WO2022164044A2 - K-hangul for speech recognition - Google Patents

Description

[Correction 28.02.2022 according to Rule 26]　K Korean for voice recognition

The present invention relates to the world's easiest and fastest language for conversational speech recognition (hereafter 'K-Hangul'), which selects syllables with clear sound distinctions from among Hangul, collects characters that match the character and pronunciation, to create an edible language.

For this, only 16 vowels with clear sound division are used, and only 7 vowels with correct pronunciation division are used to compose ' basic phonetic characters ' so that only one pronunciation is produced per character.

In addition, phonemic fluctuations may occur when conversing by connecting syllables and syllables of basic phonetic characters. Forms a matching ' transliteration character '.

In addition, proper nouns such as honorifics, numbers, place names, and names, as well as search words (commands) for machine operation, are recognized by using specific syllables <e.g., Ning, Zhu, Shuang...> to make it easier to distinguish them from general words. delimiter '.

In addition, it is equipped with a ' word separator ' without a support so that it can simultaneously perform a research role and a clear distinction between words and words that cannot be expressed by voice.

And, to make it easier to memorize words, the ' Integrated Recognition Character ' blocks the errors in speech recognition by unifying the expression of the conjugative word ending syllables such as adjectives, adverbs, nouns, and verbs into specific syllables <eg: zu, tsu> to make it easier to memorize them. ' to form

In addition, the first syllable of a basic word with the same or similar meaning is unified into the same syllable, and the ' word recognition character ' is constructed so that the unified first syllable does not change even when the basic form is used.

In addition, the 'safi assembly character ' is formed by assembling the 'safi (verbal verb, passive verb)' with specific syllables with clear pronunciation distinction <eg, king, ping>.

And, in order to express the tense as an assembly, a ' tense assembly character ' is composed of a specific syllable with clear pronunciation <eg: Neng, Nang, Naeng>.

In addition, in order to clearly distinguish the meaning and meaning of a sentence, a ' sentence assembly character ' is provided, which uses a specific syllable <eg: ka, keo, ba...> in the last syllable of the sentence by classifying the nature of the interrogative sentence. This is to create an error-free Hangul for speech recognition.

Hangeul has the best conditions to lead voice recognition in the digital age with sound letters, so if you standardize Hangul and create a new one, you can get a huge productivity boost.

However, the current Korean language is not suitable for voice recognition and digital age for the operation of artificial intelligence robots.

In particular, most syllables cannot be pronounced exactly as they sound, they must be used with phonetic symbols, and the classification of sentence types is ambiguous and unclear, so punctuation marks must be used. It is inappropriate to apply Hangul to speech recognition.

In addition, since Hangeul uses a language centered on characters and grammar despite being a phonetic alphabet, there are too many complicated phonemic changes that are linked, and the use of irregularly changing verbs and respectful words as well as expressions of the same expression are vaguely and complexly expressed. It is quite difficult to apply to voice recognition.

Moreover, because Hangul has many single-syllable words (e.g., has, many) with very short syllables, which is unfavorable for speech recognition, not only does the meaning change even if the pronunciation is high, low, long, short, slow, fast, or resting, but also words and words, The distinction between sentences and sentences is also often vague.

Also, there are many words with the same pronunciation but different characters or different meanings, and there are many errors in speech recognition when trying to match these to speech recognition. happening is a reality.

In addition, all languages can communicate by considering the front and back situations. Even in the same language, the voice has the characteristic of changing in various ways depending on the surrounding environment, the state of pronunciation, the emotions, facial expressions, and gestures of the speaker.

Therefore, communication with a machine that communicates only with human voice and text and a person who communicates by comprehensively judging all situations with voice cannot be properly communicated.

And, the 'continuous speech recognition' method currently being applied uses an inefficient method of storing a huge amount of sentence data in a large-capacity online server and dropping them one after another.

