Ganes Kesari的动态

Is the angry mother-in-law improving your TV show's viewership ??? How sad should your protagonist be this week ??? This was the question our client, a broadcaster was trying to answer. We tried 2 approaches to answer this using #data and AI. 1. Use #computervision algorithms to identify character emotions by "looking" at them. 2. Use #textanalytics to analyze dialogues from the scripts used for dubbing, by "reading" them. Then, bring in the viewership to see what emotion works. Both approaches were useful but gave just moderate results. While #DeepLearning vision techniques have covered a lot of ground, they get it wrong when characters are poker-faced or if faces aren't clearly visible. We wanted to improve the results. Our team wondered whether the background score of scenes can give away the emotions. Particularly in soaps and movies. ?? A face could display many emotions, but the audio BGM tells viewers whether they should be enraged or sympathetic. We can use DL to "hear" and decide. The Gramener team experimented on some Youtube TV shows. The results were impressive! In this blog, Murali Manohar Kondragunta walks through the approach using interesting examples and a quick demo. Check out the link in the comments and let us know what you think. #data #analytics