JP2573565B2 - Conversion device for word spelling to phonetic symbols - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Summary] For each spelling in an input word, matching is performed with an accent rule read from an accent rule storage unit to add accent identification information to a spelling having an accent. For each unconverted spelling string to which accent identification information has been added, the spelling character string is converted into a phonetic symbol based on the pronunciation rule read from the pronunciation rule storage unit. By doing so, it is possible to convert an accented word into a correct phonetic symbol.

[Industrial applications]

The present invention relates to a method for converting word spelling to phonetic symbols, and more particularly, to a method for converting word spelling to phonetic symbols in which accent identification information is involved in mediation when converting words to phonetic symbols.

In speech synthesis, there is a technique using a phonetic symbol of a given word. These phonetic symbols are derived from the spelling of the input word by looking at the spelling, but no matter what procedure such a technique takes, the phonetic symbols that are derived are correct. It is needless to say that it is desirable.

[Conventional technology]

The longest matching method is considered as a general method used for converting word spelling into phonetic symbols as described above. This method focuses only on the spelling, and converts the spelling into phonetic symbols according to the longest pronunciation rule applicable from the left of the word.

[Problems to be solved by the invention]

FIG. 12A shows an example in which the longest match method is applied to the German word stehen. As can be seen from the example shown in this figure, the longest match method is effective for German and the like.

However, it is inappropriate to determine the phonetic symbols of languages whose pronunciation changes depending on the presence or absence of accents, such as English, using the longest matching method. For example, the word author [ ] And authority [ ] When the longest match method is applied to the two words
As shown in Figure (B), author has the first syllable, authority has the second syllable, and the first accent.
The pronunciation of the syllable and the second syllable are different. Therefore, in the longest match method, no accent information is given.
I can't get the correct phonetic symbols. Even in such a case, if a correct phonetic symbol is to be obtained by the longest match method, there is no other way than to have rules for the words themselves, and an enormous increase in rule dictionary capacity is inevitable.

The present invention has been created in view of such problems,
It is an object of the present invention to provide an apparatus for converting a word spelling into a phonetic symbol, which can give a correct phonetic symbol of a word by giving accent identification information to a spelling having an accent in the word.

[Means for solving the problem]

FIG. 1 shows a principle block diagram of the present invention. In this figure, 1 is a word input means for inputting a word composed of a required number of spellings. 4 is an accent rule storage unit. Numeral 2 is accent identification information adding means, which reads an accent rule from the accent rule storage unit 4 for each spelling character string after each word spelling character in the input word, and reads the spelling character string and the read out accent rule. And adds accent identification information to the spelling having an accent.

Reference numeral 5 denotes a pronunciation rule storage unit. Numeral 3 is a spelling / pronunciation symbol conversion unit which stores pronunciation rules from the pronunciation rule storage unit (5) for each word spelling string to which unconverted accent identification information has been added from the beginning of the word spelling to which accent identification information has been added. The word spelling string to which the read and unconverted accent identification information is added is converted into phonetic symbols based on the read pronunciation rules.

[Action]

The word input from the word input means 1 is obtained by matching the word spelling character string with the accent rule read from the accent rule storage unit 4 for each word spelling character string after the word spelling character. Taken at 2,
Accent identification information is added to an accented word spelling. The word to which the accent identification information is added is matched with the pronunciation rule in the spelling / diamond symbol conversion means 3.

Matching in the spelling / pronunciation conversion means 3 is performed for each word spelling string to which unconverted accent identification information has been added from the beginning of the word spelling to which accent identification information has been added. Matching between the spelling character string and the pronunciation rule read from the pronunciation rule storage unit 5 is performed, and the word spelling character string to which unconverted accent identification information is added is converted into a phonetic symbol.

In this manner, prior to conversion of a word into a phonetic symbol, accent identification information indicating that fact is added to the accented spelling of the word, and the word is passed to the spelling / phonetic symbol conversion means 3. Thus, it is possible to automatically obtain a phonetic symbol in which the accent of the word is taken into account when converting the word into a phonetic symbol.

Therefore, correct phonetic symbols can be obtained by the present invention.

〔Example〕

FIG. 2 shows an embodiment of the present invention. In this figure,
Reference numeral 10 denotes an accent position determining unit that receives a spelling of a word for which a phonetic symbol is to be obtained and outputs a character string to which an accent identification character is added as described later. Reference numeral 11 denotes an accent rule storage unit, which gives the accent position determining unit 10 an accent rule used for controlling the addition of accent identification characters in the accent position determining unit 10. Reference numeral 12 denotes a spelling and phonetic symbol conversion unit that converts a character string output from the accent position determining unit 10 into phonetic symbols, and supplies pronunciation rules used in the conversion to the spelling and phonetic symbol conversion unit 12. This is the rule storage unit 13. Accent rule storage
The accent rules and the pronunciation rules stored in the pronunciation rule 11 and the pronunciation rule storage unit 13 are configured in a format as shown in FIG. The format shown in this figure indicates that if there is a character string containing XAY, A is replaced with B, if there is a character string containing XY, A is inserted between X and Y, and XAY is replaced with If there is a character string containing it, it means that A is deleted, and XXXAY, XXAY or XAY
Indicates that if there is a character string containing A, A is replaced with B. "(", ")" In the expression represents arbitraryness.