So, if applied to speech recognition in the current language that requires comprehensive judgment of all situations and environments, many sentences are required and the larger the amount of sentence data, the more entangled it becomes, making it more difficult to implement conversational voice recognition. The reality is that there is no

The 'continuous speech recognition' method that recognizes the entire sentence in this way is quite difficult and complicated to implement interactive voice recognition because there are proper nouns and numerous specific words that cannot be entered in advance of the data itself, and inevitably there are many errors. It is a structure that is bound to happen.

Therefore, in order to solve such a problem of speech recognition, the need for a new language is required.

The present invention was developed to solve the above inconvenience.

The present invention is an easy-to-learn language that is equipped with nine body recognition safety devices, so that all words of K-Hangul are memorized simply by listening, and at least six words are automatically memorized by assembling only one word. will make

In addition, K-Hangul, equipped with nine voice recognition safety devices (Table 1), is a digital language that enables error-free interactive voice recognition even when humans and robots as well as robots and robots exchange questions and answers for a long time. .

The present invention enables error-free communication during conversation through conversational Hangul for conversational voice recognition made so that only one pronunciation exists in one character, so that there is no error in conversational voice recognition between humans and machines as well as machines and machines The purpose is to make the implementation of

Therefore, it is widely applied to all fields such as economy, industry, society, education, medicine, culture, and service. It can be applied to any place where questions and answers are exchanged, such as counseling.

In particular, it will be possible to realize not only a keyboard-free computer and a mobile phone that operates only with voice, but also a pen meeting where singers and actors meet and talk with a pen virtually, and optimization of the metaverse, which combines the real world and the virtual world.

In addition, the present invention does not need to memorize each spelling separately, you can memorize all the words by just listening, and anyone can easily and quickly learn Hangul with 100 points for dictation.

In addition, it is possible to memorize words as well as honorific words, sapiens, tenses, sentence types, etc. without having to memorize them separately, so if you memorize only one basic word, at least 6 related words are automatically memorized. Anyone in the world can learn Hangul for recognition quickly and easily, regardless of age.

In this way, since K-Hangul is composed of syllables with clear pronunciation, communication is accurate and clear, so communication between humans and robots is possible without errors.

So, in preparation for the coming AI era, a language for speech recognition is needed.

The present invention is a language that can memorize words only by listening without having to memorize each spelling of all words separately, and enables interactive voice recognition without errors even in long conversations.

The composition of K-Hangul is shown in Table 1 below.

[Table 1] Basic composition of K-Hangul

In detail, in Hangul, vowels with unclear sound divisions are eliminated, and only vowels and consonants with clear sound divisions are selected so that letters and pronunciations match, and only one pronunciation sound can be produced per syllable.

To this end, the present invention uses 19 Hangul consonants as they are, and unifies vowels ㅐ, ㅒ, ㅔ, and ㅖ with the same or similar pronunciation as ' ㅐ ', and ㅚ, ㅙ, and ㅞ are pronounced as ' ㅚ '. , so that only 16 vowels are used.

In addition, double consonants and double consonants that cannot be pronounced literally are not used, and only the 7 consonants ㄱ, ㄴ, ㄹ, ㅁ, f, ㅅ, ㅇ are used for the consonants according to the syllable ending rules.

In this way, if vowels and consonants are arranged so that only one pronunciation exists in one syllable, the letters and pronunciations are the same and all words can be assembled, so K-Korean is easily learned 10 times faster than any other language in the world, and errors This will enable conversational voice recognition without the need for it.

Table 2 below shows the composition of the ' basic phonetic characters ' of Hangul for voice recognition.

[Table 2] Composition of basic phonetic characters

Accordingly, Hangeul with clear voice recognition is composed of about 11,000 syllables that have clear sound division and only one pronunciation per syllable.

In this way, during conversation between people, phonemes and spellings are separately explained in order to convey accurate meaning. If there is only one pronunciation in , no further explanation is necessary.

