Speech signal filtering by artificial neural networks

The improvement of quality of speech signal by subtracting noise from the useful signal has a broad application fields: Improved performance of voice communication equipments in a sound noisy environment such as air navigation, video conferencing or the environment of mobile phone, etc. In this paper, we discuss the use of a Multilayer Artificial Neural Network of feed-forward type (FF-ANN). The network is subjected to a training algorithm using the Dual Extended Kalman Filter (DEKF) to remove non-stationary and colored noise from the speech signal. We also show the efficiency of this method.