JPH09244869A - Document reading-aloud system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To read a KANA(Japanese syllabary)-KANJI(Chinese character) mixed document aloud as a document creator intends by analyzing processing operation that tag information detected in the KANAKANJI mixed document indicates and reading the KANA-KANJI mixed document aloud according to the processing operation. SOLUTION: A tag retrieval part 12 of a tag information processing part 28 retrieves tag information in the object KANA-KANJI mixed document 11 to be read aloud and a tag information analysis part 13 decides the kind of the retrieved tag information and performs the processing corresponding to the kind of the tag. The read-aloud buffer 21 of read-aloud software (reading- aloud means) 20 stores document data to be read aloud and a text processing part 22 uses a Japanese dictionary 24 where reading information and accent information on words are recorded according to the processing operation that the tag information analyzed by the tag information analysis part 13 to convert the inputted KANA-KANJI mixed document 11 into a pronunciation symbol sequence; and a speech synthesizing process part 23 converts the pronunciation sequence into a speech waveform, and a speech output part 26 outputs the speech waveform.

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 system in which a text mixed with kanji and kana is subjected to language processing, and the result is voice-synthesized to read as a voice.

[0002]

2. Description of the Related Art In recent years, a technology relating to a text-to-speech system which reads aloud as a voice has been put into practical use by performing language processing on a sentence mixed with Japanese kanji and kana and performing voice synthesis processing on the processing result. .

Conventionally, in this type of reading system, language processing is carried out by referring to a word dictionary in which reading information, accent information, etc. for words in a sentence containing kanji and kana are registered.
The results were read aloud.

FIG. 11 is a block diagram showing a configuration of a conventional example of the above-mentioned reading system.

In the conventional example shown in FIG. 11, a sentence 1 containing kanji and kana to be read, a Japanese dictionary 4 in which reading information and accent information of words are recorded, and a Japanese dictionary 4 are used. , A text processing unit 2 for converting a sentence 1 mixed with Kanji and Kana into a phonetic symbol string, voice data 5 for recording a phonetic symbol string as a voice waveform, a voice synthesis processing unit 3 for converting a phonetic symbol sequence into a voice waveform, and a voice. The audio output unit 6 outputs a waveform.

To explain the operation of the configuration shown in FIG. 11, a sentence 1 containing kanji and kana is input to the text processing unit 2. The text processing unit 2 carries out a process of confirming the reading of words. In this word reading determination process, the word reading is determined by referring to the word dictionary 4.

Further, the reading of the word is fixed by acquiring and analyzing the special word information embedded in the sentence.

Further, the word reading is fixed by extracting the word with ruby (furigana).

Further, the reading delimiter character in the page is extracted and read out in the current page, whereby the sentence and the sentence read out in the next page are classified.

In the text processing section 2, after the pronunciation is confirmed, the text processing section 2 converts it into a phonetic symbol string and outputs it to the voice synthesizing section 3.
The voice synthesizing unit 3 converts the phonetic symbol string into a voice waveform using the voice data 5, and outputs the voice waveform at the voice output unit 6.

Some reading methods use a reading system to display a reading sentence simultaneously with reading. In this type, the sentence 1 mixed with kanji and kana is output to the text processing unit 2 and is also output to the display unit, and the display unit corresponds to the processing operation in the text processing unit 2 and one sentence of the sentence 1 mixed with kanji kana is displayed. It is configured to display one by one.

Words in a mixed Kanji / Kana sentence may include proper nouns or words that require special reading. If such a word is not registered in the above word dictionary, the sentence Sometimes the author could not read what he intended.

As a conventional example for solving the above-mentioned problems, for example, as disclosed in Japanese Patent Laid-Open No. 4-331998, reading, accent, etc., of a word that is not registered in a word dictionary are There is a method in which special word information that defines information is embedded in a sentence or described outside a sentence like a footnote and used to read aloud.

In addition, ruby may be shaken for a word that is difficult to read or a word that causes a special reading. In such a case, ruby itself is read after reading the word. I sometimes read aloud.

As a conventional example for solving the above-mentioned problems, for example, JP-A-5-189194 and JP-A-1 are available.
As disclosed in Japanese Unexamined Patent Publication No. 300334/1993, a method of not reading a word with ruby, reading only ruby, and when this word appears thereafter, a method of reading according to the reading specified by ruby is used. is there.

