About a decade ago, researchers developed a way to recover sound from video footage. Lightweight objects in the video move in response to acoustic waves bouncing off of them, and image processing can identify sub-pixel vibration of said objects and reconstruct those acoustic waves.
The downside is that you need high video frame rates to capture the acoustic frequencies that matter for human beings. Ordinary HD video is about 30 frames per second, so you can only recover frequencies less than 15 Hz; if you want to recover intelligible speech sounds, you need a camera that can record at several thousand frames per second.