Computer-based generation of word forms
A database-driven system with symmetry-based classification and optimized syllable encoding addresses inefficiencies in speech-to-text technologies, enhancing speed and scalability while reducing resource consumption.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- MEZO TIBOR SANDOR
- Filing Date
- 2025-04-04
- Publication Date
- 2026-06-11
Smart Images
Figure HU2025000002_11062026_PF_FP_ABST
Abstract
Description
[0001] RECORD COPY PCT / HU2025 / 0 0 0 0 0 2 PATENT SPECIFICATION
[0002] TITLE OF INVENTION:
[0003] Computer-based Generation of Word Forms
[0004] Description
[0005] Subject of the invention: A computer-implemented method for generating word forms, comprising the creation and use of a database of words constructed from syllables, particularly for use in speech-to-text applications.
[0006] Brief indication of the field of application of the invention:
[0007] 1. In speech-to-text systems, the invention is primarily applicable in the development of speech-based text recognition systems requiring rapid and accurate text conversion.
[0008] 2. In database technology, the method utilizes structured databases and query mechanisms, thus serving as an efficient tool in data management systems, such as the development of language models or dictionaries.
[0009] 3. In language technology and natural language processing (NLP), through the use of word and syllable types as well as symmetry-based classification, the invention contributes to various NLP applications, such as machine translation, automatic correction tools, or language-learning software.
[0010] 4. In audio processing systems, the technology enabling the separation of speech signals and classification of phonemes is suitable for developing voice-based systems operating in noisy environments.
[0011] 5. In cloud computing, executing the method on servers or cloud-based systems enables the scalability of applications, making it suitable for serving large user bases, such as in smart assistants or customer-service chatbots.
[0012] 6. In linguistic research, the database coding system and symmetry-based classification method contribute to linguistic and phonological studies, particularly in the analysis of syllable and word structures.
[0013] 7. In educational applications, the invention facilitates the development of systems that assist language learners in acquiring correct pronunciation and accentuation.
[0014] 8. In Al and machine-learning-based systems, the structure supporting large-scale data processing and parallel search processes is also beneficial for training Al models, particularly in processing linguistic data.
[0015] 9. In mobile and loT devices, the optimized encoding and database management technology can be employed to enhance the efficiency of speech-based applications. Description of Solutions Closest to the Invention:
[0016] The solutions closest to the invention are technologies and systems related to speech recognition, database management, and language modelling.
[0017] Typing (spelling) via a keyboard is time- and resource-intensive. The reason for this is that our concept of words is very archaic: it relies on linguistic signs based on a set of symbols (letters of the alphabet, numbers, characters, etc.), fixed in numerical sequence within words (1 st, 2nd, 3rd, 4th, 5th, 6th, 7th letters, etc.). Meanwhile, the sequence of syllables following each other within words is not classified, nor are instances of words optimized according to the syllabic sound sequence. (On average, words consist of fewer than three syllables.)
[0018] The unique aspects of the solution presented in the invention, such as the symmetry-based classification method and optimized encoding of the syllable database, further enhance existing technologies, particularly in the fields of NLP and Speech-to-Text systems. The solutions mentioned below partially achieve the objectives of the invention, but the symmetry-based database design represents a unique feature.
[0019] I will present the relevant solutions:
[0020] 1. In Speech-to-Text systems, the Google Speech-to-Text API is a machine learning-based system capable of converting speech into text with high accuracy. While it does not emphasize a database-based operation, it relies heavily on acoustic signal processing and NLP. IBM Watson Speech to Text is a speech recognition solution specifically optimized for cloud-based applications, capable of separating speech signals from noisy backgrounds, thus supporting accurate text generation.
[0021] 2. In database management solutions, SQL-based indexed databases relate to the “structured queries" and "index structure" mentioned in the claims, referring to SQL and NoSQL systems such as PostgreSQL or MongoDB, which enable fast data access. Redis is an in-memory database that provides high speed for parallel processing of large volumes of data. It is relevant in the context of the "parallel search processes" mentioned in the invention.
[0022] 3. In the classification of phonemes and syllables, the CMU Pronouncing Dictionary is a lexicon that contains phonemic and syllabic classifications of English words. Although it is not symmetry-based, it performs similar functions to the database described in the invention. The HTK (Hidden Markov Toolkit) can be applied in speech recognition systems, where the encoding of phonemes and syllables is performed based on acoustic patterns.
[0023] 4. In database and encoding solutions for speech recognition, the Kaldi Speech Recognition Toolkit is an open-source framework well-known in the field of database-driven speech recognition and acoustic modelling. It supports fast data retrieval and parallel processing. DeepSpeech is a machine learning model used for processing speech signals and generating text output. While it places less emphasis on database optimization, it focuses strongly on accurate text generation.
[0024] 5. In cloud-based solutions, AWS Transcribe is a speech recognition service provided by Amazon that processes acoustic input into structured text and is capable of handling large volumes of data. Azure Cognitive Services Speech, Microsoft's solution, is built on a cloud-based architecture and enables database-driven speech-to-text transformation. 6. In language models and NLP, BERT and GPT models primarily focus on text-based tasks but can be integrated with speech recognition systems to enhance database-driven processing. Tesseract OCR is suitable for integration with linguistic databases for text recognition, although it is primarily used for optical character recognition.
[0025] Explanation of the Terms Used;
[0026] Phonometrics: Phonometrics is a complex, artificially created technical term derived from the words “phoneme” (the smallest meaning-distinguishing unit of speech sound) and “metrics” (measurement, structure, rhythm). Phonometrics thus refers to a procedure or method that measures, classifies, or organizes speech sounds, syllables, or their structures— particularly for purposes related to language technology, phonology, or database construction. In the context of the invention, phonometrics serves to analyze and classify the structure of word forms and syllables, especially based on symmetry principles, thereby enabling more efficient speech processing. It is a subfield of linguistic description, examining how vowels and consonants form syllables, and how syllables combine to form words.
[0027] Syllable font: The term szotagfont (in English: syllable font) refers to a specially designed typeface or character system where the basic visual units represent syllables instead of individual letters. In contrast to traditional alphabetic fonts — where each character stands for a single phoneme or letter — a syllable font encodes common syllables as single graphical units. This can improve readability, facilitate faster text input or output (especially in speech technology), and optimize language processing systems by aligning more closely with spoken language units. Such fonts may be used in educational tools, speech-to-text interfaces, or experimental writing systems where syllabic representation is more efficient or intuitive than traditional alphabetic writing. The study of syllabic writing systems.
