Solving speech separation with Lars Bramsløw

There is a well-established problem within the world of audiology; the issue of speech separation in competing voices. We sat down with Advanced Algorithms research engineer and project manager, Lars Bramsløw, to understand this problem, and how to combat it. For people without hearing impairment, separating speech can be a challenge, but our ears are amazing. Our ears and our brain work together to separate sound input from a number of sources. To determine who is speaking at any given time, and how relevant the information is to us as listeners, the brain uses – for example – differences in location, tonality, intonation, and context. For hearing aid users, this issue is much more complex due to the impaired ear. And today’s hearing aids cannot reliably sort between these factors.

The field of speech separation is, fortunately, moving ahead at breakneck speed. In several combined experiments with the Tampere University of Technology, Finland, we have seen very promising results when attempting to tackle these issues, and so we pursue and research further development. Particularly we are looking at the field of deep neural networks, a very strong contender in separating known voices in hearing aids. However, even these algorithms do not know which talker the listener wants to hear.

Currently, an attractive addition to speech separation, is the field of Electrooculography, i.e. using the attention of our eyes to determine which speaker we are looking at, and thus, which voices to enhance. This can guide the hearing aid and its algorithms to the preferred target. While both methods have a great potential, together they may offer an even better solution; to find a way to enhance the co-operation between the brain and the hearing aids, so the process is natural and automatic. Essentially, in Lars’ words, to have half an ear on both sides of the head, and being able to share your attention between people if you so wish.

It is our hope that soon, we will have technology capable of tackling this problem, and moving people with hearing impairment one step closer to normal hearing, with all the auditory stimuli that come with it.