Real-Time Subtitle Generation for Live Videos Using AI and Machine Learning

Real-time subtitling of live video streams is still a major challenge because of the difficulties involved in processing spontaneous speech, dealing with multiple speakers, and coping with varying audio conditions with low latency. This research compares the performance of an end-to-end automatic subtitling system that integrates state-of-the-art automatic speech recognition (ASR) with sophisticated natural language processing. Processed 500 hours of live broadcast content in news, sports, and entertainment segments with a hybrid architecture that combined bidirectional LSTM networks and transformer models. Results indicate a 94.2%-word accuracy rate with an average latency of 1.2 seconds, which is a 27 % improvement over current systems. The system was especially strong in dealing with domain-specific vocabulary and subtitle synchronization during rapid speaker changes. Our results show that deep learning-based methods can attain near-human quality in real-time subtitle generation and can fulfill broadcast industry latency demands, rendering automatic subtitling ever more feasible for live shows.