[0028] Syllable: The syllable is the natural, and possibly metrical, unit of words. In Hungarian, a syllable is minimally composed of a vowel (very rarely a consonant), which functions as the syllable nucleus. This vowel may be preceded by consonants (syllable onset) and / or followed by consonants (syllable coda). For example: 6, s, ez, 16, mar, konyv, stressz, etc. As a metrical unit in Hungarian, a syllable is characterized by duration (temporal value), stress, or sometimes both.
[0029] The word form: The word form is a word marked with a sign (or signs) and / or an affix (such as a suffix), whose structure is related to the relational meanings it conveys within a sentence.
[0030] The dictionary word (lexeme: The dictionary word is a conventional word known and used by a language community, in the case of inflected words, it is an abstract unit derived from the inflectional paradigm, representing the entire paradigm — that is, all word forms — in the dictionary.
[0031] The word stem: The word stem is the element of a compound word or word form that typically carries the lexical meaning. PCT / HU2025 / 0 0 0 0 0 2 The word instance (word occurrence): The word- instance (word occurrence) is each individual occurrence of a word form of a dictionary word within a specific corpus (a closed body of text).
[0032] Symmetry: Symmetry, also known as reflection, is a general term for certain classes of geometric transformations (mappings). Symmetry is an isometric— that is, distance-preserving — and involutory mapping. Because it preserves distances, it also preserves straight lines and angles.
[0033] Phonological planning: Levelt and colleagues designed the production process of word forms, for example, for the word labda ("ball") (Tar 2017: 27; see also Figure 7). References Used: PCT / HU2025 / 000002 1. A magyar nyelv ertelmezo szotara l-VII. 1959-1962. Edited by the staff of the Linguistic Institute of the Hungarian Academy of Sciences under the direction of Geza Barczi and Laszlo Orszagh. Akademiai Kiado, Budapest.
[0034] 2. Jakab, Laszlo — Bolcskei Andras 2000. Balassi-szotar. Debrecen. Szami'togepes Nyelvtorteneti Adattar 8.
[0035] 3. Papp, Ferenc 1969, 1994. A magyar nyelv szovegmutato szotara. Akademiai Kiado, Budapest.
[0036] The Reverse Dictionary of the Hungarian Language (A magyar nyelv szovegmutato szotara) lists all 58,323 headwords from The Explanatory Dictionary of the Hungarian Language (A magyar nyelv ertelmezo szotara), arranged according to their endings. I noticed that the authors of the Balassi Dictionary (Balassi-szotar) present the headwords in their inflected forms, so I began my Balassi research by assigning a code to each letter symbol of every word form. During this process, I discovered the syllable as a suitable unit of analysis and shifted my focus to indexing syllables. A valuable resource proved to be the Reverse Dictionary of the Hungarian Language, from which I broke down the words into syllables and thereby built a word and syllable database. Subsequently, I applied the newly identified syllabic structure to a real-world example, which led to the creation of the Textbook of the Old Hungarian Lament of Mary — at that point, I was also using Tolcsvai’s Grammar. My key insight into syllable structure was that syllables within word forms are organized in multiple symmetrical patterns — for example, CVC: A2'(<bl>), and in the word bal-la-da, the initial-medial-final syllable structure: < A2< B(1)> C2>. This pattern had not been identified by Levelt, Eva Tar, or anyone else before me. Although a Khmer syllable database was developed by Vietnamese researcher Tran Van Nam and his colleagues, my insight remains unrecognized by them as well. This is because under the term “syllable types,” they refer to units larger than words (such as simple words, compound words, or phrases), and they do not describe the internal word-level structural system I use— namely [CVC: A2]: [A2< B(1)> C2]. Furthermore, although Wayit Abliz and his colleagues developed a syllable-based technique for text compression in the Uyghur language, their method considers only the construction of a word form from syllable types during the compression process. However, the word forms and syllables are not classified into syllable types, and the symmetry principle is not taken into account in the generation of word forms. Therefore, their approach differs from the solution intended to be protected.
[0037] References
[0038] 1. Imrenyi, Andras — Kugler, Nora — Ladanyi, Maria— Marko, Alexandra— Tatrai, Szilard— Tolcsvai Nagy, Gabor 2017. Nyelvtan. Edited by Gabor Tolcsvai Nagy. Osiris Kiado, Budapest.
[0039] 2. LEVELT, Willem J. M. — Roelofs, Ardy — Meyer, Antje S. 1999. A theory of lexical access in speech production. Behavioral and Brain Sciences 22. 1-75.
[0040] 3. Mezo, Tibor 2021. Omagyar Maria-siralom szövegkönyve. {Textbook of the Old Hungarian Lament of Mary) Novum publishing, Ausztria. Tar, Eva 2017. Fonologiai fejlodes, variabilitás, beszedhanghibak. ELTE Eotvos Kiado, Budapest.
[0041] Tran, Van Nam —Nguyen, Th i Hue — Phan, Huy Khanh 2017. Building a syllable database to solve the problem of Khmer word segmentation. International Journal on Language Computing. 2017.02.28. DOI:10.5121 / ijnlc.2017.6101
[0042] Wayit Abliz — Hao Wu — Maihemuti Maimaiti — Jiamila Wushouer — Kahaerjiang Abiderexiti — Tuergen Yibulayin — Aishan Wumaier 2020. A Syllable-Based Technique for Uyghur Text Compression.2020.03.23. doi.org / 10.3390 / info11030172 Indication of the problem to be solved by the invention:
[0043] The objective of the invention is to increase the accuracy, speed, and scalability of the speech-to-text conversion process by developing a database-driven system that applies symmetry-based classification for the processing and generation of word forms.
[0044] The tasks to be solved by the invention, based on the claims, can be summarized as follows: 1. Creation and Optimization of a Database for Word Form Generation
[0045] 1.1. Task: To create a database that enables fast search and query capabilities for syllables and words, particularly for speech recognition applications.
[0046] 1.2. Problem to be solved: Optimization of the database’s index structure and coding system for efficient access.
