A team of Amazon researchers proposes a machine learning model that's able to localize multiple sound sources from microphone data alone.