JPS63216100A - Voice synthesizer - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Purpose of the Invention (Industrial Application Field) The present invention provides a speech system that can synthesize speech with good naturalness by simple processing of input word information given as a symbol string code. This relates to a synthesis device.

(Prior Art) Recently, prosodic parameters and phonological parameters of input word information converted into symbol strings are determined, and speech is regularly synthesized according to these metrical parameters and phonological parameters. This type of speech synthesis, together with speech recognition processing technology, has become an important technology for realizing highly natural man-machine interfaces. However, in such rule-based synthesis of speech, in order to obtain highly natural synthesized speech, it is important to accurately reproduce the accent specific to the word.

Therefore, in conventional speech synthesis devices, the accent type unique to each simple word and the accent information of compound words consisting of a combination of multiple simple words are registered as an accent dictionary, and the input is input to this accent dictionary. It looks for an accent that is unique to that input.

However, such conventional speech synthesis devices have the disadvantage that they cannot process alphanumeric characters and the like.

(Problems to be Solved by the Invention) As described above, the conventional speech synthesizer has a problem in that it cannot process special codes representing alphanumeric characters and the like.

The present invention was created in consideration of these problems, and its purpose is to process not only normal character codes but also special codes representing alphanumeric characters, etc., and to efficiently perform comparisons with accent dictionaries. An object of the present invention is to provide a speech synthesis device that can output synthesized speech with higher intelligibility and naturalness than before by performing the following steps.

[Structure of the Invention] (Means for Solving the Problems) In order to achieve the above object, the present invention provides an input means for inputting word information consisting of a special code and a character code converted into a symbol string, and a special code. a conversion means that converts the character code sandwiched between the special codes into a predetermined character code string as a group, and converts the accent information of the character code string converted by this conversion means for each character code string. The present invention is characterized by comprising means for determining prosodic parameters and phonological parameters from accent information and input word information, and means for generating synthesized speech according to the prosodic parameters and phonological parameters.

(Operation) The special code inputted from the conversion means number is converted into a predetermined character code string. Furthermore, the character codes sandwiched between the special codes are converted as a group into a predetermined character code string. Accent information is obtained for each character code string obtained in this manner. That is, certain accent information is given to one special code, and certain accent information is given to character codes sandwiched between special codes. Prosodic parameters and phonological parameters are determined from the accent information and input word information obtained in this way, and synthesized speech is generated.

Therefore, in the present invention, since the special code and the character code sandwiched between the special codes are processed separately, it is possible to easily and accurately obtain word-specific accent information indicated by the converted character code string.

(Example) Hereinafter, an example of the present invention will be described in detail based on the drawings.

FIG. 1 is a block diagram of a speech refinement device according to an embodiment of the present invention.

As shown in the figure, this speech synthesis device includes a symbol string input section 1, a code conversion section 2, a conversion table 3, and a word matching section 4.
, an accent dictionary 5, a compound word accent test section 6, a prosodic parameter generation section 7, a phoneme sequence test section 8, a speech segment file 9, a phoneme parameter generation section 10, and a speech synthesis section 11.

The symbol string input section 1 inputs word information in the form of a symbol string consisting of a special code and a character code.The code conversion section 2 converts the special code and character code input from the symbol string input section 1 into a predetermined character code string. It is converted into . in this case,
One special code is converted into one character code string, and character codes sandwiched between special codes are converted as a group into one character code string. Furthermore, there is a gap between the character code string converted from the special code and the character code string converted as a group from the character code sandwiched between the special codes.
” (space).

In the conversion table 3, character code strings corresponding to special codes and character codes are registered. FIG. 2 shows the special codes in this conversion table 3 and the character code strings corresponding to the special codes. For example, if the special code is "28", the corresponding character code string is "brackets".

Similarly, when the special code is "80", the corresponding character code string is "r Kabushikigaisha".

The word matching section 4 matches the character code string obtained from the code conversion section 2 with the accent dictionary 5 to obtain accent information corresponding to each character code string. At this time, if there is a part separated by "1" in the character code string, the word matching unit 4 performs matching with the accent dictionary 5 separately.
The accent information obtained in the compound word accent test section 6
, the prosodic parameter generator 7 and the phonetic sequence tester 8.

The compound word accent testing section 6 tests the accent in the compound word according to the type of the word and the accent information that has been matched and detected, when the input word that has been word-matched constitutes a compound word. This test provides information on accents that change when multiple simple words form a compound word.

The prosodic parameter generation section 7 determines the prosodic parameters of the character code series subjected to the accent test in this manner, and the prosodic parameter sequence is provided to the speech synthesis section 11. Also, the code obtained from the special code converter 2 is “
1'', the silence parameter is given to the speech synthesis section 11 for a certain period of time.

