JP2002207494A - Speech synthesizer, method of synthesizing speech, and computer-readable storage medium with speech synthesizing program recorded thereon - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve the problem that proper reading according to contents of a text is impossible heretofore without detecting a keyword. SOLUTION: This speech synthesizer for outputting a synthesized tone corresponding to an inputted text is provided with a means to read in character information corresponding to the inputted text, a means to detect a characteristic of a format of the inputted text according to a character code or position information in the read in character information, and a means to generate the synthesized tone according to the characteristic of the detected format.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a speech synthesizer, a speech synthesis method, and a computer-readable recording medium on which a speech synthesis program is recorded. In particular, the present invention relates to a speech synthesizer, a speech synthesis method, and a computer-readable recording medium storing a speech synthesis program for detecting a feature related to a format of an input sentence and performing appropriate reading aloud according to the feature.

[0002]

2. Description of the Related Art In a conventional speech synthesizer, voice quality and prosodic features are uniform no matter what text is read. Japanese Patent Application Laid-Open No. 9-62286 discloses a speech synthesizer capable of providing a synthesized sound having an appropriate prosody according to an input sentence. The speech synthesizer detects a keyword from an input sentence, determines a field to which the input sentence belongs, and generates prosody information based on the determined field, thereby generating a synthesized sound having a different prosody for each field. ing.

[0003]

However, in the above-mentioned speech synthesizer, it is necessary to prepare a keyword for each field in advance, and it cannot be applied to a sentence that does not include a word that matches the keyword.

SUMMARY OF THE INVENTION It is a main object of the present invention to provide a speech synthesizer capable of appropriately reading aloud according to the contents of a sentence by detecting and analyzing features relating to the sentence format without detecting a keyword. And

[0005]

According to the present invention, there is provided a speech synthesizer for outputting a synthesized sound corresponding to an input sentence, wherein the format of the input sentence is determined based on a character code or position information in the read character information. Means for detecting features;
Means for generating the synthesized sound based on the detected characteristics of the format.

According to the present invention, it is possible to detect a feature of the format of an input sentence and generate a synthesized speech based on the feature, so that it is not necessary to prepare a keyword for each field in advance.
Also, a sentence that does not include a word that matches the keyword can be appropriately read aloud according to the format of the sentence.

Further, the present invention further comprises means for judging the type of sentence of the input sentence based on the detected characteristics of the format, and the generating means comprises a means for judging the type of sentence. A speech synthesizer characterized by generating the above-mentioned synthesized sound is provided.

According to the present invention, the type of sentence meaning can be determined from the characteristics of the format of the input sentence, and a synthesized speech can be generated based on the result of the judgment. It is.

Further, the present invention is characterized in that the generation means selects a prosody generation rule corresponding to the type of sentence meaning, and generates the prosody information of the synthesized speech based on the rule. Provide equipment.

According to the present invention, the prosody information is generated according to the type of sentence meaning, so that the input sentence can be read aloud according to the prosody corresponding to the sentence meaning of the input sentence.

Further, the present invention provides a speech synthesis apparatus, wherein the generating means selects a dictionary corresponding to the type of sentence meaning and performs a linguistic analysis of the input sentence based on the dictionary. I do.

According to the present invention, the dictionary corresponding to the type of sentence is preferentially used, so that a more natural reading can be performed according to the sentence of the input sentence.

[0013] Further, the present invention further comprises means for determining a break position of the input sentence based on the detected feature, and the generating means generates the synthetic sound based on the break position. A speech synthesizing apparatus is provided.

According to the present invention, an appropriate delimiter position is determined from the characteristics of the format of an input sentence to generate a synthesized speech, so that the input sentence can be read aloud smoothly.

Further, according to the present invention, when the input sentence is a text of an electronic mail and the difference in the number of characters in each line is equal to or less than a predetermined value, the determining means determines that the line feed code of each line in the input sentence is Provided is a speech synthesizing apparatus characterized in that an input sentence is determined so as not to be divided at an existing position.

