JP2667553B2 - Speech synthesizer - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizer for converting an input sentence composed of an arbitrary character / symbol string into speech.

[Prior Art] Generally, a speech synthesizer identifies each word by performing morphological analysis of an input character / symbol string using a dictionary, and obtains grammatical information such as part of speech of the identified word and accent information from the dictionary. , Set clauses, accents and pauses for exhalation paragraphs when words are linked from syntactic structure.
Then, when a compound word is included in the character symbol string, it is impossible to input all the compound words in the dictionary in advance, so that the prosody as the compound word is determined from the information of each word constituting the compound word. .

Next, an example of a conventional speech synthesizer will be described with reference to FIG.

First, a character / symbol string is input to the character / symbol string input unit 18 as an input.
Sent to 19.

Subsequently, the morphological analysis unit 19 performs a morphological analysis of the input character symbol string using the dictionary 20, identifies words, and extracts grammatical information including syntax structures such as parts of speech of the identified words and accents. .

Further, in order to control the prosody, the synthesized speech parameter generation unit 21 extracts, from the accent and syntax structure of each word identified by the morphological analysis in the morphological analysis unit 19, And set the pose. Further, a parameter time series of a duration, a pitch pattern, a power pattern, and a phoneme feature parameter for a synthesis unit corresponding to the uttered voice is extracted.

Next, the speech synthesizing unit 22 generates an actual synthesized speech waveform based on the above-described time series of parameters for speech synthesis. Then, the generated synthetic speech waveform is input to the synthetic speech output unit 23, and the synthetic speech output unit 23 outputs the synthetic speech waveform.

In the above-described conventional speech synthesizer, when determining the accent of a compound word in which nouns in a character / symbol string are continuously formed, the synthetic speech parameter generation unit 21 uses the compound word as a whole.
Assuming that only one accent is required, accents are combined based on rules using the accent type and the number of syllables of each word constituting the compound word, and one accent nucleus, that is, the position of the accent is set. Determines the accent of compound words.

[Problems to be Solved by the Invention] However, a compound word does not always have one accent position, and there is a compound word having a plurality of accent nuclei. For example, a compound word “arranged tansu” composed of the words “arrangement” and the word “tansu” has one accent core, while a compound word “arranged arranging” composed of the words “arrangement” and the word “arranged” is Since the compound word has two accent nuclei, processing based on the assumption that a compound word has only one accent nucleus will result in utterance with an incorrect accent.

The above-described conventional speech synthesizer has a problem that when a compound word having a plurality of accent nuclei is input, it is not possible to accurately add an accent. In order to solve this problem, it is conceivable to register all compound words having multiple accent nuclei in the dictionary, but it is difficult to realize it because the dictionary capacity is greatly increased. Occurs further.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a speech synthesizer capable of controlling a prosodic combination of adjacent words in a compound word formed by combining a plurality of words and determining an accent position.

[Means for Solving the Problems] The object of the present invention is to provide a classification means for inputting a character symbol string, dividing the word into a plurality of words, and classifying a compound word composed of the divided words, A compound word in which a plurality of words having a specific morpheme are continuously formed in the compound word is input, and the prosodic combination of adjacent words is controlled by the information of each word to control the prosody of the compound word. Determining means for determining an accent position; generating means connected to the classifying means and the determining means for generating parameters for speech synthesis based on the divided words; and connected to the generating means. An output unit that outputs a synthesized speech waveform corresponding to the generated parameter, wherein information of a word having a specific morpheme is semantic information and grammatical information, and the meaning of the adjacent word is The voice synthesizing apparatus is characterized in that the determining means determines that there is no accent connection when the taste information and the grammatical information match, and that the determining means determines that there is accent connection when the taste information and the grammatical information do not match.

[Operation] The classification means divides the input character / symbol string into a plurality of words, and converts a compound word composed of the divided words into a compound word in which a plurality of words having a specific morpheme are consecutively formed. When the determining means inputs a compound word in which a plurality of words having the specific morpheme are formed consecutively and the semantic information and grammatical information of each word match, accent determination is performed. It is determined that there is no word, and when there is a mismatch, it is determined that there is an accent connection, thereby controlling the prosodic connection of adjacent words to determine the accent position of the compound word, and connecting to the classification means and the determination means. The generated generating means generates a parameter for speech synthesis based on each divided word of the compound word output from the classifying means or the determining means, and the output means connected to the generating means generates the voice synthesizing parameter. parameter Output a synthesized speech waveform corresponding to the data.