 Hereinafter, the processing steps according to the present invention will be described step by step.

FIG. 4 shows a processing flow in the accent position determination unit 10 for receiving a word for obtaining a phonetic symbol. In step S1, a spelling pointer (spell)
Is set as the first spelling of the input word, and the routine proceeds to step S2, where one accent rule is read from the accent rule storage unit (dictionary) 11. Then, in step S3, it is determined whether or not all the accent rules of the dictionary have been read. If all the words have been read, the processing for the word ends.

If the determination in step S3 is negative, a rule pointer (rule) is set at the head of the read accent rule (step S4), and the matching process in step S5 (the details are shown in FIGS. 5 to 8). Is performed, and if the matching is successful (T in step S6), a character conversion process is performed in step S7 based on the accent rule, and the process returns to step S2.

If the matching in step S5 fails (F in step S6), the process returns to step S2 to perform the same processing. Examples of these processes are shown in FIG. 9 and FIG.
Both figures show only rules that can be matched. All other rules are those that cannot be matched in the middle or finally. As is clear from the processing examples shown in FIGS. 9 and 10, as the accent identification information inserted into the character string based on the accent rules, for example, numbers are used as shown in the figure.
1 is used for identifying the first accent, 2 is used for identifying the second accent, and 3 is used for identifying the weak sound.

Next, the processing in step S5 will be described with reference to FIGS.

In FIG. 5, the matching is performed by shifting the spelling one character to the right (S105), calling match (S103) to try to match the spelling with the accent rule, and if the matching is successful, the success is called a match. Return (S
107). If the number of remaining characters in the spelling is less than the number of characters in the accent rule, a failure is returned (S106). The fifth
In the figure, spell is a pointer to spelling, rule is a pointer to accent rule character (symbol), spell.le
n is the number of characters in spelling, and rule.len is the number of characters in accent rules.

Step S5 in FIG. 5 comprises processing steps as shown in FIG. In FIG. 6, spell and rule mean pointers, respectively, as in FIG.
* rule indicates the character pointed to by the pell, and * rule indicates the character pointed to by the rule. Further, ON of the flag indicates the inside of the parentheses, and OFF indicates the outside of the parentheses.

In step S200 of FIG. 6, the matching process is started. In step S201, parenthesis processing is performed, and in step S202, matching processing for each character (sub.matc) is performed.
Perform h). Details of these two steps will be described with reference to FIGS. 7 and 8. After step S201, it is determined in step S250 whether or not the rule pointer points to the end of the rule. If the determination is affirmative, success is returned (step S251).
The process proceeds to step S202 and subsequent steps described below.

If the matching process for each character is successful in the matching process in step S202 (T in step S203), match calls itself and passes the next character (step S204). If the match returns success (T in step S205), the calling match also returns success (step S206). If a failure is returned in the matching in step S204 (F in step S205), and the pointer at that time is outside parentheses (T in step S207), a failure is returned (step S208). Conversely, if it is inside the parentheses, it goes out of the parentheses (step S209), and the match is called again (step S210). If the matching process (step S210) in the called match is successful (T in step S211), the process returns success (step S212), but if it fails (F in step S211), it returns failure (step S212). ).

In addition, in the matching process in step S202, a failure occurs in matching for each character (F in step S203), and if the pointer at that time is outside parentheses (T in step S214), the failure is returned (step S215). Conversely, if it is inside the parentheses (F in step S214), it goes out of the parentheses (step S216) and calls match again (step S21).
7). If the matching process (step S217) in the called match is successful (T in step S218), success is returned (S219), but if it is unsuccessful (F in step S218), failure is returned.

The processing of parentheses in the above-described matching processing is as follows (see FIG. 7). Rule, in FIG.
* The meanings of rule and flag are the same as in FIG. If the processing of the parentheses starts in step S300 and the rule pointer points to "(" (T in step S301),
The flag is turned on (step S302), the rule pointer is moved right by one character (step S303), the processing of parentheses is called again (step S304), and the value of the rule pointer is returned (step S305).

When the rule pointer is advanced rightward and points to “(” (T in step S306), the flag is turned OFF (step S306).
307), Step S30 to advance the rule pointer one character to the right
Start the process after 3.

If the rule pointer does not point to "(" or ")", that is, if the determination in steps S301 and S306 is F, the value of the rule pointer is returned (step S308).

FIG. 8 shows a matching process of receiving one character at a time from the spelling and the rule, which is performed subsequent to the processing of the parentheses in step S201 in FIG. In FIG. 8, C1 is one character of spelling, and C2 is one character of an accent rule.

When such a one-character matching process starts (step S400), first, in step S401, "C2" is a character that matches "+", that is, a vowel (a, e, i, o, u, y). Is determined. The judgment is true (T)
If "C1" is a vowel character (T in step S402), success is returned (step S403), but if "C1" is not a vowel character (F in step S402), failure is returned (step S404). ).

