JPH0244080B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a graphite-phoneme conversion device that converts a character string of a Japanese sentence into a phoneme sequence.

[Technical background of the invention and its problems]

2. Description of the Related Art Conventionally, so-called speech synthesis devices have been known that convert input Japanese text into speech and output the converted speech. As a device of this type, a system was devised in which a phoneme corresponding to an input character code is extracted from a phoneme dictionary stored in units of phrases or less and outputted as voice. This method had the disadvantage that the output speech did not contain accents, was monotonous, had no rhythm, and was difficult to interpret. Therefore, it was necessary to analyze the character string and convert it into a phonological sequence that included accent information, etc.
In the past, this was done by humans. In other words, humans have created phoneme sequences of character strings and input them into speech synthesis devices.

Therefore, when converting a character string into a phonological sequence, it is necessary not only to change the kanji part of the character string to the reading (phonetic), but also to correctly test the accent in order to convert it into a highly natural sound. be. In this case the hiragana string is either an adjunct or
Also, the accent position differs depending on whether it is an independent word, so even if the same hiragana character string is used, it is important to correctly distinguish whether it is an attached word or an independent word depending on the situation in which it is used. It was hot. Conventional devices do not have a function to correctly identify whether an input hiragana character string is an attached word or an independent word. Therefore, humans had to distinguish between these accents and input accent information into a speech synthesis device based on the results, but this method was burdensome for humans in terms of time and effort.

[Purpose of the invention]

SUMMARY OF THE INVENTION An object of the present invention is to provide an apparatus that converts character strings of Japanese sentences into phonological sequences, and in particular, that makes it possible to test the accent of hiragana character strings with high accuracy.

[Summary of the invention]

The present invention divides a character string containing kanji and kana into phrase units based on changes in character types, and when the first character of this phrase is a kanji, the phrase character string is checked against a kanji independent word dictionary, and on the other hand, when the first character of the phrase is a hiragana When the hiragana independent word dictionary is checked, and if the conjugated word ending is a conjugated word or a conjugated word, the remaining hiragana character string after removing the independent word and its conjugated ending from the clause is found to be a conjugated word or a conjugated word. Test whether a specific adjunct string that can be connected to the body shape is at the beginning, and if it is a specific adjunct string, match the above hiragana string with the adjunct dictionary; If it is not, it is checked against the Hiragana independent word dictionary above, and if the independent word checked from the first letter of the clause above is other than a predicate, or even if it is a predicate, it is not a conjunctive form or an adjunctive form, the independent word from within the clause is checked. By removing the conjugated endings and comparing the remaining hiragana character strings with an adjunct word dictionary, the above character strings containing kanji and kana are converted into a phoneme sequence.

In particular, this test accurately identifies accents, which are important when converting sentences into speech, based on correctly identifying whether hiragana character strings are independent words or attached words.

〔Effect of the invention〕

According to the present invention, character strings of Japanese sentences can be converted into phoneme sequences, which is especially important in converting sentences into highly natural speech by correctly identifying independent words and attached words for hiragana character strings. Since the accent test can be performed with high precision, the quality of the speech synthesizer can be improved, and the burden on humans in terms of time and labor is reduced.

[Embodiments of the invention]

Hereinafter, one embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram of the apparatus of this embodiment. First, a character string containing kanji and kana to be converted into speech is stored in the input text storage device 1. The bunsetsu cutting circuit 2 converts the input character string into bunsetsu units by detecting a change in character type, such as a change from "kana" to "kanji", for the sentences stored in the input text storage device 1. and input it into the input phrase register 3.
Set to . For each character string of a phrase set in the input phrase register 3, the following graphophonemic conversion is performed.

The clause first character verification circuit 4 detects whether the first character of the clause is a kanji or a hiragana, and performs dictionary matching from the beginning of the clause using the kanji independent word matching circuit 5 (signal line 16), or detects whether the first character of the clause is a hiragana character or a hiragana character. It is determined whether the independent word matching circuit 10 is used (signal line 17). The kanji independent word dictionary 6 stores words starting with kanji along with information on their parts of speech, and also stores the phonological sequence and accent type of the words in correspondence with the words.
The hiragana independent word dictionary 11 is a dictionary that registers words starting with hiragana, and has the same format as the kanji independent word dictionary 6 described above.

When the first character of a clause is a kanji, the kanji independent word matching circuit 5 converts the first character string of the clause into a kanji independent word dictionary 6.
Check with Among the character strings detected as matches through this matching, the longest character string, that is, the longest matching word, is stored as a candidate word in a temporary storage device provided in the dictionary independent word matching circuit 5. If the independent word of the longest matching word is not a predicate, its phoneme sequence and accent information are transferred to the phoneme sequence register 14 through the signal line 18 and set therein. If the independent word is a predicate, its phoneme sequence and accent information are sequentially sent to the circuits shown below.

