KR0175247B1 - Text conversion method for speech synthesis - Google Patents

Abstract

In the text conversion method for speech synthesis, text is input and classified according to KS2 regulations for numbers, English, Chinese characters, symbols, Korean, and Japanese, and converted into codes for the classified texts, and the coded text processed above Phonically process

Description

Text conversion method for speech synthesis

1 is a system diagram according to the present invention.

2 is a flow chart in accordance with the present invention.

3 is a flowchart of a numerical processing concrete of FIG.

4 is a flowchart of an English language processing concrete of FIG.

5 is a flowchart illustrating a Chinese character processing of FIG.

6 is a symbol processing specific flowchart of FIG.

7 is a flowchart of a Hangul process of FIG.

8 is a flowchart illustrating the Japanese language processing of FIG.

9 is a phonological processing specific flow chart of FIG.

The present invention relates to a text conversion method for speech synthesis that can convert symbols and numbers or alphabets and Chinese characters into codes and code them into codes as they are sounded based on phonological rules.

As the use of computers becomes more common and universal, the importance of human and computer communication is increasing. Until now, it has been mainly dependent on visual devices such as monitors, printers, and plotters, but thanks to the development of microprocessors, man-machine communication between computers and humans has recently been developed by audible methods through speech synthesis. It is applied in all of you.

In particular, as the information industry such as office automation, factory automation, and home automation develops, the demand for speech synthesis increases, and research is being actively conducted to synthesize the languages of the world. In the case of English, Vortax's Telesensory Prose2000 system with 95% accuracy has been developed. In the case of Japanese, a synthesis system was developed based on the advantages of simple language system and low number of syllables.

In the case of Korean, voice signal processing research has been actively conducted in recent years. However, as for speech synthesis, no research on a system composed entirely of hardware and software has been published. The study on Korean speech synthesis system is based on linguistic research on quantitative phonetics of Korean language. However, the research is weak and it is difficult to systematically study Korean speech synthesis. Meanwhile, a limited function voice synthesis IC has been developed in Korea.

Speech synthesis is divided into restricted word synthesis and unlimited word synthesis according to the range of vocabulary that can be synthesized.

Restriction word synthesis is a method of storing only a limited vocabulary in a database and outputting synthesized sounds, and unlimited word synthesis is a method of storing a basic unit in a database and outputting a random synthesis sound by combining basic units.

Restricted word synthesis has a smaller amount of database and relatively simple synthesis method than unlimited word synthesis. Therefore, it is applied to simple sentence-to-speech system such as time signal guidance and checking bank balance by phone. However, for unlimited word synthesis, there have been no studies showing satisfactory results with both Korean and Korean database quality, and the research methods that have been attempted were simulations, and no implementation of real-time Korean speech synthesis through hardware has been published.

On the other hand, the waveform encoding method and the parameter encoding method are divided according to a method of obtaining data constituting a database of speech synthesis. The waveform encoding method encodes and stores the speech waveform by PCM (pulse code modulation), ADPCM (adaptive PCM), and then generates a synthesized waveform.The parameter encoding method extracts and stores specific parameters from the speech waveform and then synthesizes the waveform. Create In the former case, the sound quality of the synthesized sound is clean, but one basic unit database file is so large that it is suitable for speech synthesis of limited words, but not for unlimited word speech synthesis.

Since the parameter coding method has a slightly lower sound quality but a small amount of database, various attempts have recently been made to suit unlimited word speech synthesis. Speech synthesis using parameter coding includes speech synthesis using formants, speech synthesis using linear predictive coding (LPC) coefficients, speech synthesis using multi-pulse LPC (MPLPC), and the like.

Accordingly, an object of the present invention is to provide a method for converting text composed of arbitrary characters into text composed of Korean characters and converting the phonetic sounds as human reading.

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

1 is a system diagram according to the present invention, a text input processing unit 11 for input processing text data, a character recognition device 13 for recognizing characters in an output value of the text input processing unit 11, and the text input. A display unit 12 for displaying the processing result of the processing unit 11 and the character recognition device 13, a first dictionary 14 having Korean, Chinese, English, and Japanese coding data, an assistant mark, a symbol table, and an English letter The second dictionary unit 17 having the notation data and the recognition coded data are received through the character recognition device 13 and the syntax is provided according to the dictionary data provided through the first and second dictionary units 14 and 17. And a code storage unit 16 for storing the coded data processed through the control unit 15.

FIG. 2 is a flow chart according to the present invention, which comprises a text inputting a first step of classifying numbers, English, Chinese characters, symbols, Korean, and Japanese according to the KS2 byte code regulation, and a code for each text classified in the first step. And a third process of phonologically processing the code text processed in the second process.

FIG. 3 is a detailed flowchart of the numerical process 2c of FIG. 2, which includes an assistant searching process for checking whether an assistant is present and reading the input numerical value by the first rule when there is no assistant in the assistant searching process. In the first reading process and the assistant search in the assistant search process, search for the three-digit number when there is an assistant, and read the three-digit number or less by the assistant table, and read the second or above three-digit assistant by the first rule. And a conversion process of converting the values read in the first and second reading processes into Korean codes and writing them to a table.

FIG. 4 is a flowchart illustrating the English process of FIG. 2, which is a process of searching an English dictionary to determine the existence of a corresponding code, accessing the corresponding code when present in the process, and converting the code into Korean code, and not present in the process. When not in the process of converting the alphabet into Hangul code.

FIG. 5 is a detailed flowchart of the Chinese character processing of FIG. 2, which includes an indexing process of indexing a Chinese character conversion table, a first writing process of converting a Korean code into a table when the method of reading Chinese characters in the indexing process is less than two; When there are two or more methods of reading Chinese characters, a second process of searching for a dictionary of Chinese characters, searching for their existence, and writing them to a table is performed.