[0047] 2. Encoding of Syllables and Words
[0048] 2.1. Task: Assigning unique codes to syllables and words based on their type and category, applying a symmetry-based classification principle.
[0049] 2.2. Problem to be solved: Developing a coding system that ensures accurate mapping and efficient database maintenance.
[0050] 3. Conversion of Spoken Input into Text
[0051] 3.1. Task: Transforming acoustic input (e.g., speech recorded by a microphone) into accurate textual output.
[0052] 3.2. Problem to be solved: Separating sources of speech signals to generate precise text output even in noisy environments.
[0053] 4. Generation of Symmetrical Word Forms
[0054] 4.1. Task: Designing a symmetrical schema for word forms that supports structured processing of syllables.
[0055] 4.2. Problem to be solved: Integrating database structure and encoding methods to enable symmetrical word form generation.
[0056] 5. Database Maintenance and Expansion
[0057] 5.1. Task: Encoding new words and syllables using the symmetry-based classification method.
[0058] 5.2. Problem to be solved: Ensuring efficiency in expanding the database and maintaining compatibility with existing data.
[0059] 6. Processing of Large Data Volumes and Parallel Searches
[0060] 6.1. Task: Structuring the database to support parallel processing and search of large datasets.
[0061] 6.2. Problem to be solved: Ensuring scalability and parallelization to improve overall system performance.
[0062] 7. Hardware and Software Integration
[0063] 7.1. Task: Implementing the procedure on a system that includes:
[0064] 7.1.1 A database management system for executing structured queries. PCT / HU2025 / 0 0 0 0 0 2 7.1.2 Acoustic input devices such as microphones.
[0065] 7.1.3 Cloud-based or server-based infrastructure.
[0066] 7.2. Problem to be solved: Integrating and optimizing system components for fast processing and generation of word forms.
[0067] 8. Optimized Integration Between Syllable Repository and Database
[0068] 8.1. Task: Improving the connection between the syllable repository and the database for faster search and processing.
[0069] 8.2. Problem to be solved: Enhancing database synchronization and accelerating searches across datasets to improve user experience.
[0070] Results Arising from the Solved Tasks:
[0071] 1. Efficiency Enhancement: Current methods of text input, such as manual typing, are highly time- and energy-consuming and rely on the letters of the alphabet and their sequential order. The syllable-based approach of the invention drastically reduces the amount of data to be processed (5,062 syllables vs. 191,559 syllables), enabling faster and more energy-efficient text input, especially in acoustic systems.
[0072] 2. Increased Data Processing Speed: The new structural configuration, through syllable formation and encoded architectures, significantly accelerates data processing. This is particularly advantageous for high-performance systems, such as Al-powered platforms or applications requiring extensive speech recognition. The current state of the art does not present a similarly optimized solution.
[0073] 3. Broadened Applicability: The invention can be applied in operating systems, smart devices, accessibility solutions, and speech recognition technologies. The pronunciation fonts described in the current state of the art are primarily limited to static textual content and simpler pronunciation functions, whereas the invention opens new pathways in the field of dynamic and adaptive text processing.
[0074] 4. Resource Savings: The invention’s optimized syllable-based approach requires less memory and computational capacity. In the current state of the art, font management methods (e.g., caching, APIs) focus on quick access to fonts but do not offer a specific energy-efficient solution that utilizes optimized syllable encoding.
[0075] 5. Research and Technological Innovation: The invention combines the sciences of phonology and computer science, resulting in a deeper integration and new research opportunities compared to systems presented in the current state of the art — particularly in the f ield of speech recognition technologies. The invention has the potential to support the development of a new generation of operating systems and applications with enhanced capabilities. The most general solution to the stated problem, following the main claim:
[0076] The objective of the invention is to provide an efficient, scalable, accurate, and easily expandable solution for the generation of word forms, particularly for speech recognition (speech-to-text) applications, while ensuring the optimization of databases and processing procedures. It combines database management, the conversion of spoken input into text, and the precise encoding of words and syllables.
[0077] Key Elements
[0078] 1. In the construction and optimization of the database, a database is created that contains syllables and words, supported by an index structure for fast search. The structure of the database enables structured queries and the rapid processing of large volumes of data.
[0079] 2. In the encoding system, each syllable and word is assigned a unique code, taking into account the characteristics of syllable types and categories. The encoding is based on a symmetry-based classification method, which supports the maintenance and expansion of the database.
[0080] 3. In the speech-to-text procedure, the acoustic signals of spoken input are processed and accurately mapped into text. The sources of speech signals are separated to reduce noise and interference.
[0081] 4. In the symmetrical structure of word forms, the word forms are generated from encoded syllables based on a symmetrical schema, ensuring organized processing and a unified approach.
[0082] 5. In terms of scalability and parallel processing, the structure of the database enables parallel search and processing operations, allowing large volumes of data to be handled quickly and efficiently.
[0083] 6. In hardware and software integration, the system includes a database management system, acoustic input devices (such as a microphone), as well as server- or cloud-based infrastructure. The optimization of the integration between the database and the syllable repository ensures fast processing and generation.
[0084] The Process of Editing the Syllable Repository
[0085] 1. Based on the process of editing the syllable repository, the databases are accessed in order to record relevant data related to the syllable inventory of the Hungarian language, applying a descriptive empirical research method.
[0086] The Databases Identif ied by Their File Names:
[0087] a. szovegtar-magyarszotagtar.xls ~ Entry of the Reverse Dictionary (Word Ending Repository)
[0088] b. szovegszotar-magyarszotagtar.xls - Entry of the Word-Ending Dictionary c. szotar-magyarszotagtar.xls - Entry of the Dictionary
[0089] d. magyarszovegszotar~magyarszotagtar.xls - Word Types (Lexemes) in the Hungarian Word-Ending Dictionary
[0090] e. magyarszotagtar-magyarszotagtar.xls - Syllable Types in the Hungarian Syllable Repository (Morphemes: Syllable, Metrical Foot) PC / " / HU2025 / 0000° 2 f. hangzotar-magyarszotagtar.xls - Phoneme Repository, Phoneme Types
[0091] (Phonemes)
[0092] . From the dictionary section of The Reverse Dictionary of the Hungarian Language, the headwords are syllabified, and the syllables are entered side by side into separate cells, forming the basis for their classification into syllable types. A total of 58,301 word forms are listed, along with 3 prefixes and 22 suffixes (in total: 58,326 indexed entries).