According to the present invention, when performing linguistic analysis of a sentence, the sentence is often divided at the position of a line feed code. However, when the sentence of the electronic mail has little variation, that is, the number of characters in each line is substantially the same. In the case of, it is determined that the user has forcibly entered a line feed to match the number of characters on each line, and by determining that the sentence is not separated at the position of the line feed code on each line, an unnatural sentence The input sentence can be read out smoothly without interruption.

Further, according to the present invention, in a speech synthesizing method for outputting a synthesized sound corresponding to an input sentence, a step of reading a character code string corresponding to the input sentence, wherein a character code or a position in the read character information is read. A speech synthesis method, comprising: detecting a format feature of the input sentence based on information; and generating the synthesized sound based on the detected format feature.

According to the present invention, it is possible to detect a feature of the format of an input sentence and generate a synthesized speech based on the feature, so that it is not necessary to prepare a keyword for each field in advance.
Also, a sentence that does not include a word that matches the keyword can be appropriately read aloud according to the format of the sentence.

Further, the present invention provides a computer-readable recording medium on which a speech synthesis program for causing a computer to execute each step of the above speech synthesis method is recorded.

According to the present invention, it is possible to detect a feature of the format of an input sentence and generate a synthesized speech based on the feature, so that it is not necessary to prepare a keyword for each field in advance.
Also, a sentence that does not include a word that matches the keyword can be appropriately read aloud according to the format of the sentence.

[0021]

(Embodiment 1) FIG. 1 is a functional block diagram showing one embodiment of a speech synthesizing apparatus according to the present invention. Reference numeral 17 denotes a speech synthesizer according to the present embodiment.

The text in FIG. 1 is character information input to the speech synthesizer 17, for example, ASCII.
It is a data string represented by a so-called character code such as a code or a JIS code, or positional information indicating a character alignment or the like. In the present embodiment, a case will be described in which the speech synthesizer 17 is built in a general-purpose personal computer, and outputs a text such as an e-mail, a Web page, or a document in an electronic file format by speech synthesis. The present invention is not limited to this, and can be applied to all devices that perform speech synthesis on input text.

Referring to FIG. 1, the flow of processing in the speech synthesizer 17 from the input of a text to the output of a synthesized sound will be described.

The text input to the speech synthesizer 17 is subjected to a linguistic analysis of the text by a well-known morphological analysis or dictionary lookup in the language analysis unit 2, and information necessary for speech synthesis such as reading and prosody is converted to the text. Is given to Next, in the prosody information generation unit 3, prosody information is generated from information such as readings and prosody output from the language analysis unit 2,
The voice is synthesized by the voice synthesis unit 4, and a synthesized sound is output from the output unit 5.

On the other hand, the text input to the speech synthesizer 17 is input to the language analysis unit 2 as described above and also to the format analysis unit 1, and the character information corresponding to the text is read. Thereafter, it is determined whether or not the read character information matches a predetermined rule that specifies a characteristic of the format of the input sentence. If the character information matches, a characteristic of the format corresponding to the matched rule is detected. You. Then, the type of the meaning of the text is determined based on the detected characteristics of the format, and information indicating the type of the meaning is output. Here, the characteristics of the format of the input text include whether there is a bullet point included in the text, whether or not there is character alignment, whether or not there is a paragraph, the degree of variation in the number of characters in each line, and the occupation of kanji. The percentage, the percentage of the symbol, whether or not there is a quoted part, the type of end-of-sentence symbol, and whether or not there is an emphasis character. The type of sentence means that the content of the text is a news article, a report sentence, a notification sentence, an informal sentence, or a formal sentence. .

The information indicating the type of sentiment of the text analyzed by the format analyzing unit 1 is output to the language analyzing unit 2, the prosody information generating unit 3, and the speech synthesizing unit 4. , And a synthesized sound is generated. A specific example of generating a synthesized sound in accordance with the type of sentence will be described in Embodiments 2 to 4.

In the format analysis unit 1, the break position of the input text is determined based on the detected format characteristics of the input text, and the information indicating the break position is included in the language analysis unit 2 together with the information indicating the text characteristics.
Is output to A specific example of generating a synthesized sound based on a break position will be described in a fifth embodiment.

