JPH09330098A - Voice synthesizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent reading error by preferentially retrieving a provisional word dictionary by a language analyzing section before retrieving the phonetic character string data for the written data of words cut down using a standard language dictionary. SOLUTION: A kana-printed-alongside-kanji data analyzing section 5 separates and extracts kana-printed-alongside-kanji data from an inputted kanji-kana mingled section and the kana-printed-alongside-kanji data, and a temporary word dictionary 6 composed from RAM stores the kana-printed-alongside-kanji data extracted by the analyzing section 5 and the written data corresponding to this kana-printed-alongside-kanji data. And the dictionary 6 retrieves the words cut down by a language analyzing section 1 in preference to a standard word dictionary 2 and gives the results to the analyzing section 1. On the other hand, the analyzing section 1 inputted with kanji-kana mingled sentences and kana-printed-alongside-kanji data applies grammatic interpretation to the written data of the kanji-kana mingled sentences to cut down words, converts written data for each word to phonetic character string data for each word and to form the phonetic character string data of the whole kanji-kana mingled sentences.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice synthesizing device for converting text data of a mixture of kanji and kana into phonetic character string data by using a word dictionary and reading the same. , Abbreviation) for preventing erroneous conversion into phonetic character string data.

[0002]

2. Description of the Related Art In recent years, speech synthesis has been used to read out general information sent by FM multiple character broadcasting and kanji / kana mixed sentences of traffic information by VICS (Vehicle Information Communication System) as text data (notation data). Has been. A large-capacity word dictionary is required in the language analysis processing for converting this notation data into phonetic character string data for reading it aloud. That is, the word dictionary
A large number of word notation data and phonetic character string data are stored as a pair.

[0003]

However, in the conventional word dictionary, no matter how the data is enriched, it is possible to correspond to a lot of words such as newly created road names, product names, and abbreviations. Since it is impossible, there is a problem that a reading error occurs. By the way, since the above-mentioned coined words and abbreviations are apt to be misread when they are displayed, it is considered that the FM broadcast side and the VICS transmission side temporarily add ruby to them. Therefore, by using this ruby, a reading error can be eliminated even in speech synthesis.

It is an object of the present invention to provide a voice synthesizing device capable of preventing a reading error even if temporary ruby is not added to a coined word or an abbreviation.

[0005]

In order to solve the above problems, the present invention converts a notation data of words forming a kanji / kana mixed sentence into phonetic character string data and synthesizes it into a voice. In, a standard word dictionary that permanently registers the notation data of the word and the phonetic character string data, and a ruby added by phonetic character string data to a part of the notation data of the kanji / kana mixed sentence. A ruby data analysis unit that performs an analysis to extract the notation data corresponding to, a provisional word dictionary that provisionally registers the notation data and ruby data extracted by the ruby data analysis unit, and cuts out the word from the kanji-kana mixed sentence. , Prior to searching the phonetic character string data for the cut-out word notation data using the standard word dictionary, preferentially using the temporary word dictionary That includes a language analysis unit. By this means, even words that are not registered in the standard word dictionary, such as coined words and abbreviations, can be searched using the temporary word dictionary, and reading errors can be prevented.

[0006] Words registered in the temporary word dictionary by a certain amount are moved to the standard word dictionary. By this means, the frequency of re-registration of the standard word dictionary is reduced and the registration capacity of the temporary word dictionary is not increased. Of the words registered in the provisional word dictionary, the most frequently used words are moved to the standard word dictionary. By this means, the registration frequency of the standard word dictionary and the registration amount of the temporary word dictionary can be further balanced.

Of the words registered in the temporary word dictionary, the ones with low frequency of use are deleted. This means is advantageous when the provisional word dictionary is limited and the standard word dictionary cannot be re-registered. Of the words registered in the provisional word dictionary, those that are old and are not used frequently at present are deleted. By this means, older ones are deleted preferentially among comrades who are less frequently used.

The ruby data analysis section extracts ruby data from the written data of the kanji / kana mixed sentence using a code indicating the position of the ruby data. By this means, ruby data can be easily distinguished from other notation data.

[0009]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing an outline of a main part of a speech synthesizer according to the present invention. As shown in the figure, in the speech synthesizer, the language analysis unit 1 that inputs a kanji / kana mixed sentence and ruby data performs grammatical interpretation on the written data of the kanji / kana mixed sentence to cut out a word, and The notation data is converted into phonetic character string data for each word to form phonetic character string data for the entire kanji-kana mixed sentence. The standard word dictionary 2 stores phonetic character string data for reading out the notation data for the word notation data, retrieves the phonetic character string data corresponding to the word notation data from the language analysis unit 1, and returns the result. The phonetic character string data of is given to the language analysis unit 1. In this case, for the notation data of the word to which the ruby data is added, the ruby data is used as the phonetic character string data without performing a search in the standard word dictionary 4. The sound processing unit 3 connects the phonemes based on the phonetic character string data of the kanji / kana mixed sentence obtained by the language analysis unit 1 to form a speech synthesis waveform. The phoneme dictionary stores phoneme data with respect to phonetic character string data, and the sound processing unit 3
The phoneme data corresponding to the phonetic character string data from is searched and the resulting phoneme data is given to the acoustic processing unit 3.

The ruby data analysis unit 5 separates and extracts ruby data from the input kanji / kana mixing part and ruby data. The provisional word dictionary 6 is composed of a RAM, and stores the ruby data extracted by the ruby data analysis unit 5 and notation data corresponding to this ruby data. Provisional word dictionary 6
Searches the words extracted by the language analysis unit 1 with priority over the standard word dictionary, and returns the results to the language analysis unit 1
Give to.

