This document discusses using optical flow analysis of video footage to extract sound information from minute vibrations of objects captured on video. It summarizes previous research using laser vibrometry and phase-based video processing. It then describes using a steerable pyramid to link phase variations in video frames to local motions, and how local motion signals can be aligned to reconstruct sound. Experiments showed sound could be recovered from everyday objects filmed at 240 FPS, with SNR from -0.49 to 1.96 dB and similarity coefficients from 0.21 to 0.64. Higher frame rates and larger object motions improved reconstruction quality.