JP3414326B2 - Speech synthesis dictionary registration apparatus and method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice synthesizing apparatus, and more particularly to a voice synthesizing apparatus for generating synthetic speech based on text, which registers dictionary information for storing pronunciation information of words for generating pronunciation information. Dictionary registration device and method.

[0002]

2. Description of the Related Art Conventionally, a word dictionary, accent information, part-of-speech information, etc. are registered in a text-to-speech language dictionary. All of this information is entered manually to register the language dictionary.

Expanding the language dictionary is an essential task for improving the reading accuracy of the speech synthesizer. for that reason,
Many studies have been conducted on the simplification and automation of dictionary registration devices. For example, Japanese Unexamined Patent Publication No. 7-152392 discloses a word dictionary registration mechanism for identifying words from an input character string and setting pronunciation, accent type, grammatical information, etc. for each word. There has been proposed a configuration of a speech synthesizer that simplifies the dictionary registration by providing a word dictionary registration mechanism that semi-automatically generates phoneme change information from "or" or "reading" and "accent".

Further, Japanese Patent Laid-Open No. 1-303498 discloses that
When it is instructed to register the word information used for speech synthesis in the accent dictionary, the accent type given to the reading of the word to be registered is sequentially changed, and the reading information is output to the speech synthesizer in order. And an accent dictionary registration means for deciding an accent type to be registered in the accent dictionary with respect to the reading of the word in accordance with the monitor result of the synthesized speech which is synthesized by the speech synthesizer and sequentially output. A device has been proposed. This accent dictionary creating device automatically outputs the synthesized speech of the word to be registered with different accents in order, and specifies the accent information according to the monitor result of the synthesized speech, thereby simplifying the correct accent of the word. In addition, it is possible to accurately specify and register efficiently in the dictionary.

Regarding a dictionary registration device for registering an unregistered word in a language dictionary for speech synthesis, Japanese Patent Laid-Open No. 320208/1996 discloses a dictionary search corresponding to an input of a word, reading the word, accent type , Voice that estimates grammatical information, displays candidates, and automatically registers in the dictionary based on selection or correction input from the candidates, and can easily read dictionary of word notation, accent type, grammatical information, etc. in dictionary A configuration of a dictionary registration device for composition has been proposed. This dictionary registration device
The dictionary is searched according to the input of the word, the word notation, the accent type, the grammatical information, etc. are estimated to display the candidate, and the dictionary is automatically registered based on the selection or correction input from the candidate. You can easily register the reading of the notation, accent type, grammar information, etc. in the dictionary.

As described above, in the conventional apparatus, it is finally necessary to manually review and correct the automatically extracted candidates.

[0007]

By the way, as the reading information of the word dictionary for speech synthesis, it is necessary to register the utterance content itself, not the furigana written as a word reading in a Japanese dictionary, for example. For example, in the Japanese dictionary, the furigana of "voice" is described as "onsei".

However, actually, the description when a person says "voice" is closer to "onse" rather than "onsei". In Japanese, this is usually read in single kanji as "edan sequel" i "" or "odan sequel".
This is because when the same vowel sounds such as "u", "oh", and "good", there is no subsequent vowel, and instead, the preceding vowel becomes a longer sound.

Since it is necessary to describe the utterance content itself of the synthetic voice in the dictionary for the synthetic voice, the pronunciation is written as "sei" like "voice", but in reality, the preceding vowel is a long sound. In the case of uttering in the converted form, "se" was registered.

However, when a dictionary for voice synthesis is manually created, it is difficult to add all the readings that reflect the phoneme changes that occur during pronunciation, as described above.
There are many input errors.

Therefore, conventionally, when registering a voice reading, it is provided with a function of converting the read information into a synthetic voice and confirming whether or not the reading information can be correctly added.
Multiple readings were created automatically, and all of them were made to be heard by synthetic speech, and the most natural reading was selected.

However, even with such a method, the final judgment is made manually, so that the selection result varies depending on the variation among operators, the sound quality of the synthesized voice, and the utterance speed. .