The phoneme sequence testing unit 8 tests whether the nine rhymes are nasalized or devoiced from each phoneme information of the character code sequence obtained by the word matching unit 4. Then, a phoneme sequence corresponding to the character code sequence is generated according to the test result, and this is provided to the phoneme parameter generation section 10. This phoneme parameter generation section 10 refers to the speech segment file 9 to generate a phoneme parameter string corresponding to the phoneme sequence, but the code obtained from the special code conversion section 2 is "1".
In this case, a phoneme parameter string representing silence is created without referring to the speech unit file 9.

The speech synthesis unit 11 speech-synthesizes the word indicated by the input symbol string according to the created phonetic parameter string and the above-mentioned prosodic parameter string.

Next, the operation of this embodiment will be explained. Now, as shown in FIG. 3(a) in the symbol string input section 1, "80, C4, B3,
BC, CA, DE, 28, CA, D7, C01DE, 2
Assume that a symbol string "9" is input. Of this symbol string, "80", "28", and "29" are special codes, and the others are normal character codes. The code conversion unit 2 converts such a symbol string into a character code string. FIG. 3 shows the state of the conversion. In other words, the code conversion unit 2 refers to the conversion table 3 and converts it into a character code string. In other words, "80" is converted to "kabushikigaisha", "C4", "B3", rB
cJ, rcAJ, and “DE” are respectively “to” and “tsu”.
Converted to ``shi'', ``ha'', and ``''. "28" is converted to "brackets" rCAJ, "D7", "CO", rD
E.

are converted to ``ha'', ``u'', ``yu'', and ``'' respectively, and ``29'' is converted to ``bracket''. In this case, a "U" (space) is inserted between "Kabushikigaisha" and "Toshiba". Similarly, "1" is inserted between "Toushiba" and "bracket". As can be seen from the figure, the special code itself is converted into a character code string. On the other hand, normal character codes sandwiched between special codes are treated as a group of character code strings after conversion. In other words, "Toshiba" is treated as a group of character code strings.

Next, the character code string obtained by the code conversion section 2 is input to the word matching section 4, where it is checked against an accent dictionary 5 to obtain accent information. The comparison with this accent dictionary is performed for each word separated by "-". In other words, "Turnip."

The character string "Shikigaisha" is compared with the accent dictionary 5 to obtain predetermined accent information.The character string "Toshiba" is compared with the accent dictionary IF 5 and predetermined accent information is obtained. in this way"
Since accent matching processing is performed for each word separated by "1", for example, even if the character code string "Toshiba" is not registered in the accent dictionary 5, it will affect the accent verification processing of other words. There isn't.

The compound word accent testing section 6 tests the accent of a compound word according to the character code string and accent information, and provides information on the accent when a plurality of simple words constitute a compound word. Thereafter, based on the character code string and accent information obtained in this way, prosodic parameters and phoneme parameters are generated by the prosodic parameter generation section 7, the phoneme sequence verification section 8, the speech segment file 9, and the phoneme parameter generation section 10. Ru. In this case, since the character code string "-" is included, a parameter string representing a pause (for example, a silent section of 250113) is also generated between the words that are 1ljl adjacent to that (LJ,).These parameters are used in speech synthesis. The synthesized speech is inputted to the unit 11 and outputted.

In this way, in this embodiment, when a special code is included in an input symbol string, it is converted into a character code string, the character codes sandwiched between the special codes are converted into a character code string as a group, and the converted characters are Since each code string undergoes word matching processing separately, word-specific accent information indicated by those character code strings can be obtained easily and with high precision.

Furthermore, by providing a pause between a special code and an adjacent code string, great practical effects such as increasing intelligibility can be achieved.

Note that the present invention is not limited to the above-described embodiments; for example, the symbol representing a pose may be a symbol other than "0", and the correspondence between special codes and character code strings is also limited to the above-described embodiments. What is not to be said is a matter of enthusiasm.

[Effects of the Invention] As explained in detail above, according to the present invention, not only normal character codes but also special codes representing alphanumeric characters can be processed and collation with accent dictionaries can be performed efficiently, thereby improving intelligibility. and can output highly natural synthesized speech.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of a speech synthesis device according to an embodiment of the present invention, Fig. 2 is a diagram showing an example of the structure of a conversion table, and Fig. 3 is an input symbol string and characters obtained by converting it. FIG. 3 is a diagram showing an example of a code string.

Claims (2)

[Claims] (1) An input means for inputting word information consisting of a special code and a character code converted into symbol strings, and an input means for converting the special code into a predetermined character code string and converting the special code into a character code string sandwiched between or adjacent to the special code. a conversion means for converting a group of character codes with silent intervals or separated spaces into a predetermined character code string; a means for obtaining accent information for each character code string converted by the conversion means; A speech synthesis device comprising: means for obtaining prosodic parameters and phonological parameters from information and input word information; and means for generating synthesized speech according to the prosodic parameters and phonological parameters. (2) The conversion means converts character code strings as a group from character code strings converted from special codes and character code strings sandwiched between or adjacent to the special codes with silent intervals or delimiters. 2. The speech synthesis device according to claim 1, wherein a voiceless section is inserted between the voice synthesis sections.