Extracting Sound from Silent Video
Halls department, Hall 4
Thursday, 29 December 2016
14:00 - 15:00
When sound waves propagate in air, they cause small vibrations when they hit objects. It has been shown that sound waves can be extracted by measuring these small vibrations in a video. In a recent work, we show that certain pixels in a video can record sound similar to a microphone. By studying local vibrations in detail, we isolate locations with higher sound quality and significantly improve sound quality. We also estimate incoming sound direction for the first time by measuring the delay between local sound readings. Finally, we reach real-time performance in 20KHz video by limiting computation to only a few localities.
Mohammad Amin Sadeghi completed his PhD at the University of Illinois at Urbana-Champaign. His research in Computer Vision and Machine Learning has been cited over 600 times. He has received the best student paper award at CVPR, which is the highest ranking conference in computer vision. He has recently started as a faculty member at the university of Tehran.