Therefore, the present invention has been made in view of the above problems, and its purpose is not to register a reading describing the utterance content itself when creating a word dictionary for synthetic speech. Provide a voice synthesis dictionary registration device and method for automatically registering a pronunciation for voice synthesis by inputting a pronunciation in a furigana format as used in a Japanese dictionary for each kanji of a word to be registered. Especially.

[0014]

Means for Solving the Problems A speech synthesis dictionary registration device of the present invention which achieves the above object refers to a text analysis means for analyzing input text and a word dictionary from the result of the text analysis means. The above-mentioned voice synthesizing apparatus comprising: a reading information generating unit that generates reading information; and a voice waveform generating unit that performs voice synthesis by referring to the word dictionary based on the reading information generated by the reading information generating unit. A device for creating a word dictionary, wherein:
A furigana registration means for registering the reading of a word with a standard furigana described in a Japanese dictionary, and a kanji delimiter for inserting a kanji delimiter in the furigana if the registered word contains kanji An insertion unit, a dictionary information registration unit that generates reading information for voice synthesis from the furigana in which the Chinese character delimiter is inserted, and the word to be registered are provided.

According to the present invention, instead of the kanji delimiter insertion means, a single kanji dictionary in which readings of single kanji are registered, and kanji in the kana kana registered by the kana kana registration means using the single kanji dictionary. It may be configured to include a kanji delimiter automatic insertion means for automatically inserting a delimiter.

[0016]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described. The present invention, in its preferred embodiment, is a text analysis means (2) for analyzing an input text.
And a word dictionary (5) from the result of the text analysis means (2)
Reading information generating means for generating reading information by referring to (3)
And a speech waveform generation means (4) for performing speech synthesis by referring to the word dictionary (5) based on the reading information generated by the reading information generation means (3). A device for creating a word dictionary (5), and a furigana registration means (10) for registering the reading of a word in the word dictionary (6) with a standard furigana described in a Japanese dictionary or the like, When the word contains a kanji character, a kanji delimiter insertion means (11) for inserting a kanji delimiter in the furigana, the furigana in which the kanji delimiter is inserted, and the registered word are used for voice synthesis. Dictionary information registration means (12) for generating reading information of and registering it in the word dictionary.

In the embodiment of the present invention, a furigana registration means (10) for registering the reading of a word in the word dictionary (6) with a standard furigana written in a Japanese dictionary, and a reading of a single kanji character. A single kanji dictionary (21) in which is registered, and a kanji delimiter automatic insertion means (22) for automatically inserting a kanji delimiter in the furigana registered by the furigana registration means using the single kanji dictionary (21). And dictionary information registration means (1) for generating reading information for voice synthesis from the furigana in which the Kanji delimiter is inserted and the word to be registered.
2) and may be provided.

In the embodiment of the present invention, the furigana registration means (10), the kanji delimiter insertion means (11), the dictionary information registration means (12), the kanji delimiter automatic insertion means (22) are on a computer. The processing and functions are realized by the program executed in. In this case, the present invention can be implemented by reading the program from a recording medium recording the program or a communication medium and executing the program on a computer.

In a preferred embodiment, the method of the present invention for registering a word in a word dictionary for speech synthesis used for speech synthesis from input text information is (a)
As the reading of the word to be registered, a step of inputting standard reading information corresponding to a furigana written in a Japanese dictionary or the like, and registering the word and the furigana information in the dictionary in the dictionary format, ( b) If the word contains more than one kanji, insert kanji delimiters at the boundaries of each kanji reading to divide the reading of the word into individual kanji readings,
Outputting the reading information of the word to which the kanji delimiter is added together with the corresponding kanji; and (c) dividing the reading information of the word into the kanji delimiters and performing a predetermined conversion for each kanji. If the rule is met, a step of generating peculiar reading information for voice synthesis according to the conversion rule and registering it in the word dictionary.

Further, in the present invention, a single-kanji dictionary in which readings of single-kanji are registered in advance is provided, and (a) as a reading of a word to be registered, a standard kanji corresponding to a furigana written in a Japanese dictionary or the like Inputting the reading information and registering the word and the furigana information in the word dictionary in the form of a dictionary; (b) if the word includes a plurality of kanji, refer to the single kanji dictionary and Using kanji reading and furigana information of the word, a kanji delimiter is automatically inserted in the furigana reading at the boundary of each kanji reading of the word,
Outputting the reading information of the word to which the kanji delimiter is added together with the corresponding kanji; and (c) dividing the reading information of the word into the kanji delimiters and performing a predetermined conversion for each kanji. If the rule is satisfied, a step of generating peculiar reading information for voice synthesis according to the conversion rule and registering the reading information in the word dictionary is included.

