EMERGING TECHNOLOGIES
Software Enhances Audio Processing
iStock illustration
WASHINGTON, D.C. — A new AI-powered audio processing tool from South Carolina-based software company Wave Sciences could enhance military communications by replicating human hearing.
The GLIMPSE engine can pick out a single voice in an audio recording of a crowded room, solving a pressing dilemma with a lighthearted name: the “cocktail party problem.” It’s “where you have lots of voices, sources, something to reflect the sound around, and you end up with a babble,” said Keith McElveen, Wave Sciences president and founder.
The fact that humans can make sense of that babble, but not advanced technology, has been puzzling, McElveen said in an interview at the National Training and Simulation Association’s Capitol Hill Modeling and Simulation Expo.
In a recording, if someone’s voice is “at or below the background noise — deep learning, nothing else, is going to pull it out,” he said. “But you, as a human, if I go and stand right there … you could still, with just two ears, figure it out.”
In human hearing, “your brain automatically builds a refocusing filter — just like you’ve got depth perception in your eyes,” McElveen said. “That’s how you’re able to pick out one voice from another and know where that voice is coming from.”
To replicate this — rather than using more advanced audio capture technology — the GLIMPSE engine relies on a machine learning algorithm, which “models the room on the fly,” he said. “It’s like portrait mode on your camera on your smartphone.”
The solution allows users to hone in on a space or conversation in a room, making it sharp and blurring out everything else, he said. They can also toggle between different versions of the same conversations, he added. “Our filter says, ‘Are you interested in maybe refocusing on that voice, or this voice, or that voice?’”
The technology has already been used by the Justice Department to enhance audio in court proceedings and has the potential for further government and defense applications, including military communications and drone detection, McElveen said.
Comments (0)