What’s the problem?
Hearing loss accounts for a larger share of global disability than almost any other condition. It cannot yet be cured and existing technology often fails to help. Current hearing aids are limited in their ability to provide real-world benefit: speech in background noise remains difficult to understand, multi-talker environments are hard to parse, and music is hopelessly distorted. The key challenge in hearing aid design is fundamental: hearing loss is a complex nonlinear problem and our current understanding of it is too superficial to provide a basis for hand-designed solutions.
What’s the opportunity?
Fortunately, there is hope: recent advances in deep learning and auditory neuroscience have opened up the opportunity to solve the problem empirically. Hearing relies on the information about sound that is encoded in the brain’s neural activity patterns. If hearing is impaired by hearing loss, it is because the details of these activity patterns have been distorted. The ideal hearing aid would correct these distortions by transforming incoming sounds such that, when processed by the impaired ear, they elicit the same neural activity patterns as the processing of the original sounds by a healthy ear. If this ideal can be achieved, hearing will be restored to normal. We can therefore reframe the design of hearing aids as an optimization problem in which the goal is to find the sound input that produces a desired auditory experience by eliciting the required neural activity.
This is, of course, easier said than done. But we have spent the past several years developing a unique capability for recording large-scale neural activity data with high spatiotemporal resolution. We are now ready to use this resource to train deep learning models that link sound to perception via neural activity and to develop them into transformative applications. We are supported by the UK’s medical and engineering research councils (MRC and EPSRC) and are working in partnership with the Royal National ENT Hospital and Perceptual Technologies, a startup formed to bring our technology to market.
You can read more about our plans and the research behind it at lesicalab.com.
Who are we looking for?
We are looking for experts who are interested in applying deep learning methods to large-scale neural data to develop the next generation of hearing technologies. We are recruiting for multiple postdoc positions to join a team that will work together on different aspects of the problem. Each team member will have significant autonomy in contributing to a radically new approach to sensory device design.
What will you do?
What will you bring?
Other qualifications such as knowledge of neuroscience or auditory processing, or experience working with time-series data, are desirable but not essential.
What are we offering?
Please get in touch if you would like to discuss these opportunities (lesica@xxxxxxxxx). And please forward this announcement to anyone else who might be interested.
Nicholas A. Lesica, Ph.D.
Professor of Neuroengineering
Ear Institute
University College London