Visual processing-inspired fern-audio features for noise-robust speaker verification