JPH01321557A - Text voice synthesizing device - Google Patents

Abstract

PURPOSE:To obtain the correct reading even in the case an identified unregistered word is equal to a Japanese word by using a KANJI (Chinese characters) dictionary where the information on the single KANJI are registered and selecting the reading out of the KANJI dictionary in the case a detected unregistered word includes KANJI. CONSTITUTION:The inputted Japanese sentences undergo the necessary analyses for the morpheme, the syntax structure, the meaning, etc., through a Japanese word processing part 1, and is extracted the required bit of information for preparing the rhythm. When a word unregistered in a dictionary 4 is identified, the proper reading is given to the word for each KANJI. For this purpose, a KANJI dictionary 5 containing the reading is prepared and plural types of reading are selected based on a fixed rule. The selection of the KANJI reading is decided based on the description of the detected unregistered word, the character description of the preceding and next words, the morpheme information, the meaning and the origin of KANJI of the detected unregistered word, the syntax structure information, the context information, etc. Thus the proper reading can be given to the unregistered KANJI character strings.

Description

[Detailed Description of the Invention] The present invention relates to a text-to-speech synthesis device, more specifically, a process for assigning an appropriate pronunciation to a kanji character string of an unregistered word in an unregistered word process of a text-to-speech synthesis device. Regarding.

When performing sentence analysis of stolen Japanese text, the sentence is generally first divided into words, which are searched by matching with words registered in a dictionary in advance. However, it is impossible to register all the words that appear in a given text in a dictionary, and there are always unregistered words, which require identification processing.

Previously, the reading was selected based on judgments such as if the unregistered word string had at least two consecutive kanji characters in the onyomi, or if an isolated kanji and the next character were hiragana, it would be the kunyomi. However, with this method, when the identified unregistered word is a Japanese word, it is not possible to give the correct reading.

Purpose The present invention was made in view of the above-mentioned circumstances.
In particular, it converts any Japanese text into audio.

When a Japanese text-to-speech synthesizer that reads aloud detects a word that is not registered in the dictionary during word segmentation related to Japanese text analysis (morphological analysis, syntactic analysis, semantic analysis), the appropriate pronunciation for the detected word is detected. This was done for the purpose of giving.

Configuration In order to achieve the above object, the present invention is a text-to-speech synthesizer that converts any Japanese text into speech, and which converts an input Japanese text into speech by matching the input Japanese text with words in a dictionary created in advance. In the process of dividing into words, it has an unregistered word detection unit that detects unregistered word character strings that did not exist in the dictionary, and a kanji dictionary that registers information about single kanji, and processes the detected unregistered words. When a kanji is included, a kanji reading is selectively assigned from a kanji dictionary by a kanji reading assignment process. Hereinafter, the present invention will be explained based on examples.

FIG. 1 is a block diagram for explaining one embodiment of the present invention, in which 1 is a Japanese language processing unit, 2 is a prosody generation unit, 3 is a speech synthesis unit, 4 is a word dictionary, and 5 is a kanji character A dictionary, 6 is a kanji reading assignment unit, 7 is a dictionary search unit, 8 is an unregistered word processing unit, 9 is a morphological analysis unit, 10 is a syntactic analysis unit, 11 is a semantic analysis unit, and the present invention is capable of reading any Japanese language. In a text-to-speech synthesis device that converts sentences into speech, when a word that is not registered in the dictionary is identified, a kanji dictionary in which the readings are registered is used to give the word an appropriate reading for the kanji in the string. is prepared, and the user selects from among multiple readings according to certain rules.

In FIG. 1, an input Japanese sentence is subjected to necessary analysis such as morphological, syntactic, and semantic analysis in a Japanese processing unit 1 to extract information necessary for prosody generation. Reading, accent, morphological information, syntactic information.

Semantic information and the like are input to the prosody generation unit 2, and this information is used to generate accents at the sentence level.

Prosodic information such as stress, pause position, pause length, intonation, etc. is generated and output to the speech synthesis unit 3, which then processes to generate a speech signal. Kanji reading assignment is performed on detected and identified unregistered In the embodiment shown in Fig. 1, this process is performed every time an unregistered word is detected, but after the detection process is completed, the process shifts to prosody generation. It may be added at any stage up to that point.

FIG. 2 is a diagram for explaining an example of kanji reading assignment processing, and in the figure, 21 is a character notation form extraction unit.

