JP3284976B2 - Speech synthesis device and computer-readable recording medium - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizer, and more particularly, to a speech synthesizer for reading an arbitrary text.

[0002]

2. Description of the Related Art When a text is read aloud by speech synthesis, it is desired that the text be read aloud according to the content of the text. For example, if the text "123-4567" means a telephone number, it is desirable to read out "Ichi ni Nissan no Yogo Ogana", and if the text means a mathematical formula, "Hyaku Niju Sanyo Yosui" It is hoped that it will be read out. Also, if there is a table in the text,
"-", "|", Etc., which are used as ruled lines in the table, have only a visual meaning and are not preferably read out.

[0003] The easiest way to realize the reading according to the content of the text is to have the user specify the reading, that is, the reading mode. An example is shown in FIG. This speech synthesizing apparatus comprises a speech synthesizing unit 4, an input unit 1 for inputting text, and a mode selecting unit 6 for designating a speech mode instruction from a user to the speech synthesizing unit 4. The text synthesis unit 4 reads the text in accordance with the reading mode specified by the mode selection unit 6.

[0004] For example, "math formula reading" and "telephone number reading"
In the state where the user has designated “read phone number” by the mode selecting means 6 and the text “123-4567” is input by the input means 1, the voice synthesizing means 4 In accordance with the instruction of "phone number reading" specified by the mode selection means 6, the numbers are read as sticks ("123" is read as "ichini-san"), and "-" is read as "no", and "ichi-ni-san-no-go-roku". What? " On the other hand, in a state where the user designates “reading of mathematical formula” by the mode selecting means 6, “123−4567 = −444” is inputted by the input means 1.
When the text "4" is input, the voice synthesizing means 4
In accordance with the instruction of “formula reading” specified by the mode selection means 6, the number is digit reading (“123” is read as “hypnotic”), and “−” is “manas”.
Say, "Hyaku-ninju-san-sen-sen-kyo-hyaku-kuro-juna-in-no-kyo-kun-sui-sen-yon-yon-hyaku-kun-juyon."

As described above, it is possible to correctly read out the same character string according to the meaning of the text by selecting the reading mode.

However, in the above-described speech synthesizer in which the user specifies the reading mode, the user needs to know the contents of the text to be read in advance, that is, before the text is read out. Therefore, it cannot be applied when the text to be read out is unknown to the user. Further, since it is necessary to provide the mode selecting means 6 for inputting the instruction of the reading mode from the user on the voice synthesizing device side, necessary information is obtained by voice from the voice synthesizing device installed at a remote place through a telephone line, for example. It is difficult to apply in such cases.

To solve such a problem, it is necessary for the speech synthesizer itself to automatically determine the reading mode according to the contents of the text. A conventional example of such a technique is described in JP-A-8-36395.

According to the technique described in the publication, when a numeric character string of a predetermined special word or special pattern is detected in the text, a numeric character string or a numeric character string of a special pattern existing before and after the special word. Then, the reading according to the rules corresponding to the special word or the special pattern is applied to automatically determine the reading mode according to the content of the text.

[0009]

In the above-mentioned prior art, a special character string that is a special reading is focused on a point that includes a special word or appears as a special pattern. This is a useful technique as one of techniques for automatically determining a reading mode in response.

However, for a special pattern such as "numerical-numerical", when using the rule that the numeral is read as a bar and "-" is read as "no", "123-4567" is used.
The text says, "I'm not happy with Ichinini-san."
However, the text "123-4567 = -4444" is also read as "Ichinii-san's not afraid." To improve this, new special patterns and rules need to be added, not just the special patterns and rules described above. In this example, the number of special patterns and rules to be added may not be so large, but when automatically determining the mode in which the ruled lines of the table are not read out, "-", "|" , Etc.,
In other words, since special patterns vary widely, it is necessary to register a huge number of special patterns, which makes practical application difficult.

SUMMARY OF THE INVENTION An object of the present invention is to provide a speech synthesizing apparatus capable of automatically determining an appropriate reading mode according to the contents of text without detecting a special pattern.

[0012]

SUMMARY OF THE INVENTION A speech synthesizer according to the present invention determines a reading mode based on the appearance frequency of each character in a text. More specifically, input means for inputting text, and text of characters in the input text
The apparatus includes frequency extracting means for extracting the appearance frequency in the whole, mode determining means for determining a reading mode based on the frequency information, and speech synthesizing means operating in accordance with the determined reading mode.

[0013] In the speech synthesizer of the present invention configured as above, enter the text to be read aloud input means, the frequency extraction means, character Te in the text
The appearance frequency in the entire text is extracted, the mode determining means determines the reading mode based on the extracted frequency information, and the voice synthesizing means reads the text in the determined reading mode.

Further, the speech synthesizing apparatus of the present invention divides an input text so that even if sentences in various formats are mixed in a text, each can be read out in an appropriate reading mode. Then, processing is performed for each divided text. More specifically, an input unit for inputting a text, a dividing unit for dividing the input text, a frequency extracting unit for extracting, for each divided text, an appearance frequency of a character in the text in the entire text , A mode determining unit that determines a reading mode for each of the divided texts based on the frequency information, and a speech synthesis unit that operates in accordance with the determined reading mode for each of the divided texts.