The following specific example will be described. It is assumed that the following kanji and kana mixed sentences and ruby sentences are received as the notation data. "Kyoshin Expressway 10km uphill traffic jam" Here, "Kyoshin Expressway" is the name of the new road, and (Kyoshin Expressway) is the ruby sentence and the correct reading. Actually, the ruby is displayed in lower case on the Kyoshin Expressway. Such a position display code is different from the codes of other notation data. Utilizing this difference, the ruby data analysis unit 5 extracts this ruby data (Kyoshin Kosoku) and the corresponding notation data "Keijin Expressway", and the temporary word dictionary 6 reads (Only displayed)
To store.

Notation data Phonetic character string data ... Keijin high speed Kyoshin Kosoku ... The standard word dictionary 2 stores words (only a very small part is displayed) as follows.

Notation data Phonetic character string data ……… Upward Nobori 10 Juu km Kyokyo 〃 Keijin Ko 〃 Shin High Speed Kosoku ……… If you search the notation data "Keishin High Speed" in the standard word dictionary 2, Since there is no "Kyojin Expressway" in Dictionary 2,
For example, since the language analysis processing is performed by dividing into “K”, “God”, and “High speed”, the probability of reading errors increases. Language analysis unit 1
Then, even if there is no ruby added temporarily, it is possible to search using the provisional word dictionary 6, so that a reading error is eliminated.

FIG. 2 is a flow chart for explaining a series of operations of the speech synthesizer of the present invention. In step S1, notation data is input. In step S2, the ruby sentence is extracted from the kanji / kana mixed sentence and the ruby sentence that form the notation data. In step S3, it is determined whether there is ruby data in the extraction process. If this determination is "NO", the flow proceeds to step S6.

In step S4, the judgment is "YE
If it is "S", it is judged whether or not it is already registered in the temporary word dictionary 6.
If this determination is "YES", the process proceeds to step S6. In step S5, if the registration has not been completed yet, the notation data corresponding to the coined word, the abbreviation, etc. to which the ruby is temporarily added are written together with the phonetic character string data given by the ruby, and the temporary word dictionary 6
Register with.

In step S6, the phonetic character string data of the word is first searched using the temporary word dictionary 6. In step S7, the phonetic character string data is searched using the standard word dictionary 2 except for the words searched in the temporary word dictionary 6. As a result, the display data that has been input is "Kyoshin Kosoku", "Nobori",
It is converted into phonetic character string data such as “ju”, “kilometer”, and “jutai”.

In step S8, acoustic processing is performed. In step S9, it is output as voice synthesis waveform data. Therefore, according to the present invention, even if a ruby is temporarily added to a coined word or an abbreviation and thereafter it is not added, since it is registered in the temporary word dictionary 6 during the addition, a reading error is eliminated.

When the number of words registered in the temporary word dictionary 6 reaches a certain amount, the words may be collectively registered in the standard word dictionary 2. Although it is complicated to register each word in the standard word dictionary 2, the complexity is reduced because it is performed collectively, and the capacity of the temporary word dictionary 6 can be reduced. Also, the words that are frequently used among the registered words in the temporary word dictionary 6 are moved to the standard word dictionary 2. This is to balance the storage capacities of the temporary word dictionary 6 and the standard word dictionary 2.

Further, among the registered words in the provisional word dictionary 6, words having a low frequency of use may be simply deleted and words having a high frequency of use may be left. This is advantageous when the storage capacity of the temporary word dictionary 6 is limited. Further, among the registered words in the provisional word dictionary 6, ones that have been registered for a long time and that are currently used less frequently may be deleted. For the same reason as above.

[Brief description of drawings]

FIG. 1 is a diagram showing an outline of a main part of a speech synthesizer according to the present invention.

FIG. 2 is a flowchart illustrating a series of operations of the speech synthesizer of the present invention.

[Explanation of symbols]

 1 ... Language analysis unit 2 ... Standard word dictionary 5 ... Ruby data analysis unit 6 ... Provisional word dictionary

Claims (6)

[Claims] 1. A speech synthesizer for converting notation data of words constituting a kanji / kana mixed sentence into phonetic character string data and synthesizing it into a voice, wherein the notation data of the word, the phonetic character string data and permanent A standard word dictionary to be registered dynamically, a ruby data analysis unit for performing an analysis to extract the ruby and the corresponding notation data given by phonetic character string data to a part of the notation data of the sentence containing the kanji and kana, and A provisional word dictionary for temporarily registering the notation data and the ruby data extracted by the ruby data analysis unit, cutting out the word from the kanji / kana mixed sentence, and the phonetic character string data for the notation data of the cut out word, A speech synthesizer comprising a language analysis unit that preferentially searches the temporary word dictionary before searching using the standard word dictionary. 2. The speech synthesizer according to claim 1, wherein the words registered in the temporary word dictionary by a predetermined amount are moved to the standard word dictionary. 3. The voice synthesizing apparatus according to claim 1, wherein words that are frequently used among words registered in the temporary word dictionary are moved to the standard word dictionary. 4. The speech synthesizer according to claim 1, wherein words having a low frequency of use are deleted from the words registered in the temporary word dictionary. 5. The speech synthesis apparatus according to claim 1, wherein among words registered in the temporary word dictionary, words that have been registered for a long time and are currently infrequently used are deleted. 6. The speech synthesis apparatus according to claim 1, wherein the ruby data analysis unit extracts ruby data from the written data of the kanji / kana mixed sentence using a code indicating the position of the ruby data. .