Table 3 below compares the example sentences using K-Hangul and Hangeul.

[Table 3] Example sentences in K-Hangul

Since the above 'why, other, and we' have the same pronunciation, K-Hangul is used as 'other'.

Also, 'ㅐ' and 'ㅔ' in 'Recurrence and Please' are unified as 'ㅐ' because the vowels are different but the pronunciation is the same.

In addition, even if there is only one pronunciation in one syllable, phonemic fluctuations may occur when syllables are connected due to the characteristics of the language, so the configuration of ' transliteration prevention characters ' is necessary.

For example, since 'relapse' and 'please' have different characters, there is a problem when applied to voice recognition.

So, in order to clearly distinguish the words 'recurrence' and 'please', 'please' is to form a new word with '<bar' by adding the word recognition character 'ba'.

Also, if you pronounce 'even though the pronunciation is the same', a phonological change occurs to 'barmun gatna'.

In addition, the basic form of 'same as' is a one-syllable word 'same', so the recognition probability of speech is low. It is to change the word itself into a two-syllable word with the same meaning but different characters.

In this way, since the recognition rate of a word is lower as the number of syllables is shorter, K-Hangul does not use single-syllable words if possible, and the length of the sentence is longer than that of Korean, but shorter than that of English.

And, to distinguish proper nouns such as general words, honorifics, numbers, search words, and names, a ' recognition delimiter ' is constructed.

For example, if you change the example sentence in Table 3 " Is n't it? It becomes an easy-to-learn and easy-to-understand structure.

In addition, if the recognition division ' ju ' is added to the preceding syllable of all numbers, ordinary words and numbers are distinguished and easily recognized.

In addition, if you look at an example of applying a mobile phone to a search term, you can turn on the mobile phone with <Turn on and turn it off>, turn it off, and use < Geomchi sap > for a search. To prevent voice recognition errors that may occur, compose a search word with a command of at least 3 syllables or more.

And, it forms a ' word separator ' without a consonant (allowing a consonant 'ㅇ' without a phonetic change) to distinguish between words and words that are difficult to express verbally.

In detail, proverbs with a suffix such as 'Eun, Eul, Man' in the current Hangul are not used in K-Hangul.

Because, if the word following the word delimiter (proposition) with a backing is a word that starts with 'ㅇ or ㅎ', a phonological change must occur.

For example, if you pronounce ' just to listen' quickly, 'listen' to 'dekima naeseo' is changed to 'deki', and 'ㄴ' goes backwards, causing phonological change from 'hae' to 'me'. .

So, K-Hangul uses word separators (subjects) of one or two syllables or more, so that words It constitutes a word delimiter that does not cause phonological change in the division.

In this way, there is no problem even if the pronunciation is long, short, loud, or low, and even if you pronounce it quickly or slowly, you can easily recognize between words. Table 4 below shows examples of word delimiters that do not cause any errors in recognition.

[Table 4] Example sentences of word classification

In addition, too many expressions of Hangeul are one of the reasons that make Hangeul difficult.

Therefore, by integrating various expressions of conjugation types such as adjectives, adverbs, nouns, and verbs, the ' integrated recognition character ' is formed so that the variety of expressions is maintained but Hangeul is easy.

In K-Hangul, the first syllable of the basic form does not change in any case, so there is no need to memorize words separately for conjugation forms such as adjectives, adverbs, nouns, and verbs.

Because the basic form is already used for general words, sapi-dong and tense words, you can take the basic form and use it for adjectives, adverbs, nouns, and verbs.

For example, if a word such as steep, long, enormous, softly, quietly has a different syllable for each word such as 'reun, ran, nan, si, hi', it becomes difficult to learn Hangul and recognize voice that much. will be.