If the determination in step S401 is false (F),
In step S405, it is determined whether or not "C2" is a character that matches "-", that is, a consonant (an alphabet other than a vowel). If the judgment is true (T), "C
If "1" is a consonant (T in step S406), success is returned (step S407), but if "C1" is not a consonant (F in step S406), failure is returned (step S408).

If the determination in step S405 is false (F),
In step S409, “C2” is changed to “＄”, that is, a number (0 to 0).
A determination is made as to whether or not the character matches 9). If the determination is true (T) and “C1” is a number (T in step S401), success is returned (step S411).
However, if “C1” is not a number (F in step S410),
Return failure (step S412).

If the determination in step S409 is false (F),
In step S413, it is determined whether or not “C2” is “C1”. If the determination is true (T), return success (step S415), and if false (F), return failure (step S416).

Examples of words that have undergone such matching processing between word spelling and accent rules are shown in FIG. 9 and FIG. The upper part of these figures shows this, and the accent rules show only those accent rules that match the words shown in them. The matching process between these words and the accent rules proceeds in accordance with those shown in FIGS. 5 to 8, and the details of the matching process are omitted.

The word spelling character string to which the accent identification character has been added as described above is sent from the accent position determining unit 10 to the spelling / phonetic symbol converting unit 12, and is subjected to the following conversion processing to the accent position determining unit 10. The phonetic symbols for the input word are output from the spelling / phonetic symbol conversion unit 12.

FIG. 11 shows the conversion processing flow. Step S5
At 00, the spelling pointer is set at the beginning of the input spelling. If the spelling pointer points to the last character of the input spelling (T), the process ends. However, this is not the case (F), so the pronunciation rule is read from the pronunciation rule storage unit (dictionary) 13 (step S502). ). The longest pronunciation rule matching the character string to the right of the character pointed to by the pronunciation rule and the spelling pointer is searched for (steps S503, S504).
If a match between them can be obtained (step
(F in S504), the spelling is converted into a phonetic symbol, but the spelling itself is not converted, and the converted phonetic symbol is stored separately from the spelling in a storage unit in the spelling / phonetic symbol conversion unit 12 (step S504). S505).

Next, the spelling pointer is advanced rightward by the number of converted characters (step S506), and the process returns to step S501. Then, the processing enters the matching search loop of steps S502 to S504 again, and the spelling pointer is shifted to the right from the character pointed to by the advanced spelling pointer. The matching process is performed on the character string in the same manner as described above.

When the spelling pointer reaches the last character of the spelling in this way, the process of converting the input spelling character string with accent identification characters into phonetic symbols is completed, and the following similar conversion process is started.

Examples of such a conversion process are shown in FIGS. 9 and 10. The lower part of these figures shows this, and the pronunciation rules only show the pronunciation rules that match the spelling string with the accented identification characters in the example.

〔The invention's effect〕

As described above, according to the present invention, it is possible to automatically convert a word into a phonetic symbol by effectively using accent identification information for converting the word into a phonetic symbol. Since the means that uses accent identification information is used for automatic conversion of words to phonetic symbols, correct conversion of words to phonetic symbols used in languages whose pronunciation changes depending on the presence or absence of accents Can be.

[Brief description of the drawings]

1 is a block diagram showing the principle of the present invention, FIG. 2 is a diagram showing an embodiment of the present invention, FIG. 3 is a diagram for explaining the format of accent rules and pronunciation rules, and FIG. FIG. 5 is a diagram showing a processing flow, FIG. 5 to FIG. 8 are detailed diagrams of a matching step in the processing flow of FIG. 5, FIG. 9 and FIG. 10 are diagrams showing processing examples, and FIG. FIG. 12A and FIG. 12B are views for explaining the drawbacks of the conventional longest matching method. In FIG. 1, 1 is word input means, 2 is accent identification information adding means,
Reference numeral 3 denotes a spelling / phonetic symbol conversion unit. In FIG. 2, reference numeral 10 denotes an accent position determination unit, 11 denotes an accent rule storage unit, 12 denotes a spelling / phonetic symbol conversion unit, and 13 denotes a pronunciation rule storage unit.

Claims (1)

(57) [Claims] 1. A word input means (1) for inputting a word composed of a required number of spellings, an accent rule storage unit (4), and a spelling character string after each word spelling character in the input word. An accent identification information adding means (2) for reading an accent rule from the accent rule storage section (4), matching the spelled character string with the read accent rule, and adding accent identification information to the spelled character having an accent; A pronunciation rule storage unit (5), and a pronunciation rule from the pronunciation rule storage unit (5) for each word spelling string to which unconverted accent identification information has been added from the beginning of the word spelling added with accent identification information. And a spelling / pronunciation conversion unit that converts a word spelling string to which unconverted accent identification information has been added into phonetic symbols based on the read pronunciation rules. (3) An apparatus for converting word spelling into phonetic symbols, characterized by comprising (3).