Restoring Reality with Spatial Audio and Generative Models