In the speech synthesizing apparatus according to the present invention, the text to be read aloud input by the input means is divided by the dividing means, and the frequency extracting means outputs the text for each divided text. The entire text of the middle character
Extract the frequency of appearance, based on the mode determination unit the frequency information determines a reading mode for each divided text, read text in reading mode speech synthesis means is the decision.

The type and number of reading modes to be automatically switched by the present invention are arbitrary. As an example, the present invention can be applied to mutual switching between a mode in which specific symbol characters such as “−”, “|”, and “*”, which are frequently used for ruled lines in a table, are not read, and a mode in which they are read. In this case, the frequency extracting means particularly extracts the frequency of appearance of the specific symbol character, and the mode determining means determines whether the mode of reading the symbol character is not read or the mode of reading the symbol character based on the frequency of appearance of each symbol character. To decide.

As another example, the present invention can be applied to switching between a mode in which an alphabet character string is read out as Roman characters and a mode in which it is read out as English. In this case, the frequency extracting means particularly extracts the appearance frequency of the alphabet vowels and consonants, and the mode determining means reads the alphabet character string as a Roman character based on the appearance frequency of the alphabet vowels and consonants, and reads out as English. Determine one of the modes. Then, in a more preferred embodiment, the voice synthesis means determines that a mode in which the alphabet character string is read out as English is determined.
The numbers and symbols in the text are read aloud in English.

As another example, the present invention can be applied to a mode of reading a numeric character string including a specific symbol character as a telephone number and a mode of reading a numeric character string as a mathematical expression.
In this case, for example, the frequency extracting means may be “−”, “+, =,
The mode determination means extracts the frequency of occurrence of a specific symbol character such as “×, ÷”, and the mode determining means reads out a numeric character string as a telephone number or a mode to read as a mathematical expression based on the frequency of occurrence of the specific symbol character. To decide.

[0019]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, embodiments of the present invention will be described in detail with reference to the drawings.

(1) First Embodiment FIG. 1 is a block diagram of a first embodiment of the present invention.
The speech synthesizing apparatus according to the present embodiment includes an input unit 1 for inputting a text, a frequency extracting unit 2 for extracting a frequency of appearance of a character in the input text, and a reading mode determined based on the frequency information. The apparatus includes a mode determining unit 3 and a voice synthesizing unit 4 that operates according to the determined reading mode.

The input means 1 inputs an arbitrary text to be read out, such as a text input from a keyboard (not shown) or a text stored in a hard disk or the like, and passes the text to the frequency extracting means 2 and the speech synthesizing means 4. .

The frequency extracting means 2 extracts the appearance frequency of the character in the text received from the input means 1 and passes the frequency information to the mode determining means 3. The appearance frequency is, for example,
It is extracted for each character such as "A" and "I" and for each character type (category) such as "symbol" and "Kanji". Further, it is not always necessary to extract the appearance frequencies of all characters and types. Which character or type of appearance frequency is to be extracted is determined in advance according to the type of the reading mode to be automatically switched.

The mode determining means 3 determines a reading mode based on the frequency information received from the frequency extracting means 2 based on a predetermined rule. Which rule is used is determined in advance according to the type of the reading mode to which the speech synthesizer is to be automatically switched.

The voice synthesizing means 4 reads out the text provided from the input means 1 in accordance with a reading mode instruction from the mode determining means 3. FIG. 2 shows a configuration example of the voice synthesizing means 4.

The voice synthesizing means 4 of this example includes a text analyzing means 41 for analyzing text and generating phonetic symbols, a mode storing means 42 for storing a reading mode, a word dictionary 43 for storing words and their phonetic symbols. And a speech dictionary 44 for storing speech waveforms in phoneme units and speech waveform generation means 45 for producing speech waveforms from phonetic symbols.

The mode storage means 42 stores a reading mode instruction from the mode determination means 3. Word dictionary 43
Stores a word, its pronunciation, and phonetic symbols that are information on accents. For words that need to be read separately in each reading mode, their phonetic symbols (including those not read) are stored in the word dictionary 43 for each reading mode.
The text analysis unit 41 divides the text provided from the input unit 1 into words and converts the text into phonetic symbols according to the contents of the word dictionary 43 and the mode storage unit 42 for each word. The waveform dictionary 44 stores a speech waveform for each phoneme and its phoneme label. The speech waveform generating means 45 searches for, extracts, and edits a necessary phoneme unit speech waveform from the waveform dictionary 44 using the phoneme label as a key based on the phonetic symbols output from the text analysis means 41. Generates an audio waveform.

Next, examples of the present embodiment will be described in detail with reference to the drawings.

(A) First Example FIG. 3 is a flowchart of a first example of the present embodiment, and FIGS. 4 and 5 show text examples. The voice synthesizing apparatus according to the present embodiment has a function of automatically selecting a mode in which the symbol character is not read out or a mode in which the symbol character is read out based on the appearance frequency of the symbol character.