[Embodiment] An embodiment of the speech synthesizer according to the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of the speech synthesizer of this embodiment.

In FIG. 1, a sentence mixed with Japanese / Kanji characters as a character / symbol string is input to a character / symbol string input unit 10 and sent to a morphological analysis unit 11 as a classification means.

Next, the morphological analysis unit 11 refers to the dictionary 12, and uses the longest match method or the phrase number minimum method of selecting words so that the number of phrases in the text is minimized. Divide into words.

Then, in the word string as a compound word composed of the divided words, a portion where a noun as a specific morpheme continues is sent to a compound word accent determination unit 13 as determination means.

Further, the compound word accent determination unit 13 refers to the word accent combination table 14 to determine whether or not the accesses of the adjacent words are connected or not based on the grammatical information and the semantic information as the information of the words constituting the compound word. When determining and combining, word accents are combined using the accent combination rule. When they are not combined, each word is given an independent accent.

Next, the synthetic speech parameter generation unit 15 as a generation means.
Then, from the accent of each word, the setting of the phrase accent and the setting of the phrase and the pause of the exhalation paragraph when the attached word etc. are chained are performed, and the duration time for the synthesizing unit corresponding to the phrase, Generation of frequency pattern and power pattern parameters is performed.

In addition, since the compound word accent determination unit 13 has already processed the part where the noun is linked, the synthetic speech parameter generation unit 15 does not set the accent of the part where the noun is linked.

Then, the speech synthesis unit 16 as an output unit generates an actual synthesized speech waveform based on the time series of the speech synthesis parameters, and the synthesized speech output unit 17 outputs the synthesized speech generated by the speech synthesis unit 16. Output a waveform.

Next, the compound word accent giving process performed by the compound word accent determination unit 13 will be described in detail with reference to FIGS.

First, a character symbol string is input from the character symbol string input unit 10 and is divided into words constituting the character symbol string by the morphological analysis unit 11, and only the substrings in which the nouns are continuous in the word string are determined by compound word accent determination. Input to the part 13.

Each of the words that make up the compound word has the information searched in the dictionary 12, and the grammatical information
The word accent combination table 14 is used to determine whether or not the accents are combined using the part-of-speech information and the semantic information.

The above-mentioned word accent combination table 14 indicates, when nouns are consecutive, information on whether the accent of the preceding word and the accent of the following word are combined to form one accent phrase. In order to perform a search, a category of a word located first and a category of a word located later are classified and arranged for each item, and a combination thereof is tabulated. The word category mentioned above is a combination of the word meaning category and grammatical information.

Next, the compound word accent determination unit 13 uses the semantic category number and grammatical information of the word located first and the word located later as search items, and
With reference to 14, it is determined whether accents are combined.

As a result, when the accents are combined, the accent combining process is performed according to the conventional rule.

Hereinafter, the above operation will be described in the case of the compound word "organization".

This compound word is composed of two words, “arrange” and “sort”. It should be noted that the compound word of "organization" is 2
The fact that it is composed of one word is determined by the morphological analysis unit 11 described above.

The determined compound word “arranged and organized” is converted into the semantic information (indicating that the semantic category number is “259a”) and grammatical information (here, “259a”) associated with the word “arranged” by the compound accent determiner 13. It means that it is a "sahen noun")
And the semantic information (in this case, indicating that the semantic category number is "259a") and grammatical information (indicating that it is "sa-noun") associated with the word "tidy" as search items Search Binding Table 14. In this example, when the word located first and the word located later are such a combination of the semantic category number and the grammatical information, it is determined from the word accent combination table 14 that no accent combination occurs. .

Therefore, from the compound word accent determination unit 13, the word dictionary
Accent as searched in 12, ie, "Sort" is 1
The type (accent at the beginning of the word) and “tidy” are type 0 (flat accent), and each is determined to have an independent accent.

Next, a case where the compound word is “arranged closet” will be described.

The morphological analysis unit 11 determines that this compound word is composed of two words, “arrange” and “door cabinet”.

