Generating audiovisual content using AI for the Vyslovuj.cz platform

The thesis presents the design, pilot implementation, and validation of a technical solution for automated generation of speech-therapy educational videos using artificial intelligence, developed for the educational platform Vyslovuj.cz. The main objective is to determine whether audiovisual content for Czech-language speech-therapy exercises can be generated automatically while meeting linguistic and visual quality requirements, and at the same time remaining scalable and economically efficient, given that current manual production—based on recording actors and post-processing—is both time-consuming and costly. The work compares two technological approaches: the use of commercial AI platforms available via API, and the development of an in-house solution based on open-source models. Both approaches are evaluated in terms of technical feasibility, operational demands, output quality, and suitability for integration into the platform’s workflow. Pilot testing on a set of Czech words focused on speech intelligibility, naturalness of the synthesized voice, accuracy of visual articulation, and overall process stability. The results show that commercial tools are the most practical option for rapid concept validation and MVP development, while the open-source approach offers long-term potential for greater control and flexibility, though at the cost of substantially higher compute requirements and technical complexity. The thesis thus provides an analytical and methodological framework for deciding between external and internal technological approaches in the field of AI-generated educational content.