A modification to the process described at http://www.phys.huji.ac.il/~springer/DigitalNeedle/ could do it, but you'd get like a fraction of a second of decoded audio from the images on that page, so not that interesting.