Noise-canceling headphones have come a long way with their transparency modes, allowing users to hear real-world noise even when listening to something on their headphones. However, they don’t let you control what you hear. Now, researchers have created a set of AI noise-canceling headphones that will let you select what noises you want to hear so you can filter out unnecessary sounds but still keep an ear out for important things.
The new headphones use a system that the researchers call “semantic hearing,” it allows the headphones to stream captured audio to a connected smartphone, letting it cancel out any environmental sounds that the user doesn’t want to let through.
The system works through both voice commands and the smartphone app, and allows the wearer to select between 20 different classes of sounds that they want to let through the filters. These classes include sounds such as sirens, baby cries, speech, vacuum cleaners, and even bird chirps. A video detailing the system is embedded below.
The team presented its findings for the AI noise-canceling headphones during UIS ‘23 in San Francisco. The researchers say that they hope to release a commercialized version of the headphones down the line, allowing everyday users to take advantage of it.
Tech. Entertainment. Science. Your inbox.
Sign up for the most interesting tech & entertainment news out there.
By signing up, I agree to the Terms of Use and have reviewed the Privacy Notice.
Being able to let certain sounds through noise-canceling filters is important, especially if you wear your headphones out in public. Through the use of AI, these headphones let you control what you hear so that you can still cut yourself off from the world, but you’ll also be able to hear sounds that might indicate danger approaching or other important noises, like babies crying.
The team tested the headphones in multiple types of environments and found that the semantic hearing system works well, and that 22 participants even rated the target sound as being high quality compared to the original recording. In some cases, the system did struggle to distinguish between vocal music and human speech, but the researchers hope that more real-world data will help iron out those kinks and improve the outcomes.