Further, in a sentence reading system for reading a sentence on a page-by-page basis, if a word spans pages, even if the word is registered in a dictionary, it will be treated as a word having a different meaning when performing language processing. It was processed and sometimes read unnaturally.

Therefore, the reading delimiter that appears at the end of the page (for example, symbols such as punctuation marks and parentheses)
It is conceivable to read out the sentences up to and combine the rest of the sentences with the next page, and then read them out when the page breaks. In addition, a method may be considered in which the part where the kanji and hiragana are switched last in the page is determined as the break.

[0018]

DISCLOSURE OF THE INVENTION Problems to be Solved by the Invention
In the method disclosed in Japanese Laid-Open Patent Publication No. 31998, the special word notation information, the phoneme, and the temperament information are separated by a colon (:), and the entire special word information is sandwiched by a semicolon (;), and a sentence containing a kanji kana to be processed It is supposed to be embedded in the above, but a symbol that indicates special word information (here, a semicolon)
If the same symbol is described in the text of a sentence mixed with kanji and kana, the symbol is processed as being judged as special word information, which causes a problem such as a reading error.

Further, in the methods disclosed in Japanese Patent Laid-Open Nos. 5-189194 and 1-330034, only the control for prohibiting repeated reading of a word to which ruby is added is performed, but otherwise. Because the word is read aloud, if a symbol character such as parentheses or “○” is registered in the word dictionary, it may be read aloud as well. There was a problem that it was not a thing.

Further, in the reading delimiter detection method when reading aloud on a page-by-page basis, when the delimiter character is judged to be a reading delimiter, if the delimiter character does not appear near the end of the page, the next page is read. There are many strings to carry over. In this case, there is a problem that a large amount of memory is required. Furthermore, in the case of the one provided with the display unit, the page to be displayed is switched to the next page and then read out. Therefore, there is a problem that the user feels uncomfortable.

When it is determined that the place where the kanji and hiragana are switched last is a reading delimiter, when the word is composed of kanji and hiragana such as "deep", "deep"
Since it is divided into "i" and read out as "shin" and "i", there is a problem that the reading of words is impaired.

The present invention has been made in view of the problems of the above-described conventional techniques, and it is possible to read a sentence according to the reading intended by the sentence creator and read the sentence in page units. It is to realize a text-to-speech system that allows reading without any discomfort.

[0023]

According to the text-to-speech system of the present invention, a text containing kanji and kana with tag information indicating that a predetermined processing operation is to be performed is used as a text to be processed. A sentence reading method for reading a sentence containing the kanji and kana mixed for each input, a tag information search means for searching tag information in the sentence containing the kanji and kana, and a processing operation indicated by tag information detected as a result of the search. It is characterized by comprising tag information analyzing means for analyzing and reading-out executing means for reading out the sentence containing the kanji and kana input according to the processing operation indicated by the tag information analyzed by the tag information analyzing means.

In this case, the tag information is composed of a first symbol indicating the start of the tag information, a second symbol indicating the end of the tag information, and an identifier indicating the type of processing operation sandwiched between the symbols. The tag information search means may search the tag information by searching the first and second symbols, and the tag information analysis means may analyze the processing operation indicated by the tag information by the identifier.

Further, there is a process for designating the reading as the type of the processing operation indicated by the tag information, and the tag information includes the first tag information to which the identifier for designating the reading and the phoneme / temperament are added, and the first tag information. And the second tag information in which the third symbol indicating the end of the processing range is further added to the identifier attached to the first tag information. The character string may be read aloud as a phoneme or temperament in the first tag information.

Further, there is a process of prohibiting reading aloud as a type of processing operation indicated by the tag information, and the tag information includes third tag information to which an identifier for prohibiting reading aloud is added, and the third tag information. And the fourth tag information in which the third symbol indicating the end of the processing range is further added to the identifier attached to the reading identifier, and the reading execution means is not limited to the character string between the third and fourth tag information. You may not read it aloud.

Further, there is tag information indicating a reading break as the type of processing operation indicated by the tag information, and the fifth tag information with an identifier indicating the reading break is used as the tag information and read out. The execution means does not continuously read the character strings after the fifth tag information,
It may be read aloud together with a sentence containing a mixture of kanji and kana to be input next.