In the case of this embodiment, the frequency extracting means 2 extracts the frequency of appearance of a character in a text. In particular, for a symbol character having only a visual meaning in a text, the frequency of occurrence is determined for each symbol character. Extract. The target symbol characters are “−”, “|”, “*” and the like. Further, the mode determining means 3 is preset with a rule that, when the appearance frequency of the symbol character is higher than a predetermined frequency, the mode is set to a mode in which the symbol character is not read out, and otherwise, a mode in which the symbol character is read out is set.

The operation of this embodiment will be described below with reference to FIGS.

First, the input means 1 inputs a text and passes it to the frequency extracting means 2 and the speech synthesizing means 4 (step S).
1). Next, the frequency extracting means 2 calculates the frequency of appearance of the characters in the text. At this time, the frequency of occurrence of predetermined symbol characters such as “−”, “|”, “*” is determined for each symbol character. Is obtained (step S2). Next, the mode determining means 3 determines the magnitude of the appearance frequency of each symbol character (step S3), and adds the symbol character having a remarkably high frequency to a list of unreadable characters (step S4). The list thus generated is stored in the mode storage unit 42 of the speech synthesis unit 4 as the information on the reading mode. If there are no unread characters, the list is empty.

The text analysis means 41 of the speech synthesis means 4
When the above list is stored in the mode storage means 42,
Start processing the input text. First, the symbol characters in the list stored in the mode storage unit 42 are deleted from the text (step S5). Then, the text is divided into words (step S6), and the word dictionary 4
3 is converted into phonetic symbols for each word (step S
7). Next, the audio waveform generation unit 45 generates an audio waveform by acquiring and editing a necessary phoneme unit audio waveform by referring to the waveform dictionary 44 based on the phonetic symbols output from the text analysis unit 41. (Step S8). And
Whether to output the audio waveform electric signal directly to a speaker or the like,
Output to another place through the telephone line (step S9).

Since the speech synthesizer of this embodiment operates as described above, for example, when a text as shown in FIG. 4 is input, one symbol character "*" and one other character
7 characters and "*" because the frequency of occurrence of "*" is low.
Is not erased, and is therefore read aloud as "usually used when writing tomorrow." On the other hand, in the case of the text shown in FIG. 5, for example, there are 36 symbol characters “*” and 26 other characters,
Since the appearance frequency of “*” is extremely high, it is deleted from the text, and only the sentences in the frame drawn with “*” are read out.

In this embodiment, the text analysis means 4
Before the word reading stage in No. 1, the unread characters were deleted from the text so that the unread characters were not read aloud. However, "-",
Regarding predetermined symbol characters such as “|” and “*”, at the stage of reading words, if the symbol characters are present in the list, no phonetic symbol is added, so that reading is not performed.

(B) Second Example FIG. 6 is a flowchart of a second example of the present embodiment, FIGS. 7 and 8 show examples of text, and FIG. 9 shows an example of reading out the text of FIG. The speech synthesizer of the present embodiment
Based on the frequency of appearance of alphabet vowels and consonants, it has a function to automatically select a mode for reading alphabetic character strings as Roman characters and a mode for reading English characters as English.

In the case of this embodiment, the frequency extracting means 2 extracts the appearance frequency for each of the alphabet vowels (a, i, u, e, o) and the alphabet consonants. For example, when the value of the appearance frequency of an alphabet vowel divided by the appearance frequency of an alphabet consonant is larger than a predetermined value, the mode determining means 3 sets a mode in which the alphabet character string in the text is read out as Roman characters. Is set in advance so that the mode is set to the mode of reading out as English.

The operation of this embodiment will be described below with reference to FIGS. 1, 2, and 6 to 9.

First, the input means 1 inputs a text and passes it to the frequency extracting means 2 and the speech synthesizing means 4 (step S).
11). Next, the frequency extracting means 2 accumulates the number of appearances for each of the alphabet vowels and consonants, and obtains the appearance frequency of each category (step S12). Next, the mode determining means 3 calculates the value of the appearance frequency of the alphabet vowel divided by the appearance frequency of the alphabet consonant, and if the value is larger than a predetermined value, determines the Roman reading mode. The reading mode is determined and set in the mode storage unit 42 of the speech synthesis unit 4 (steps S13 to S15).

Text analysis means 41 of speech synthesis means 4
Starts processing on the input text when the mode is set in the mode storage means 42. First, the text is divided into words (step S16). Next, if the Roman character reading mode is set in the mode storage means 42, the alphabet in the text is regarded as Roman characters and converted into Hiragana (steps S17, S19). For example,
“Tanaka” is converted to “tanaka”. If the English reading mode is set in the mode storage means 42, the alphabet part in the text is left as it is. However, for numbers and symbols in the text,
Converted to English reading (steps S17, S1
8). For example, "24" is "twenty-four"
Converted to Next, the text analysis unit 41 refers to the word dictionary 43, converts each word into a phonetic symbol, and outputs it to the speech waveform generation unit 45 (step S20).