The compound word accent determination unit 13 generates a semantic category number “259a” and grammatical information “sa-variable noun” attached to the word “arrangement”, and a semantic category number “95” attached to the word “door cabinet”.
2a ”and grammatical information“ ordinary noun ”are used as search items to search the word accent combination table 14. With this combination,
From the word accent combination table 14, it is determined that accent combination occurs. The semantic information attached to a word is the meaning of a concept in which each word comes in a common category,
As shown in FIG. 2, “arrangement” and “ordering” are categories that mean the same concept “clearing”, and “door cabinet” is another category other than “clearing”. The meaning category number is a number assigned to various categories.

Accordingly, the compound word accent determination unit 13 combines the accents of the two words according to a conventional method, and determines that the “arranged cabinet” has a type 4 (“to” accent) accent.

As described above, with respect to a compound word composed of two nouns, the case where there is a combination of accents and the case where there is no combination of accents have been described based on two examples.

In the present embodiment, the speech synthesizer has been described as determining the combination of accents between two adjacent nouns.
The speech synthesizer of the present invention can be applied to determine the accent of a compound word in which three or more nouns are chained.

[Effects of the Invention] A classification means for inputting a character string and dividing it into a plurality of words and classifying a compound word composed of the divided words, and a specific morpheme among the classified compound words Determining means for inputting a compound word in which a plurality of words are formed in succession and controlling the prosodic combination of adjacent words by the information of each word to determine the accent position of the compound word; Means for generating speech synthesis parameters based on each of the divided words connected to the means and the determination means, and a synthesized speech waveform connected to the generation means and corresponding to the generated parameters. Output means for outputting a word having a specific morpheme is semantic information and grammatical information, and a decision is made when the semantic information and grammatical information of the adjacent word match. The determining means determines that there is no accent connection, and in the case of a mismatch, the determining means determines that there is an accent connection, so that accenting of compound words can be performed accurately and uniquely. It is possible to omit the compound word accent combining process for the compound word of the noun chain, which is performed in the voice parameter generation unit.

[Brief description of the drawings]

FIG. 1 is a block diagram showing the configuration of one embodiment of a speech synthesizer according to the present invention, FIG. 2 is a detailed explanatory diagram of a compound word accent determination unit and a word accent combination table in FIG. 1, and FIG. 3 is a block diagram showing the configuration of the speech synthesizer of FIG. 10,18 …… Character / symbol string input section, 11,19 …… Morphological analysis section,
12,20 …… Dictionary, 13 …… Compound accent determination unit, 14…
... Word accent combination table, 15, 21... Synthesized speech parameter generation unit, 16, 22... Speech synthesis unit, 17, 23.

Continuation of the front page (56) References JP-A-2-93499 (JP, A) JP-A-1-20598 (JP, A) JP-A-2-201643 (JP, A) JP-A-1-321557 (JP) , A) Miyabe, Sakai et al. "PIVOT: J-E: Japanese Compound Word Analysis" Proc. Of the 37th Annual Conference of IPSJ No.3. 2 PP. 937-938 (1988) Miyazaki, Masahiro "Automatic Extraction of Compound Word Accent Phrases Using Semantic Associations between Words" Transactions of Electronics and Communications, Vol. J68-D No. 1, PP. 25-32 (1985) NHK ed. "Japanese Pronunciation Accent Dictionary Revised New Edition" Japan Broadcast Publishing Association

Claims (1)

(57) [Claims] 1. A classification means for inputting a character symbol string and dividing it into a plurality of words, and classifying a compound word composed of the divided words, and a specific morpheme among the classified compound words. Determining means for inputting a compound word in which a plurality of words are formed in succession and controlling the prosodic combination of adjacent words by the information of each word to determine the accent position of the compound word; Means for generating speech synthesis parameters based on each of the divided words connected to the means and the determination means, and a synthesized speech waveform connected to the generation means and corresponding to the generated parameters. Output means for outputting a word having a specific morpheme, the information having the specific morpheme is semantic information and grammatical information, and when the semantic information and grammatical information of the adjacent word match. Said determining means determines that there is no accent bond, said determining means when the semantic information and grammatical information of said word adjacent the discrepancies speech synthesis apparatus characterized by determining that there is an accent bind to.