KR970003093B1 - Synthesis unit drawing-up method for high quality korean text to speech transformation - Google Patents

Abstract

Context dependent unit editing method for the text-to-speech converter about Korean, in order to increase the intelligibility and the naturalness of composite speech. The said method comprising the steps of: selecting the syllable type and the syllable head of the selected context dependent unit according to the analysis of rhythm type and the restriction of phoneme connection and phoneme type about Korean; deciding the type of detail using part according to the appearance environment of syllable, syllable numbers, rhythm, and intonation type; and displaying the phoneme row in the voice wave of the context dependent unit to enable to adjust the pitch length of phoneme or syllable.

Description

How to Write Synthetic Unit (CDU) for High Quality Korean Sentence Conversion

1 is an overall flowchart of a CDU creation method,

2 is a flow chart of the synthesis unit selection,

3 is a detailed usage decision flow chart,

4 is a detailed use segment and notation flowchart.

The present invention relates to a method of preparing a synthesis unit (CDU) for Korean sentence-to-speech conversion.

An object of the present invention is to provide a method for constructing a new synthesis unit that can greatly improve the understanding and naturalness of a synthesized speech in an unlimited text-to-speech converter.

In order to achieve the above object, the present invention provides a first step of selecting a syllable type and syllable item of a synthesized unit selected by Korean phonological structure type, phoneme connection constraint, and rhythm type analysis, and the appearance of syllables in each synthesized unit. The second step of determining the shape of the detailed use part determined by the environment, the number of syllables, the length and shortness, and the intonation type, and the detailed use part in the speech waveform of the synthesized unit, but the same form is the syllable after the connection of the detailed use unit of the same symbol and the synthesized unit. Or a third step of marking a phoneme sequence to enable length pitch adjustment of a phoneme unit.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

In order to create a synthesis unit, a device for analyzing and displaying the waveform, spectrograph, pitch, and energy of the voice is required by attaching the voice A / D and D / A converters from the main processor with the CPU and the auxiliary memory.

FIG. 1 is a flow chart of a method for preparing a synthesized unit for high quality Korean sentence speech conversion. The process comprises 1) selection of syllables for use 2) determination of detailed use units 3) segmentation and marking of use units in synthesized unit speech waveforms. In the first syllable selection, the syllable and rhyme environment of Korean are analyzed and the synthesis unit necessary for synthesis is selected. In the second syllable decision, the detailed usage part is decided according to the environment within the syllable. In the third process, the segmentation and the notation of the use unit for the synthesized unit speech waveform form the synthesis units (CDU: Context Dependent Units) according to the environment by using the segmentation and the notation of the detail use unit on the actual speech waveform of the used syllable.

2 is a flowchart for selecting syllables to be used in Korean sentence / voice conversion. In phonological structure analysis, Korean phonetic dictionary (about 60,000 words) is used, and the phonological structure of Korean is composed of (consonants) (half vowels) vowels (consonants) in the case of one syllable, and in the case of the second syllable ( Consonants) (half vowels) Vowels (consonants) (consonants) (half vowels) Here, () denotes an optional condition. Consonant groups between vowels and vowels dominate the consonants of the next consonant. Three syllables or more can be composed of a combination of one-syllable structure and two-syllable structure.

Rhythmic form analysis was performed using a Korean rhyme database containing rhythmic forms of sentence reading. The accent form of the accent was divided into rising type, flat type and descending type. The rising type and the falling type change the characteristics of the vocal cord vibration corresponding to the sound source, and the characteristic of the form changes.

Therefore, it is difficult to implement a rising voice and a falling voice by adjusting only a pitch in a sound having a normal height. Existing synthetic units do not consider this rhyme environment. As a complementary method, the synthesizing units of rising type and falling type should be included.

The following is a syllable type selected by analyzing the phonological structure of Korean, the constraints of phoneme connection, and the form of rhyme.

