Oticon More™ - Trained on More than 12M Real-life Sounds
Why is the inclusion of a Deep Neural Network (DNN) in new Oticon More™ such an important step forward?
With Oticon More™, Oticon challenges the hearing industry to embrace new frontiers in technology - inclusion of DNN within the hearing aid. Speech is a highly complex acoustic signal, with information embedded in the detail of the waveform in different, often subtle ways. Competing sounds in everyday environments are also complex sources of acoustic energy but in ways that can differ from speech signals. DNNs approach signal processing from a different direction. Oticon More works to create a full, but rebalanced sound scene, allowing for improved performance but without creating an unnatural, restricted listening experience.
What are DNNs?
DNNs are signal processors that learn to detect patterns in data that the conscious human mind may not be able to recognize or are difficult to capture in a discrete set of processing steps. Modeled after the natural way the brain develops, they learn, not by rules, but through repeated experience. Classic signal processing in our field, and in many other fields, is based on algorithms -- someone writing down the rules that explain the data, allowing for sorting, predictions, etc. DNNs are different. DNNs learn by experience, not by explicitly being told the rules. DNNs are most effective in situations where it is difficult to write all the rules that allow us to make sense of input data or stimulation.
Were there challenges to the inclusion of DNNs in a hearing aid?
DNNs require an extremely high level of signal processing resources that have made DNNs too large and too power hungry to be able to be integrated directly into a hearing aid – until now. With Oticon More, we have customized a DNN to meet the space and power challenges of modern hearings aids. That in and of itself is a significant technical achievement. But it took more than just having a DNN on board, we had to train the DNN in a unique manner to address the specific signal processing need that we had.
How do you train DNN for speech understanding?
We provided 12 million samples of sound, including both clean speech and a wide variety of competing sounds. We told the DNN which samples contained clean speech and which samples contained competition. We then set it to work to find all the dimensions that it could that differentiate speech from other sounds, including the way both speech and noise spread across frequency and over time. The learnings from that training process are embedded in the DNNs on-board each Oticon More hearing aid. Every sound that passes through the device is compared to the results of what was discovered in the learning phase. Clearer contrasts are created as speech signals are protected and competition is reduced. This allows the device to process speech in noise more like the human brain does – not based on a limited set of rules that we wrote but rather based on extensive experiential learning.
What more can we look forward to with Oticon More?
The PolarisTM platform is the backbone of Oticon More, purpose-built for hearing aids. This integrated intelligent microchip, Oticon’s most intelligent and powerful ever, supports Oticon More’s life-changing technologies. We anticipate more innovations and technologies will emerge in the future and we have allowed for platform upgradability and additional memory to support this.
Where can we learn more about Oticon More?
You can learn more about Oticon More at www.Oticon.com/More. Oticon is also conducting a series of educational webisodes that take a deeper dive into DNN and the audiology of Oticon More. Each webisode is approved for 1 hour of continuing education by ASHA, AAA and IHS. The series consists of five Webisodes with the first premiering on February 2. Hearing care professionals can register here: www.oticonnext.com/webisodes.