“Our study found that acoustic environments contain enough visual cues to generate highly recognizable streetscape images that accurately depict different places.”
UT Austin trained a model on 10-second YouTube clips and generated street scenes from 100 recordings. Human judges matched the right image 80% of the time when picking from three. The pictures even caught whether it was sunny, cloudy or night. Every voice memo and phone call now leaks a rough picture of where you are standing.