5 easy steps for understanding DNN and its benefits for your clients

At Oticon, we’re committed to changing the lives of people with hearing loss. One of the ways we do this is through the development and use of leading technologies. This is why we’ve worked with a DNN, or “deep neural network”, for our newest hearing aid, Oticon More™.

This type of technology is new in the field of audiology and you probably have some questions about it. In this blog post, we will provide you with a simple explanation of DNN, plus its benefits and how to explain them to your clients.

You might not know DNN – or deep neural network - by its name, but you’ve most likely used it already without realizing. It’s been used for a variety of everyday tasks, like language translation. It’s even been used for medical diagnosis – UCLA trained a DNN to detect cancer cells! Now, we’re using it for sound processing and balancing in Oticon More. But what exactly is DNN and how does it work?

The general idea of a DNN is that it learns through repetitive action from a collection of samples, like 1,000 pictures of different dogs, as opposed to a rigid set of rules, like “a dog has a black nose and floppy ears.” In this way, a DNN learns in the same way the human brain does – through practice and making mistakes.

Here’s how it works:

1. A computer is given a piece of information, like an image or a sound. Let’s say in this example it’s given a trumpet sound. Unlike you or I, a computer wouldn’t know what this is.
   
   
2. The computer passes this sound through its DNN, recognizing what it can and sorting elements of it – like a high pitch or a low pitch sound.
   
   
3. When it reaches the end of this process it decides if the sound is a trumpet or not.
   
   
4. It’s given feedback on this answer – a yes or a no – which the computer uses to strengthen its decision making
   
   
5. The process is repeated over and over with lots of different trumpet sounds, until the computer can learn to recognize it instantly. Just like a brain would.

Traditional hearing aids rely on theoretical models and strictly defined rules on how to best enhance speech and reduce background noise. But this can cause an inflexibility to changing environments, as they don’t catch all the nuance of sound and can make mistakes. After training a DNN with 12 million real-life sound scenes, like family gatherings, restaurants, busy roads and public transport, we built it into our new hearing aid, Oticon More. The hearing aid can then utilize the intelligence of the DNN to mimic the way the brain works when prioritizing and balancing sound.

The benefits of DNN and how to explain it to your client

It gives the hearing aid user access to a full and precisely balanced sound scene – the DNN has been trained to process a great variety of sound scenes, so it can give the user access to all relevant sounds in a clearer and more balanced manner.

In fact, our study has shown that it makes the full sound scene 60% clearer to the user

Clearer sound helps users follow conversation – This clearer sound information improves the brain’s ability to track the most important sounds, all while maintaining an openness to other sound sources. For example, following a friend’s conversation while seated in a busy restaurant.

It helps to deliver a good neural code that thebrain needs to work optimally – signal processing like a DNN mimics how the brain learns. This supports the hearing centre by giving the brain what it needs to work optimally – which can also help maintain its health.

Learn more about how sound supports the brain’s health here.

See what our test group had to say about Oticon More

From rediscovering forgotten sounds to no longer being held back by noisy situations, Oticon More’s revolutionary use of DNN could have a great impact on your clients’ lives.

Find out more about Oticon More here.

If you’d like to learn more about DNN, read our technology page.

^[1] Bahram Jalali, Claire Lifan Chen, and Ata Mahjoubfar, University of California, Los Angeles (UCLA)

https://www.mathworks.com/company/newsletters/articles/cancer-diagnostics-with-deep-learning-and-photonic-time-stretch.html

^[2] Santurette, S., Juul Jensen, J., Ng, E.H.N. , Man, K.L.B (2020) Oticon More(TM) Clinical Evidence. Oticon Whitepaper