[Operation] In the present invention configured as described above, since the reading execution means performs the reading according to the processing indicated by the tag information, various reading processing can be performed.

The start and end of the tag information are the first symbol and the second symbol.
The type of processing of tag information is performed by an identifier, so it is possible to confirm whether the first symbol and the second symbol in a sentence mixed with Kanji and Kana have an identifier. it can.

Examples of the processing indicated by the tag information include reading specification, reading prohibition, reading breaks, and the like.
When reading is specified or reading is prohibited, a sentence containing kanji kana characters between the first tag information and the second tag information, or a sentence containing kanji kana characters between the third tag information and the fourth tag information However, even if the same symbol as the above tag information is described in a sentence containing kanji and kana, the symbol is not mistaken for the above tag information, so that a malfunction can be prevented.

Further, since it is possible to avoid misinterpretation of a difficult-to-read word or a word to be read specially, the voice synthesis is performed according to the reading intended by the sentence creator.

Further, it is possible to read aloud a word to which ruby is added repeatedly, and to prohibit reading of symbol characters such as parentheses and "○", so that speech synthesis that is easy to hear is performed.

Further, even when a word spans a page, the reading delimiter can be designated in advance at an optimum position, so that the control can be performed with a small memory, and the reading can be performed easily without being erroneously read.

[0034]

Next, embodiments of the present invention will be described with reference to the drawings.

First, the format of the tag information used in this embodiment will be described with reference to FIGS.

In this embodiment, the "<" character is used as the first symbol indicating the start of tag information. In the tag information, a character string representing the type of tag information is described after the “<” character.

As the type of information indicated by the tag information, reading designation information (content identifier: T shown in FIG. 5 is used.
CHG), reading prohibited information (content identifier: TNOP) shown in FIG. 6, and reading delimiter information (content identifier: TEOT) shown in FIG.

When the tag information specifies how to read the word shown in FIG. 5, the word reading information and accent information are described after the "<" character. Next, the ">" character as the second symbol indicating the end of the tag information is described.

Further, when the tag information is the tag information for designating the reading of the words shown in FIG. 5 and the tag information for designating the prohibition of reading the words as shown in FIG. 6, a character string indicating the kind of the tag information ( The tag range in which the third symbol "/" is added before the content identifier) is used to specify the range of words to be processed.

In the case of the tag information for designating the reading of words shown in FIG. 5, it is necessary to indicate the phoneme and temperament indicating the reading of words, but the tag for designating the prohibition of reading of words shown in FIG. In the case of information, it is not necessary to simply read the processing range. Further, the reading delimiter information shown in FIG. 7 is used like a page break, and tag information added with the "/" character is not required because there is no processing range. In this way, the contents of the tag information required depending on the processing also differ, and the following tag information is used in this embodiment.

First tag information: Tag information indicating the start of the range of reading designation Second tag information: Tag information indicating the end of the range of reading designation Third tag information: Tag information of the beginning of the reading prohibited range Tag information: Tag information at the end of the reading prohibited range Fifth tag information: Tag information for reading delimiter The tag information search in the present embodiment is performed by searching for the "<" character. When the "<" character is detected, it is then checked whether or not the following character string is each of the above content identifiers, and if they are different, it is recognized as the "<" character in the text of the kanji-kana mixed sentence. Also, the end of the tag information is confirmed by searching for the second symbol “>” character that appears immediately after.

FIG. 1 is a block diagram showing the configuration of this embodiment, in which a sentence 11 mixed with kanji and kana to be read is read.
And a tag information processing unit 28 that processes tag information in the kanji-kana mixed sentence 11 and display software 14 that determines the display position and size of the character string in the kanji-kana mixed sentence 11.
A sentence display unit 19 for displaying the processed character string, reading software 20 (reading execution means) for analyzing the sentence 11 mixed with Kanji and Kana and converting it into a voice waveform, and a voice output unit 26 for outputting the voice waveform. , Carry-over buffer 27 that stores the character string prepared when reading the next page
Consists of

The tag information processing unit 28 is composed of a tag searching unit 12 for searching the tag information and a tag information analyzing unit 13 for discriminating the kind of the searched tag information and performing a process according to the kind of the tag. .