[0021]

Embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a diagram showing a configuration of a first exemplary embodiment of the present invention.

Referring to FIG. 1, the apparatus according to the first embodiment of the present invention uses an input terminal 1 for inputting a text and a text analysis for dividing an input text into morphemes using part-of-speech information of a word dictionary 6. Based on the result of the means 2 and the text analysis means, the reading information generation means 3 for generating the reading information for the input text using the reading information stored in the word dictionary 6 and the reading information generation means 3 are generated. Based on the reading information of the text, a voice synthesis device 9 having a voice segment data 7 and a voice waveform generation means 4 for generating a synthetic voice using the prosodic pattern data 8 and dictionary information are registered in the word dictionary 6. To provide a furigana registration means 10, a kanji delimiter insertion means 11, a dictionary information registration means 12, a furigana input terminal 13, a kanji delimiter input terminal 14, and a dictionary information input terminal 15. It has been made.

From the furigana input terminal 13, the word dictionary 6
Input the pronunciation of the word to be registered in to in the furigana format as in a Japanese dictionary.

The input furigana is registered in the kanji delimiter insertion means 11 through the furigana registration means 10 in the dictionary format of the word to be registered and furigana information, and output to the kanji delimiter insertion means 11.

The kanji delimiter input terminal 14 uses a word and furigana information to insert a kanji delimiter into the furigana information,
It is output to the dictionary information registration means 12.

The dictionary information registration means 13 converts the words and the kana characters into which kanji characters are inserted into the reading information of the word dictionary 6 and registers them.

Further, from the dictionary information input terminal 15, part-of-speech information and accent information to be registered in the word dictionary 6 are input, and the dictionary information registration means 12 puts them together into the data format of the word dictionary 6.

Next, the operation of the embodiment of the present invention will be described with reference to FIG.

When registering a word in the word dictionary 6 used in the speech synthesizer 9, the furigana of the word to be registered is input from the furigana input terminal 10 to the furigana registration means 10.

For example, when "Tokyo" is registered in the dictionary, "Tokyo" is input.

The furigana registration means 10 registers "Tokyo" as a reading of "Tokyo" and outputs it to the kanji delimiter insertion means 11.

The Kanji delimiter inserting means 11 inserts a Kanji delimiter into "Tokyo" through the Kanji delimiter input terminal 14. The kanji delimiter is a symbol inserted to divide the reading of a word consisting of a plurality of kanji into the readings of each kanji, and is represented by using "#", for example.

When a kanji delimiter is inserted in "Tokyo" as an example,
It looks like "Today # today". That is, it is inserted at the boundary between the readings of “east” and “Kyo”.

The reading information of words with kanji delimiters is
It is output to the dictionary information registration means 12 together with the corresponding kanji.

The dictionary information registration means 18 divides the reading information of a word into Kanji delimiters, and generates reading information for speech synthesis for each Kanji according to a rule (conversion rule) shown below as an example.

Rule 1: "Edan" and "I" → "I" are converted into phonographic symbols.

Rule 2: “Odan” and “U” → “U” are converted into phonographic symbols.

Rule 3: Continuation of the same vowel → convert subsequent vowels into prolongation symbols.

[0039] For example, reading "Tokyo""To#Kyo"
Then, first, it is divided into the reading "to" for "east" and the reading "kyo" for "kyo."

Next, the reading of "east" is converted into "to" according to (2) of the above rule.

Subsequently, "Kyo" is also converted into "Kyo" according to rule (2).

Finally, the divided readings are joined together to generate the reading information for the synthetic voice "TOKYO".

In this way, even after dividing into kanji readings, the above rule is applied to each divided kanji.
There is no conversion error in which "veal" is changed to "koshi" or Ookayama is changed to "oo kayama", and only the reading of single kanji can be lengthened.