So, if you take the basic form as it is and attach the integrated recognition character <z> to the back of the basic form, like steep <Garparz>, long <Gidaz>, enormous <Umchij>, quietly <Salgozu>, quietly <Joyongz>, will become

As another example, looking at 'breathing exercise', if 'breathe' of the basic form 'breathe' is taken as it is and unified into 'breath breathing exercise', it becomes possible to distinguish the words 'z' from 'breath' and 'exercise'.

In addition, words with the same or similar meaning are unified using the same syllable or consonant in the first syllable, and the unified first syllable constitutes a ' word recognition character ' that is configured not to change even when the basic form is used.

For example, the basic K-Hangul form of Hangul 'Boda 《see》 uses <Sibota>, a combination of Korean and English. becomes easier, the search and classification of words becomes faster, and it is possible to prevent speech recognition errors.

In other words, since the first syllable of a word has a relatively accurate pronunciation, only a small number of words with the same or similar meaning are found and recalled at the moment of pronouncing <poetry>, thereby enabling simple and accurate speech recognition without errors.

In addition, since the vowels 'a, ㅐ, ㅑ, ㅓ, ㅕ, ㅗ, ㅛ, TT, ㅠ, ㅡ, ㅣ' have relatively clear pronunciation, it is difficult to memorize the words by assembling them as in the example sentences in Table 4 below. it will be easy

[Table 5] Example sentences of word recognition characters

In this way, the language <element>, letter <jagul>, and word <jadan> in Table 5 are different from each other due to phonological fluctuations, so the word itself is used interchangeably in K-Hangul.

So, if word recognition characters are applied, learning Korean becomes easier and faster even if you learn K-Hangul and current Korean at the same time.

This is because, if you add 'word recognition' to the current Hangul, it becomes K-Hangul, and if you subtract it, it becomes the current Hangul, because it is possible to distinguish between K-Hangul and Hangeul with a very simple and convenient structure.

For example, in the case of <geo family, house>, if the word recognition character <geo> is subtracted, 'family, house' becomes the current Korean alphabet.

In addition, as shown in Table 5 above, K-Hangul can be assembled by using the Hangul sequence such as 'A, I, DA...', 'Dog, Nae, Dae...'.

By applying the order, arrangement, and shape of the Hangul alphabet, consonants, vowels, etc., it is possible to assemble almost all words such as animals and plants, as well as colors, classes, microorganisms, and elements.

However, when searching for words and sentences using the old word arrangement method for voice recognition, numerous words with completely different meanings and meanings are called up, and a large amount of data must be repeatedly searched for.

This will slow down the search speed and increase the number of voice recognition errors.

However, if the first syllable does not change using 'word recognition characters', a small number of words with the same meaning are grouped together to find the word, so you can find the word very accurately, quickly, and easily, and reduce speech recognition errors. will be.

In addition, the basic form of Hangeul 'come' is a one-syllable word with a low probability of voice recognition, so K-Hangul uses a two-syllable word <Keota> mixed with English.

Therefore, the first syllable of <Kuota>, <keo>, is structured so that it does not change under any circumstances, using the active verb <Keo king ta>, the passive verb <Keo ping ta>, and the clearly distinct syllable 'king, ping'. It forms a ' safi assembly character '.

[Table 6] Safi Assembly and Example Sentences of 'Keota'

As such, since there are not many words that can directly express the expression of safidong in all languages of the world, most words are expressed indirectly by bringing other words or sentences.

However, in K-Hangul, if only 'king' is inserted in the basic word, the correct verb can be expressed.

Therefore, K-Hangul can be easily assembled by inserting king and ping into the basic form of all words, so the distinction between verbs and passives is clear and clear, and accurate communication is possible without errors.

Also, looking at the tense application example of the basic form "eat <mom-muk-da>" in which the first syllable does not change, a specific syllable with a clear pronunciation distinction into 'past <mom-mook- neng -da>, future <mom- muk nang -da>, and present , 'Neng, Nang, Naeng' make up ' Tense Assembly Characters '.