Vowels (1-8): 8

Half vowel + vowel (9-20): 12

Consonants + Vowels (21-164): 144

Consonant + half vowel + vowel (165-326): 162

Vowel + Consonant (327-466): 140

Vowel + Consonant (467-530): 64

Vowels + Half Bars + Vowels (531-546): 16

Vowels + Consonants + Vowels (547-698): 152

Vowel (E) + Half Bar + Collection (699-710): 12

Vowel (+) + Consonant + Collect (711-843): 133

Vowels (+) + Consonants + Half Bars + Vaults (844-1053): 210

Vowel + nasal / ㅁ / + consonant (ㄸ) + vowel (ah) (1054-1073): 20

Vowel (N) + Rain / ㅁ / + Composite Consonant + Collect (1074-1081): 8

Vowel + navel / b / + consonant (ㄸ) + vowel (ah) (1082-1101): 20

Vowel (N) + Non / B / + Composite Consonant + Collect (1102-1109): 8

Vowel + navel / o / + consonant (ㄸ) + vowel (ah) (1110-1129): 20

Vowel (N) + Rain / ㅇ / + Combined Consonant + Collect (1130-1137): 8

Vowel + Yumin / D / + Consonant (ㄸ) + Vowel (A) 1138-1157): 20

Vowel (Y) + Yoon / ㄹ / + Consonant (ㄸ) + vowel (1158-1164): 7

Vowels + Yoon / ㄹ / + Yoon / ㄹ / + vowels (e) (1165-1184): 20

Vowel (E) + Yoon / ㄹ / + Yoon / ㄹ / + Collection (1185-1204): 20

Rising accent [collection] (1205-1212): 8

Rising accent [half vowel + vowel] (1213-1224): 12

Rising accent [vowel + meteor consonant] (1225-1256): 32

Descent accent [collection] (1257-1264): 8

Descent accent [half vowel + vowel] (1265-1276): 12

Descent accent [vowel + meteor consonant] (1277-1308: 32

Pause (1309): 1

1309

3 is a process of determining the appearance environment of syllables, the number of syllables, the length and length of the syllable, and the type of intonation in each synthesis unit. The shape of the detail use part is determined by the presence or absence of leading syllables and trailing syllables, apical vowels, and intonation. The detailed use part is divided into long sound single type, single sound single type, long sound introduction type, connection type, ascending termination type, flat termination type and descending termination type.

Existing synthetic units do not consider the appearance of syllables, the number of syllables, the length and length of the syllable, and the intonation type.

4 shows a process of marking a detailed use part by referring to a spectrogram, pitch, energy, etc. on the speech waveform of the synthesized unit. According to the type of unit used for each synthesis unit, each part that is the connection point in speech synthesis is described in detail considering the connection of formant, pitch, energy, etc. Duplicate symbols are not used in the unit of synthesis, and the same form is designated by the same symbol, and the connecting parts are designed to be connected by the same symbol. In addition, the phoneme string is marked so that the length pitch of syllables or phoneme units can be adjusted after connection of the detailed unit of the synthesis unit.

Accordingly, the present invention, which is performed and operated by the above-described processing procedure, enables the development of a high quality Korean unlimited sentence / voice conversion apparatus with significantly improved comprehension and naturalness of synthesized speech, and media conversion and speech for the disabled in information and communication. It can be utilized in a reader, a voice calibrator, etc., thereby enabling high quality media conversion in information and communication, and facilitating social adaptation of the disabled.

Claims (1)

A first step of selecting syllable types and syllable items of the synthesized unit selected by the analysis of the Korean phoneme structure form, the constraints of the phoneme connection, and the form of rhyme; A second step of determining the shape of the detailed use part determined by the appearance environment of the syllables, the number of syllables, the length and length, and the intonation type in each synthesis unit; And a third step of displaying a phoneme sequence so that a detailed use part is indicated on the speech waveform of the synthesized unit, but the same form is used to control the length pitch of the syllable or phoneme unit after connection of the same symbol, the detailed use unit of the synthesized unit. How to write a synthesis unit (CDU) for high-quality Korean sentence-to-speech, characterized in that for.