The voice waveform generation means 45 generates a voice waveform by acquiring and editing a required phoneme unit voice waveform by referring to the waveform dictionary 44 based on the phonetic symbols output from the text analysis means 41. Then, an electric signal of an audio waveform is directly output to a speaker or the like, or is output to another place through a telephone line (step S22).

Since the speech synthesizing apparatus of this embodiment operates as described above, for example, when an English text as shown in FIG. The frequency information that the consonant appearance frequency = 36 characters is obtained, and the frequency of the consonant is higher than that of the vowel, so that the mode determining means 3 determines the English reading mode. As a result, the text in FIG. 7 is read out as it is in English. Further, although the numbers and symbols do not exist in the text of FIG. 7, if they exist, they are read out in English. On the other hand, for example, when the Roman text shown in FIG.
The frequency information that 52 characters and the appearance frequency of alphabet consonants = 46 characters is obtained, and the frequency of vowels is higher than consonants, so that the mode determining means 3 determines the Roman character reading mode. As a result, the text of FIG. 8 is read out as shown in FIG.

In this embodiment, the Roman alphabet reading mode and the English reading mode are determined based on whether the value of the appearance frequency of alphabet vowels ÷ the appearance frequency of alphabet consonants is greater than a predetermined value. The difference between the frequency of occurrence of
It is also possible to determine the mode based on the ratio of the appearance frequency of alphabet vowels or consonants to the appearance frequency of alphabets (vowels and consonants).

(C) Third Example FIG. 10 is a flowchart of a third example of the present embodiment, FIGS. 11 and 13 are text examples, FIG. 12 is a text-to-speech example of FIG. 11, and FIG. 14 illustrates an example of reading out the text of FIG. 13. The voice synthesizing apparatus according to the present embodiment extracts a frequency of appearance of a specific symbol character, and based on the frequency of occurrence, a mode in which a numeric character string including the specific symbol character is read as a telephone number, and a mode in which the character string is read as a mathematical expression. Has a function to select automatically.

In the case of this embodiment, the frequency extracting means 2 extracts the frequency of occurrence of a specific symbol character in a text. The target symbol character is, in the case of this embodiment,
“=, +, ÷, ×”. Further, the mode determining means 3 sets a mode in which, when the frequency of occurrence of the specific symbol character is larger than a predetermined value, a numeric character string including the specific symbol character is read as a mathematical expression, and otherwise, the mode is read as a telephone number. Is set in advance.
Further, the word dictionary 43 stores “Manasu” in the case of the mathematical expression reading mode and “No” in the case of the telephone number reading mode as pronunciation symbols of the symbol character “−”.

The operation of this embodiment will be described below with reference to FIGS. 1, 2, and 10 to 14.

First, the input means 1 inputs a text and passes it to the frequency extracting means 2 and the speech synthesizing means 4 (step S).
31). Next, the frequency extracting means 2 outputs the symbol characters “=, +,
The number of appearances in the text “{, ×” ”is accumulated, and the overall appearance frequency is obtained (step S32). Next, when the appearance frequency is larger than a predetermined value, the mode determining means 3 determines the mode as the numerical expression reading mode, and otherwise, determines the mode as the telephone number reading mode. (Steps S33 to S3)
5).

The text analysis means 41 of the speech synthesis means 4
Starts processing on the input text when the mode is set in the mode storage means 42. First, the text is divided into words (step S36). Next, if the formula reading mode is set in the mode storage means 42, the numbers in the text are converted to digit reading (steps S37, S37).
38). For example, “24” is converted to “Nijyo”. If the telephone number reading mode is set in the mode storage means 42, the numbers in the text are converted into stick readings (steps S37, S39). For example, "24" is converted to "Niyon". Next, the text analysis means 41
Is converted into phonetic symbols for each word with reference to the word dictionary 43 and output to the speech waveform generating means 45 (step S4).
0). At this time, "-" in the text is converted to a phonetic symbol of "Manai" in the mathematical expression reading mode and "No" in the telephone number reading mode.

The voice waveform generating means 45 generates a voice waveform by acquiring and editing a required voice element-based voice waveform by referring to the waveform dictionary 44 based on the phonetic symbols output from the text analysis means 41. Then, an electric signal of an audio waveform is directly output to a speaker or the like, or is output to another place through a telephone line (step S42).

The speech synthesizing apparatus of this embodiment operates as described above. For example, when a text in which names and telephone numbers are listed as shown in FIG. Frequency information that the appearance frequency of “=, +, ÷, ×” = 0 is obtained, and the mode determination unit 3 determines the telephone number reading mode. As a result, the text in FIG. 11 is read out as in FIG. On the other hand, for example, when the text of the mathematical formula shown in FIG. 13 is input, the frequency extraction means 2 obtains the frequency information of the appearance frequency of the symbol character “=, +, ÷, ×” = 7. The mode determining means 3 determines the telephone number reading mode. As a result, the text in FIG. 13 is read out as shown in FIG.

