This paper deals with the enhancement of noisy speech signals for mobile radio applications with a view to the transmission, in order to get a signal more intelligible and pleasant to listen to. We give a synthesis of different methods we have proposed and tested on real signals recorded by one or two microphones. After a brief recall on methods achieving poor performances, we present four methods which appear efficient. First of all, two improvements are brought to the spectral subtraction algorithm : the noise over-estimation factor is controlled and a segmentation algorithm is applied to the noisy speech signal. In a second method, we improve the signal estimation using an iterative Wiener filtering and some spectral constraints. Then, the case of two microphones is investigated. One approach is based on the coherence function. In the other one, a noise cancellation technique is realized on each channel to get new observations having a better signal-to-noise ratio. An identification between the new observations is performed to give an estimation of the desired speech signal. Some points of comparison between these methods are given.
Cite as: Faucon, G., Bouquin, R.L. (1992) Synthesis on noise cancelling methods for mobile radio applications. Proc. ETRW on Speech Processing in Adverse Conditions, 135-138
@inproceedings{faucon92_spac,
author={G. Faucon and R. Le Bouquin},
title={{Synthesis on noise cancelling methods for mobile radio applications}},
year=1992,
booktitle={Proc. ETRW on Speech Processing in Adverse Conditions},
pages={135--138}
}