JPH02234198A - Text voice synthesizing system - Google Patents

Abstract

PURPOSE:To obtain a natural utterance generated by desired reading and an accent by providing a function for dividing a word or editing a result of analysis of language. CONSTITUTION:A Japanese language sentence is inputted from a character input part 1, a character-string to be uttered is sent to a morpheme analyzing part 2, and the morpheme analyzing part 2 divides the character-string into word units, adds form information such as a part of speech, an inflective form, etc., and an accent type, a reading symbol, etc., to them, respectively and outputs them. An analysis result editing part 3 corrects these information in accordance with necessity and a desire through input/output devices 9, 10 and an input/output control part 8, and thereafter, a combination analyzing part 4 extracts a combined relation, syntax information, a degree of inter-sentence/ clause separation, etc., a meter symbol generating part 5 generates a meter symbol train, and a voice synthesizing part 6 synthesizes a voice. In such a way, a composite voice by desired reading and an accent can be obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a text-to-speech synthesis system, and more particularly, to a text-to-speech synthesis system having functions for word segmentation and editing of language analysis results. Generally speaking, text-to-speech synthesis systems perform language analysis processing to divide input strings into word units in order to generate pronunciations, accents, pauses, and intonation.
The pronunciation, accent, part of speech information, etc. of those words are extracted, and phonological and prosodic symbols are generated from this information. However, incorrect analysis often occurs in the language analysis process. Therefore, in order to give the correct pronunciation, accent, and intonation, the following methods have traditionally been used. (1) Correct utterance is performed by correcting a part of the phonological and metrical symbol string obtained as a result of language analysis and phonological and metrical symbol generation.

(2) An input sentence creation word processor dedicated to text-to-speech synthesis, which simultaneously separates words and phrases during Kana-Kanji conversion.
Determine the pronunciations and determine their accents through language analysis processing. In addition, in text-to-speech synthesis systems, as mentioned above,
In order to generate pronunciation, accent, pause, and intonation, we perform language analysis processing to divide the input string into words, extract the pronunciation, accent, part of speech information, etc. of each word, and then extract phonological and prosody information from this information. Since symbols are generated, it is impossible to register every word in the word dictionary used for this language analysis, and it is usually around several hundred thousand words, and abbreviations and proper nouns may be unregistered words. There is. Therefore, in general, unregistered word processing is performed during language analysis processing to determine the word spelling, pronunciation, accent, etc. However, this non-g:. Errors often occur during word recording processing. Also, in Japanese there are homographs (
Single i has the same notation but different reading! ? l), and it is not easy to correctly judge these and decide on a single word, and errors often occur. therefore,
A means for making the desired utterances is required.

The above-mentioned conventional technology has the following problems.

In (1), the phonological and metrical symbols are often difficult to understand for people who are not familiar with them, and it takes time to find the phonological symbol that corresponds to one word, so it is difficult to correct the phonological and metrical symbols. is not easy. In (2), when outputting audio, a dedicated word processor must be used, and other text files cannot be read as is, and phonological and prosodic symbols are still used to specify accents. Must be corrected.

The present invention has been made in order to solve the above-mentioned drawbacks, and by providing a function to edit word segmentation or language analysis results, it is possible to specify the correct or desired pronunciation or accent. The purpose of this project is to provide a text-to-speech synthesis system that can produce more natural speech.

