20.500.12556/RUL-172581 Sistem za interakcijo z virtualnimi liki na podlagi velikih jezikovnih modelov Solution for Interaction with Virtual Characters Based on Large Language Models Cilj diplomskega dela je bil razviti sistem za pogovor z virtualnimi liki na osnovi razpoznave govora, generacije odgovorov in sinteze govora. Pri izvedbi je bilo treba pregledati obstoječa orodja umetne inteligence (AI/Artifical Inteligence), orodja za generacijo besedila in govora ter jih preizkusiti in povezati v celoto, ki nam omogoča uporabo sistema, tj. govorno interakcijo z virtualnim likom. Izbor orodij je temeljil na kriterijih kakovosti, izvedljivosti, izhodnega rezultata, hitrosti odziva in združljivosti z načrtovano arhitekturo sistema. Ključna je bila storitev velikih jezikovnih modelov (LLM/Large Language Model), saj predstavlja osrednji del sistema in definira odzive na vhodna vprašanja. Sledila je implementacija za pretvorbo besedilnih odgovorov v govor, integracijo prepoznave govora in njegovo pretvorbo v besedilo. Rezultat diplomskega dela je delujoč sistem, ki združuje napredne storitve velikih jezikovnih modelov in omogoča govorno interakcijo z virtualnim likom. Kot dodatek smo razvili tudi predvajalnik, ki je sicer primarno namenjen predvajanju zvoka, lahko pa vanj vstavimo poljuben videoposnetek, da se lažje vključimo v interakcijo z likom. Sistem tako predstavlja osnovo za nadaljnji razvoj, ki bo speljan v smeri naprednega video prikazovanja. The goal of the thesis was to develop a system for conversation with virtual characters based on speech recognition, response generation, and speech synthesis. During implementation, it was necessary to review existing artificial intelligence (AI) tools, including those for text and speech generation, test them, and integrate them into a system that enables us to use the system, i.e., speech interaction with a virtual character. The selection of tools was based on the criteria of quality, feasibility, output result, response speed, and compatibility with the planned system architecture. The key was the service of large language models (LLMs/Large Language Models), as they represent the central part of the system and define responses to input questions. This was followed by the implementation of converting text responses into speech, integrating speech recognition, and its subsequent conversion into text. The result of the thesis is a working system that combines advanced services of large language models and enables speech interaction with a virtual character. As an addition, we have also developed a player that is primarily intended for playing audio, but can also be used to insert any video clip to facilitate interaction with the character. The system thus represents the basis for further development, which will be directed towards advanced video display. uporabniški vmesnik veliki jezikovni modeli analiza govora sinteza govora virtualni lik user interface large language models speech analysis speech synthesis virtual character true false false Slovenski jezik Angleški jezik Diplomsko delo/naloga 2025-09-09 11:20:01 2025-09-09 11:20:10 2025-10-16 03:46:44 0000-00-00 00:00:00 2025 0 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 63138 250354691 Sorsek_Matej_-_Sistem_za_interakcijo_z_virtualnimi_liki_na_podlagi_velikih_jezikovnih_modelov.pdf Sorsek_Matej_-_Sistem_za_interakcijo_z_virtualnimi_liki_na_podlagi_velikih_jezikovnih_modelov.pdf 1 F4F23C31C736242D049393E8F6D0A073 fb046d9de55e2b3a2480ce265aa7b30b18c7e001b1e345288ad436bf570771dd 08093092-8d5e-11f0-9328-0050569b8976 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=216483 Fakulteta za elektrotehniko 0 0 0