Kannada Subtitle Generation with Speaker Diarization: A Comprehensive Solution for Regional Language Visual Media Accessibility

Due to the rapid advancements in technology and the broad availability of high-speed internet, the landscape of digital media consumption is changing quickly, and this has increased the requirement for inclusive language representation in speech recognition and subtitle generation. In order to meet this demand, the paper presents a specialized approach for generating subtitles in the Kannada language, which is mostly spoken in the State of Karnataka. The methodology involves the extraction and pre-processing of audio from the input video, followed by segmentation and identification of the speakers using diarization techniques. This distinctive diarization process serves as a key component of our study. Following the speaker diarization phase, our system transitions to a speech recognition phase, which generates the initial. SRT (SubRip Subtitle) file. The model then refines and further processes this file adding accurate punctuation, translating it into Kannada, and finally creating a Kannada. SRT file th at is re ad y synchronized with the original video. In the last phase, the method integrates these Kannada subtitles into the original video content using VLC Media Player. The model achieves a good accuracy with audio pre-processing compared to without audio pre-processing. The performance is better than State-Of-The-Art (SOTA) techniques. In addition to accuracy, various other important metrics such as Word Error Rate (WER), Levenshtein Distance (Edit Distance), and BLEU score are evaluated. The use of diarization in Kannada subtitling is an effective and novel approach for segmenting and identifying two speakers, and generating subtitles for the video samples. In conclusion, the implementation of Kannada subtitling using diarization techniques serves as a crucial tool, benefiting the hearing-impaired and enabling native Kannada speakers to access information present in English videos.