OJS Tashkent State University of Economics
Not a member yet
554 research outputs found
Sort by
СИНТЕЗ УЗБЕКСКОЙ РЕЧИ НА БАЗЕ ГИБРИДНОЙ НЕЙРОСЕТЕВОЙ АРХИТЕКТУРЫ
No full textВ данной статье рассматриваются особенности разработки систем преобразования текста в речь, а также описываются ключевые этапы реализации TTS-системы. Основной целью работы является описание и определение составных модулей, на основе которых была реализована TTS-система узбекского языка, использующая гибридную нейросетевую модель Tacotron2+ParallelWaveGAN. В рамках исследования проведены эксперименты и осуществлена оценка модели синтеза речи, которая составила 4,3 балла из 5, тогда как естественная речь была оценена на 4,7 балла по системе оценивания MOS. Основные результаты показывают, что сочетание моделей Tacotron2+ParallelWaveGAN генерирует более естественные звуковые аудио, в отличие от классической модели Tacotron2+WaveNet. Данные выводы и разработанная TTS-система могут быть использованы в ассистивных системах для людей с ограниченными возможностями. В завершении статьи обсуждаются решения, которые могут положительно повлиять на разработанную систему с точки зрения улучшения естественности и просодики генерируемых аудиофайлов
INSON KINEMATIKASINING SOHALARDA QO’LLANILISHI
No full textInsonlar harakatlarini tahlil qilish tibbiyot, biomexanika, robototexnika kabi turli sohalarda ishlatiladi. Masalan, signalni qayta ishlash, mashinali o‘qitish va kompyuter ko‘rish usullarini qo‘llash orqali identifikatsiya qilish, tanib olish va tahlil qilish. Bunda insonning tana harakati xususiyatlari aniqlash asosida tahlil qilish amalga oshiriladi. Ushbu maqolada inson harakatining xususiyatlari tahlil qilingan. Tadqiqot shuni ko‘rsatdiki, inson tanasining yuqori segmenti yurish tahlili uchun yaxshiroq ma’lumot manbai hisoblanadi
MASHINALI O‘QITISH ALGORITMLARI YORDAMIDA HUJJATLARNI TASNIFLASH USULLARI TAHLILI
No full textMatnli hujjatning avtomatik tasnifi onlayn matnli ma’lumotlarning payda bo‘lganidan beri matn tahlili sohasidagi tadqiqotda ahamiyatli hisoblanadi. Raqamli kutubxonalar, elektron pochtalar, bloglar va boshqalar kabi manbalar raqamli davrda matnli hujjatlarning tez o‘sishini ta’minlaydi. Umuman olganda, matnli hujjatning toifalari ma’lumot olish, mashinani o‘qitish va tabiiy tilni qayta ishlash kabi bir qancha sohalarini o‘z ichiga oladi. Ushbu maqola matnli hujjatlar to‘plamini oldindan belgilangan toifa belgilariga tasniflash uchun ham nazorat ostida va nazoratsiz mashinali o‘qitish usullaridan foydalanadigan tadqiqotlar tahlil qilingan
TABIIY TIL QOLIPLARINI N-GRAM METODI VOSITASIDA ANIQLASH
No full textN-gram metodi – matn ma\u27lumotlaridagi qolip va munosabatlarni aniqlash uchun tabiiy tilni qayta ishlash (Natural Language Processing, NLP)da qo\u27llaniladigan matnni tahlil qilish usuli. Ushbu metod matnni n-gram deb ataladigan kichikroq birliklarga bo\u27lish va matn ma\u27lumotlari haqida tushunchaga ega bo\u27lish uchun ushbu n-grammlarning chastotasi va ularning korpusda tarqalishini tahlil qilishni o\u27z ichiga oladi. N-gramlar so\u27zlar, belgilar yoki boshqa har qanday mazmunli matn birliklaridan iborat bo\u27lishi mumkin. Til korpusidahi N-gramlar tahlili muhim ahamiyatga ega bo`lib, u matn ma\u27lumotlarini tahlil qilish va ma\u27lumotlar ichidagi qolip va munosabatlarni aniqlashning sodda, ammo samarali usulini taklif qiladi. N-gram metodi tilni modellashtirish, matnni tasniflash va his-tuyg`ularni tahlil qilish kabi turli xil NLP ilovalarini ishlab chiqish uchun foydali bo\u27lishi mumkin. N-gram tahlili tilni modellashtirishda matn ma\u27lumotlaridagi qolip va munosabatlarni aniqlash hamda tabiiy tilni qayta ishlash vazifalari uchun bashoratli modellarni yaratish uchun qo`llaniladi. Shungdek, N-gram tahlili matnni tasniflashda matnning asosiy xususiyatlarini aniqlash va matnni oldindan belgilangan toifalarga ajratish uchun ishlatiladi
Selection of features in the problems of personal identification by keystroke dynamics
No full textIn this article examines the problem of identifying features when authenticating the identity of a user of computer systems based on keyboard handwriting. To solve this problem, a feature extraction method is proposed. The main idea of this method is to search for a set of representative features. In this case, the search for representative features is carried out in two stages. At the first stage, time and frequency features are determined. At the second stage, the following are determined: 1) a subset of strongly related features 2) a set of representative features. Experimental studies have been conducted to assess the performance of the proposed method. The results of the experimental study showed that the proposed method of feature extraction showed high accuracy in solving the problem of personal authentication by keyboard handwriting
