The Human Speechome Project is an effort to observe and computationally model the longitudinal language development of a single child at an unprecedented scale. To achieve this, we are recording, storing, visualizing, and analyzing communication and behavior patterns in over 200,000 hours of home video and speech recordings. The tools that are being developed for mining and learning from hundreds of terabytes of multimedia data offer the potential for breaking open new business opportunities for a broad range of industries�from security to Internet commerce.