(2) In order to achieve the above object, the present invention provides a text-to-speech synthesis system that converts character information into speech and outputs it, which includes a language analysis section, a language analysis result editing section, and a phonological/prosodic symbol generation section. , a regular speech synthesis section, and is capable of modifying, adding, and deleting language analysis results; and furthermore, the language analysis result editing section is capable of editing information such as word breaks, word pronunciations, accents, parts of speech, etc. Furthermore, the linguistic analysis result editing section can edit prosodic information based on the results of the linguistic analysis, This feature is characterized in that if a isomorphism is detected, the corresponding isomorphism can be selected. The present invention will be explained below based on examples. FIG. 1 is a block diagram for explaining one embodiment of the text-to-speech synthesis system according to the present invention. In the figure, 1 is a character input section, 2 is a morphological analysis section, 3 is an analysis result editing section, and 4 is a related part. 5 is a prosodic symbol generation section, 6 is a speech synthesis section, 7 is a receiving analysis section;
is a word dictionary, 8 is a display input control unit, 9 is a display,
10 is a keyboard, and 11 is a speaker. A Japanese sentence is input from the character input section 1, character conversion is performed if necessary, and the character string to be uttered is sent to the morphological analysis section 2. The morphological analysis unit 2 divides the character string into words and outputs each word with morphological information such as part of speech and conjugation, accent type, reading symbol, etc. added. The analysis result editing unit 3 modifies this information as necessary and desired via the input/output device and the input/output control unit, and then the dependency analysis unit 4 changes the dependency relationship, syntactic information, and the degree of separation between clauses. Extract etc. From the various information obtained in this way, the prosodic symbol generation unit 5 calculates the reading, accent,
A prosodic symbol string including pauses, intonation, etc. is generated, and the speech synthesizer 6 synthesizes speech accordingly. The character input section 1, the morphological analysis section 2, the dependency analysis section 4, the prosodic symbol generation section 5, and the speech synthesis section 6 can be realized by conventional techniques. When editing the analysis results, it is sufficient to be able to display and modify at least one extracted piece of utterance-related information. For example, the results are displayed on the display 9 as shown in Figure 2, and corrections can be made by moving the cursor (square dotted line) to the location. This can be achieved with an editor that allows you to move and rewrite the text (shown in ). When displaying such information, pronunciations can be displayed in hiragana, katakana, phonetic symbols, phonetic symbols, phonetic numbers, etc., and parts of speech, accents, etc. can be displayed as codes and classification names based on the classification in the system. Furthermore, if a homomorphic word is detected, multiple candidates can be stored and selected in the analysis result editing department, making it possible to give the desired pronunciation and accent. Furthermore, if the word division position is incorrect and it is corrected, suitable candidates are displayed from multiple analysis results stored in advance, or the subsequent character strings are reanalyzed and a new one is displayed. By doing so, the correction work can be reduced. Furthermore, the analysis result editing section may be placed next to the dependency analysis shown in FIG. 1, so that the results of the dependency analysis can also be edited. By doing so, the pose position, pose length,
You can also modify the exhalation paragraph and voice to suit your needs. Furthermore, by outputting the analysis results and editing results as a file, and reading the file to start generating prosodic symbols and synthesizing speech, it is possible to save the desired vocal form and always obtain the desired synthesized speech. Can be done. As is clear from the above explanation, according to the present invention, even if the language analysis result of the text-to-speech synthesis system is incorrect or unregistered, it can be easily corrected, and as a result, the desired pronunciation and accent can be obtained. You can obtain synthesized speech using

[Brief explanation of the drawing]

FIG. 1 is a block diagram for explaining an embodiment of a text-to-speech synthesis system according to the present invention, and FIG. 2 is a diagram showing a display. 1...Character input section, 2...
- Morphological analysis unit, 3... Analysis result editing unit, 4... Dependency analysis unit, 5... Prosodic symbol generation unit, 6... Speech synthesis unit, 7... Word dictionary, 8...・Display input control section,
9...Display, 10...Keyboard, 11.
... Subika. l Figure 1

Claims (1)

[Claims] 1. A text-to-speech synthesis system that converts text information into speech and outputs the same, comprising a language analysis section, a language analysis result editing section, a phonological/prosodic symbol generation section, and a regular speech synthesis section. A text-to-speech synthesis system characterized in that it is capable of modifying, adding, and deleting language analysis results. 2. The text-to-speech synthesis system according to claim 1, wherein the language analysis result editing unit is capable of editing word divisions, word pronunciations, accents, parts of speech, and the like. 3. The text-to-speech synthesis system according to claim 1 or 2, wherein the language analysis result editing section is capable of editing prosodic information. 4. The language analysis result editing unit according to claim 1, 2 or 3, is capable of selecting a homograph when the existence of a homograph is detected as a result of the language analysis. Text-to-speech synthesis system.