[0093] . Based on The Reverse Dictionary of the Hungarian Language, the words are listed in reverse order and syllabified accordingly, while their grammatical information is encoded.. The words are arranged in alphabetical order. The grammatical codes of the words are preserved.
[0094] . The words are classified into word types based on their syllable count: D word, A word, B1 word, B2 word, B3 word, B4 word, B5 word, B6 word, B7 word, B8 word.
[0095] For example:
[0096] D word = sztrajk (monosyllabic word)
[0097] A word = labda (two-syllable word)
[0098] B1 word = ballada (three-syllable word)
[0099] B2 word = levendula (four-syllable word)
[0100] etc.
[0101] 6. The alphabetical ordering is based on the reverse dictionary. The words are counted, which also allows us to infer the number of syllables.
[0102] Results Report (Word Type, Syllable, Syllable Count, and Number of Words)
[0103] SzavakT: ■ szdma I. i;
[0104] Bffl j BtM B(8) 1
[0105] 1 1 2 L 3 3 4 5 > 6 | 7 8 9 10 — -J
[0106] i 1721 5539,15211 300 89 58.326 ]
[0107] 12734 2OT42i 15608 5539 1521 j 300 59
[0108] >20842115608 5539Ti52TrWT 59
[0109] 15608 553911521 [ 300 "{ 59*
[0110] 5539 j 1521) 300 T 59
[0111] i1521 I 300 59
[0112] 1 300 [ 59
[0113] T’ 59
[0114]
[0115] 25468 52526 62432 27695 9126 2100 472 9 10 191,559
[0116] • Derived from the word types, the syllables are classified into, for example, 55 syllable types:
[0117] - D syllable.
[0118] [The term D syllable refers to a specific syllable type derived from the classification system based on word types. In this context, “D” typically represents monosyllabic words or syllables— those that consist of a single syllable. These syllables often serve as the structural foundation for analyzing more complex word forms and for building syllable-based databases in linguistic research or speech processing applications.] - A1 syllable, A2 syllable, A3 syllable, A4 syllable, A5 syllable, A6 syllable, A7 syllable, A8 syllable, A9 syllable;
[0119] [These labels refer to the systematic classification of word-initial syllables based on specific phonological or structural features. In this system: The prefix " A" indicates that the syllable appears at the beginning of a word. The numbers 1 through 9 distinguish subtypes of initial syllables, with classification based on the total number of syllables in the word. For example:
[0120] A1: initial syllable of a two-syllable word
[0121] A2: initial syllable of a three-syllable word
[0122] A3: initial syllable of a four-syllable word, and so on.
[0123] This labelling system allows for a structured and fine-grained categorization of initial syllables, aiding in phonological analysis, database construction, and speech technology applications.]
[0124] - B1 syllable, B(2-1 ) syllable, B(2-2) syllable, B(3-1) syllable, B(3-2) syllable, B(3-3) syllable, B(4-1) syllable, B(4-2) syllable, B(4-3) syllable, B(4-4) syllable, B(5-1) syllable, B(5-2) syllable, B(5-3) syllable, B(5-4) syllable, B(5-5) syllable, B(6-1) syllable, B(6-2) syllable, B(6-3) syllable, B(6-4) syllable, B(6-5) syllable, B(6-6) syllable, B(7-1 ) syllable, B(7-2) syllable, B(7-3) syllable, B(7-4) syllable, B(7-5) syllable, B(7-6) syllable, B(7-7) syllable, B(8-1) syllable, B(8-2) syllable, B(8-3) syllable, B(8-4) syllable, B(8-5) syllable, B(8-6) syllable, B(8-7) syllable, B(8-8) syllable.
[0125] [These labels refer to a systematic classification of medial syllables (syllables that occur within the body of a word, between the first and last syllable), according to their position and the total number of syllables in the word. In this system: The prefix " B” indicates that the syllable is a medial (middle) syllable. The number in parentheses reflects the total number of syllables in the word and the specific position of the syllable within that word. For example:
[0126] B(1): the first (and only) medial syllable in a three-syllable word
[0127] B(2-1 ): the first medial syllable in a four-syllable word
[0128] B(3-2): the second syllable in a five-syllable word
[0129] B(5-3): the third syllable in a seven-syllable word
[0130] B(8-6): the sixth syllable in an ten-syllable word
[0131] This fine-grained classification enables precise tracking, encoding, and analysis of syllables in complex word forms, supporting advanced phonological modeling, syllable-based databases, and speech technology systems.]
[0132] - C1 syllable, C2 syllable, C3 syllable, C4 syllable, C5 syllable, C6 syllable, C7 syllable, C8 syllable, C9 syllable.
[0133] [These labels represent a systematic classification of final syllables in words, based on their position and the total number of syllables in the word. In this system: The prefix “C” indicates that the syllable is a final (word-ending) syllable. The numbers 1-9 refer to the syllable’s position in the word, which also reflects the word's total syllable count. For example:
[0134] C1: final syllable of a two-syllable word
[0135] C2: final syllable of a three-syllable word
[0136] C3: final syllable of a four-syllable-word
[0137] C9: final syllable of a ten-syllable word
[0138] This classification allows for the detailed analysis and organization of word-final syllables, contributing to linguistic research, speech synthesis, and database-driven speech-to-text technologies.] In summary, the naming convention is based on the position of the syllable within the word: initial (A), medial (B), final (C), and standalone syllables representing one-syllable words (D). For initial (A) and final (C) positions, the total number of syllables in the word is indicated with ascending Arabic numerals (e.g., A1, C2, etc.). For medial (B) positions, the notation is enclosed in parentheses. The first number refers to the total number of medial syllables in the word, and the second number refers to the position of the given syllable within the medial sequence. Example:
[0139] B(4-2) indicates the second medial syllable in a six-syllable word, which contains a total of four medial syllables.
[0140] This system allows for precise classification and indexing of syllables based on their structural role and position within words, providing a foundation for advanced phonological analysis and computational language processing.
