next up previous
Next: Neuromorphic and Digital Hybrid Up: 1st European Workshop on Previous: A Paced Analog Silicon

A Multi-Microphone Sub-Band Adaptive Speech Enhancement System Employing Artificial Neural Network Based Non-Linear Filters

Amir Hussain, Douglas R. Campbell, Dept of Electronic Engineering and Physics, University of Paisley, Paisley PA1 2BE, Scotland In this paper, a general class of single-hidden layered linear-in-the-parameters, feedforward Artificial Neural Network (ANN) based adaptive non-linear filters is proposed for processing band-limited signals in a multi-microphone sub-band adaptive (MMSBA) speech enhancement system. The conventional MMSBA system which is based conceptually on the human auditory system offers the possibility of performing ``binaural unmasking" outwith the human body, providing signals of improved Signal to Noise Ratio (SNR) to the human ear, than a conventional aid or cochlear implant processor. Initial comparative results achieved in simulation experiments demonstrate that the proposed MMSBA system employing ``biologically inspired" ANN based sub-band processing is more capable of outperforming conventional noise cancellation schemes.

Dr L S Smith (Staff)
Tue Dec 2 14:23:49 GMT 1997