Communications of the ACM
Mitsubishi Quiets Car Noise With Machine Learning
Mitsubishi Electric says its new noise suppression technology will aid hands-free phone calls in the car and elsewhere.
Credit: John Boyd
Mitsubishi Electric says its new noise suppression technology can improve the quality of hands-free phone communication in cars.
The technology is designed to filter out almost all unwanted ambient sound that enters a far-field microphone while speaking, including passing cars, windshield wipers, and turn signals.
"Previously, only stationary noises such as road noise or the sound of the air conditioner were really dealt with, because the noise mixed with the speech could be easily predicted from past observations when the driver was not talking," says Mitsubishi Electric Research Labs researcher Jonathan Le Roux.
Researchers are developing speech-enhancement systems to better distinguish human speech from other sounds. The systems employ machine-learning methods based on deep neural networks, and are trained to distinguish and suppress noise and retain clear speech using large amounts of noise-contaminated speech data.
In tests, the technology was able to cancel out 96 percent of the ambient noise compared to 78 percent achieved by conventional methods.
The researchers say the technology could work in other hands-free command-and-control situations such as call centers and when using Apple's Siri or Google's Voice Search in smartphones.
From IEEE Spectrum
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found