[0141] Within each syllable type, the syllable pairs are numbered and listed one below the other. Subsequently, in a separate column, the syllables are arranged in alphabetical order, and their frequency is counted. As a result, we obtain three key metrics:
[0142] - the total number of syllable occurrences (e.g., in the A3 syllable type: 15,608 syllables),
[0143] - the number of distinct syllables (e.g., in the A3 syllable type: 1,398 different syllables),
[0144] - and the frequency of individual syllables (e.g., in the A3 syllable type: the syllable “ad" occurs 5 times).
[0145] The frequency of individual syllables is represented on two separate worksheets. The syllables are entered as records (rows), and the syllable types are entered as fields (columns).
[0146] The intersection of the rows and columns allows the corresponding cell to display both the name of the syllable type and the frequency of the syllable.
[0147] As a result, we obtain the following:
[0148] - the number of syllable occurrences across different syllable types: 25, 194 instances,
[0149] - the total number of syllable occurrences in the examined vocabulary set: 191,559 instances,
[0150] - the total number of unique syllables (syllabic inventory) in the examined vocabulary set: 5,062 syllables.
[0151] In summary, for acoustic text input in a computer system, it is sufficient to program the pronunciation of only 5,062 syllables in Hungarian, for example. Compared to the total number of syllable occurrences — 191,559 — this is a relatively small quantity. The novelty of the invention lies in demonstrating that, contrary to expectations of large volumes, the syllabic inventory remains manageable within a syllable pronunciation system. The invention will be described in more detail below with reference to exemplary embodiments illustrated in the accompanying drawings, in which the figures are listed with their respective subject matter as follows:
[0152] 1. Figure 1 - Symbolic structure of the word asztal
[0153] 2. Figure 2 - Symmetrical schematic structure of the syllable [stra:jk]
[0154] 3. Figure 3 - Labeling of the word sztrajk
[0155] 4. Figure 4 - Diagram of the encoded structure of the word labda
[0156] 5. Figure 5 - Diagram of the encoded structure of the word ballada
[0157] 6. Figure 6 - Diagram of the encoded structure of the word bal
[0158] 7. Figure 7 - Word form production according to Levelt et al.
[0159] 8. Figure 8 - Process of speech formation from sounds to words
[0160] Figure 1 shows the symbolic structure of words according to the invention, applied to the word asztal (meaning table in English). The technical content arises from both the meaning of the word asztal (a piece of furniture consisting of a horizontal surface supported by legs) and the circuit arrangement of the device carrying the sound sequence (i.e., the computer system in the field of information technology). The sound sequence is referred to as the phonological pole, while the meaning is referred to as the semantic pole. Computers encode, store, and make available sound sequences (letters) by arranging them into fonts. As a guide to this process, Figure 1 breaks down the word [ostol] into the syllabic level: an initial syllable (labeled A1) and a final syllable (labeled C1 ). At the center of each syllable is the vowel (represented by the letter A), while the consonants (such as SZ, T, and L) are positioned on the periphery, visually separated by concentric circles to indicate structural hierarchy. The combination of syllables in the word asztal is indicated by the '+' symbol.
[0161] Figure 2 illustrates a concentric circle and a schematic grid, which is a square layout symmetrically divided (mirrored) along its vertical axis with a thin line. In the centre of the concentric structure lies the vowel of the syllable — for example, the letter A in the syllable of sztrajk (meaning strike in English).
[0162] • Before the A, the consonant R is placed;
[0163] • Before the T, the consonant SZ is placed— each in its own circle (layer);
[0164] • After the A, the consonants J and K are positioned, again each in a separate peripheral circle.
[0165] This centre-periphery model allows for the segmentation of any Hungarian syllable into phonemes. For example, the syllable [stra:jk] can be represented as: <sz<t<rj>k>, demonstrating a symmetrical structural schema. In this system, a one-syllable word is marked with the label D.
[0166] [In the context of the invention, sztrajk is a monosyllabic word and is analyzed for its phonological structure and symmetrical syllable schema. Phonemic breakdown (IPA):
[0167] [ftrɔ.jk] → represented orthographically as: sztrajk.
[0168] Syllable type:
[0169] • Since it consists of a single syllable, it is categorized as a D syllable (one-syllable word type).
[0170] Symmetrical schematic segmentation (Figure 2 model):
[0171] Using the concentric circle / centre-periphery model:
[0172] • Central vowel (nucleus): A
[0173] • Onset consonants (before the nucleus): SZ → T → R
[0174] • Coda consonants (after the nucleus): J → K
[0175] Visual structure:
[0176] <sz<t<rj>k>
[0177] This notation captures the symmetrical nature of the syllable, where phonemes are distributed around the vocalic core (A), with consonants arranged in mirrored positions moving outward from the centre.]
[0178] Figure 3 shows the labelling of the word sztrajk.
[0179] Syllable labelling refers to the analytical process in which the phonemes within a syllable are assigned group properties — either vowel or consonant— or, more precisely, only these group-level properties are taken into account when constructing the phonetic structure of the syllable. In this classification system:
[0180] • V = vocalis (Latin for vowel)
[0181] • C = consonans (Latin for consonant)
[0182] Applied to the syllable sztrajk, the phonemes are labelled as follows:
[0183] • SZ→ C
[0184] • T→ C
[0185] • R → C
[0186] • Á→ V
[0187] • J→ C
[0188] • K→ C
[0189] Thus, the full phoneme labelling for the syllable is:
[0190] C-C-C-V-C-C PCWJ2W 00 0 00 2 This abstraction focuses not on the individual letters, but on their functional phonological roles, enabling a structured and generalized syllable representation for computational and linguistic applications.
[0191] Figure 4 shows the encoded structure of the word labda (cf. Figure 7 in the representation by Levelt and colleagues, as referenced in Tar 2017: 27). In this context, the word labda (ball in English) is analyzed using the invention’s symmetry-based encoding model, where syllables are structurally segmented and labelled according to their phonological roles and positional types. Syllabification: lab-da
[0192] • Initial syllable (A1 ): lab
[0193] • Final syllable (C2): da
[0194] Encoding structure:
[0195] • lab→ C-V-C
[0196] • da→ C-V
[0197] The phonemes are mapped according to their consonant (C) or vowel (V) function, possibly using concentric structural segmentation as described in earlier figures. The focus is on how these syllables symmetrically mirror each other around vowel nuclei and contribute to the word’s encoded template.