On the other hand, when the first character of a bunsetsu is a hiragana, the hiragana independent word matching circuit 10 matches the first character string of the bunsetsu with the hiragana independent word dictionary 11, performs the same process as the kanji independent word matching circuit 5, and connects the signal line to the hiragana independent word matching circuit 10. 19
The phoneme sequence and accent information are sent to the phoneme sequence register 14 through the phoneme sequence register 14.

Next, when the independent word matched by the above-mentioned Kanji independent word matching circuit 5 or Hiragana independent word matching circuit 10 is a predicate (signal lines 20, 21), the conjugated ending verification circuit 7 checks the hiragana character string following the independent word. Conjugation ending table 8
A conjugated ending test is carried out by comparing the conjugated endings with ). If it fails, the second complementary word is taken out from the temporary storage provided in the above-mentioned Kanji independent word matching circuit 5 or the Hiragana independent word matching circuit 10, and the second complementary word is extracted from the above-mentioned process for the longest matching word. Do the same thing.

In the conjugated ending testing circuit 7, after the conjugated ending is verified, the independent word and the conjugated ending are removed from the input phrase character string. When the conjugated ending that has been verified is in the conjunction form or the conjunction form, the remaining character string is sent to the hiragana character string verification circuit 9 (signal line 23). The hiragana character string verification circuit 9 verifies whether the first character string of the remaining character strings is a specific character of two or more hiragana characters as shown below by comparing it with the specific character table 15. If the characters are specific characters, there is a high possibility that they are adjunct words, so they are sent to the adjunct word matching circuit 12 via the signal line 24.

After the conjunction form, "masu", "tari", "tsutsu"
``Even'', ``Nagara'', etc. After the adjunctive form,
``kara'', ``sae'', ``shika'', ``only'', ``but'', etc.

On the other hand, if it is not a specific character as mentioned above, it goes to the hiragana independent word matching circuit 10 through the signal line 25, and the remaining character strings are sent to the hiragana independent word dictionary 1.
Check with 1. If the matching is successful here, the hiragana independent word matching circuit 10 converts the accents of the above-mentioned adjunctive form or adjunctive form into the accent type of these words (registered in the above-mentioned Kanji independent word dictionary or hiragana independent word dictionary). The accent information is verified based on the conjugated form (distinction between the adjunctive form and the adjunctive form), and the accent information is sent to the phoneme series register 14. (Signal line 19) The adjunct word matching circuit 12 verifies the adjunct word by comparing the character string starting with the above-mentioned specific character with the adjunct word dictionary 13. The attached word dictionary 13 contains attached words and compound forms of attached words (e.g. "only because")
are used as headwords, and the conditions under which they are connected to independent words, as well as information regarding the shift of accent when they are combined with independent words, are registered. The adjunct word matching circuit 12 converts the adjunct word that has been matched into a phoneme sequence and sends it to the phoneme sequence register 14 (signal line 26), and also converts the accent of a clause in which an independent word and an adjunct word are combined into a phoneme sequence using the above adjunct word. The test is performed based on the information registered in the dictionary 13, and the clause accent position information is also sent to the phoneme sequence register 14.

In addition, when the independent word matched by the kanji independent matching circuit 5 or the hiragana independent word matching circuit 10 is other than a pragmatic word, or even if it is a pragmatic word, it is verified by the conjugated ending test circuit 7. When the conjugated ending is other than the adjunctive form or the adjunctive form, the remaining character string that includes the verified independent word and its conjugated ending from the clause is sent to the signal line 27,
It is sent to the attached word matching circuit 12 through 28 and 29. Here, the remaining character strings are compared with the adjunct word dictionary 13, and when the adjunct words are matched, the phoneme sequence and accent information are sent to the phoneme sequence register 14, as described above. Furthermore, if the phrase character string remains even after removing the collated adjunct, the remaining character string is sent to the hiragana character string verification circuit 9 through the signal line 30, and the same process as above is performed.

As described above, according to this embodiment, by testing a specific character string in the hiragana character string verification circuit 9 and changing the collation order in the hiragana independent word verification circuit 10 and the attached word verification circuit 12 based on the verification,
It is possible to correctly distinguish and test whether a hiragana character string following a conjunctive form or a conjunctive form is an attached word or an independent word.

For example, if the input string is "speak",
When "Kashi" is verified as a conjunctive form, if the remaining character string "masu" is checked directly in the hiragana independent word matching circuit 10 as in the past without passing through the hiragana character string verification circuit 9, it becomes independent. Since the word "masu" (increase) is tested, the accent is also tested as "masu" (flat type). But needless to say, this is wrong. In this embodiment, since "masu" is verified as a specific character by the hiragana character string verification circuit 9, in this case, the verification of "masu" is performed by the adjunct word verification circuit 12, so that it can be verified as an auxiliary verb of the adjunct word. Accent of phrase “――――‐ Talk
'is obtained.