In FIG. 1, the dotted line from the format analyzer 1 indicates that the information generated based on the characteristics of the format of the input text is the language analyzer 2, the prosody information generator 3, or the speech synthesizer 4. Or a combination thereof.

FIG. 2 is a detailed functional block diagram of the format analyzer 1. The format analysis unit 1 reads character information corresponding to the input text (not shown), and the feature detection unit 15 matches the character information with a predetermined rule for specifying the characteristics of the format of the input sentence. Then, if they match, a feature corresponding to the matching rule is detected. Then, the format classification unit 16 determines the type of the meaning of the text based on the feature, and outputs information indicating the type of the meaning.

The feature detecting section 15 includes an item detecting section 6, a character shift detecting section 7, a paragraph detecting section 8, a character number variation detecting section 9, a kanji appearance rate calculating section 10, a symbol appearance rate calculating section 11, and a quoting section detecting section 12. , A sentence end symbol detection unit 13 and an emphasized character detection unit 14. In each unit, rules for detecting format features from the character information of the input text are set in advance. Note that the feature detection unit 15
It is not necessary to constitute all of the above-mentioned parts, and it may be constituted by a part thereof.

Next, the rules for detecting format features in each section of the feature detection section 15 will be described.

The bullet detecting section 6 detects whether or not the text has bullets. The bullets are detected, for example, as follows. That is, when the first characters of a plurality of lines are the same symbol (symbol “•” in FIG. 3), when the first characters of the plurality of lines are numbers, and when the arrangement of each number is a continuous number (FIG. 4) If the first character of a plurality of lines is the same symbol, the character following the first character is a number, and the sequence of each number is a consecutive number (FIG. 5), This is the case, for example, when the leading character strings are evenly allocated, and immediately after that, there is the same number of blanks (FIG. 6). In the above example shown in FIGS. 3 to 6, each of a plurality of lines indicating a bullet point is shown in units of one line, but each line may be a plurality of lines. Immediately after the equally assigned character string in FIG. 6 is not a blank but may be a symbol such as “:”.

In the character shift detecting section 7, the character string of each line is
It is detected whether there is a so-called right-justified, left-justified, or center-justified character alignment, or whether the aligned character string is equally allocated. Character alignment detection is
For example, it is performed based on the position information indicating the character alignment in the character information of the input text.

The paragraph detecting section 8 detects whether a paragraph exists in the text. A paragraph is determined to have a paragraph when a blank line is inserted between sentences.

The number-of-characters variation detecting section 9 counts the number of characters in each line and detects the variation in the number of characters in each line. When the number of characters in each line is substantially the same, it is determined that the degree of variation is low. Conversely, if the number of characters in each line is significantly different, it is determined that the degree of variation is high. That is,
The difference in the number of characters in each line is calculated, and if the difference value is equal to or less than a predetermined value, the degree of variation is determined to be low, and if the difference value is equal to or greater than the predetermined value, the degree of variation is determined to be high.

The kanji appearance rate calculation unit 10 extracts kanji from the entire text of the text and calculates the kanji appearance rate represented by the number of kanji in the total number of characters. Kanji is extracted by considering characters other than Hiragana, Katakana, numbers, and symbols as Kanji.

The symbol appearance rate calculation unit 11 extracts symbols from the entire text of the text and calculates the symbol appearance rate represented by the number of symbols in the total number of characters.

The quoted part detection unit 12 detects a quoted part of the electronic mail text. To detect a quote, the first character of each line is
For example, the determination is made based on whether the character is a preset quotation mark such as “>”.

The end-of-sentence symbol detection unit 13 detects whether the last character of each line is a period mark symbol or another symbol. The detection of the end-of-sentence symbol is determined based on whether or not the last character of each line is a symbol other than a predetermined period, such as “!”, “?”, “…”.

The emphasized character detecting section 14 detects whether or not there is an emphasized character in the text. Detection of emphasized characters
Judgment is made based on whether or not an underline or bold character is added to characters in the text.