[0198] Symmetrical phonological template example:
[0199] • labda -* <lb> + <d<a»
[0200] This representation shows the central vowels (a) as phonological nuclei and the surrounding consonants symmetrically placed at the periphery, aligned with the system used to process syllables and word forms in the invention.
[0201] We start from the sound sequence [labda], meaning ‘ball’, which is divided into types (the label A1 means “initial syllable in two-syllable words”; the label C1 means “final syllable in two-syllable words”), and then further into categories (CVC; CV). These form the syllable types and syllable forms.
[0202] The syllable type classifies the syllable based on its position within the word (initial, medial, or final). The syllable form organizes the syllable according to the group properties of phonemes, based on its internal structure of vowels and consonants (onset, nucleus, coda — or beginning, core, and closure).
[0203] Thus, each syllable receives two codes, separated by a colon, for example: CVC: A1 or CV-. C1. At the bottom level are the segmented phonemes (sounds): / I / , / o / , / b / , / d / , / o / . So far, in the explanation, we have moved top-down on the left side of Figure 4. Now, on the right side of Figure 4, we proceed bottom-up, due to the way the computer’s processing unit operates. At the syllable form level, the syllables are combined (KMZ + KM = <lab + da>), while at the syllable type level, the combined syllables are matched with the word's sound sequence: <labda> = [labda], in order to ensure the output is accurate. Figure 5 presents the encoded structure of the word ballada, which differs from the word labda shown in Figure 4 in that the number of syllables increases from two to three at the syllable type level.
[0204] This increase requires a modification of the syllable type labels:
[0205] • A2 = initial syllable in three-syllable words,
[0206] • B(1 ) = medial syllable in three-syllable words,
[0207] • C2 = final syllable in three-syllable words.
[0208] Naturally, in words with four to ten syllables, the number of syllables— and thus the labeling— continues to rise accordingly.
[0209] The encoded syllable structures of the word ballada are as follows:
[0210] • CVC: A2
[0211] • CV: B(1)
[0212] • CV: C2
[0213] Figure 6 presents the encoded structure of the word bal (meaning left in English), which differs from the word labda shown in Figure 4 in that, at the syllable type level, the number of syllables is reduced to one. These represent monosyllabic words. The designation for a one-syllable word is: D. The encoded syllable structure of the word bal is: CVC: D.
[0214] Figure 7 illustrates how Levelt and his colleagues designed the process of word form production. Their model is based on the use of linguistic signs, which rely on a predefined symbol set (the letters of the alphabet), and the word forms are encoded in a fixed numerical sequence— that is, the letters of a word are ordered by position (1st, 2nd, 3rd, 4th, 5th, etc.).
[0215] Figure 8 illustrates the process of speech formation from sounds to words, providing a summary of the previously described elements in the form of a flowchart.
[0216] At the top level, words consist of a sound sequence and meaning, based on the principle of association.
[0217] At a lower level, syllable types follow the principle of combination, where syllables are assembled to form word forms:
[0218] initial syllable + medial syllable + final syllable = word form
[0219] Below that, at an even deeper level, the principle of symmetry governs the formation of fundamental syllable structures from groups of sounds — typically:
[0220] consonant + vowel + consonant
[0221] These structures are categorized in the syllable repository as:
[0222] • full syllables, • open syllables,
[0223] • closed syllables,
[0224] • reduced syllables,
[0225] • or non-structured syllables.
[0226] At the bottom layer of the flowchart are the letters of the alphabet (ABC), numbers, characters, etc.
[0227] (Note: The flowchart is also naturally interpreted in an ascending direction, from phonemes upward to fully formed words.) Examples Supporting the Scope of Protection:
[0228] 1. "Each syllable and word is assigned distinct codes based on their syllable and word types and categories," e.g.:
[0229] a. Words are classified into word types based on their number of syllables:
[0230] D word, A word, B1 word, B2 word, B3 word, B4 word, B5 word, B6 word, B7 word, B8 word.
[0231] For example:
[0232] D word = sztrajk (monosyllabic word)
[0233] A word = labda (two-syllable word)
[0234] B1 word = ballada (three-syllable word)
[0235] B2 word = levendula (four-syllable word)
[0236] etc.
[0237] b. The syllables derived from the word types are classified into 55 syllable types:
[0238] - D syllable;
[0239] - A1 syllable, A2 syllable, A3 syllable, A4 syllable, A5 syllable, A6 syllable, A7 syllable, A8 syllable, A9 syllable;
[0240] - B1 syllable, B(2-1) syllable, B(2-2) syllable, B(3-1) syllable,
[0241] B(3-2) syllable, B(3-3) syllable, B(4-1) syllable, B(4-2) syllable, B(4-3) syllable, B(4-4) syllable, B(5-1 ) syllable, B(5-2) syllable, B(5-3) syllable, B(5-4) syllable, B(5-5) syllable, B(6-1) syllable, B(6-2) syllable, B(6-3) syllable, B(6-4) syllable, B(6-5) syllable, B(6-6) syllable, B(7-1) syllable, B(7-2) syllable, B(7-3) syllable, B(7-4) syllable, B(7-5) syllable, B(7-6) syllable, B(7-7) syllable, B(8-1) syllable, B(8-2) syllable, B(8-3) syllable, -B(8-4) syllable, B(8-5) syllable, B(8-6) syllable, B(8-7) syllable, B(8-8) syllable. - C1 syllable, C2 syllable, C3 syllable, C4 syllable, C5 syllable,
[0242] C6 syllable, C7 syllable, C8 syllable, C9 syllable.
[0243] For example:
[0244] D syllable = sztrajk- A1 syllable = lab- A2 syllable = bal- A3 syllable = le- B1 syllable = la- B(2-1) syllable = ven- B(2-2) syllable = du¬ C1 syllable = da- C2 syllable = da- C3 syllable = la- etc.
