A Web-Based Video Player with an In-Built ASRBased Subtitle Generator for English Videos

Audio transcription, which bridges the gap between text-based accessibility and spoken communication, is becoming increasingly important in today's digital world. It enhances accessibility for those with hearing impairments, ensures accurate spoken information documentation, and makes audio content easier to find and analyse. For accurate speech-to-text transcription on low-end devices, we present in this work a revised Wav2Vec2-based Automatic Speech Recognition (ASR) system trained with the Hugging Face Transformer. The system preprocesses audio by tokenising, normalising, and resampling it into feature vectors. A transformer encoder is then used to process these feature vectors. A transformer encoder is then used to process these feature vectors. Frame-level annotations are not required because Connectionist Temporal Classification (CTC) loss is used to associate text sequences with audio attributes. Word Error Rate (WER) evaluation showed that the model performed well, which makes it appropriate for webbased automatic subtitle generation. The system incorporates a web-based video player to improve accessibility and user experience. The subtitles are automatically synchronised and shown when the correct video plays after they have been created and saved. This improves content accessibility for a variety of audiences by guaranteeing that viewers can view accurate subtitles without any delays.