In this embodiment, the symbol characters "=, +,
When the appearance frequency of “÷, ×” is larger than a predetermined value, the formula reading mode is set. Otherwise, the phone number reading mode is set. However, instead of the symbol characters “=, +, ÷, ×”, the symbol characters “ The appearance frequency of “−” may be determined, and the phone number reading mode may be determined when the appearance frequency is greater than a predetermined value, and the mathematical expression reading mode may be determined otherwise.

(2) Second Embodiment FIG. 15 is a block diagram of a second embodiment of the present invention. The speech synthesizing apparatus according to the present embodiment includes an input unit 1 for inputting a text, a dividing unit 5 for dividing the input text, and a frequency extraction for extracting an appearance frequency of a character in the text for each divided text. Means 2, mode determining means 3 for determining a reading mode for each divided text based on the frequency information, and speech synthesizing means 4 operating in accordance with the determined reading mode for each divided text. I have.

The input unit 1 inputs an arbitrary text to be read out, such as a text input from a keyboard (not shown) or a text stored in a hard disk or the like, and passes the text to the dividing unit 5.

The dividing means 5 divides the text into units such as sentences, paragraphs, etc., using a space, line feed, "." The divided text is a frequency extracting means 2
And passed to the speech synthesis means 4.

The frequency extracting means 2 extracts the appearance frequency of the character in each of the divided texts received from the input means 1, and passes the frequency information to the mode determining means 3. The appearance frequency is, for example, for each character such as "a", "i", "symbol",
It is extracted for each character type (category) such as "Kanji". Further, it is not always necessary to extract the appearance frequencies of all characters and types. Which character or type of appearance frequency is to be extracted is determined in advance according to the type of the reading mode to be automatically switched.

The mode determining means 3 determines the reading mode for each of the divided texts based on the frequency information received from the frequency extracting means 2 based on a predetermined rule. The rule to be used is determined in advance according to the type of the reading mode to be automatically switched.

When the text-to-speech mode is determined by the mode determining means 3, the speech synthesizing means 4 reads out the text given from the dividing means 5 according to the instruction of the text-to-speech mode from the mode determining means 3 for each divided text. The voice synthesizing means 4 is configured as shown in FIG. 2, for example, as in the first embodiment.

FIG. 16 is a flowchart of the present embodiment. First, the input unit 1 inputs a text to be read out and passes it to the dividing unit 5 (step S51). Next, the dividing unit 5 divides the text, and passes each divided text to the frequency extracting unit 2 and the speech synthesizing unit 4 (step S52). Next, the processing by the frequency extracting means 2, the mode determining means 3, and the speech synthesizing means 4 is repeated for all the divided texts (steps S53, S54).

In the embodiment in which the mode of not reading out the symbol character or the mode of reading out the symbol character is automatically selected based on the frequency of appearance of the symbol character, for example, the content of the processing for one divided text in step S54 is as follows. This corresponds to the processing of steps S2 to S8 in FIG. Further, in the embodiment in which the mode in which the alphabet character string is read out as Roman characters and the mode in which it is read out as English are automatically selected based on the appearance frequencies of the alphabet vowels and consonants, the processing in steps S12 to S21 in FIG. Equivalent to. Further, based on the frequency of occurrence of a specific symbol character, a mode in which a numeric character string including the specific symbol character is read as a telephone number and a mode in which a numerical character string is read as a mathematical expression are automatically selected. Step S32 of 10
To S41.

Finally, the voice synthesizing means 4 outputs the electric signal of the voice waveform generated for each of the divided texts directly to a speaker or the like, or to another place through a telephone line (step S54).

Next, examples of the present embodiment will be described in detail with reference to the drawings. As an example, an embodiment will be described in which a mode in which the symbol character is not read out and a mode in which the symbol character is read out are automatically selected based on the appearance frequency of the symbol character. The text shown in FIG. 17 is used as the text to be read. This text is a text of an e-mail, and is composed of three parts: a normal text, a table, and a sender information field called a signature. Of these, "-" and "|" used as the ruled lines of the table, and "*" which is the signature frame have only a visual meaning, so it is desirable not to read them out.

In the case of such an embodiment, the frequency extracting means 2
Extracts the frequency of appearance of characters in the segmented text, but extracts the frequency of appearance of each symbol character especially for symbol characters having only a visual meaning. The target symbol character is
"-", "|", "*" And the like. Further, the mode determining means 3 is preset with a rule that, when the appearance frequency of the symbol character is higher than a predetermined frequency, the mode is set to a mode in which the symbol character is not read out, and otherwise, a mode in which the symbol character is read out is set. Note that step S5 in FIG.
In step 4, the processes of steps S2 to S8 in FIG. 3 are executed.

In operation, first, the input means 1 is operated as shown in FIG.
7 and input it to the dividing means 5 (step S
51). Next, the dividing means 5 divides the text into three based on blank lines, namely, the first to third lines, the fifth to eleventh lines, and the thirteenth to sixteenth lines (step S52). The divided text is passed to the frequency extracting means 3 and the speech synthesizing means 4 in order from the head of the text.

