Making machines that can listen
AI Summary: Smaragdis pursued music technology at Berklee and MIT Media Lab, where he shifted from signal‑processing to machine‑learning approaches for teaching computers to understand sound. At Mitsubishi Electric Research Labs he applied audio‑based AI to detect events in video streams, enabling highlight extraction and integration into consumer devices and traffic monitoring systems. At Adobe he founded an audio research team that built accessible sound‑manipulation tools for creators while identifying deeper technical challenges. As a professor at UIUC, his group now incorporates neural‑network models to separate individual sources from mixed audio recordings.