Speech enhancement based on neural networks applied to cochlear implant coding strategies
Federico Bolner, Tobias Goehring, Jessica Monaghan, Bas van Dijk, Jan Wouters, Stefan Bleeck
Abstract
Traditionally, algorithms that attempt to significantly improve speech intelligibility in noise for cochlear implant (CI) users have met with limited success, particularly in the presence of a fluctuating masker. In the present study, a speech enhancement algorithm integrating an artificial neural network (NN) into CI coding strategies is proposed. The algorithm decomposes the noisy input signal into time-frequency units, extracts a set of auditory-inspired features and feeds them to the NN to produce an estimation of which CI channels contain more perceptually important information (higher signal-to-noise ratio, SNR). This estimate is then used accordingly to retain a subset of channels for electrical stimulation, as in traditional n-of-m coding strategies. The proposed algorithm was tested with 10 normal-hearing participants listening to CI noise-vocoder simulations against a conventional Wiener filter based enhancement algorithm. Significant improvements in speech intelligibility in stationary and fluctuating noise were found over both unprocessed and Wiener filter processed conditions.
BibTeX
@inproceedings{icassp2016_speechenhancemen,
title = {Speech enhancement based on neural networks applied to cochlear implant coding strategies},
author = {Federico Bolner and Tobias Goehring and Jessica Monaghan and Bas van Dijk and Jan Wouters and Stefan Bleeck},
booktitle = {ICASSP 2016},
year = {2016}
}