FYI.

This story is over 5 years old.

Tech

Why Robots Need Ears

Researchers are developing a microphone device to help machines distinguish sound as well as humans.
The Tarzan robot with the microphone. Image: Glyn Dewis

Machine vision—the tech used to give our computers and robots sight—has come a long way, with some computers even able to read emotions from facial expressions. But while our future robot workers and companions might have pretty good visual skills, we'd probably become better friends with them if they could actually hear us too.

Enter researchers from Imperial College London's department of electronic engineering, who are working on a microphone that could help equip robots with a functional set of super-human "ears."

Advertisement

"The idea is to get machines to be as good at using sound as human beings are," Mike Brookes, a reader in signal processing at Imperial College London, told me. He said that human-robot interactions would only seem more natural when they could actually hear us properly.

The new microphone is a spherical system that contains 32 microphones and a camera that can zoom in on conversations in noisy rooms and hear sounds directed at it from all angles. It detects the minute differences in how long different sounds take to reach the microphones around the device and compares all this information using an algorithm. This helps it work out which sounds are coming from where, and which ones the robot should focus on.

Tarzan at the Royal Society's Summer Exhibition. Image: Emiko Jozuka

"Being able to listen selectively, focusing on one person, is vital for human communication," said Patrick A Naylor, also at Imperial, in a press statement. "Until artificial intelligence can listen to different parts of the soundscape going on around it and pick out important conversations, AI will never properly be able to interact and converse with humans in noisy real-world situations."

Humans can process certain sounds and voices in a noisy room as our brains pick out the important sounds from background noise. Yet according to the the researchers, robots and other devices using speech recognition only recognise that what they hear is your voice, whether it's quiet, or you're talking really closely into the microphone.

Advertisement

Sounds constantly reverberate around space when they bounce off different surfaces. And this is one of the biggest challenges facing researchers designing speech recognition interfaces.

To demonstrate just how hard it can be for a robot to pick out the sound of your voice, the researchers from Imperial brought along a Nao robot dubbed Tarzan to the Royal Society's Summer Exhibition. They placed it in a soundproof box where different environments from everything to open space, a bathroom, a theatre, and a stadium could be simulated. Visitors can go inside the box and ask Tarzan a set of preprogrammed questions in each of the different environments.

Researcher Alistair Moore with Tarzan at the Royal Society. Image: Emiko Jozuka

When I asked Tarzan how old he was in a simulated theatre soundscape, we had no problems interacting owing to the lack of reverberations. He told me he was two and said hello. But once I asked it the same question in a simulated stadium setting, the poor bot got confused by my echoey voice and couldn't really respond to my question.

"The energy from our voice is localised in time so by analysing the sound each time we can get a sequence of features that tell us what this speech is," Alistair H Moore, a research associate at Imperial's Communications and Signal Processing Group, told me as we stood inside the soundproof box. "However, when we add reverberation, the energy spreads out in time, which means that the features become confused."

In the future, the researchers aim to apply the technology they're developing to robotic assistants in hospitals and busy environments, so that they can listen and take instructions in the same way that a human would.

"If we can improve the way that machines use sound, and get them to approach how well humans hear, it will allow robots to interact with people in a much more natural way and effortless way," said Brookes. "You could talk to a machine and it will know that you're talking to it. It doesn't actually have to be able to see you directly."

Worried about killer robots? Maybe you should be. Check out Motherboard's documentary Inhuman Kind.