FIG. 6 is a detailed flowchart of symbol processing of FIG. 2, which includes a mode selection process for searching for abbreviations, punctuation marks, slashes, equations, and unit existing processing modes, and a Hangul code for converting a mode selected value to a Hangul code during the mode selection process. It is a conversion process.

FIG. 7 is a detailed flowchart of the Hangul process of FIG. 2, which separates the stems from the stems and endings according to the notation of the Korean word (survey, ending, suffix), and records the stems and endings separated in the splitting process. The recording process takes place.

FIG. 8 is a detailed flowchart of the Japanese processing procedure of FIG. 2, which includes indexing a Japanese table, converting a Japanese table into Korean code according to the indexed table, and storing the converted code in a table.

9 is a detailed flowchart of the phonological process of FIG.

Therefore, when a specific embodiment of the present invention is described in detail with reference to FIGS. 1 to 9, the text input through the text input processing unit 11 is recognized by the character recognition device 13 and the ( It is input in step 2a). The text input in the step (2a) is referred to the KS2 byte code, and the classification is performed. If it is a number (2c) process is performed, it will be described in detail with reference to FIG. Check if there is an assistant for the text entered in step (3a). When there is no assistant in step (3a), it reads in step (3b) according to rule 1 as follows.

[Rule 1]

x, xxx, xxx, xxx, xxx, xxx

Thousand one hundred twenty one billion one hundred thousand one hundred and ten days (where x is the position of the number)

If x is 0, the part is not read.

If the position does not match when there is,, read the number with ',' as a unit.

However, when there is an assistant in step (3a), it checks whether the number is more than three digits in step (3c). In the process (3c), three digits or more are read according to Rule 1, and three digits or less are read by the following assistant checker.

Assistant Assistant

[Reading table]

1. spirit one two three four five six seven eight nine ten

2. One two three four five six seven eight nine ten twenty thirty forty fifty sixty seventy forty eighty ninety

3. One two three four five six seven eight nine ten twenty thirty forty fifty sixty seventy forty eighty ninety

4. One two three four five six seven eight nine ten twenty thirty forty fifty sixty forty seventy eighty ninety

5. One two three four five six seven eight nine ten twenty thirty forty fifty sixty seventy seventy eighty

The read value is converted into a Korean code in step 3g and recorded in a table in step 3h. In the case of English processing in step (2d), it is specifically executed as shown in FIG.

In step 4a, the input text value is compared with the English dictionary of the first dictionary 14. When the input text value exists in the first dictionary 14, the corresponding code is accessed in step 4c and converted to Korean code in step 4d.

If it is confirmed in step (4b) that the first dictionary 14 is absent, in step (4e) the alphabet is converted into Hangul code one by one. The code writes the converted value to the table in step (4f).

In the case of the Chinese character processing in the process (2e), it is specifically executed as shown in FIG. In operation 5a, the Chinese character conversion table is indexed by the first dictionary 14. In step (5b), check whether there are two or more ways to read Chinese characters. If it is not more than two, convert to Korean code in (5h) and record it in a table in (5i).

When there are two or more ways of reading Chinese characters in the process (5b), when the Chinese dictionary is searched in the first dictionary 14 in the process (5c) and converted into a Korean code in the process (5f) and not present (5e) The word that does not exist in the process is determined and stored in the table.

It is executed in the symbol processing date and time (2f), which will be described in detail with reference to FIG.

Symbol processing consists of abbreviated marks, feelings, questions, and quotation marks, quotation marks, slashes (/), modifiers, and word symbols (6a)-(6e), which are listed gradually in the process (6a)-(6e). After processing in the process, determine the reading method in the case of slash without generating code in case of shortening or punctuation in (6f)-(6i) and indexing in case of formula and word symbol. After conversion, write to table in (6k).

In step (2g), one process is shown in FIG.

Divide the Korean word into stems and mothers in step (7a) and record the stems and mothers separated in step (7b).

In the process (2h), Japanese processing is the same as that of FIG.

Index the Japanese table in step (8a), convert it to Korean code in step (8b), and record it in step (8c).

The code processed and stored in each process is phonologically processed in step 2i. The phonological process in step (2i) is performed as shown in FIG.

As mentioned above, by converting all texts composed of Hangul, Hanja, numbers, symbols, and English into Hangul code, the machine acquires the way of reading the text by humans. This has the advantage.

Claims (4)

In the speech synthesis method, a first process of classifying text by numbers, English, Chinese characters, symbols, Korean, and Japanese according to KS2 regulations, and a second process of converting the text into codes for each text classified in the first process And a third step of phonologically processing the coded text processed in the second step. The method of claim 1, wherein the number processing process includes: an assistant search process for checking whether an assistant is present; a first reading process of reading the input numeric value by a first rule when there is no assistant in the assistant search process; In the assistant search process, when there is an assistant, search for three digits or less and read three digits or less by the assistant table, and read a second digit or more by the first rule. 2, a text conversion method for speech synthesis, characterized in that the conversion process of converting the value read in the reading process into a Hangul code and write to the table. The method according to claim 1, wherein the English processing process searches for an English dictionary to determine the existence of the corresponding code, accessing the corresponding code when present in the process, converting the code into Korean code, and when not present in the process. Text conversion method for speech synthesis, characterized in that the process consisting of converting the alphabet to the Hangul code. The method of claim 1, wherein the Chinese character processing comprises: an indexing process of indexing a Chinese character conversion table, a first writing process of converting a Korean code into a table when the method of reading Chinese characters in the indexing process is less than two; And a second process of searching for a Chinese character dictionary when there are two or more reading methods, searching for the existence of a Chinese word, and writing them in a table.