Speech segregation using Deep Neural Networks (DNN)

Many hearing aid users lose their ability to segregate voices and this complicates social situations. Researchers from Eriksholm Research Centre aim at solving this problem in future hearing aids by using Deep Neural Networks (DNN).

What is DNN?

Deep Neural Networks (DNN) is a trainable algorithm that enables a computer to solve many human-like tasks like separating sounds, faces, and voices. As an example, the technology is already used in Smartphones to identify people in photographs and in automatic speech recognition.
DNN works like large networks of simple ‘neurons’ inspired by the way the brain works. Each neuron is a simple element that receives multiple inputs and performs a series of simple actions which generate an output. When many neurons are stacked and densely interconnected, they can handle more complex input, and such neuron-layers can learn to perform very difficult tasks. This means that they learn from examples, and apply their knowledge to different, but similar, tasks. As an example, they can learn to analyse, recognise and separate specific sound sources.

Learn more in the video below.

Why use DNN in hearing aids?

New technology will soon enable hearing aids to recognize and segregate voices: a well-known challenge for hearing aid users. This requires hearing aids that can separate competing voices. Today's hearing aids are unable to do it but the help comes from DNN. Since 2012, we have been working closely together with Professor Tuomas Virtanen and his research team from Tampere University of Technology in Finland, trying to solve the challenge of segregating voices in hearing aids.

Read the interview with Tuomas Virtanen, where he explains how hearing aids will get much smarter in the future.

Further reading:

Bramsløw L, Naithani G, Virtanen T. Deep Neural Networks for speech enhancement in noise. Conference proceedings of Speech in Noise SPIN, 2022.

Naithani G, Barker T, Parascandalo G, Bramsløw L, Pontoppidan NH, Virtanen T (2017). Evaluation of the benefit of neural network based speech separation algorithms with hearing impaired listeners. In J. Barker, J. Culling, J. Hansen, A. Hussain, P. Nordqvist, et al, eds. Proceedings of the 1st International Workshop on Challenges in Hearing Assistive Technology. Stockholm, pp. 35–36.

Naithani G, Barker T, Parascandolo G, Bramsløw L, Pontoppidan NH, Virtanen T(2017). Low-latency sound source separation using convolutional recurrent deep neural networks. In 2017 IEEE Workshop on Application of Signal Processing to Audio and Acoustics. New Paltz, NY: IEEE, pp. 1–5.

Naithani G, Parascandolo G, Barker T, Pontoppidan NH, Virtanen T (2016). Low-Latency Sound Source Separation Using Deep Neural Networks. In 2016 IEEE Global Conference on Signal and Information Processing. IEEE, pp. 272–276.