The display software 14 includes a display buffer 15 for storing the text data to be displayed, a text shaping means 16 for determining the position and size of a character string in the text data,
It is composed of font data 18 for recording the shape of a character string and dot data creation means 17 for creating dot data for display using the font data 18.

The reading software 20 includes a reading buffer 21 for storing text data to be read, a Japanese dictionary 24 for recording reading information and accent information of words,
A text processing unit 22 for converting a sentence 11 mixed with Kanji and Kana into a phonetic symbol string using a Japanese dictionary 24, voice data 25 in which the phonetic symbol string is recorded as a voice waveform, and a phonetic symbol string for converting into a voice waveform. It is composed of a voice synthesis processing section 23. The processing in each of the above units is performed by a control device (not shown).

FIG. 2 is a flow chart collectively showing the processing operation in the present embodiment, FIG. 3 is a flow chart showing the processing procedure in the tag information search processing A3 in FIG. 2, and FIG.
6 is a flowchart showing a processing procedure in tag information analysis processing A4 in FIG.

First, the overall processing procedure in this embodiment will be described with reference to FIG.

The text 11 mixed with kanji and kana is input to the tag information retrieval unit 12 as document data for each page (step A1). Next, the input document data for one page and the document data in the carry-over buffer 27 described later are combined (step A2). Then, the tag information search unit 12
It is searched whether or not the tag information is present in the combined text data in (step A3), and then it is confirmed whether or not the tag information is detected in step A3 (step A4).

When it is confirmed that the tag information is detected in step A4, the type of the tag information detected by the tag information analysis unit 13 is determined, and the sentence data is analyzed according to the type. (Step A5), Step A after analysis
3 and the above operation is repeated.

When it is confirmed that the tag information is not detected in step A4, the display position and the display size of the character string of the sentence data stored in the display buffer 15 by the text shaping means 16 of the display software 14. And so on. Further, the dot data creating means 17 creates a character string for display using the font data 18 and displays it on the display unit 19 (step A6).

Next, in the text processing means 22 of the reading software 20, the sentence data stored in the reading buffer 21 is converted into a phonetic symbol string using the Japanese dictionary 24, and the speech synthesis processing section 23 is also used. In step S7, the audio data 25 is used to convert the audio waveform into an audio waveform, and the audio output unit 26 outputs the audio (step A7).

When the display and reading of the acquired sentence data for one page are completed, the page number is updated (step A8), and then it is confirmed whether the updated page number exceeds the final page of the sentence containing the kanji and kana (step A8). A
9). As a result of the confirmation in step A9, if it is confirmed that the last page is not exceeded, the process returns to step A1 and the above operation is repeated, and if it is confirmed that the last page is exceeded, display and The reading process ends.

Next, the tag information search processing performed in step A3 will be described in detail with reference to FIG.

In the tag information search processing in step A3, first, the symbol "<" indicating the start of tag information in the text data is searched (step B1), and it is confirmed whether the search was successful (step B2).

When the symbol "<" indicating the start of tag information is detected, the character string before the tag information is displayed in the display buffer 1.
5 (step B3) and further confirms whether the next character is a content identifier (step B4). If the following character is not an identifier, the detected symbol “<” is not a symbol that indicates the start of tag information, but is determined to be the “<” character described in the text and the “<” character is displayed. The data is stored in the buffer 15 (step B7).

As a result of the confirmation in step B4, if it is confirmed that the following character is not "<", the character string before the tag information is stored in the reading buffer 21 (step B
5) Set the return value to "I was able to search" (step B
6), the tag information search process ends.

In step B2, when the "<" character cannot be retrieved, the character strings are stored in the display buffer 15 and the reading buffer 21 (step B8, step B9), and the return value is set to "cannot be retrieved". (Step B10), the tag information search process ends.

Next, the tag information analysis processing in step A5 will be described in detail with reference to FIG.

In the tag information analysis processing in step A5, first, the type of tag information is determined (step C1), and different processing is performed depending on the type of tag information.

As shown in FIGS. 5 to 7, there are three types of tag information: reading designation, reading prohibition, and reading delimiter.
There are types.

When the tag information is the tag information designating the reading of the word, the reading information and accent information (phonological / temperal information) of the word described in the tag information is acquired (step C2), and the word is read. The reading information and the accent information of are stored in the reading buffer 21 (step C3). Next, the tag information indicating the end of the range of the word processed by this tag information is detected (step C4), and the character string surrounded by this tag information and the end tag information of the range is displayed / buffered. It stores in 15 (step C5), and ends the tag information analysis process.