As described above, the feature detecting section 15 detects a feature related to the format of the input text based on the character code in the read character information. That is,
For example, when the item detection unit 6 includes, in the read character information, a character code pattern in which the first character of a plurality of lines, which is a rule predetermined by the item detection unit 6, is the same symbol, The feature of the format is identified as having a bullet, and the information is detected.

The feature detecting section 15 detects features related to the format of the input text based on the position information in the read character information. That is, for example, in the character shift detection unit 7, in the read character information,
When the position information indicating that there is right alignment is included, the format feature of the input text is identified as having the right alignment, and the information is detected.

For the detection of each character of the input text in each section of the feature detection section 15, for example, to detect a ">" character, a character code corresponding to the character may be detected, and these processes are well known. This can be done with software technology.

The format classifying section 16 converts the input text based on the result detected by each section of the feature detecting section 15 into a well-written sentence such as a letter (e-mail) exchanged between friends, emphasized characters and character alignment. Judgment type such as a notice sentence that informs the schedule and the like, a report sentence that makes heavy use of bullets, and a news article, etc., is classified and classified, and information indicating the type of sentence Is output. In the above classification, for example, if the input text includes bullets, character alignments, and paragraphs and the number of characters in each paragraph does not vary, it is determined to be a notification because the format is well-formed. For example, the classification condition may be set based on an empirical rule.
As an example, FIG. 7 shows a flowchart illustrating the processing of the format classification unit 16 when the input text is an e-mail.

The processing of the format classification unit 16 will be described below with reference to FIG. If the quote section is detected by the quote section detection section 12, it is determined that the probability of exchanging emails between friends is high, and the text is classified into unreadable sentences (S201). If it is determined in S201 that there is no quoted portion and no paragraph is detected by the paragraph detection unit 8, it is determined that there is a high probability that friends are exchanging e-mails, and the text is classified into plain text. (S202). Even if a paragraph is detected in S202 and the value calculated by the kanji appearance rate calculation unit 10 is equal to or less than a predetermined value (K%), friends exchange emails because kanji is not frequently used. It is determined that the probability is high, and the text is classified into unreadable sentences (S2).
03). If it is not less than the predetermined value (K%) in S203 and the symbol appearance rate is larger than the predetermined value (Y%), the text is classified as a news article (S204). S2
04 is equal to or less than a predetermined value (Y%),
If the last character of the line is a symbol other than a punctuation mark, for example, "!" Or "?", It is judged that there is a high probability that a friend has exchanged mail, and the text is classified into plain text. Is done. In this case, it may be determined that at least one of the plurality of lines of the text satisfies the condition, or that the plurality of lines satisfies the condition more than a predetermined number. You may make it determine. If the last character of the line is a punctuation mark in S205 and the emphasized character is detected by the emphasized character detection unit 14 or the character shift is detected by the character shift detection unit 7, the text of the line is the schedule of the event. The notice is classified as a notice that emphasizes the layout relatively (S206). If both the emphasized character and the character alignment are not detected in S206, and the item detection is detected by the item detection unit 6, the text is classified into a formal sentence report (S207). S
If no bullets are detected at 207, the text is classified as plain text.

In the example shown in FIG. 7, the meaning of the input text is classified into four types. For example, the input text is classified into two types: a plain text such as an e-mail and a formal text such as a report text. In addition, adding somewhat formal text such as news articles to it and classifying it into three types,
The number and types of classification are not limited to this. Further, the example of FIG. 7 shows an example when the type classification of the input text is determined by the format classification unit 16, and it is not necessary to use all the conditions shown in the flowchart of FIG. The determination may be made using only the division.

(Embodiment 2) FIG. 8 is a functional block diagram showing another embodiment of the speech synthesizer according to the present invention. The same components as those in FIG. 1 are denoted by the same reference numerals, and the same description will be omitted.