CREATION OF A MULTI-FUNCTIONAL DEVICE FOR REAL-TIME MONITORING IN EVALUATION OF THE BASIC CHARACTERISTICS OF WATER
No full text This article is about finding an effective solution to the problems of determining the temperature, acidity PH and salinity TDS of groundwater. The spectral-impedance method proposed in the article allows for the creation of devices for measuring all the main quality indicators of water . Increasing the accuracy of remote monitoring devices using artificial intelligence.The purpose of this paper is to improve the accuracy of well water quality assessment using the IoT measurement method. For this purpose, a device in the form of a "poke" floating on the water surface with many electrodes (sensors) is lowered into the well, and the main indicators of the water are monitored using wireless radio waves. It was found that the measurement results from the multi-electrode sensor correspond to the quadratic approximation. The proposed method, in comparison with the currently used methods, increases the reliability of the estimation of the composition of groundwater based on the simultaneous measurement of the spectral impedance of the ions contained in the hydrogen indicator.  
MODELING AND STORING DATA IN GRAPH DATABASES
No full textThis paper presents the study of graph databases and their application in the context of social networks. It describes a data model that represents users, communities, and the relationships between them as a graph, where nodes represent objects and edges represent their relationships. The data structure in graph databases is compared to relational databases, emphasizing the freedom and flexibility to create and modify relationships between nodes without strict restrictions. A method for storing graph data structures is discussed, including the possibility of storing them in SQL tables using JSON, and the use of specialized graph DBMSs such as Neo4j. The specifics of data storage in Neo4j are highlighted, including caching for improving read/write performance and optimizing graph traversal. The paper emphasizes the Cypher query language, which is specifically used in Neo4j to work with graph data. Example queries with explanations are provided, demonstrating the capabilities of the Cypher language for working with data in graph databases. The conclusion discusses the application areas of graph databases, including fraud detection and supply chain mapping, how graph databases provide flexible options for storing information, and highlights their wide range of applications in various fields
EXPLAINABILITY OF THE SVM CLASSIFICATION MODEL FOR SENTIMENT ANALYSIS TASK OF UZBEK LANGUAGE
No full textThis paper investigates the integration of local model-agnostic explanations with support vector machine models to enhance explainability in sentiment analysis for the Uzbek language. While SVM models are effective for classification tasks, they often function as black-box models with limited transparency. To address this, we used LIME, which perturbs input data and observes changes in the model\u27s output, revealing the text features that most influence classification. This approach improves transparency and trust in AI systems. Our case study focuses on sentiment analysis in the low-resource Uzbek language, showing how LIME aids in understanding SVM model decisions
OPTIMAL SOLUTIONS FOR DETERMINING THE DISTANCE TO AN OBJECT IN AN AUTONOMOUS MOBILE DEVICE FOR PEOPLE WITH DISABILITIES
No full textThis article describes an analysis of methods for measuring the distance to obstacles and objects for a device that helps people with disabilities move independently inside the house and find the necessary objects, as well as the results of the process of selecting the most optimal method for this device. The article also presents the results of an analysis of the capabilities of the methods used for this purpose
ALGORITHMS FOR RECOGNITION OF HISTOLOGICAL IMAGES BASED ON THRESHOLD RULES
No full textThe article discusses various histological image recognition algorithms to improve the accuracy of pathology diagnostics. Particular attention is paid to the proposed threshold rule-based method, which demonstrates higher classification accuracy and training speed compared to popular machine learning methods such as SVM, Random Forest, and XGBoost. During the experiments, the proposed method showed high accuracy with minimal training and classification time. The results of a comparative analysis by key metrics are presented, confirming the effectiveness of the proposed approach for automating diagnostics in digital pathology