Technology for Automated Transformation of Lecture Materials into Educational Textbooks Using Neural Speech Recognition and Large Language Models

Abstract

The paper addresses the automated transformation of lecture materials into structured educational documents using neural speech recognition and large language models. An architecture of a human–machine intelligent system is proposed, based on the functional distribution of tasks among the instructor, an intelligent processing pipeline, and contextual knowledge sources. A formal problem statement is developed; the input and target information objects, transformation requirements, and a sequence of specialized processing stages are defined, including automatic transcription, contextual correction, written-speech normalization, and semantic structuring of the material. The architecture is implemented as a local software prototype and verified through a two-stage study comprising experimental evaluation on university lectures and independent functional testing after deployment in a different computing environment. The results confirmed the operability of the proposed architectural approach, the feasibility of generating structured educational documents, a substantial reduction in the labor required to prepare teaching materials, and the reproducibility and portability of the software implementation. The limitations of the research prototype and the main directions for further development of the proposed architecture are identified