Source: by Stuart McDill for Thomson Reuters, June 18
A skype call in a noisy cafe is a challenge for humans let alone digital devices. Now artificial intelligence can help.
Cambridge start-up AudioTelligence say their machine learning can get rid of the background noise and allow your digital assistant to hear you in even the noisiest environments.
“Effectively, our technology is analogous to a zoom lens in the camera world, so it allows you to zoom in on the sound that you’re interested in to the exclusion of all of the background noise that was really getting in the way of you understanding what was being said,” said Ken Roberts, Chief Executive of AudioTelligence.
The company says a typical smart speaker only hears 22 words out of hundred when it’s noisy – rising to 94 with their signal separation technology.
“We’re moving to a time when our homes will be automated and the way we interact with our home will be through speech technology. You’ll want to be able to walk into your home, your lounge for example and be able to turn the lights on and turn the heating up and put the TV on without having to answer the same question five times,” explained Roberts.
Typical smart speakers use several carefully calibrated microphones inside a speaker to cancel out background noise – also cancelling out certain frequencies. Now existing smart speakers could be upgraded without needing any calibrating or training.
It’s really challenging for Alexa to understand what was being said when you’re in the kitchen and the kettle’s boiling and the washing machine’s going and of course you’ve got the kids screaming, but AudioTelligence’s solution can remove all of that background noise so that Alexa can understand you. However, it can’t entirely remove the children, so whether you’ll be able to actually hear the answer or not is a different matter.