“We launched WIDEX EVOKE with SoundSense technology to put users back in control of the most difficult hearing situations,” says Jens Brehm Nielsen, Data Science & Machine Learning Architect at Widex. “And we can see that EVOKE users have taken the opportunity to do that and, in the process, are helping us understand more about them. That information will help us to make the EVOKE and future hearing aids even better.”
SoundSense Learn is an Artificially Intelligent system, because AI refers to systems that solve tasks humans are inherently good at – such as driving a car, doing the dishes etc. SoundSense Learn expands into entirely new applications by helping end users adjust their hearing aids precisely in the moment, something no humans can replicate to the same degree of accuracy.
The SoundSense Learn smartphone app is connected to the EVOKE hearing aids and uses machine learning to guide the users in optimizing the settings to their exact needs. The app gathers a variety of anonymous data such as how often they turn the volume up or down, which sound presets they use and how many custom settings they create – including those made with SoundSense Learn.
What’s in a tag?
Tagging of custom settings has proved to be one of the interesting pieces of data generated by EVOKE.
“We found that many people have created a setting and tagged it with, for instance, ‘Work’ which suggests that it is something that our end-users need and want. And from SoundSense Learn we already have an idea of how they like the settings.”
Most hearing aids give users the ability to customise their sound experience by adjusting frequency bands to boost or cut bass, middle or high tones. Adjusting frequencies works well in many situations once the initial settings have been set by a skilled audiologist. However, some situations are so complex that hitting the right combination of adjustments can be difficult.
“Widex hearing aids are well-known for the quality of their sound,” says Jens. “But SoundSense Learn has added an extra layer of quality sound on top of that by using a machine learning algorithm together with reinforcement learning – the two key ingredients in state-of-the art AI algorithm – that enables the algorithm to learn in the moment.
“The algorithm learns an optimal setting every time a user finds the sound to be a little below expectations in a given sound environment. It learns these settings by simply asking the user to compare two settings that are carefully picked by the algorithm. This allows it to learn an optimal setting in a new environment very fast.”
By collating and analysing the anonymous data WIDEX EVOKE will continue to become even smarter as time passes.