Spectral Domain Texture Analysis for Speech Enhancement