eduzhai > Applied Sciences > Engineering >

Lightweight Online Noise Reduction on Embedded Devices using Hierarchical Recurrent Neural Networks

  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Deep-learning based noise reduction algorithms have proven their successespecially for non-stationary noises, which makes it desirable to also use themfor embedded devices like hearing aids (HAs). This, however, is currently notpossible with state-of-the-art methods. They either require a lot of parametersand computational power and thus are only feasible using modern CPUs. Or theyare not suitable for online processing, which requires constraints likelow-latency by the filter bank and the algorithm itself.In this work, we propose a mask-based noise reduction approach. Usinghierarchical recurrent neural networks, we are able to drastically reduce thenumber of neurons per layer while including temporal context via hierarchicalconnections. This allows us to optimize our model towards a minimum number ofparameters and floating-point operations (FLOPs), while preserving noisereduction quality compared to previous work. Our smallest network contains only5k parameters, which makes this algorithm applicable on embedded devices. Weevaluate our model on a mixture of EUROM and a real-world noise database andreport objective metrics on unseen noise.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×