to select another word candidate. .

(S27) In order to cut out a sentence starting from the next character of the word candidate (if the word is extracted from the ruby format data, the word is in the ruby format),
Call the word cutout process recursively.

(S28) If the result of S27 is success, S
In step 29, if it fails, the selected word candidate is rejected, the process returns to step S24, and the next word candidate is processed.

(S29) This process is terminated on the assumption that the words have been cut out successfully.

(S30) This process is terminated on the assumption that the extraction of words has failed.

(S31) This process is performed when it is determined that the data is ruby format data as a result of the determination in step S22. The notation, reading, accent type, mora number, and part of speech are extracted from the ruby format data to form a set of word data, which is added to the work buffer and the process proceeds to S32. However, a flag is set in the word information to indicate that it has been extracted from the ruby format, so that the grammatical connection check, which is required in the latter stage, will always succeed.

(S32) User dictionary addition designation information (see 93 in FIG. 9) is extracted from the ruby format data, and it is determined whether or not this information is 1. This result is Yes
If so, the process proceeds to S23, and if No, the process proceeds to S27.
As a result, only the words in the ruby format designated by the user are registered in the user dictionary.

(S33) Word information specified by the ruby format is added to the user dictionary 143, and the process proceeds to S27. Immediately after the processing of S33 ends, the registered word specified by the ruby format becomes valid.

According to the above processing, the input text containing the ruby format data is divided into words, the reading and accent information of each word is acquired, and finally converted into an intermediate language. Also, at the same time as processing the ruby format data, the word specified in the ruby format can be automatically registered in the user dictionary, so that the user can register the user word without having to be aware of the operation mode of the device. become.

Further, according to the present embodiment, the user dictionary addition designation information is provided in the ruby format data, this information is acquired, and addition / non-addition to the user dictionary is switched according to the acquired value. Therefore, it is possible to add only the words in the ruby format specified by the user to the user dictionary.

Third Embodiment [Structure] As shown in FIG. 5, the text-to-speech conversion unit 142 in the third embodiment analyzes the words of the input text using the main dictionary 144 and the user dictionary 143. hand,
A test analysis unit 31 for converting into an intermediate language, a voice synthesis unit 32 for converting the intermediate language output from the text analysis unit 31 into a voice waveform, and a ruby for checking ruby format data in the input text. The format processing unit 33 and the data recognized as the ruby format data are received from the ruby format processing unit 33, and the word information necessary for registering the user dictionary such as grammatical information and pronunciation information is acquired and stored in the user dictionary 143. The user word addition unit 36 for adding and the user dictionary addition designation information from the ruby format data, and according to this information, the ruby selection unit 34 that determines whether or not to add the ruby format data to the user dictionary, and A key for determining whether or not the ruby format data has already been registered in the main dictionary 144. The data check portion 35 is provided.

[Operation] Next, referring to FIG. 5 and FIG.
The operation of the entire apparatus will be described. First, in the “text-speech conversion mode”, text data is treated as “read-aloud data”. The control unit 141 sends the input text data (see FIG. 10) to the text-to-speech conversion unit 142. The text-to-speech conversion unit 142 analyzes the input text data and converts it into an intermediate language. At this time, if the text data includes ruby format data (see FIG. 9), the ruby format processing unit 33, the ruby selecting unit 34, the user word adding unit 36, and the duplicate data checking unit 35.
To update the user dictionary 143. At this time, only when the ruby format data has been designated for registration by the ruby selection unit 34 and the same word is not registered in the main dictionary 144 by the duplicate data inspection unit 35,
The word data specified by the ruby format is registered in the user dictionary. After that, the acquired intermediate language is sent to the voice synthesis unit 32, converted into voice waveform data, and output from the speaker 145.

Next, the operation of the text analysis unit 31 will be described in detail. The text analysis unit 31 receives the text data, and uses the user dictionary 143, the main dictionary 144, the ruby format processing unit 33, the ruby selection unit 34, the duplicate data search unit 35, and the user word addition unit 36 to perform a word segmentation process. Then, the input text is converted into word segmentation data (see FIG. 11). Then, based on the information attached to the extracted words, the extracted words are put together in phrase units and an intermediate language is output (see FIG. 12).