The prosody generation rule selection unit 18 in the prosody information generation unit 3 selects a prosody generation rule corresponding to the information indicating the type of sentence output from the format analysis unit 1. The prosody generation rule is a parameter for generating a prosody such as a pitch length, a pitch variation width, a utterance speed, and a pause length which are set in advance according to the type of text. For example, information indicating the type of sentence When the sentence is a loose sentence, the pitch fluctuation width is set to be large, and is controlled so that the intonation is clearly generated. Also, if the information indicating the type of sentiment is informal, it contains a relatively large number of relatively simple words and is easy to hear, so the utterance speed is set faster and the pause length is set shorter. And is controlled to read out at the rate of human conversation. If the information indicating the type of sentence is a well-formed sentence such as a report sentence, the pitch length is set so as to be read out with a calm tone. If the information indicating the type of sentence is a sentence with a large number of bullets, such as a notification sentence, the utterance speed is set to be slow and the pause length is set to be long so that it is easy to hear. The above rule is an example. Instead of setting the rule in advance, the user may be allowed to set the rule freely.

As described above, the type of the sentence is classified from the format of the text, and the prosody information of the synthesized sound is generated in accordance with the type of the sentence. Therefore, the text is read out in the prosody corresponding to the sentence of the text. It becomes possible. In particular, when the input text is an e-mail and a plurality of e-mails are read aloud in succession, the content of the sentence can be grasped from the prosody since each e-mail is read out according to the prosody corresponding to the content. is there.

(Embodiment 3) FIG. 9 is a functional block diagram showing another embodiment of the speech synthesizer according to the present invention. The same components as those in FIG. 1 are denoted by the same reference numerals, and the same description will be omitted.

The dictionary selecting section 19 in the language analyzing section 2 selects a dictionary corresponding to the information indicating the type of sentence output from the format analyzing section 1. The selected dictionary is used in the language analysis unit 2. For example, if the information indicating the type of sentence is a plain text, a dictionary for spoken language is selected, and the language analysis unit 2 uses the dictionary for spoken language so as to be used in preference to a dictionary normally used. Control. Since the text of an e-mail exchanged between friends is often written in spoken language, selecting a spoken language dictionary enables more accurate morphological analysis.

If there is a symbol (such as "-" or "-") at the end of the sentence, which characterizes the spoken language, such as "○○ ー ー" or "XX Nanda ~" By setting the prosodic rules suitable for the symbols in the spoken language dictionary in advance, it becomes possible to generate a prosody closer to the spoken tone.

When it is determined that the item is an itemized item, the first character representing the itemized item (for example, the symbol "." Shown in FIG. 3) is set in the spoken word dictionary in advance so as not to be read out. You may do it.

In spoken sentences such as an e-mail sent between friends, a part of a word may be written in katakana, for example, "write XX and put squares." By setting such katakana in the spoken language dictionary so as to correct the hiragana, correct linguistic analysis can be performed in the linguistic analysis unit 2, and more natural reading aloud becomes possible.

(Embodiment 4) FIG. 10 is a functional block diagram showing another embodiment of the speech synthesizer according to the present invention.
The same components as those in FIG. 1 are denoted by the same reference numerals, and the same description will be omitted.

The voice quality selecting section 20 in the voice synthesizing section 4 selects a voice quality corresponding to the information indicating the type of sentence output from the format analyzing section 1. This selected voice quality
Used by the synthesized speech unit 4. The voice quality selection unit 20 prepares a plurality of voice qualities in advance, and for example, if the information indicating the type of sentence is a formal sentence such as a notification sentence or a report sentence, a low voice is used. Appropriate voice quality is set in advance in accordance with.

As described above, by analyzing and classifying the text format, it is possible to select a voice quality according to the content of the text, and it is possible to read out various texts with different voice quality. .

(Embodiment 5) FIG. 11 is a functional block diagram showing another embodiment of the speech synthesizer according to the present invention.
The same components as those in FIG. 1 are denoted by the same reference numerals, and the same description will be omitted.

The format analyzer 1 includes a feature detector 15
The information indicating the characteristics of the text detected in each part and the information indicating the position in the text are output to the delimiter position determination unit 21. That is, when a bullet is detected by the bullet detection unit 6 of the feature detection unit 15, information indicating that the bullet is detected and its positional information are output. If found,
Information indicating that the character misalignment has been detected and position information thereof are output. When a plurality of features are detected by each unit of the feature detection unit 15, all the information is output.

