Reduction of Musical Residual Noise Using Hybrid Median Filter

Musical residual noise is a major problem for a speech enhancement system. This noise is annoying to the human ear and seriously deteriorates the quality of enhanced speech. In this paper, we employ a three-step-decision gain factor to enhance a noisy speech, enabling background noise to be efficiently reduced. Hence the enhanced speech is post-processed by a hybrid median filter to significantly reduce residual noise. In the case of a noise-like spectrum, block median filtering is performed to reduce the spectral variation. A musical tone is then reduced. In the case of a speech-like spectrum, directional median filtering is performed to slightly reducing the residual noise, while a harmonic spectrum is not destroyed. Experimental results show that the proposed approach can significantly improve the performance of a speech enhancement system by significantly reducing musical residual noise.