Artificial Intelligence (AI) for Hearing Aids
Despite significant progress in hearing aid technology, many customers are no longer satisfied with their hearing experiences in loud places. Understanding speech-in-noise remains the main challenge for those with hearing loss. How can this be changed? Might Artificial Intelligence (AI) help? Yes, there are DNNs for that!
Basics of the Deep Neural Network (DNN) – the AI tireless workhorse
AI is a rapidly advancing field of computer science. It is said that AI (Artificial Intelligence) imitates HI (Human Intelligence). In order to do so, scientists and engineers create Deep Neural Networks (DNNs). These are highly complicated computer software models, which are meant to perform highly complex processing tasks, including superior sound processing in noisy situations.
Inspired by the structure and function of the human brain, DNNs, have many layers of interconnected computing nodes referred to as “neurons.” And so, DNNs are often described as “Artificial Intelligence (AI) Brains.” That said, DNNs are NOT exact replicas of the hugely complicated and adaptable human brain. They are modelled after the human brain. They are types of AI Software Models that are designed to “learn” from data and that are “trained” to perform complicated procedures by mimicking certain human brain functions related to their task(s).
DNN learning & training happens under the direction of sets of specific instructions known as algorithms. Although DNNs become the working units or workhorses of AI, they are not all the same. With different designs for different purposes, the range of their abilities depends on their intended use and the depth and breadth of their “schooling.”
The AI Hearing-assistive Age has arrived
AI-enabled hearing aids are already on the market. Although they vary by manufacturer, AI DNNs can greatly improve hearing aid performance in noisy places. Their designs and education are fast becoming the new trade secrets among hearing aid manufacturers.
For those with hearing loss, DNNs can be turned into super-efficient sound processors. Well-trained hearing assistive DNNs draw on information of how the human brain’s Central Auditory Processing System works with sound and deals with hearing-in-noise problems. And there is a lot to learn. By the time DNNs graduate, they know how to carry out important procedures related to sound.
They detect and localize sound. They identify the type of sound environment, look for patterns and analyze the sound qualities, such as volume, pitch, tone, inflection etc. They separate speech from noise, lower background din and echoing while they raise the speech signal, thereby improving the SNR (Speech-to-noise ratio) etc. They may even direct hearing aid microphone orientations and adjust hearing aid programs to the specific sound environment. They can do all of this automatically and at lightning speed, in real time in the actual venue where we are. Ideally, once in AI-mode, there is no more need to fidget with apps or click hearing aid buttons!
Although the degree and type of hearing loss still present DNNs with challenges, the human brain receives the clearest sound and speech signals possible under the circumstances. Such refined signals are easier for the brain to recognize and interpret, which improves understanding and makes “hearing” less stressful. Listening Fatigue is reduced, and people may well regain the confidence to venture into places that they have avoided due to the dreaded speech-in-noise hearing issues.
More changes and some worries on the horizon
Beyond tackling speech-in-noise, the obvious next steps for AI are to become better at compensating for the effects that inner ear and nerve damage have on AI-polished sound signals. Some say that eventually AI DNNs may personalize listening experiences by learning from the client’s reactions to sound and from their choices of sound environments. Wow! Also, future customers might be able to tweak their hearing aid DNNs on their own. Who knows?
Seemingly limitless and somewhat spooky AI possibilities raise questions about privacy issues and what people might feel encouraged to reveal about themselves to their ear-DNN. How about the hacking potential? There is obviously a lot that customers and audiologists alike must learn about AI – its advantages but also its limits and possible risks. Maybe hearing aids will be offered with various “degrees” of AI assistance and people can choose how far they want to take the AI experience.
For now, hearing assistive DNNs can help a lot. But they do not “fix” hearing loss and they do not replace the human brain. The best advice is still to respect and preserve the natural hearing that we have because so far Mother Nature does it best.