Robust speech processing using local adaptive non-linear filtering

A local adaptive non-linear algorithm for robust speech processing is proposed. The algorithm is based on calculation of the rank-order statistics of an input speech signal over a sliding window. The algorithm is locally adaptive because it can vary the size and contents of a sliding window signal as well as an estimation function employed for recovering a clean speech signal from a noisy signal. The algorithm is able to improve the quality of a speech signal preserving its intelligibility and introducing only imperceptible musical noise. The performance of the adaptive algorithm for suppressing additive, impulsive and mixed noise in an input test speech signal is compared with that of existing speech enhancement algorithms in terms of several objective metrics.