[Table 7] Prototype assembly and example sentences of 'eat'

As such, in K-Hangul, the basic form never changes no matter what words such as 'sapija' or 'tense' are used.

This makes it easier to assemble and memorize words, and interactive voice recognition becomes possible.

And, ambiguous sentence type expression is the most important cause of speech recognition errors.

So, in order to communicate accurately by giving and receiving questions and answers without errors with the artificial intelligence robot, it is necessary to accurately and clearly express the nature of the sentence to be communicated.

In this way, error-free communication is possible.

[Table 8] Door bell assembly and usage

For example, if you currently say "Let's go home" in Korean, it is impossible to know whether it is a question sentence, a statement sentence, a request sentence, or a command sentence, without judging by the situation before and after.

So, for a clear expression of the ambiguous "Let's go home", based on the K-Hangeul basic form < Gogata >, the question asking for an answer is < Gogata >, and doubts about a situation that does not need an answer or an inner thought alone uses <expensive off >.

And the answer is <Gogata>, and if you want to accompany you, you can use <Gogapa>, and if you are angry and say "Go home" out loud, you can express it as <Gogapa>.

In addition, the expression of sentence types such as K-Hangul <Ka, Ta, Pa..> is not smooth because the pronunciation is strong.

And, since the recognition probability of a one-syllable word is low and the pronunciation is not smooth, the two-syllable <Gaka, Data, Bappa...> can be appropriately selected from two methods depending on the situation.

In this way, K-Hangul is a digital language that is equipped with 9 types of voice recognition safety devices to prevent speech recognition errors, and to learn K-Hangul and Korean at the same time easily and quickly.

Claims (2)

1. When conversing by selecting consonants and vowels with clear sound divisions in Hangul, and combining the basic phonetic characters and the ' basic phonetic characters ', which are composed of characters that have only one pronunciation per syllable while matching the character and pronunciation, In order to prevent phonological changes occurring due to the connection of syllables and syllables, the syllable is replaced with a non-phonological character or replaced with a new word, and the division of commands, honorifics, numbers, etc., such as names and places. To facilitate recognition of proper nouns, ' recognition delimiters ' using specific characters <e.g.: Ning, Zhu, Xiap...> A ' word delimiter ' consisting of silver, is) and nima (man), and an ' integrated recognition character ' that unifies the expression of the end syllables of adjectives, adverbs, verbs, and nouns into specific syllables <e.g., zu, tsu> ‘ Word recognition character ’, which unifies the first syllable of a basic word with the same or similar meaning into the same syllable, and forms the unified first syllable so that it does not change even when the basic form is used : ' Sapija Assembly Character ', which assembles 'ping' into specific syllables, ' Tense Assembly Character ' that assembles tense into specific syllables <eg: Neng, Nang, Naeng>, and complex consonants with strong pronunciation in ambiguous sentence types ' Assembling the door bell' is a method of assembling sentence types with characters <e.g.: k, k, k, k, k, k, k, k, p, p, ja> K-Hangul for interactive voice recognition equipped with 9 voice recognition safety devices with the character '.
2. In claim 1, Korean vowels ㅐ, ㅔ, ㅖ, and ㅒ with the same or similar pronunciation are unified as 'ㅐ', and ㅚ, ㅙ, and ㅞ with the same pronunciation are unified as 'ㅚ'. Double consonants and double consonants are not used, and only 7 consonants are used so that only one pronunciation occurs per character.

   For this purpose, 16 Hangul vowels are selected, with distinct sound divisions α, ㅐ, ㅑ, ㅓ, ㅕ, ㅗ, ㅛ, TT, ㅠ, ㅡ, ㅣ, ㅘ, ㅚ, ㅝ, ㅟ, and ㅢ. Hangul for speech recognition, consisting of ㄱ, ㄴ, ㄹ, ㅁ, ㅇ, ㅅ, ㅇ with clear and clear syllable division.