In the above example, after testing the conjunction form, the hiragana independent word matching circuit 10 is replaced by the attached word matching circuit 1.
By going to step 2, a correct test result can be obtained for the hiragana character string, but an error will also occur if the test always goes to the adjunct matching circuit 12 after testing the conjunction form. For example, the input clause character string is "easy to speak"
In the case of , after ``Kashi'' is verified as a conjunction form, if the remaining character string ``Tasu'' goes to the adjunct matching circuit 12 unconditionally, the character ``ya'' is verified as a particle of the adjunct. After that, the remaining character string "Sui" after removing "ya" is converted into the independent word "Sui" by the Hiragana independent word matching circuit 10.
Because the accent is separated into two, it becomes ``――――‐ talk and □sui.'' Needless to say, this is a mistake; the correct answer should be ``――――-Easy to talk.'' In this embodiment, after testing the adjunctive form or the adjunctive form, the remaining character examples are sent to the hiragana character string testing circuit 9 to identify whether they are attached words or independent words. There is no adjunct word matching or independent word matching unconditionally, and the accent can be detected correctly without making the error as in the above example.

[Brief explanation of drawings]

FIG. 1 is a schematic diagram showing an apparatus according to an embodiment of the present invention. 1... Input sentence storage device, 2... Bunsetsu cutting circuit, 3
…Input clause register, 4…Bunsetsu first character verification circuit, 5…Kanji independent word collation circuit, 6…Kanji independent word dictionary storage device, 7…conjugation ending verification circuit, 8…conjugation ending table storage device, 9…hiragana character string Verification circuit, 1
0... Hiragana independent word matching circuit, 11... Hiragana independent word dictionary storage device, 12... Adjunct word matching circuit, 13
...Adjunct word dictionary storage device, 14...Phonological sequence register, 15...Specific character table storage device, 16-30...Signal line.

Claims (1)

[Claims] 1. A bunsetsu cutting means for dividing a character string of a Japanese sentence into clauses based on changes in character types, a bunsetsu first character verification means for testing whether the first character of the bunsetsu outputted from this bunsetsu cutting means is a kanji character or a hiragana character, When the first character of the phrase is verified as a kanji by the first character verification means, an independent word included in the beginning of the phrase is detected, it is verified whether or not this independent word is a term, and the independent word is removed from the phrase. Kanji independent word matching means outputs the first remaining character string and phonological accent information of this independent word, and when the first character of the clause is verified as hiragana by the clause beginning character verification means, the beginning part of this clause is verified as hiragana. A hiragana independent word that detects an independent word included in the clause, tests whether or not this independent word is a pragmatic word, and outputs a first remaining character string obtained by removing this independent word from the clause and phonological accent information of this independent word. When the independent word is verified as a term by the collation means and the Kanji independent word collation means or the Hiragana independent word collation means, the first
A second character string that is obtained by testing whether the conjugated ending of the independent word contained in the remaining character string is an adjunctive form or an adjunctive form, and removes this conjugative ending from the first remaining character string.
a conjugated ending verification means for directly outputting the remaining character strings, the phonological accent information of the conjugated ending, and the phonological accent information of the independent word outputted from the Kanji independent word matching means or the Hiragana independent word matching means; If the testing means verifies that the ending of the conjugated word is a conjunctive form or an adnominal form, it is determined whether or not the beginning part of the second remaining character string is a specific character string that can be connected to a conjunctive form or an adnominal form. hiragana character string verification means for outputting the second remaining character string and the independent word and its conjugation ending phonological accent information outputted from the conjugation ending verification means as they are to the hiragana independent word matching means when the character string is not a character string; and if the independent word is determined to be not a predicate by the Kanji independent word matching means or the Hiragana independent word matching means, the first remaining character string and the phonological accent information of the independent word, The phonological accent information of the second remaining character string and the independent word and its conjugated ending when the conjugated ending is determined to be neither the adjunctive form nor the adjunctive form; When the beginning part is verified to be a specific character string, the second remaining character string, the independent word, and the phonological accent information at the end of its conjugated word are input, and the beginning part of the first or second remaining character string is input. The phonological accent of the entire character string consisting of the independent word, the conjugated ending, the attached word, or the independent word and the attached word is determined based on the information when this attached word is connected to an independent word or a conjugated word. What is claimed is: 1. A graphite-phoneme conversion device comprising: an adjunct collation means for determining information; and a phoneme series register for storing phonetic accent information of the entire character string output from the adjunction collation means.