Blind speech dereverberation with multi-channel linear prediction based on short time fourier transform representation.