A Noise-Robust Front-End Based on Tree-Structured Filter-Bank for Speech Recognition