Solving the cocktail party problem

29 October 2018

Interview with

Ken Roberts, Audiotelligence


Anyone with hearing aids will be able to tell you that having a conversation in a busy restaurant is just about impossible. As hearing loss affects almost everyone at some point in their life, one Cambridge based company has set out to solve what's termed the cocktail party problem. To find out more, Georgia Mills went to visit the chief executive of Audiotelligence, Ken Roberts to test out his technology in a slightly noisy pub...

Ken - If you talk to people who wear hearing aids you’ll find very often that actually they’ll turn them off in a noisy pub, or even take them out because actually it’s easier to try and focus in on a single person speaking than working with hearing aids that are really high quality amplifiers; but they amplify all of the sounds that you don’t want to listen to as well as the sound that you do.

Georgia - But we have come here armed with some very snazzy looking, bleeping and flashing technology. What is your secret weapon?

Ken - Basically, what we have on the table here is a mini PC. On top of the mini PC is a small circuit board with 8 microphones. It’s roughly around the same size and shape as a mobile phone, which is intentional. Our software running on the mini PC actually uses the information from those 8 microphones to figure out where the individual sources are in the acoustic scene around us. And by doing that we can separate out the one that we’re interested in and get rid of all the background noise that we’re really not interested in. And that’s really the difference between what we do and what a hearing aid does. A hearing aid amplifies all of that acoustic scene, but by doing that it actually amplifies the background noise, whereas we separate the source that you’re interested in from the background noise.

Georgia - And how this works at the moment is with a little visualisation of the sound in the room on an iphone. It looks like a doughnut with flashing lights where people are speaking and if you want to focus on a particular person, you select the corresponding flashing light. And here’s a comparison of Ken speaking first without, and then with, the technology being turned on.

Ken - Incomprehensible muttering.

Georgia - I’m going to have to turn it back on because I can’t hear properly. Lets get it back on yes.

Ken - Yeah. I think that’s a good example of the difference that the technology delivers.

Georgia - Does it have any other applications?

Ken - Yes, absolutely. One of the biggest growth areas in the world today is home assistance and speech recognition. And one of the big challenges that automatic speech recognition software has is being fed a clean signal. About 30 percent of errors that speech recognition software has are due to noisy signals, so by inserting our technology ahead of your speech processor, we can feed a clean denoised signal. So again, just like the environment here, we could get rid of all the background noise and greatly improve the recognition rates for speech recognition technologies.


Add a comment