Jago T. Reed-Jones
Paul Fergus
Davis L. Ellis
John Marsland
Karl O. Jones
School of Engineering, Liverpool John Moores University
United Kingdom
e-mail: j.t.reedjones@2019.ljmu.ac.uk
Abstract:
This work tests the addition of BiLSTM layers to convolutional neural networks for Binaural Sound Source Localisation, with emphasis on improving the system’s robustness to reverberation and noise, in the context of full-horizontal plane localisation.
This is tested through comparison of two sytems trained on identical datasets.
Results from this show that the BiLSTM improved performance, along with improved robustness to reverberation and noise.
Key words:
binaural sound localization
convolutional recurrent neural network
sound source localization
Section:
The full text of the report is included in IEEE InfoTech-2024 eProceedings
and will be available on the IEEE Xplorer DL website
https://ieeexplore.ieee.org/xpl/conhome/1828024/all-proceedings
and will be available on the IEEE Xplorer DL website
https://ieeexplore.ieee.org/xpl/conhome/1828024/all-proceedings