Spectral Subtraction using Modified Cascaded Median based Noise Estimation for Speech Enhancement

A modified version of the cascaded median based noise estimation for speech enhancement using spectral subtraction has been proposed. The proposed method involves less storage as compared to the original cascaded median based method reported earlier. In this method, the connection of median block is in cascaded but operation of that blocks are different from original cascaded median based method. Here, single median is calculated for each frame from samples of noisy spectra. It is simulated on MATLAB with 16 kHz sampling frequency, half-overlapped hamming window and tested through Perceptual Evaluation of Speech Quality (PESQ) measure. The perceptual speech quality gets improved with the proposed method and at the same time, simulation time is marginally less than the cascaded median based method.