At this time, if the ruby format data is present in the input text, the word cut-out process is performed on the corresponding portion so as to obtain the designated reading without collating with the main dictionary or the user dictionary. Furthermore, at this time, if the user dictionary registration designation in the ruby format data is 1,
The notation, reading, accent type, mora number, and part of speech are acquired from the ruby format data (see FIG. 13), and these are added to the user dictionary as a set of word data.

As a result, the ruby format data will be valid only for the words for which the user dictionary has been designated to be registered, so that when words of the same notation appear, the reading registered in the user dictionary can be converted into an intermediate language. Becomes

Next, referring to FIG. 6, the text analysis unit 31
The word cutout processing, particularly the processing of ruby format data, will be described in detail.

When the text is input, the process of cutting out the word starting from the first character of the text is started. This process is a recursive process, and the extracted words are
It is stored in the working buffer and its initial state is empty.

(S41) If the beginning is the end of a sentence, it is determined that the word extraction has succeeded, and the process proceeds to S49. If not the end of sentence, S42
Go to.

(S42) Ruby format processor 33
Determines whether ruby format data exists at the beginning of the input sentence. The ruby format processing unit collates with a predetermined ruby format rule, determines whether or not the format is satisfied, and returns the result to the text analysis unit. In the text analysis unit, the determination result of the ruby format processing unit is Yes.
If s, the process proceeds to S51, and if No, S4.
The process proceeds to 3.

(S43) The main dictionary and the user dictionary are searched to acquire all the words starting from the beginning of the sentence, and these are set as a word candidate group.

(S44) If the number of word candidates in the acquired word candidate group is 0, it is determined that the word cutting has failed, and the word cutting process is terminated, so that the process proceeds to S50. Number of word candidates is 1
If the number is greater than or equal to the number, the process proceeds to S45.

(S45) One word candidate is selected from the acquired word candidate group.

(S46) Check the grammatical connection relation between the selected word candidate and the previous word stored in the work buffer. As a result of the check, if the grammatical connection relation is allowed, the word candidate is stored in the work buffer and the process proceeds to S47. If the combination is not allowed, the process returns to S44 and another word candidate is selected. .

(S47) To cut out a sentence starting from the next character of the word candidate (if the word is extracted from the ruby format data, the word is in the ruby format),
The word cutout process is recursively called.

(S48) If the result of step S47 is successful, the process proceeds to S49. If the result is unsuccessful, the selected word candidate is discarded, and the process returns to S44 to process the next word candidate.

(S49) Assuming that the words have been cut out successfully, this process ends.

(S50) This process is terminated because it is determined that the cutting of the word has failed.

(S51) This process is performed when it is determined that the data is ruby format data as a result of the determination in step S42. The notation, reading, accent type, number of mora, and part-of-speech are obtained from the ruby format data to form a set of word data, which is added to the work buffer and the process proceeds to S52. However, a flag is set in the word information to indicate that it has been extracted from the ruby format, so that the grammatical connection check, which is required in the latter stage, will always succeed.

(S52) The user dictionary designation information (see 93 in FIG. 9) is extracted from the ruby format data, and it is determined whether or not this extension information is 1. If Yes (if 1), proceed to S53, and if No, S4.
Proceed to 7.

(S53) The same notation as the acquired word information,
It is determined whether or not the words having the same reading are already registered in the main dictionary. When the main dictionary is searched and a word having the same notation and the same reading as the word exists, it is determined that there is duplicate data, and the process proceeds to S47. If not, the process proceeds to S54.

(S54) The acquired word information is added to the user dictionary 143, and the process proceeds to S47.

According to the above processing, the input text containing the ruby format data is divided into words, and the reading and accent information of each word can be acquired and finally converted into an intermediate language. In addition, since the word in the ruby format can be automatically registered in the user dictionary at the same time as the ruby format is processed, the user can perform the user word registration without having to be aware of the operation mode of the device.