Next, the divided texts in the first to third lines are processed first (steps S53 and S54). Specifically, the frequency extracting means 2 extracts the frequency of appearance of each character in the divided text, and particularly for predetermined symbol characters such as "-", "|", "*", for each symbol character. The appearance frequency is obtained (step S2 in FIG. 3). Next, the mode determining means 3
The magnitude of the appearance frequency is determined for each symbol character (step S3), and the symbol character with an extremely high frequency is added to a list of unreadable characters (step S4). 17 of FIG.
In the case of the divided text on the third line, the specific symbol character is "-"
The list is empty because there is only one and of course less frequently. This list is stored in the mode storage unit 42 of the speech synthesis unit 4 as information on the reading mode of the first divided text.
Is stored.

The text analysis means 41 of the speech synthesis means 4
When the above list is stored in the mode storage means 42,
Start processing for the first split text. First, the symbol characters in the list stored in the mode storage unit 42 are deleted from the text (step S5). In this case,
Since the list is empty, no symbol characters are deleted. And
The text is divided into words (step S6), and converted into phonetic symbols for each word with reference to the word dictionary 43 (step S7). Next, the audio waveform generation unit 45 generates an audio waveform by acquiring and editing a necessary phoneme unit audio waveform by referring to the waveform dictionary 44 based on the phonetic symbols output from the text analysis unit 41. (Step S8).
The generated voice waveform is stored until the time of voice output.

When the processing for the first divided text is completed, the divided texts in the fifth to eleventh lines in FIG. 17 are processed (steps S53 and S54). First, frequency extraction means 2
Calculates the appearance frequency in the divided text for each predetermined symbol character such as "-", "|", "*" (step S2 in FIG. 3) as in the previous case. Each appearance frequency is compared with a predetermined value (step S
3) If the value is larger than the value, the symbol character is added to the list of unread characters (step S4). 5-1 in FIG.
In the case of the divided text on the first line, frequency information of, for example, 2 Chinese characters, 6 alphabets, 30 numbers, 107 symbols “−”, 14 symbols “|”, and 4 symbols “,” is obtained. Because “−” and “|” are frequent,
"-" And "|" are determined to be unreadable characters and are added to the list. The list thus generated is stored in the mode storage unit 42 of the speech synthesis unit 4 as information on the reading mode of the second divided text.

The text analysis means 41 of the speech synthesis means 4
When the above list is stored in the mode storage means 42,
The processing for the second divided text is started, and the symbol characters "-" and "|" in the list are deleted from the divided text (step S5). Then, the text is divided into words (step S6), and converted into phonetic symbols for each word with reference to the word dictionary 43 (step S7). next,
The voice waveform generating means 45 generates the waveform dictionary 4 based on the phonetic symbols.
4 to obtain and edit a necessary phoneme unit speech waveform to generate a speech waveform (step S8). The generated voice waveform is stored until the time of voice output.

When the processing for the second divided text is completed, the divided text on the thirteenth to sixteenth lines in FIG. 17 is processed (steps S53 and S54). First, frequency extraction means 2
Calculates the appearance frequency in the divided text for each predetermined symbol character as in the previous case (step S2 in FIG. 3), and the mode determination means 3 compares each appearance frequency with a predetermined value. (Step S3) If the value is larger than the value, the symbol character is added to the list of unread characters (Step S3).
4). In the case of the divided texts in the 13th to 16th lines, frequency information of 7 characters of kanji, 15 characters of the alphabet, 36 characters of the symbol "*", and 3 characters of other symbols is obtained. Since the frequency of "*" is high, "*""Is judged as an unreadable character and is added to the list. The list generated in this manner is stored in the mode storage unit 42 of the speech synthesis unit 4 as information on the reading mode of the third divided text.

The text analysis means 41 of the speech synthesis means 4
When the above list is stored in the mode storage means 42,
The processing for the third divided text is started, and the symbol character "*" in the list is deleted from the divided text (step S5). Then, the text is divided into words (step S6), and converted into phonetic symbols for each word with reference to the word dictionary 43 (step S7). Next, the voice waveform generating means 45 generates a voice waveform by acquiring and editing a required phoneme unit voice waveform by referring to the waveform dictionary 44 based on the phonetic symbols (step S8). The generated voice waveform is stored until the time of voice output.

As described above, the processing for all the divided texts of the input text has been completed, and the speech synthesizing means 4 converts the stored electric signal of the speech waveform into the first divided text,
The third divided text is output to a speaker or the like in the order of the third divided text (step S55). Thus, the text in FIG. 17 is read out as in FIG. In this case, "-" on the third line in FIG. 17 is read out, but "-" on the 5-11th line is not read out. That is, according to the second embodiment, in addition to the effects of the first embodiment, even if sentences in various formats are mixed in a text, an effect of correctly reading out aloud is exerted.

As an example of the second embodiment, there has been described an example in which a mode in which the symbol character is not read out and a mode in which the symbol character is read out are automatically selected based on the frequency of appearance of the symbol character. Based on the frequency of occurrence, a mode in which the alphabet character string is read out as Roman characters, a mode in which the text is read out as English is automatically selected, or based on the frequency of occurrence of a specific symbol character, a numeral containing the specific symbol character It is needless to say that the present invention can also be applied to a case where a mode in which a character string is read as a telephone number or a mode in which a character string is read as a mathematical expression is automatically selected.

