Robust Speech Recognition Based on Mixed Histogram Transform and Asymmetric Noise Suppression