Speech Enhancement Using Spectral Subtraction Based on a Modified Noise Minimum Statistics Estimation

In this paper, we use the spectral subtraction method to reduce non-stationary angle grinder noise of speech signal. Motor based devices generate acoustic noise that often contains a wide-band part plus some limited number of dominant sinusoidal signals. This noise disturbs the audibility of speech signals used for ordinary daily communication. The spectral subtraction requires an estimate of the noise, here, we present a novel approach based on a multi-band minimum statistics. The performance of the proposed algorithm is tested under various working conditions of an angle grinder. Change in angle grinder working condition changes noise statistics,hence, the noise becomes non-stationary. Objective and subjective measures reveal that our proposed algorithm can eliminate the noise well enough, resulting in a minimum speech distortion and musical noise in the processed speech signal.