FIG. 19 is a block diagram showing an example of a computer system to which the present invention is applied. The computer system of this example includes a computer main body 100 including a CPU, a main memory, and the like, a keyboard 101 connected thereto,
Auxiliary storage device 102, audio output device 103 such as speaker
And a recording medium 104. The recording medium 104 is a machine-readable recording medium such as a CD-ROM, a semiconductor memory, and a magnetic disk. The program recorded in the recording medium 104 is installed in the computer main body 100 when the computer system starts up. Of the input means 1, the frequency extracting means 2, the mode determining means 3 in the first embodiment described above by controlling the operation of
And the voice synthesizing means 4 and the input means 1, the frequency extracting means 2, the mode determining means 3, the voice synthesizing means 4 and the dividing means 5 in the second embodiment.

Although the embodiments of the present invention have been described above, the present invention is not limited to the above embodiments, and various other modifications can be made. For example, a function to select a mode that does not read out the symbol character based on the frequency of appearance of the symbol character, a mode to read out, and a mode that reads the alphabet character string as Roman characters based on the frequency of occurrence of alphabet vowels and consonants, as English Multiple functions such as a function to select a reading mode and a mode to read a numeric character string including the specific symbol character as a telephone number based on the frequency of occurrence of the specific symbol character, and a function to select a mode to read as a mathematical expression Is also considered as another embodiment of the present invention. In this case, words that need to be read separately for each reading mode may be stored in the word dictionary 43 with their phonetic symbols for each combination of reading modes. For example, the symbol "-"
The reading of "phone number reading" and "Romaji reading" is "no", "Formula reading" and "English reading" or "Romaji reading" is "Manasu", "phone number reading" and "English reading" In the case of "", "" (not read) may be stored.

[0073]

As described above, according to the present invention, the reading mode is determined based on the appearance frequency of the character in the text. For this reason, even in the case of a reading mode of a type that is difficult to automatically switch in the conventional technology by detecting a special pattern, such as determining a mode in which a ruled line of a table is not read or determining a mode between a Roman character reading and an English reading, The correct mode can be automatically determined without a mode selection by the user. Of course, it is necessary to have a condition that a description such as a telephone number is present in the text at a certain frequency, but it is not possible to automatically switch between a mode in which a numeric character string is read out as a telephone number and a mode in which it is read out as a mathematical expression. It also has the flexibility of being applicable.

Further, in a configuration in which a text is divided based on blank lines or the like, even if sentences in various formats are mixed in the text, each can be read out in an appropriate reading mode.

[Brief description of the drawings]

FIG. 1 is a block diagram of a first embodiment of the present invention.

FIG. 2 is a block diagram illustrating a configuration example of a speech synthesis unit.

FIG. 3 is a flowchart of a first example according to the first embodiment of the present invention.

FIG. 4 is a diagram showing an example of a text used in the first example according to the first embodiment of the present invention.

FIG. 5 is a diagram showing another example of the text used in the first example of the first embodiment of the present invention.

FIG. 6 is a flowchart of a second example according to the first embodiment of the present invention.

FIG. 7 is a diagram showing an example of a text used in a second example according to the first embodiment of the present invention.

FIG. 8 is a diagram showing another example of the text used in the second example according to the first embodiment of the present invention.

FIG. 9 is a diagram showing an example of reading out the text of FIG. 8;

FIG. 10 is a flowchart of a third example according to the first embodiment of the present invention.

FIG. 11 is a diagram showing an example of a text used in a third example according to the first embodiment of the present invention.

FIG. 12 is a diagram showing an example of reading out the text of FIG. 11;

FIG. 13 is a diagram showing another example of the text used in the third example according to the first embodiment of the present invention.

FIG. 14 is a diagram showing an example of reading out the text of FIG. 13;

FIG. 15 is a block diagram of a second embodiment of the present invention.

FIG. 16 is a flowchart of an example of the second embodiment of the present invention.

FIG. 17 is a diagram showing an example of a text used in one example according to the second embodiment of the present invention.

18 is a diagram showing an example of reading out the text of FIG. 17;

FIG. 19 is a block diagram illustrating an example of a computer system to which the present invention has been applied.

