AI Based Regulatory Breach Detection in TV & Radio Content for a media and communications regulator in European region

Challenges.

• Analyzing content across 40+ TV and 5+ radio channels 24/7 requires a large team of analysts, resulting in significant personnel expensesTraditional clustering using KYC details and behavior patterns 
• Manual analysis is prone to fatigue, subjectivity, and inconsistencies in interpretation, potentially leading to missed violations or false positives. 
• Adding new channels or expanding monitoring hours is difficult with a manual approach due to resource constraints. 

Solution.

• Implemented an end to end AI solution on top of Data Lake to process large volumes of unstructured content (Video, Images and Audio), and make detections from content in a highly configuration driven manner. 
• Languages Covered (both vocal as well as written – for sub-titles) – English, Bangla, Urdu, Modern Arabic and Punjabi. 
• Solution has been designed in a manner that it can process multiple hours of Videos and audio in few minutes. 
• Technologies used: Python, NiFi, OpenCV, Spleeter, Tesseract, Pydub, PyTorch, Cognitive APIs (Google and Azure), BERT, Azure Data Lake Storage Gen 2, Azure SQL Database, R, Shiny  

The Impact.

Significantly reduced manual effort (by more than 60%) in reviewing Video and Audio content for locating/flagging harmful content, and with an accuracy of more than 85%.