Foul Language Censored Social Media Video Generation Using Audio Censoring Model
Brindha Subburaj, Uma Maheswari Jayachandran, Siya Bansal, Vedansh Kumar · Advances in social networking and online communities book series · 2025
Advancements in communication medium and the proliferation of social media led to increasing amount of content shared often without moderation. The offensive are raising significantly, posing threat to mental health, and degrades the user experience. Current study focuses on identifying and censoring audio clips with offensive and foul words. We propose a novel audio censoring method using OpenAI's whisper model. The audio files are extracted from video and forwarded to whisper model. The text transcripts of the audio file is obtained through the whisper model. Keywords from the text are compared with a dictionary of offensive and foul words to detect such terms in the input audio. The matched foul words are identified and respective audio are censored and beep tone is inserted. Finally, the censored audio is integrated and video is generated. The proposed system is tested using dataset retrieved from youtube and the has shown significant performance and achieves accuracy of 93.81%, shows the robustness of proposed foul word identification and censoring system.