[0245] c. The syllables are classified into 5 syllable categories (syllable forms), V = symbol for vowel, C = symbol for consonant:
[0246] • Reduced syllable (V)
[0247] • Non-structured syllable (C; CO; CCC; CCCC)
[0248] • Open syllable (CV; CCV; CCCV)
[0249] • Closed syllable (VC; VCC; VCCC)
[0250] • Full syllable (CVC; CVCC; CVCCC; CCVC; CCVCC; CCVCCC;
[0251] CCCVC; CCCVCC) Compare with examples:
[0252] • V = a- • C = s- • CC = hm- « CCC = pszt- • CCCC = sscc- • CV = ka- • CCV = sta- « CCCV = stra- • VC = asz- • VCC = ing- • VCCC = inst- • CVC = tal- • CVCC = rend- • CVCCC = monst- • CCVC = kris- « CCVCC = sport- • CCVCCC = szkunksz-
[0253]
[0254] • CCCVC = skrib- • CCCVCC = strand- Compare with the notation of syllable fonts, illustrated with examples.
[0255] Here is the notation of syllable fonts with detailed phonological structure and anglebracket segmentation, illustrating how each syllable is encoded using consonantvowel structure and symmetry-based positioning:
[0256] 1. Reduced syllable (V):
[0257] Example: a- Font:
[0258] 2. Non-structured syllables:
[0259] C = s- Font: <s>
[0260] CC = hm- Font: <hm>or<h<m>
[0261] CCC = pszt- Font: <pszt> or <p <sz> t>
[0262] CCCC = sscc- Font: <sscc> or <s <sc> c>
[0263] 3. Open syllables:
[0264] CV = / ca- Font: <ka>or<k
[0265] CCV = sta- Font: <sta> or <s <t
[0266] CCCV = stra- Font: <stra> or <s <t <r
[0267] 4. Closed syllables:
[0268] VC = asz- Font: <asz> or sz>
[0269] VCC = ing- Font: <ing> or n> g> VCCC = inst- Font: <inst> or n> s> t>
[0270] 5. Full syllables:
[0271] CVC = tal- Font: <tal> or <t l>
[0272] CVCC = rend- Font: <rend> or <r <e> n> d>
[0273] CVCCC = monst- Font: <monst> or <m <o> n> s> t>
[0274] CCVC = kris- Font: <kris> or <k <r s>
[0275] CCVCC = sport- Font: <sport> or <s <p <o> r> t>
[0276] CCVCCC = szkunksz- Font: <szkunksz> or <sz <k n> k> sz>
[0277] CCCVC = sknb- Font: <skrib> or <s <k <r b>
[0278] CCCVCC = Strand- Font: <strand> or <s <t <r n> d>
[0279] This notation reflects the centre-periphery model described earlier, where vowels (V) are central and consonants (C) surround them in symmetrical layers. It supports linguistic modelling, phonological analysis, and database structuring for applications such as speech-to-text systems.
[0280] " The coding system of the database is developed using a symmetry-based classification method," cf. Figures 2 and 3 in the centre-periphery model: C< C< C< V> C> C.
[0281] " During classification and encoding, the position of phonemes within syllables is taken into account," e.g.:
[0282] a. The position of the consonant located at the periphery of the syllable:
[0283] CVC-1 -the first consonant of the full syllable (appears before the vowel) CVC-2 - the second consonant of the full syllable (appears after the vowel) CVCC-1 - the first consonant of the full syllable (appears before the vowel) CVCC-2-the second consonant of the full syllable (appears after the vowel) CVCC-3 - the third consonant of the full syllable (appears after the vowel) etc.
[0284] b. Encoded notation of syllable forms in the case of vowels, e.g.:
[0285] CVCC: D, CVC: D, CV: D, V: D, VC: D - syllables in monosyllabic words (type D) CVC: A1, CV: A1, V: A1, VC: A1 - initial syllables in two-syllable words (type A1 ) CVCC: C1, CVC: C1, CV: C1, V: C1, VC: C1 – final syllables in two-syllable words (type C1)
[0286] CVC: A2, CV: A2, V: A2, VC: A2 - initial syllables in three-syllable words (type A2) CVCC: B(1 ), CVC: B(1 ), CV: B(1 ), V: B(1 ), VC: B(1 ) - medial syllables in three-syllable words (type B(1 ))
[0287] CVC: C2, CV: C2, VC: C2 - final syllables in two-syllable words, (type C2) CV: A3, V: A3 - initial syllables in four-syllable words (type A3)
[0288] CVC: B(2-1), CV: B(2-1), V: B(2-1) - first medial syllable in four-syllable words (type B(2-1))
[0289] CVC: B(2-2), CV: B(2-2), VC: B(2-2) - second medial syllable in four-syllable words (type B(2-2))
[0290] CVC: C3, CV: C3- final syllables in four-syllable words (type C3)
[0291] etc. PCT / HU2025 / 0 0 0 0° 2 This format — syllable structure: syllable type — provides a compact and highly systematic way to classify syllables by both internal composition and positional role within the word.
[0292] E.g.;
[0293] » CVCC: D - a full syllable in a monosyllabic word form
[0294] » CVC: A1 - a full syllable as the initial syllable in a two-syllable word form » CVC: A2 - a full syllable as the initial syllable in a three-syllable word form » CVC: C1 - a full syllable as the final syllable in a two-syllable word form
[0295] » CVC: C3 - a full syllable as the final syllable in a four-syllable word form
[0296] » CV: C1 - an open syllable as the final syllable in a two-syllable word form » VC: A2 - a closed syllable as the initial syllable in a three-syllable word form » V: D - a reduced syllable in a monosyllabic word form
[0297] » CVCC: B(1 ) - a full syllable as the medial syllable in a three-syllable word form » CVC: B(2-1 ) - a full syllable as the first medial syllable in a four-syllable word form
[0298] » CVC: B(2-2) - a full syllable as the second medial syllable in a four-syllable word form
[0299] etc.
[0300] This system enables a precise and scalable classification of syllables based on both phonological structure and positional function within the word.
