JPH067341B2 - Speech synthesis method - Google Patents

Description

The present invention relates to speech synthesis, and more particularly to a speech synthesis method suitable for synthesizing speech from an input composed of special character strings of arbitrary contents including symbols, alphabets, special symbols and the like. .

[Background of the Invention]

A method of synthesizing a voice by inputting an arbitrary character string is called rule synthesis or law synthesis, and various techniques have been developed. However, in the technology so far, the input method is only a kana character string, a roman character string, or a mixed kanji / kana sentence, and a method of handling alphabets and symbols appearing in everyday sentences has not been developed. Furthermore, when outputting to the synthesis unit in the same format as that of a computer terminal or the like, various terminal control symbols are mixed, and it is difficult to convert the input code string as it is to speech. Moreover, since a dictionary cannot prepare a new abbreviation newly defined by the user, it cannot be processed by the method using the dictionary.

[Object of the Invention]

It is an object of the present invention to provide a method for converting a sentence of arbitrary content represented by a code string including kana, kanji, alphabet, various symbols, and control symbols into voice.

[Outline of Invention]

To achieve the above object, in the present invention, a symbol or alphabet string is used as a headword and a dictionary having its reading as a content is provided before the conventional rule synthesis processing unit, and the input symbol or alphabet string is read. Is converted to a character string (for example, a kana string) and then passed to the rule composition processing unit. However, some symbols, such as commas, are used as notation symbols in sentences, and these need to be passed as they are to the rule synthesis processing unit, and they are excluded from the conversion. In order to make these symbols themselves to be read, an escape symbol is prepared, and when surrounded by escape symbols, they are converted to the reading. In principle, symbols are processed in units of one symbol, and only when an escape symbol appears, a series of strings is processed. For the alphabet, dictionary processing is performed in units of a series of alphabets. Abbreviations and foreign words that have a habit of reading such as NATO / Nato / are converted to their readings (Kana string, etc.) in the alphabet string, but the abbreviations defined by the user are read by the user. Unless is registered in the dictionary, each character forming the alphabet string is converted into reading. Special characters such as Greek letters and numbers can be handled in the same way.

Example of Invention

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an apparatus for explaining this embodiment.
In FIG. 1, a kana-kanji mixed sentence including special characters such as symbols and alphabets is input to the control unit 2 from the input terminal 1. According to a predetermined procedure, the control unit extracts information necessary for synthesis from the voice synthesis control information memory 3, creates control information for the synthesis unit 4, sequentially sends it to the synthesis unit 4, synthesizes it into a voice waveform, and outputs it from the output terminal 5. Output. For the procedure for converting the kana-kanji character string into the synthesis control information, see, for example, S82- in Reference (1) Material of Acoustical Society of Japan.
The technology disclosed in 08 "Study of Japanese text conversion" or the like can be used. A technique for synthesizing a voice from synthesis control information is described in, for example, the literature (2) IEICE Transactions Vo. J61-D, No.11 pp.858-86
5 (78/11) “Voice synthesis method using PARCOR-VCV” and can be easily realized.
Therefore, here, it is sufficient to show only the conversion procedure from the kana-kanji mixed input including symbols and alphabets to the kana-kanji mixed sentence. This conversion is executed using the control unit 2 and various conversion dictionaries 6. For simplicity, enter the
Although it is assumed to be composed of kanji, symbols and alphabets, the processing of Greek characters and numeral strings can be realized by adding the same processing as the processing of alphabets and so will be omitted. FIG. 2 is a flowchart showing the procedure of this conversion. In FIG. 2, when a kana-kanji mixed sentence including various symbols and alphabets is input from the input terminal 21, the input code is sequentially determined by the determination unit 22 and assigned to each process. The symbols are sent to the text notation symbol determination unit 23, the Kana and Kanji are sent to the rule processing unit 33 that creates the voice synthesis control information from the Kana / Kanji input sentence, and the alphabet is sent to the alphabet character string creation unit 28 via the path 27. To be In the sentence notation symbol determination unit 23, commas (,), dots (.), Parentheses (“,”), etc. are sent to the escape processing determination unit 24, and other symbols are symbol reading dictionary processing unit 25.
Send to. If the input symbol is surrounded by the escape symbol, the escape processing determination unit 24 determines the symbol reading dictionary processing unit 25.
, And others are sent to the rule processing unit 33 as they are. The symbol reading dictionary processing unit 25 looks up the symbol reading dictionary in the various conversion dictionaries 6 of FIG. 1 and uses the reading corresponding to the symbol as Kana notation.
It is replaced with the input symbol and sent to the rule processing unit 33. Escape symbols are treated as unreadable.

The alphabetic character string creating unit 28 determines that the alphabetic character string has ended when the input code string changes from the alphabet to another code, and the alphabetic character string reading dictionary processing unit 29 causes the various characters shown in FIG. The dictionary for reading alphabetic character strings in the conversion dictionary 6 is drawn and converted into a kana character string for the reading. If it is not in the dictionary, it is converted into alphabet reading for each character in the alphabet string,
The processing unit 32 processes it as a Kana character string, and the rule processing unit 33
Send to. The rule processing unit 33 performs the process as in the above-mentioned document (2), converts it into the control code of the synthesizing unit, and outputs it.

All code strings other than kana are treated like kanji, converted to kana with a batch dictionary containing kanji, and only those that are not in the dictionary are converted to kana characters by the same process as the reading conversion process 32 for each alphabetic character. A similar effect can be obtained in the conversion procedure.

〔The invention's effect〕

As described above, according to the present invention, it is possible to synthesize a voice in an easily audible format from an arbitrary sentence including kana, kanji, and symbols other than the symbols on the sentence notation.

[Brief description of drawings]

FIG. 1 is a block diagram for explaining an embodiment of the present invention, and FIG.
The figure is a flow chart for explaining the process of the present invention. 1 ... Input end, 2 ... Control part, 3 ... Voice synthesis control information memory, 4 ... Synthesis part, 5 ... Output end, 6 ... Various conversion dictionaries.

Claims (1)

[Claims] 1. A kana character and a kanji character as well as a sentence including at least an alphabet and a symbol are inputted, and each input code in the inputted sentence is sequentially judged by the control means,
In a voice synthesizing method of giving a reading corresponding to an input code by using a dictionary, synthesizing the reading regularly by a synthesizing means to synthesize a voice, when the determined input code is kana or kanji, the dictionary Is used to give a reading corresponding to the kana or kanji character string, and the readings are combined in a rule. When the determined input code is a symbol, it is determined whether the symbol is a predetermined sentence notation symbol. However, (1) when the symbol is a predetermined text notation symbol, it is further determined whether or not the symbol is enclosed by an escape symbol. If the symbol is enclosed by the escape symbol, the dictionary is provided as the dictionary. Using the symbol reading dictionary provided, the escape symbol is given a reading corresponding to the symbol as having no reading, and the reading is rule-synthesized, and when not enclosed by the escape symbol, (2) When the symbol is not a predetermined text notation symbol, the symbol corresponding to the symbol is given using the symbol reading dictionary, and the reading is rule-synthesized. When the determined input code is an alphabet, an alphabetic character string is created with a series of the alphabetic characters as a unit, and whether or not there is a reading corresponding to the alphabetic character string in the alphabetic character string reading dictionary provided as the dictionary. If it is determined that the alphabetic character string is in the alphabetic character string reading dictionary,
The reading is subjected to rule synthesis, and when the alphabet character string is not in the alphabet string reading dictionary, it is converted into alphabet reading for each character of the alphabet, and the reading is rule-synthesized. Synthesis method.