The delimiter position determining unit 21 uses the language analyzing unit 2 based on the information indicating the characteristics of the text and the position information.
Determines the breakpoints between sentences and words in the text analyzed by.

First, an example of a sentence separation rule for determining a position for separating sentences will be described.

For example, if the input text has a bullet, the contents of the sentence are divided for each bullet, so that the line of the bullet is determined to be divided at the line feed position. You can read each line of the bullet smoothly. That is, when detecting the information indicating that the text has bullet points, the break position determining unit 21 instructs the linguistic analysis unit 2 to break the sentence at the position of the line feed code for the bulleted line. It is good to instruct. Similarly, in the case where the text has a character alignment, the content of the sentence is different between the line of the character alignment and the next line. Therefore, the delimiter position determination unit 21 detects the information indicating the characteristic of the character alignment in the text. In this case, the language analysis unit 2 is instructed to separate the sentence at the position of the line feed code for the line of the character alignment. By doing so, it becomes possible to smoothly read the bulleted line and the next line.

In the case where the number of characters in each line of the input text is small in an electronic mail or the like, the user may input a line break in order to make the plurality of lines easy to see as a whole by adjusting the number of characters in each line. Since there are many cases, if the line break is determined not to separate sentences, the line and the next line can be read out smoothly. That is, when the degree of variation is low, that is, when information indicating a characteristic that the difference in the number of characters in each line is equal to or smaller than a predetermined value is detected, the break position determining unit 21 instructs the language analysis unit 2 to insert a line feed code in each line. And ignore the sentence at that position.

Also, when there is a paragraph in the input text in an electronic mail or the like, since the content of the sentence is often different for each paragraph, if it is determined that the sentence is not delimited by a line feed in the paragraph, You can read sentences smoothly. That is, when detecting the information indicating the feature that the text has a paragraph,
What is necessary is just to instruct the linguistic analysis unit 2 to ignore the line feed code in the paragraph and not to break the sentence at that position. If there is a line with an extremely small number of characters in one paragraph, there is a high possibility that the content of the sentence has changed at that position, and the sentence may be separated by a line feed code at the end of the line.

If the input text includes a quoted part in an electronic mail or the like, it is highly likely that a line feed code was forcibly inserted by a mail server or the like when replying from the other party. If you decide not to break the sentence in the line feed, you can read the sentence smoothly.
That is, when detecting the information indicating the feature that the text has a quoted portion, the break position determining unit 21 ignores the line feed code in the quoted portion and separates the sentence at the position. Just tell them not to.

As in the above example of the sentence separation rule, when the sentence separation is determined from the characteristics of the format of the input text, a symbol indicating a sentence separation such as a period is detected and the sentence is separated at that position. In comparison with, for example, it is possible to suppress erroneous detection of a sentence delimiter in a case where there is no punctuation in a line in which a headline of a sentence is described, or a case where a line feed is mistakenly made in the middle of a sentence. Note that the above-described example of the sentence separation rule may be combined with a technique using a symbol indicating a sentence separation to determine the sentence separation of the input text.

Next, a description will be given of a word division rule when there is a space (a so-called space) between characters. In this word separation rule, in a character string in which a space is inserted between characters, it is determined whether to delete the space and combine characters before and after the space.

For example, if the input text has a bullet,
If the character strings are evenly aligned, blanks are deleted from the uniformly allocated character strings, and characters before and after the blanks are combined. That is, the break position determining unit 21
If the text contains information indicating that there is an itemized bullet and information indicating that the text has an equal assignment, the space is deleted from the language analysis unit 2 and the characters before and after the space are deleted. Order to join. By doing so, the character string can be analyzed as a word in the language analysis unit 2 without separating the character string by a blank.

As in the above-described example of the word separation rule, by determining the word separation from the characteristics of the format of the input text, a morphological analysis is performed by combining sentences before and after a line feed or before and after a space. Thus, the amount of calculation can be kept low.