[0301] c. Encoded notation of syllable forms in the case of consonants, e.g.:
[0302] CVCC-1: D, CVCC-2: D, CVCC-3: D -first, second, and third consonants in a full syllable of a monosyllabic word
[0303] CVC-1: D, CVC-2: D - first and second consonants in a full monosyllabic word CV-1: D, VC-1: D - consonants in open or closed monosyllabic syllables
[0304] CVC-1: A1, CVC-2: A1, CV-1: A1, VC-1: A1 - consonants in initial syllables of two- syllable words
[0305] CVCC-1: C1, CVCC-2: C1, CVCC-3: C1, CVC-1: C1, CVC-2: C1, CV-1: C1, VC-1: C1 - consonants in final syllables of two-syllable words
[0306] CVC-1: A2, CVC-2: A2, CV-1: A2, VC-1: A2 - consonants in initial syllables of three- syllable words
[0307] CVCC-1: B(1 ), CVCC-2: B(1), CVCC-3: B(1) -first, second, and third consonants in medial syllables of three-syllable words
[0308] CVC-1: B( 1 ), CVC-2: B(1 ), CV-1: B(1 ), VC-1: B(1 ) - same, for shorter syllable forms CVC-1: C2, CVC-2: C2, CV-1: C2, VC-1: C2 - consonants in final syllables of three- syllable words
[0309] CV-1: A3 - consonant in the initial syllable of a four-syllable word
[0310] CVC-1: B(2-1 ), CVC-2: B(2-1 ), CV-1: B(2-1 ), VC-1: B(2-1 ) - consonants in the first medial syllable of a four-syllable word
[0311] CVC-1: B(2-2), CVC-2: B(2-2), CV-1:B(2-2), VC-1: B(2-2) - consonants in the second medial syllable of a five-syllable word
[0312] CVC-1: C3, CVC-2: C3, CV-1: C3 - consonants in the final syllable of a four-syllable word
[0313] etc.
[0314] This notation makes it possible to pinpoint the exact consonant’s position within a syllable and its role in the larger word structure, which is essential for precise phonological encoding, analysis, and synthesis in computational linguistics and speech technologies. The circuit arrangement makes it possible to generate word forms composed of syllables, cf. Figure 5:
[0315] CVC + CV + CV
[0316]
[0317] = <bal + la + da>; A2 + B(1) + C2 = <ballada>;
[0318] B1 word = [ballada].
[0319] This demonstrates how phonological structure (syllable forms) and positional types (syllable types) combine systematically to construct the surface form of the word.
[0320] Word forms are based on the symmetrical structural schema of syllables——for example, the schema of a monosyllabic word (D) is symmetrical, represented in concentric circles and a square grid divided along the vertical axis (cf. Figure 2).
[0321] In word forms, the encoded syllable structure is applied (cf. Figure 1 ), which illustrates the semiotic structure of words, where a word is represented as a symbolic structure consisting of a semantic pole and a phonological pole. PCT / HU2025 / 0 0 0 002 Presentation of the Advantages of the invention Compared to the Current State of the Art:
[0322] The advantages of the invention compared to the current state of the art lie in its speed, accuracy, scalability, and simplified maintenance. These benefits result in significant improvements in speech recognition technologies, particularly in real-time applications that handle large volumes of data:
[0323] 1. in faster and more accurate data processing, the index-structured database enables rapid searching of word and syllable records, significantly reducing processing time. The symmetry-based classification method ensures efficient data management, especially when working with large-scale databases.
[0324] 2. In higher-accuracy speech recognition, the classification of syllables and words based on phonemes leads to more precise text output, as it takes into account both the position of phonemes and the structure of syllables. The separation of speech signal sources reduces the impact of noise, further enhancing the quality of speech recognition.
[0325] 3. In terms of scalability and parallel processing, the structure of the database allows for parallel search operations, enabling the efficient processing of large volumes of data. This is particularly advantageous for applications that require real-time speech recognition.
[0326] 4. In terms of ease of expansion and maintenance, new words and syllables are encoded using symmetry-based classification, which simplifies the process of updating and maintaining the database without compromising system performance.
[0327] 5. In terms of better integration and optimization, the optimized integration between the database and the syllable repository ensures the fast generation and processing of word forms, enhancing the efficiency of speech-to-text applications. The system can be flexibly implemented on servers or in cloudbased environments, improving its adaptability to various technological infrastructures.
[0328] 6. In supporting user-friendly solutions, the use of acoustic input devices — such as microphones — that enable simple speech input recording makes the system more intuitive for end users. The method’s compatibility with existing speech-to- text technologies allows for wide-ranging applications.
[0329] 7. Competitive advantage in the market: The invention’s innovative methods for database management and speech processing provide a significant competitive edge in the industry, especially in areas where processing speed and accuracy are critical — such as real-time interpretation, customer service chatbots, or educational applications.
[0330]
[0331] SZTNH-100439783
Claims
RECORD COPY 24PCT / HU2025 / 000002 Patent ClaimsMain Claim:
1. The computer-implemented method for the computational generation of word forms, comprising the construction and use of a database of words created from syllables for application in speech-to-text systems, the method including the following steps:
1. creating a database that contains syllables and words in an index structure 2. assigning distinct codes to each syllable and word based on their syllable and word types, as well as their categories3. designing the database's coding system using a symmetry-based classification method4. mapping spoken input into text output using an acoustic input device, such as a microphone5. generating an encoded-word form based on the symmetrical structural schema of syllables6. implementing the method using a syllable repository stored on a server or in cloud-based systems,characterized in that the symmetrical classification takes into account the position of phonemes within the syllables, and the structured queries enabled by the architecture of the database.Dependent Claims:
2. The method according to claim 1,characterized in that the encoding of syllable types and categories optimizes the database.
3. The method according to claims 1 or 2,characterized in that it implements the separation of speech signal sources in the text output.
4. The method according to any of claims 1 to 3,characterized in that the structure of the encoded syllables used in the word forms is based on the symmetrical schema of the syllables.
5. The method according to any of claims 1 to 4,characterized in that the symmetry-based classification method enables the encoding of new words and syllables during the maintenance and expansion of the database.
6. The method according to any of claims 1 to 5,characterized in that the database structure is designed to allow the processing of large volumes of data and parallel search operations.Claims Related to Hardware and Software Requirements:
7. System for implementing the method according to any of claims 1 to 6, comprising:
1. a database management system for providing data access and executing structured queries;2. an acoustic input device, such as a microphone, for capturing spoken input; 3. a text display output device, such as a screen, for displaying the generated text; 4. a syllable repository stored on a server or cloud-based system;5. a computer processor configured to execute the method.
8. The system according to claim 7,characterized in that the integration between the database management system and the syllable repository is optimised.