As a result of the discrimination in step C1, when the type of the tag information is the tag information for which reading is prohibited, the tag information indicating the end of the range of the word to be processed by this tag information is detected. (Step C6), the character string surrounded by this tag information and the end tag information of the range is displayed in the display buffer 15
(Step C7), and the tag information analysis process ends.

When the type of the tag information is the tag information indicating the reading delimiter in the page, the character strings after this tag information are stored in the carry-over buffer 27 (step C8).
The tag information analysis process ends.

[0064]

Next, an embodiment of the present invention will be described with reference to the drawings.

Each of FIGS. 8A to 8C is a diagram showing the sentence data, the display result, and the voice output result when the tag information "<TCHG>" designating the reading of the word is used.

When the target of voice output is a technical sentence, technical terms are often omitted and written in alphabets, such as "HDD" and "GB", as shown in FIG. 8B. . However, all the technical terms in Japanese dictionary 2
Since it is impossible to register in 4 (see Fig. 1), when outputting voice, "ETCHIDEDI" or "JeeBee"
I read the alphabet itself like. Further, in the case of a symbol character such as "x", when only "Batsu" is registered in the Japanese dictionary 24,
I was reading something that didn't make sense, such as "Roppyakujyo batsu yonhyakuhachiju".

In the present invention, the use of the tag information specifying the reading as described above prevents the inconvenience described above from occurring.

As shown in FIG. 8A, "system configuration is as follows using tag information designating how to read a word. <TCHG, hard disk drive>"
HDD </ TCHG> 1 <TCHG, Gigabyte> GB </ TCHG> Memory 32 <TCHG,
Megabyte> MB </ TCHG> Mouse available
Resolution 640 <TCHG, multiply> × </ TCH
The text data is described as G> 480 ". By performing the processing according to the procedure shown in FIGS. 2 to 5 on such text data, as shown in FIG.
“How is it? Can be output as voice.

In the present embodiment, an example of specifying how to read technical terms has been described, but words and person names that are difficult to read,
Of course, it is also possible for the sentence creator to specify the reading of a word for which he wants to specify a special reading.

FIGS. 9A to 9C are diagrams showing the sentence data, the display result, and the voice output result when the tag information "<TNOP>" for prohibiting the reading of words is used.

In the sentence 11 with kanji and kana mixed (see FIG. 1), decorative characters may be used to emphasize items in the sentence. When such a decorative character is registered in the Japanese dictionary 24, a voice output of "Marukyo no Kodate" is output for the display result of "○ today's menu" as shown in FIG. 9 (b). There is. In addition, for words such as “kosho” with ruby, the word “koshokosho” may be repeatedly output as voice.

In the present invention, the use of the tag information for prohibiting the reading of words as described above prevents the above inconvenience from occurring.

As shown in FIG. 9A, by using the tag information for prohibiting the reading of words, "<TNOP> ○ <// T
NOP>Today's Menu <<< December 25th> Today, I will make a roasted turkey. <TNOP> ☆ </ TNOP
＞ Ingredients ・ Turkey ・ Salt ・ Pepper <TNOP> (kosho) </ TNOP> ・ Soy sauce <TNOP> (soy sauce)
</ TNOP)> ”, the text data is described. By performing the processing shown in FIGS. 2 to 5 on such text data, as shown in FIG. 9C, "Today, we will make a round sword that will be used for daily life.
In this embodiment, when the “<” character is used in the text as shown in FIG. 9 (a), it is described as “<<<”. Then, the text and the tag information are distinguished.

In each of FIGS. 10A to 10C, tag information "<TEOT>" for designating a page break is specified.
It is a figure which shows the text data when using, a display result, and a voice output result.

When the sentence 1 containing kanji and kana is composed of a plurality of pages, the words may extend over the pages as shown in FIG. 10 (b). When the voice output is controlled on a page-by-page basis, the word "Echigo" is divided into "Etsu" and "Rear" as shown in FIG.
"Etsu" instead of the word "Echigo"
It is interpreted as a word that is read as "nochi" and misinterpreted.

In the present invention, the use of the tag information for designating the page breaks as described above prevents the above inconvenience from occurring.