FIG. 2 is a diagram for explaining the first embodiment of the present invention, in which the reading of the word input from the furigana registration means 10 is inserted into the kanji delimiter insertion means 11 to insert a kanji delimiter. In the dictionary information insertion means 12, an example is shown in which the phoneme is changed according to the conversion rule and the reading for synthetic speech is generated.

For example, the flow of processing will be described using the "large item".

First, the reading of "large item" is input from the furigana registration input terminal 13.

In the furigana registration means 10, "Omono" is registered as the word notation "large" and its reading.

Next, a kanji delimiter "#" is inserted from the kanji delimiter input terminal 14, and the kanji delimiter insertion means 11
"Oh #mono" is registered as the reading of "big game".

The dictionary information registration means 12 separates "Oh #mono" into "Oh" and "Omono" readings before and after the kanji delimiter, and the readings that match the phoneme change rule are:
Convert according to the rules. In this case, since the reading "O" of "Large" matches the above-mentioned rule 3, it is converted into "O".

In the dictionary information registration means 12, words are once separated into readings for each single Chinese character, the readings are converted according to rules, and then the readings are connected again. At this time, the kanji delimiter is not inserted.

In this way, it is possible to generate the reading "Ohmono" of the synthetic voice dictionary for "large".

Next, a second embodiment of the present invention will be described. FIG. 3 is a diagram showing the configuration of the second exemplary embodiment of the present invention. Referring to FIG. 3, in the second embodiment of the present invention, instead of the Kanji delimiter inserting means 11 of the first embodiment described with reference to FIG. 1, a single Kanji dictionary 21 and a Kanji delimiter automatic inserting means are provided. 22 and 22.

The reading of single kanji is registered in the single kanji dictionary 21, and the kanji segment automatic insertion means 22 uses the reading of single kanji and the furigana of the word registered by the furigana registration means 10. , Automatically insert kanji delimiters in furigana.

The furigana in which the kanji delimiters are inserted is converted into reading information for voice synthesis in the dictionary registration means 12, as in the first embodiment.

FIG. 4 is a diagram showing an example of the single-kanji dictionary 21 in the second embodiment of the present invention. As shown in FIG. 4, in the single-kanji dictionary 21, on-reading and kun-reading are registered for each single kanji.

In the case of kun reading, the sending kana is enclosed by ().

Taking the registration of "English" as an example, a method of automatically inserting a kanji delimiter in a furigana will be described.

First, the furigana registration action recording means 10 registers the furigana "Eigo" of "English".

The kanji character automatic insertion means 22 separates "English" into single kanji characters, and from each reading information,
Search for Yomi that matches the furigana. in this case,
As the reading of “English”, the reading “on” of “on” and the reading “go” of “word” are selected.

Next, the selection result is associated with the "English" furigana "Eigo", and the "English" furigana "Eigo" is used.
Determines that it can be separated into the readings of single kanji, "English" and "word," and inserts a kanji delimiter between them to finally generate the reading of "ei #go".

The reading added with the kanji delimiter is converted by the dictionary information registration means 12 into the reading "Ego" for voice synthesis.

Thus, for words registered in the dictionary,
It is possible to automatically generate the reading information for speech synthesis by inputting a furigana as in a Japanese dictionary as the reading information and manually or automatically inserting a kanji delimiter in the reading information.

[0063]

As described above, according to the present invention,
By adding furigana to the words you want to register and inserting kanji delimiters, it is possible to automatically generate reading information for voice synthesis, depending on the difference between furigana and utterance content. It is possible to eliminate the input error of the reading information, and it is possible to facilitate the registration work and improve the efficiency.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of a first exemplary embodiment of the present invention.

FIG. 2 is a diagram for explaining the first embodiment of the present invention and is an example of word reading information.

FIG. 3 is a diagram showing a configuration of a second exemplary embodiment of the present invention.

FIG. 4 is a diagram showing an example of a single-kanji dictionary according to the second embodiment of the present invention.

