Softcomputing Approach to Noise Modeling

The inaccuracy of estimating the noise reference, to improve the quality of speech corrupted by additive noise, could cause additional musical artifacts to be introduced at the output of the processed speech. These musical artifacts could be so unbearable that the processed speech could be unrecognizable at the receiver end. In this paper, we introduce the use of a novel fuzzy neural network (FNN) based approach in a human auditory system, to intelligently filter the noise. Simulation results were compared using two types of hybrid fuzzy neural networks; namely: POPFNN-CRI(S) [1] and the FALCON-MART [2] against the popular Classical LMS and ASS algorithm. Although FALCON-MART produces a poorer correlation results, it has a faster on-line recall time, which is highly suitable for on-line application. Key-words:Softcomputing, Compositional Rule, POPFNN-CRI, FALCON-MART, Human Auditory System, speech enhancement, Fuzzy Neural Networks (FNN).