JPH0562356B2 - - Google Patents

Description

[Detailed description of the invention]

[Technical Field of the Invention] The present invention relates to a speech synthesis method for converting character strings into speech. [Technical background of the invention and its problems] In a speech synthesis device that converts sentences into speech,
It is necessary to analyze character strings to generate phonological and prosodic information. Of the two, information regarding the accent of words has a large weight in prosodic information. Testing the accent of a word has different levels of difficulty depending on whether the word is a simple word or a compound word, and it goes without saying that compound words are troublesome. Conventionally, regarding the test of accent of compound words,
It has been common practice to regularly determine the accent type of a compound word based on the words that make up the compound word, that is, the accent type of the constituent words (which is unique to each constituent word). However, some compound words cannot be tested for accent using the regular method described above, and in the case of such compound words,
Erroneous accent information was generated, and as a result, the intelligibility of the synthesized speech decreased, and the naturalness was also impaired. This was a problem with conventional methods. [Object of the Invention] The object of the present invention is to use a speech rule synthesis method for converting character strings into speech to test the accent of compound words with high accuracy, thereby creating synthesized speech with higher intelligibility and naturalness than ever before. The objective is to provide a method for generating . [Summary of the Invention] The present invention divides a Japanese sentence containing kanji and kana into phrases, compares the character strings of the phrases with a word dictionary, and removes the matched character strings from the phrase string. If the first character in the column is a kanji, it is processed as a compound word, whereas if it is not a kanji, it is processed as a simple word. In each process, phonological information and prosody information are generated, and phonological parameters are calculated based on this information. and prosodic parameters are created, and both parameters are sent to the synthesizer to obtain synthesized speech. However, in the compound word processing of the above two processes, if the word can be matched as a compound word, the latter constituent words of the compound word are post-connected. By reading out accent information from a word dictionary and testing the accent of compound words using that information, highly accurate testing becomes possible.This allows the correct accent to be given to compound words in synthesized speech, and improves understanding of synthesized speech. It enhances the quality and naturalness. [Effects of the Invention] According to the present invention, it is possible to test accents with higher accuracy than before for compound parts that are often used in Japanese sentences, and as a result, the naturalness and intelligibility of synthesized speech of compound words can be improved. . In particular, compound words are often used as keywords in sentences, and in such cases, the present invention has great effects. [Embodiments of the Invention] One embodiment of the present invention is shown in FIG. A character string containing kanji and kana in a Japanese sentence is divided into phrases by a phrase segmentation circuit 1 according to a change in character type (for example, from kana to kanji). The word matching circuit 2 matches the phrase character string with the word dictionary 3, and removes the words that could be matched from the phrase. When the head of the remaining character string obtained as a result is a kanji character, the composite collation circuit 4 is driven. On the other hand, if the first character is not a kanji character, the phonological information and accent type of the word that could be matched are read from the word dictionary 3, and based on that, the phonological sequence circuit 6 and the prosodic parameter generating circuit 9
to drive. The compound word matching circuit 4 matches the above-mentioned remaining character strings with the word dictionary 3, and if matching is successful, reads the phonological information of the word from the word dictionary 3, drives the phonological sequence testing circuit 6 based on it, and also Post-connection accent information is also read out from the dictionary 3 at the same time, and the compound word accent test circuit 5 is driven based on it. The compound word accent test circuit 5 tests the accent of a compound word using the subsequent accent information of the rear constituent words of the compound word. For example, in the case of a compound word [urban construction], when the front constituent word [city] is matched by the word matching circuit 2 and the latter constituent word [construction] is matched by the compound matching circuit 4, the latter constituent word [construction] is matched by the compound word matching circuit 4. When extracting the post-connection accent information, [1]
It is. This value [1] indicates that the accent nucleus of the compound word is the first mora of the rear constituent word, so the accent nucleus of the compound word [toshikensetsu] is the first mora of the rear constituent word [ke] It can be verified that In other words, the accent of the compound word is [toshikensetsu], and it can be verified that it is a so-called type 3, with the accent core in the third mora. In the same way, for the compound word ``struggle'', the post-conjunctive accent information of the last component word ``struggle'' is obtained.

[0], and this number

[0] indicates that the compound word as a whole has a flat type with no accent core in any mora, so it can be verified that the accent of this compound word is [accent-] and that it is O type. In the case of conventional accent testing methods, the accent of compound words is regularly tested based on the number of moras of the post-component words and the unique accent type when the constituent words are used as simple words. Since it does not have accent information, the accent in the above example [Struggling] is determined to be [Accent To], that is, type 5.
Therefore, the erroneous prosody information is sent to the prosody parameter generation circuit. As described above, in this method, it is possible to perform a correct test even on a compound word for which conventional accent tests are incorrect, that is, it is possible to test the accent of a compound word with high accuracy. Next, the phoneme sequence testing circuit 6 tests the presence or absence of nasalization and devoicing of the phoneme based on the phoneme information of the word, generates a phoneme sequence, and the phoneme parameter generation circuit 7 uses the phoneme file 8 to Create a sequence of phonological parameters based on the above phonological sequence,
This is sent to the speech synthesizer 10. Furthermore, the prosodic parameter generation circuit 9 generates a sequence of prosodic parameters based on the prosodic information of the simple words and compound words, and sends them to the speech synthesizer 10. The speech synthesizer 10 is driven by the above phonetic parameter string and prosodic parameter string, and outputs synthesized speech. By the above method, it is possible to obtain synthesized speech with high naturalness and intelligibility for Japanese sentences containing compound words. The present invention is not limited to generating synthesized speech of kanji-kana mixed sentences, but also generates synthesized speech of kana character strings input in units of clauses using Kana Key, by adding some modifications. , can be applied.

[Brief explanation of the drawing]

FIG. 1 is a schematic configuration diagram showing an embodiment of the present invention;
FIG. 2 is a diagram showing the format of the word dictionary. 1...Bunsetsu cutting circuit, 2...Word matching circuit, 3
...Word dictionary storage device, 4...Compound word matching circuit,
5...Compound word accent testing circuit, 6...Phonological sequence testing circuit, 7...Phonological parameter generation circuit, 8
. . . speech segment file storage device, 9 . . . prosodic parameter generation circuit, 10 . . . speech synthesizer.

Claims (1)

[Claims] 1. A bunsetsu cutting circuit that separates character strings containing kanji and kana in Japanese sentences into bunsetsu units, a word matching circuit that matches the strings of clauses with a word dictionary and extracts phonological and prosodic information, and a word matching circuit that extracts phonological and prosodic information by comparing the strings of clauses with a word dictionary. When the first character string after removing the character string from the bunsetsu is a kanji, the compound word matching circuit checks the remaining string against the word dictionary again, and the accent of the string that was matched as a compound word is a compound word accent testing circuit that tests using post-conjunctive accent information extracted from the word dictionary during the compound word matching; and a phonology that tests a phonological sequence based on the matching results of the word matching circuit or the compound word matching circuit. a sequence testing circuit, a phonological parameter generation circuit that converts the phonological sequence into a phonological parameter, a prosodic parameter generation circuit that generates a prosodic parameter based on the test result of the word matching circuit or compound word accent testing circuit, and the phonological parameter. and a speech synthesizer that generates synthesis parameters based on prosodic parameters.