Speech Enhancement Using Compressed Sensing

This paper proposes an approach based on the compressed sensing for speech enhancement. It attempts to exploit the fact that it is relatively easy to find sparse representation for speech signal but the same cannot hold true for noise. Thus for a given speech signal a sparse vector is derived which extracts the speech signal from the given noisy speech signal. Approach is further modified to ensure the speech enhancement across the varying segments of speech signal, which is present due to language, emotion and speech uttered by a person. Experimental results show that the proposed approach can be an alternative to the existing approaches for speech enhancement.