FIG. 20 is a block diagram of a speech synthesizing apparatus that allows a user to designate a reading mode.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 input means 2 frequency extraction means 3 mode determination means 4 voice synthesis means 5 division means

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-4-20998 (JP, A) JP-A-4-306766 (JP, A) JP-A-9-6379 (JP, A) JP-A-4- 253098 (JP, A) JP-A-8-36395 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 13/00-13/08 G06F 17/28

Claims (20)

(57) [Claims] An input unit for inputting a text; a frequency extracting unit for extracting a frequency of appearance of characters in the input text in the entire text; and a reading mode is determined based on the frequency information. And a voice synthesizing unit that operates in accordance with the determined reading mode. 2. An input unit for inputting a text, a dividing unit for dividing an input text, and a text of a character in the text for each divided text.
Frequency extracting means for extracting an appearance frequency in the entire text, mode determining means for determining a reading mode for each divided text based on the frequency information, and operating according to the determined reading mode for each divided text. A speech synthesizing apparatus comprising: a speech synthesizing unit. 3. The speech synthesizer according to claim 1, wherein said frequency extracting means has a configuration for extracting an appearance frequency of a character category. 4. The speech synthesizer according to claim 1, wherein said mode determining means determines one of a mode in which the symbol character is not read out and a mode in which the symbol character is read out based on the frequency of appearance of the symbol character. . 5. The mode determining means according to claim 1, wherein the mode determining means determines one of a mode in which the alphabet character string is read as a Roman character and a mode in which the character string is read as an English language, based on the appearance frequencies of the alphabet vowels and consonants. A speech synthesizer as described. 6. The voice synthesizing means, when a mode in which an alphabet character string is read out in English is determined,
6. The speech synthesizer according to claim 5, wherein numbers and symbols are read out in English. 7. The mode determining means determines one of a mode in which a numeric character string including the specific symbol character is read out as a telephone number and a mode in which it is read out as a mathematical expression, based on the frequency of occurrence of the specific symbol character. 3. The speech synthesizer according to claim 1, having a configuration. 8. An input means for inputting a text, a frequency extracting means for extracting an appearance frequency of a character in the input text, and a means for determining a reading mode based on the frequency information, comprising: A mode determining means for determining one of a mode in which the alphabet character string is read out as Roman characters and a mode in which it is read out as English based on the appearance frequency of consonants; and a voice synthesizing means which operates in accordance with the determined reading mode. A speech synthesizer comprising: 9. An input means for inputting a text, a dividing means for dividing the input text, a frequency extracting means for extracting a frequency of appearance of a character in the text for each of the divided texts, Based on the frequency of appearance of alphabet vowels and consonants, it is means for determining a reading mode for each of the divided texts. A speech synthesis apparatus comprising: a determination unit; and a speech synthesis unit that operates in accordance with the determined reading mode for each divided text. 10. The speech synthesis apparatus according to claim 8, wherein said speech synthesis means has a configuration in which, when a mode in which an alphabet character string is read out in English is determined, numbers and symbols are also read out in English. . 11. A computer, comprising: input means for inputting a text; frequency extracting means for extracting a frequency of appearance of characters in the input text in the entire text; and a reading mode based on the frequency information. A computer-readable recording medium storing a program to function as a mode determining means for determining, a speech synthesizing means operating in accordance with the determined reading mode. 12. A computer, comprising: input means for inputting text; splitting means for splitting input text; text of characters in the text for each split text;
Frequency extracting means for extracting the appearance frequency in the entire text, mode determining means for determining a reading mode for each divided text based on the frequency information, and speech synthesis operating according to the determined reading mode for each divided text. Means, a computer-readable recording medium recording a program to function as a computer. 13. The computer-readable recording medium according to claim 11, wherein said frequency extracting means has a configuration for extracting an appearance frequency of a character category. 14. A mode in which a symbol character is not read out based on the frequency of appearance of the symbol character.
13. The computer-readable recording medium according to claim 11, wherein the computer-readable recording medium has a configuration for determining one of a reading mode. 15. The apparatus according to claim 11, wherein said mode determining means determines one of a mode in which an alphabetic character string is read out as Roman characters and a mode in which it is read out as English, based on the frequency of appearance of alphabet vowels and consonants. The computer-readable recording medium according to claim 1. 16. The computer-readable recording medium according to claim 15, wherein said voice synthesizing means has a configuration in which, when a mode in which an alphabetic character string is read out in English is determined, numerals and symbols are also read out in English. 17. The mode determining means determines one of a mode in which a numeric character string including the specific symbol character is read out as a telephone number and a mode in which it is read out as a mathematical expression, based on the frequency of occurrence of the specific symbol character. 13. The computer-readable recording medium according to claim 11, having a configuration. 18. A computer, comprising : input means for inputting a text; frequency extracting means for extracting a frequency of appearance of a character in the input text; means for determining a reading mode based on the frequency information; A mode determining means for determining one of a mode in which an alphabet character string is read out as Roman characters and a mode in which it is read out as English based on the frequency of appearance of vowels and consonants; A computer-readable recording medium on which a program to be operated is recorded. 19. A computer, comprising : input means for inputting text; dividing means for dividing input text; frequency extracting means for extracting the appearance frequency of a character in the text for each divided text; A mode for determining a reading mode for each divided text based on the frequency of appearance of alphabet vowels and consonants, which determines one of a mode for reading an alphabet character string as Roman characters and a mode for reading as English. A computer-readable recording medium that stores a program that functions as a determining unit, and a speech synthesizing unit that operates in accordance with the determined reading mode for each divided text. 20. The computer-readable recording according to claim 18, wherein said speech synthesizing means has a configuration in which, when a mode in which an alphabetic character string is read out in English is determined, numbers and symbols are also read out in English. Medium.