22 is a kanji dictionary, 23 is a reading search section, 24 is a judgment rule section, 25 is a reading selection judgment section, 26 is a Kun-yomi representative reading selection section, 27 is a Kun-yomi selection section, 28 is an On-yomi selection section, and 29 is a representative reading selection section. Section 30 is an extracted pronunciation assignment section, into which character strings identified as unregistered words, notation of preceding and following words, morpheme information, syntax, context information, etc. are input. First, the format of the notation of the unregistered word string is examined, and the number of characters, the number of consecutive kanji characters, the presence of hiragana and its position, etc. are extracted.First, the kanji dictionary is searched for each kanji included in the string. Then, the reading, meaning of kanji, character origin, etc. are extracted. This information is given to the pronunciation selection determination section, and using rules, the pronunciation is instructed to select a pronunciation, the process proceeds to each selection process, and the extracted pronunciation is assigned. As shown in Figure 2, the most frequent reading of the short kanji is taken as the representative reading, and the kun reading is used as the representative reading.

One possibility is to register the on-yomi reading. In addition, types of on-yomi such as idiomatic sounds, Chinese sounds, Wu sounds, Tang sounds, Song sounds, etc. may be registered as the type.As for the rules used for reading selection judgment, for example, if the unregistered word detected character string is a kanji When there is one character and the next character is a hiragana character, it instructs to select the representative reading of kun-yomi, when the detected character string is two kanji characters and the following character is a hiragana character, it instructs to select kun-yomi, and in other cases, it selects on-yomi. It may be possible to give instructions. These judgments may be made based on rules based on other input information. For example, if the morpheme information gives priority to the selection instruction described above and the next word after an unregistered word is a Japanese word.

Instruct students to select Kun-yomi. Syntax 1 Context information, meaning of kanji,
From the glyph source, it is also possible to use word co-occurrence based on dependencies and meanings.

As is clear from the above description, the present invention provides a text-to-speech synthesis device that converts arbitrary Japanese sentences into speech, which performs word-by-word processing by matching input Japanese sentences with words in a dictionary created in advance. In the process of dividing the unregistered word into two, it has an unregistered word detection unit that detects unregistered word character strings that did not exist in the dictionary, and a kanji dictionary that registers information about single kanji, so that the detected unregistered word has a kanji. When a kanji is included, a reading is selectively assigned from a kanji dictionary through a kanji reading assignment process. Character notation, morphological information, or the meaning of a detected unregistered word.

Determined by at least one information such as syntax information 1 context information, or.

If the unregistered word detected character string is one Kanji character and the next character is Hiragana, or if the detected character string is two Kanji characters followed by at least one Hiragana character, Kun-yomi is selected, otherwise On-yomi is selected. or when the word following the unregistered word is a Japanese noun, giving priority to the selection of pronunciation based on judgment from the character notation.

Kun-yomi is added to the unregistered word detected character string, and the most frequent representative reading, on-yomi, and kun-yomi are registered in the kanji dictionary, and when the unregistered character string is a kanji-character and the next character is a hiragana. If the representative reading is a kun-yomi, select that reading, otherwise select the kun-yomi; At times, one should choose on-yomi.

Effects As is clear from the above description, according to the present invention, an appropriate pronunciation can be assigned to an unregistered Kanji character string extracted and identified by the unregistered word processing of text-to-speech synthesis.

[Brief explanation of the drawing]

FIG. 1 is a block diagram for explaining an embodiment of a text-to-speech synthesis device according to the present invention, and FIG. 2 is a diagram for explaining an example of a process for adding pronunciations of kanji. 1... Japanese language processing unit, 2... Prosody generation unit, 3...
Speech synthesis section, 4... Word dictionary, 5... Kanji dictionary, 6
...Kanji reading assignment section, 7...Dictionary search section, 8...
Unregistered word processing unit, 9...Morphological analysis unit, 1o...Syntax analysis unit, 11...Semantic analysis unit, 21...Character notation form extraction unit, 22...Kanji dictionary, 23... - Reading search unit, 24... Judgment rule unit, 25... Reading selection determining unit, 26... Kun reading representative reading selection unit, 27... Kun reading selection unit, 28... On reading selection unit, 29 ... Representative reading selection part, 3o... Extracted reading assignment part.

Claims (1)

[Claims] 1. A text-to-speech synthesis device that converts any Japanese text into speech, in which the input Japanese text is divided into word units by matching with words in a dictionary created in advance. It has an unregistered word detection unit that detects unregistered word character strings that did not exist in A text-to-speech synthesis device characterized by selectively assigning readings from a kanji dictionary.