Real-Time Subtitling in the Streaming Era

The aim of the study was to identify the characteristics of real-time automatic subtitling systems in the context of global streaming platforms. The methodology was based on a comprehensive approach that combined an overview of automatic speech recognition (ASR) technologies, neural machine translation and large language models, and an analysis of practical use cases (YouTube Live Captions, Zoom Auto-Caption, Netflix). As a result, it was found that the integration of ASR, neural machine translation, and large language models ensured the speed of speech stream processing, system scalability, and the possibility of multilingual coverage. At the same time, significant limitations were identified: reduced accuracy when working in noisy environments or with agglutinative languages, inaccurate reproduction of culturally marked expressions, and problems with synchronizing text with video. Case studies showed that YouTube experienced quality instability and the risk of incorrect subtitles, Zoom focused on organizational accessibility management, and Netflix prioritized regulatory and editorial control in accordance with its style guide. A comparative analysis of standards showed that automated systems did not ensure consistent compliance with the requirements of accuracy, readability, and cultural adaptation, even with a low error rate. The practical significance of the study lies in the possibility of using its results to improve automated subtitling systems and develop hybrid models capable of combining algorithmic solutions with editorial control to ensure compliance with international quality standards.