As shown in FIG. 10A, "<TEOT>" is described before the character string to be carried over to the next page by using the tag information for designating the page reading delimiter. By performing the processing according to the procedure shown in FIGS. 2 to 5 on this text data, as shown in FIG.
"Etsu" is carried over to the next page and output as "Echigo".

[0078]

As described above, according to the present invention, since the means for distinguishing the tag information from the text is provided, it is possible to use the word or phrase indicating the start of the tag information in the text.

Further, by using the tag information for designating the reading of the word and the tag information for prohibiting the reading of the word, it is possible to provide a sentence reading system for performing the reading intended by the sentence creator.

Further, when reading a sentence page by page, even if words are spread over pages, it is possible to control with a small memory by using the tag information for designating a page reading delimiter. Without
It is possible to provide a text-to-speech system that is easy to catch.

[Brief description of drawings]

FIG. 1 is a system configuration diagram of an embodiment of the present invention.

2 is a flow chart of the embodiment shown in FIG.

FIG. 3 is a flow chart of the embodiment shown in FIG.

FIG. 4 is a flow chart of the embodiment shown in FIG.

5 is an explanatory diagram of tag information in the embodiment shown in FIG.

6 is a diagram showing an example of tag information for designating the reading of words in the embodiment shown in FIG.

FIG. 7 is a diagram showing an example of tag information for designating word reading in the embodiment shown in FIG. 1;

8 is a diagram showing an example of tag information for designating a reading method of words in the embodiment shown in FIG.

9 is a diagram showing an example of tag information for prohibiting reading of words in the embodiment shown in FIG.

FIG. 10 is a diagram showing an example of tag information for designating page reading breaks in the embodiment shown in FIG. 1;

FIG. 11 is a system configuration diagram of a conventional example.

[Explanation of symbols]

 11 Texts mixed with Kanji and Kana 12 Tag information search section 13 Tag information analysis section 14 Display software 15 Display buffer 16 Text shaping means 17 Dot data creation means 18 Font data 19 Text display section 20 Reading software 21 Reading buffer 22 Text processing section 23 Speech synthesis Processing unit 24 Japanese dictionary 25 Voice data 26 Voice output unit 27 Carry-over buffer

Claims (5)

[Claims] 1. A kanji-kana mixed sentence to be processed is one in which tag information indicating that a predetermined processing operation is to be inserted is used, and the kanji-kana mixed sentence input every one page is read out. A sentence reading method, comprising: tag information searching means for searching tag information in a sentence containing the kanji and kana; tag information analyzing means for analyzing processing operation indicated by tag information detected as a result of the search; A text-to-speech system, comprising: a text-to-speech execution means that reads out the text containing the kanji and kana input according to the processing operation indicated by the tag information analyzed by the means. 2. The sentence reading system according to claim 1, wherein the tag information is sandwiched between a first symbol indicating the start of the tag information, a second symbol indicating the end of the tag information, and each symbol. The tag information searching means searches the tag information by searching the first and second symbols, and when the first symbol is detected, the tag information is continuously searched. The sentence reading system is characterized in that the tag information analyzing means analyzes the processing operation indicated by the tag information by the identifier when it is continuous. 3. The text-to-speech system according to claim 2, wherein there is a process of designating a reading as the type of processing operation indicated by the tag information, and the tag information includes an identifier for designating the reading and a phoneme / phonetic temperament. And the second tag information in which the identifier attached to the first tag information is further attached with a third symbol indicating the end of the processing range. A text-to-speech system characterized in that a character string between the first and second tag information is read as a phoneme / temperal in the first tag information. 4. The sentence reading method according to claim 2 or 3, wherein there is a process of prohibiting reading as a type of processing operation indicated by the tag information, and the tag information has an identifier indicating that reading is prohibited. And the fourth tag information in which the identifier attached to the third tag information is further attached with a third symbol indicating the end of the processing range. A text-to-speech system, which does not read a character string between the third and fourth tag information. 5. The sentence reading system according to claim 2, wherein there is tag information indicating a reading break as a type of processing operation indicated by the tag information, and the reading break is included as the tag information. Fifth tag information with an identifier indicating that is used is used, and the reading execution means does not continuously read the character string after the fifth tag information, and the next input kanji-kana mixed sentence. A text-to-speech system characterized by reading aloud along with it.