The first to fifth embodiments described above.
It is also possible to provide a voice synthesizing device obtained by arbitrarily combining the components of the voice synthesizing device in the above. It is apparent that the same operation and effect as described in each embodiment can be obtained even when the respective configurations are arbitrarily combined.

Further, a part or all of the processing in each of the above-described embodiments can be provided as a program (program) comprising an ordered sequence of instructions suitable for processing by a computer. In addition, the program can be provided as a computer-readable recording medium on which the program is recorded for installation, execution, and distribution of the program.

[0072]

According to the present invention, since the features of the format of the input text are detected and speech synthesis is performed based on the features, it is possible to read out the input text appropriately according to the meaning of the text.

Further, according to the present invention, an appropriate delimiter position is determined from the characteristics of the format of the input text to generate a synthesized sound, so that the input text can be read aloud smoothly.

[Brief description of the drawings]

FIG. 1 is a functional block diagram of a speech synthesizer according to Embodiment 1 of the present invention.

FIG. 2 is a functional block diagram of a format analysis unit according to the first embodiment of the present invention.

FIG. 3 is a first example of a schematic diagram showing a format of a text having an itemized list according to the first embodiment of the present invention.

FIG. 4 is a second example of a schematic diagram showing a format of a text having bullet points according to the first embodiment of the present invention.

FIG. 5 is a third example of a schematic diagram showing a format of a text having bullet points according to the first embodiment of the present invention.

FIG. 6 is a fourth example of a schematic diagram showing a format of a text having bullet points according to the first embodiment of the present invention.

FIG. 7 is a flowchart of a format analysis unit according to Embodiment 1 of the present invention.

FIG. 8 is a functional block diagram of a speech synthesizer according to Embodiment 2 of the present invention.

FIG. 9 is a functional block diagram of a speech synthesizer according to Embodiment 3 of the present invention.

FIG. 10 is a functional block diagram of a speech synthesis device according to a fourth embodiment of the present invention. You.

FIG. 11 is a functional block diagram of a speech synthesizer according to Embodiment 5 of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Format analysis part 2 Language analysis part 3 Prosody information generation part 4 Voice synthesis part 17 Voice synthesis device

Claims (8)

[Claims] 1. A speech synthesizer for outputting a synthesized sound corresponding to an input sentence, comprising: means for reading character information corresponding to the input sentence; and a character code or position information in the read character information. A speech synthesizer comprising: means for detecting a feature of a format of an input sentence; and means for generating the synthetic sound based on the feature of the detected format. 2. The apparatus according to claim 1, further comprising: a unit configured to determine a type of sentence of the input sentence based on the detected characteristic of the format. The generating unit generates the synthesized sound according to the type of sentence. The speech synthesis device according to claim 1, wherein the speech synthesis device generates the speech. 3. The method according to claim 2, wherein said generating means selects a prosody generation rule corresponding to the type of sentence meaning, and generates the prosody information of the synthesized sound based on the rule. Speech synthesizer. 4. The speech synthesizer according to claim 2, wherein said generating means selects a dictionary corresponding to the type of sentence meaning, and performs language analysis of the input sentence based on the dictionary. . 5. The apparatus according to claim 5, further comprising: means for determining a break position of the input sentence based on the detected feature of the format, wherein the generating means generates the synthetic sound based on the break position. The speech synthesizer according to claim 1, wherein: 6. When the input sentence is an electronic mail sentence and the difference in the number of characters in each line is equal to or less than a predetermined value, the determining means determines whether or not a line feed code exists in each line of the input sentence. The speech synthesizer according to claim 5, wherein the input sentence is determined so as not to be divided. 7. A speech synthesizing method for outputting a synthesized speech corresponding to an input sentence, wherein a character code string corresponding to the input sentence is read, and a character code or position information in the read character information is read. A speech synthesis method comprising: detecting a format feature of the input sentence; and generating the synthesized speech based on the detected format feature. 8. A computer-readable recording medium on which a speech synthesis program for causing a computer to execute each step of the speech synthesis method according to claim 7 is recorded.