[Explanation of symbols]

1 Text input terminal 2 Text analysis means 3 reading information generation means 4 Voice waveform generation means 5 Synthetic voice output terminal 6 word dictionary 7 Speech segment data 8 Prosody pattern data 9 Speech synthesizer 10 Furigana registration means 11 Kanji character insertion means 12 Dictionary information registration means 13 Furigana input terminal 14 Kanji separated input terminal 15 Dictionary information input terminal 21 Single Kanji dictionary 22 Automatic Kanji character insertion means

Claims (5)

(57) [Claims] 1. A text analysis unit for analyzing input text information, a reading information generation unit for generating reading information by referring to a word dictionary from an analysis result output from the text analysis unit, and the reading information generation. A device for creating the word dictionary in a voice synthesizing device comprising: a voice waveform generating unit for performing voice synthesis by referring to the word dictionary based on the reading information generated by the means; Using the standard kana kana registration means for registering the reading of words in standard kana kana, the single kanji dictionary in which the reading of single kanji is registered, and the single kanji dictionary A kanji delimiter automatic insertion means for automatically inserting a kanji delimiter in the furigana registered by the kana registration means, the furigana in which the kanji delimiter is inserted, and the registration unit A dictionary registration device for voice synthesis, comprising: dictionary information registration means for generating reading information for voice synthesis from words and registering it in the word dictionary. 2. A device for registering a word in a word dictionary for voice synthesis for voice-synthesizing from input text information, the standard being equivalent to a furigana written in a Japanese dictionary as a reading of the registered word. Inputting the reading information of, and registering the word and the furigana information in the word dictionary in the form of a dictionary, a single kanji dictionary in which readings of single kanji are registered in advance, and the word includes a plurality of kanji. When including, the kanji delimiter is automatically added to the reading of the single kanji by referring to the single kanji dictionary and the reading of the single kanji and the reading of the kana in the boundary of the reading of each kanji of the word. Means for outputting the reading information of the word to which the Kanji delimiter is added together with the corresponding Kanji, and the reading information of the word is divided for each Kanji delimiter, and a predetermined conversion predetermined for each Kanji Applies to the rules In this case, a unit for generating peculiar reading information for voice synthesis according to the conversion rule and registering it in the word dictionary. 3. A method of registering a word in a word dictionary for voice synthesis used for voice-synthesizing from input text information, comprising a single-kanji dictionary in which readings of single-kanji are registered in advance. ) As the reading of the word to be registered, a step of inputting standard reading information corresponding to a furigana written in a Japanese dictionary and registering the word and the furigana information in the dictionary format in the dictionary, b) When the word includes a plurality of kanji, refer to the single kanji dictionary to read the single kanji and use the furigana information of the word, and use the furigana of the boundary of each kanji of the word. A kanji delimiter is automatically inserted in the reading of, and the reading information of the word with the kanji delimiter is output together with the corresponding kanji, and (c) the reading information of the word for each of the kanji delimiters. Split and han When it corresponds to a predetermined conversion rule predetermined for each character, a step of generating peculiar reading information for speech synthesis according to the conversion rule and registering it in the word dictionary is included. How to register a voice synthesis dictionary. 4. An apparatus for registering words in a word dictionary for speech synthesis for synthesizing speech from input text information by a computer, comprising a single Kanji dictionary in which readings of single Kanji are registered in advance. ) As a reading of the word to be registered, a process of inputting standard reading information corresponding to the furigana written in a Japanese dictionary and registering the word and the furigana information in the dictionary in the dictionary format, b) When the word includes a plurality of kanji, refer to the single kanji dictionary to read the single kanji and use the furigana information of the word, and use the furigana of the boundary of each kanji of the word. The kanji delimiter is automatically inserted into the reading of No. and the reading information of the word with the kanji delimiter is output together with the corresponding kanji. (C) The reading information of the word is divided for each kanji delimiter. Each kanji A predetermined conversion rule defined for the above, a process of generating peculiar reading information for voice synthesis according to the conversion rule and registering it in the word dictionary; and (a) to (c) above. A recording medium recording a program for causing the computer to execute processing. 5. The conversion rule converts the "i" of "i" following the "edan" into a syllabary symbol, and converts the "u" of "odan" and "u" into a syllabic symbol. , claims the same vowels when continuously converts subsequent vowels prolonged sound of symbols including at least a rule that, it is characterized by
2. The voice synthesis dictionary registration device described in 2 .