Further, according to the present embodiment, when the same word data exists in the main dictionary, it is not added to the user dictionary, so that the user can register the word in the user dictionary without being aware of the duplication of the word data. It can be performed.

The present invention is not limited to the above-mentioned embodiments, but can be modified in various ways based on the spirit of the present invention.

For example, in the above-described embodiment, the word data designated by the ruby format is registered in the user dictionary. However, instead of the ruby format, it is designated by the format in which the notation and the intermediate language character string are paired. You may do it.

Further, in particular, by designating a person's name or a place name in the ruby format, many of these words are not registered in the system dictionary, so that it is particularly effective for reading them out as intended by the user.

[0091]

As described above in detail, according to the present invention, the system automatically registers the word designated by the ruby format in the user dictionary. The effect of being free from the hassle of repeating ruby designation, and the effect that the user can register the user words in the reading mode dictionary, so that it is not necessary to be aware of the distinction between the word registration mode and the reading mode, and the user Without introducing a dictionary editing device,
The effect is that words can be registered in the user dictionary.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a text-to-speech conversion unit according to the first embodiment.

FIG. 2 is a word cutout processing flow of a text analysis unit according to the first embodiment.

FIG. 3 is a configuration diagram of a text-to-speech conversion unit according to the second embodiment.

FIG. 4 is a word cutout processing flow of a text analysis unit according to the second embodiment.

FIG. 5 is a configuration diagram of a text-to-speech conversion unit according to the third embodiment.

FIG. 6 is a word cutout processing flow of a text analysis unit according to the third embodiment.

FIG. 7 is an example 1 of a ruby format.

FIG. 8 is an example 1 of input text.

FIG. 9 is a second example of a ruby format.

FIG. 10 is an example 2 of input text.

FIG. 11 is an example of a word cutout processing result.

FIG. 12 is an output example of a text analysis unit.

FIG. 13 is an example of additional word data.

FIG. 14 is an overall configuration diagram of a general text-to-speech synthesizer.

[Explanation of symbols]

11, 21, 31 Text analysis section 12, 22, 32 Speech synthesizer 13, 23, 33 ruby format processing unit 14, 25, 36 User word addition section 142 Text-to-speech converter 143 user dictionary 144 Main dictionary 24, 34 Ruby sorting section 35 Multiple data inspection section

Claims (5)

[Claims] 1. A text-to-speech conversion unit for converting an input text into a speech waveform, a main dictionary pre-installed in the apparatus for analyzing words in the input text, and a user dictionary in which a user can register words. In the text-to-speech synthesis device, the text-to-speech conversion unit synthesizes a speech waveform from the intermediate language and a text analysis unit that analyzes words in the input text using the main dictionary and the user dictionary and converts the words into an intermediate language. In cooperation with the speech synthesis unit and the text analysis unit, it is determined whether or not the input text includes ruby format data in which words and their readings are juxtaposed, and whether or not the ruby format is satisfied. The ruby format processing unit that determines whether or not the ruby format data recognized as the ruby format is Text-to-speech conversion system which is characterized in that a user word additional portion receives from the formatting processing unit adds the word information necessary to the user dictionary registration add to the user dictionary. 2. The text-to-speech synthesis apparatus according to claim 1, wherein the text-to-speech conversion unit further acquires extension information from the ruby format data, and adds the ruby format data to a user dictionary according to the extension information. A text-to-speech conversion device comprising a ruby selection unit for determining whether or not to perform. 3. The text-to-speech synthesis apparatus according to claim 2, wherein the text-to-speech conversion unit further determines whether or not the word information extracted from the ruby format data is already registered in the main dictionary. A text-to-speech conversion device, which is provided with a duplicate data inspection unit. 4. The text-to-speech synthesis apparatus according to claim 1, wherein the words extracted from the ruby format data are words representing a place name and a person's name. apparatus. 5. The text-to-speech synthesizer according to claim 1, wherein a format represented by a set of notation and intermediate language character string is used instead of the ruby format. Text-to-speech synthesizer.