Question
What is a DNN and how is ReSound using this technology in hearing aids?
Answer
A Deep Neural Network (DNN) is a type of deep learning that is a subset of machine learning, employing multilayered neural networks inspired by biological neuroscience (Bengio et al., 2021). To elaborate, DNNs are a specific architecture within deep learning characterized by multiple layers between the first input and last output layers, known as “hidden layers” (Kelleher, 2019). With an appropriately large dataset, the computer becomes better at processing due to increased opportunities for learning. In short, it is a decision-making technology utilizing an artificial neural network modeled after the human brain.
In the field of audiology, DNNs are being increasingly utilized in hearing aids to enhance auditory experiences. Some ways include reading soundscapes to produce clearer versions of sounds as well as organizing and balancing auditory inputs. DNNs can also isolate speech, removing background noise to deliver cleaner and clearer sound. For example, the ReSound Vivia uses DNN to both classify and remove noise from the environment, allowing the user to be more engaged in conversations.
Despite its technological advantages, the use of DNN must be judicious. Too much training of the DNN dataset is a significant concern, as DNNs can become too tailored to the training data, reducing their generalizability to slightly varied inputs that users encounter in the real world. Conversely, too little training can lead to errors in the outcome. The degree of constraints on how DNNs can adapt is also a variable that can diminish their effectiveness; DNN that is too restrictive can exclude its application in situations that were outside of its training, while DNN that is overly adaptive can result in unintended results or disadvantages for the user. Another important issue is the quality of the data being fed to the DNN and how the model interprets them, which can pose a significant challenge in deploying DNNs in real-world situations (Khoei et al., 2023). Additionally, the computation time required for training DNNs can be resource-intensive and time-consuming. All of these factors need to be weighed to find the optimal amount of training, the right degree of adaptability, integrity of the data trained, and conservation of power needed to run the operation.
To overcome these challenges and responsibly apply DNN to provide maximal benefit for hearing aid users, ReSound ensured that the most up-to-date knowledge in this field was applied when implementing DNN in the ReSound Vivia hearing aids. The system is also powerful enough to perform 4.9 trillion operations per day, meaning that the DNN architecture was optimized for efficiency and longer battery life. Perhaps most importantly, the DNN was trained with 13.5 million real-life conversations in noise, taking care not to over- or under-train the system, to allow for appropriate generalization to the world at large. Choosing these samples and parameters allows the ReSound Vivia to intelligently classify and reduce noise in the listening environment, without adding bulk to the hearing aid size or diminishing the hearing aid battery runtime for the user.
One important thing to remember when it comes to the intelligent, proper use of hearing technology with humans is that, regardless of the sophistication of the DNN, no hearing aid can ever truly predict which conversation partner or speaker the user wishes to hear best. For this reason, ReSound Vivia uses DNN in combination with beamforming directionality and provides this powerful technology in the Hear in Noise program. We understand that despite the marvels of modern DNN, it is still best practice to keep the user in the driver’s seat by allowing them to choose when to engage the most powerful noise reduction and directionality possible in modern hearing aids. Putting DNN into an automatic program runs the very real risk of amplifying speech at the expense of important noise sources that are salient or could be deemed important for the user. As such, the ReSound automatic program in ReSound Vivia, 360 All-Around, is the best of all worlds for most listening situations. The user-selectable Hear in Noise program is available for the user as a super-charged option for reducing noise in the most difficult listening environments. In our premium ReSound Vivia 960S, the Hear in Noise program engages Intelligent Focus, combining the strongest beamforming directionality and our state-of-the-art DNN system. By allowing individual users to choose this Hear in Noise program when needed, they remain in control of what they wish to hear in their daily lives.
In today’s fast-paced world of technology, more is being discovered about deep learning and DNNs at astonishing speeds. While this technology affords new opportunities to increase accessibility to everyone in widely varied fields of study, ReSound continues to explore the most advantageous ways to implement it to support the unique needs of hearing aid users. A judicious approach to DNN, optimized to provide real-world benefits for people with hearing loss while maintaining cosmetic appeal and power conservation, remains the cornerstone of how this and other new technology is implemented in ReSound hearing aids.
References
Bengio, Y., Lecun, Y., & Hinton, G. (2021). Deep learning for AI. Communications of the ACM, 64(7), 58–65.
Kelleher, J. D. (2019). Deep learning. MIT Press.
Khoei, T. T., Ould Slimane, H., & Kaabouch, N. (2023). Deep learning: Systematic review, models, challenges, and research directions. Artificial Intelligence Review. Advance online publication. https://doi.org/10.1007/s00521-023-08957-4
