AudioTelligence has been in the news again. This time it was featured on the BBC's The Naked Scientists podcast.
Georgia Mills met with the chief executive of AudioTelligence, Ken Roberts, in The Punter, a popular pub near the centre of Cambridge. Its noisy environment made the perfect setting for the interview, which was recorded with AudioTelligence’s source separation. At one point Georgia turned off the source separation and asked Ken to carry on speaking. After a few seconds of Ken's "incomprehensible babble" Georgia then said:
I’m going to have to turn it back on because I can’t hear properly.
This is an excellent example of what AudioTelligence is all about. Without the source separation, Georgia was unable to hear what was being said. When it was switched back on, the conversation was perfectly clear.
The interview focused mostly on using AudioTelligence's source separation for hearing assistance. One of the earliest signs of age related deafness is the inability to understand speech in noisy environments. This effect is traditionally known as ‘the cocktail party problem’ and affects many more people than just those who wear hearing aids. By using source separation, AudioTelligence can focus on the desired speech and allow people to hear more of what they want. This is all done with minimal processing delay, so there is no loss of lip synchronisation and conversation can carry on as normal.
AudioTelligence's source separation technology is not just limited to hearing assistance. It has application in home assistants, automotive voice control and VOIP applications as well. Dave Betts, Chief Scientific Officer notes that:
We tested our source separation technology with a well known home assistant. In noisy conditions, the home assistant only recognised 22% of commands. We then added our source separation and the recognition rate shot back up to 91%.
The full text and the audio for Ken and Georgia's interview can be found at https://www.thenakedscientists.com/articles/interviews/solving-cocktail-party-problem