A Structural Dynamics Based Visual Microphone

Visual Microphone enables passive acoustic recovery from subtle object vibrations captured in video sequences. Conventional approaches rely on multiscale phase decompositions with high computational cost. This work proposes a structural dynamics Visual Microphone framework operating directly on vectorized temporal pixel signals, avoiding phase based transformations. The method combines Principal Component Analysis and Complexity Pursuit within a temporally chunked architecture that estimates projection matrices once and reuses them throughout the sequence. Experiments on benchmark datasets achieved an 83% reduction in execution time and reduced pitch RMSE by 31% relative to traditional phase based methods, demonstrating efficient acoustic recovery from video.