Noise estimation based on time–frequency correlation for speech enhancement