Making machines that can listen

MIT News has profiled Paris Smaragdis, a professor who joined MIT in 2025 and is working to build machines that can hear and interpret sound better than people can. He holds a post in Music and Theatre Arts with a shared position between the Schwarzman College of Computing and the Department of Electrical Engineering and Computer Science. Possible uses cited in the October 28 article include locating a traffic accident from the noise of the crash and predicting when an industrial drill might fail while boring a geothermal well.

Raised in Greece, Smaragdis trained on classical guitar, then studied bass and computer-driven sound at Berklee before earning graduate degrees at the MIT Media Lab under the late Barry Vercoe. There he favored machine learning over the signal processing most peers used, arguing that computers must be taught to listen because humans cannot fully explain how hearing works. He later worked at Mitsubishi Electric Research Labs, where his audio analysis reached video recorders, traffic monitors and phones, and founded an audio research team at Adobe.

During 15 years at the University of Illinois, his group built neural networks that isolate individual sounds from mixtures. That work led to his cleaning up archival recordings for the 2021 series "The Beatles: Get Back," and his contributions helped the team win a 2022 Emmy for sound editing.

At MIT, his group is building tools that give musicians feedback, which he says should augment rather than replace them, along with signal processing methods that do not rely on fixed sampling rates or frame sizes. He says the ideal way to apply AI to sound and music has not yet been found.